Morpho 123456

Contemporary Morphology
Trends in Linguistics
Studies and Monographs 49
Editor
Werner Winter
Mouton de Gruyter
Berlin · New York
Contemporary
Morphology
Edited by
Wolfgang U. Dressler, Hans C. Luschützky,
Oskar E. Pfeiffer, John R. Rennison
Mouton de Gruyter
Berlin · New York 1990
Mouton de Gruyter (formerly Mouton, The Hague)
is a Division of Walter de Gruyter & Co., Berlin.
® Printed on acid-free paper which falls within the guidelines of the

ANSI to ensure permanence and durability.
Library of Congress Cataloging-in-Publication Data
Contemporary morphology / edited by Wolfgang U. Dressier

... [et al.].
p. cm. — (Trends in linguistics. Studies and mono-
graphs ; 49)
"Selected papers from the Third International Morphology
Meeting held, under the auspices of the International Asso-
ciation of Morphology, in Krems (Austria) from July 4 to
July 7, 1988" - Pref.
Includes bibliographical references and indexes,
ISBN 0-89925-663-5 (acid-free paper)
1. Grammar, Comparative and general — Morphology —
Congresses.
I. Dressier, Wolfgang U., 1939— , II. International Mor-
phology Meeting (3rd : 1988 : Krems an der Donau, Austria)
III. International Association of Morphology. IV. Series.
P241.C66 1990
415 —dc20 90-42745
CIP
Deutsche Bibliothek Cataloging-in-Publication Data
Contemporary morphology / ed. by Wolfgang U. Dressier ... —

Berlin ; New York : Mouton de Gruyter, 1990
(Trends in linguistics : Studies and monographs ; 49)
ISBN 3-11-012349-5
NE: Dressler, Wolfgang U. [Hrsg.]; Trends in linguistics / Stud-
ies and monographs
© Copyright 1990 by Walter de Gruyter & Co., D-1000 Berlin 30.

All rights reserved, including those of translation into foreign languages. No part of
this book may be reproduced or transmitted in any form or by any means, electronic
or mechanical, including photocopy, recording, or any information storage and retrieval
system, without permission in writing from the publisher.
Typesetting: Arthur Collignon GmbH, Berlin. — Printing: Gerike GmbH, Berlin. —
Binding: Lüderitz & Bauer, Berlin. — Printed in Germany.
Preface
This volume presents selected papers from the Third International Mor-
phology Meeting held, under the auspices of the International Association
of Morphology, in Krems (Austria) from July 4 to July 7, 1988. In
contrast to the relatively small previous meetings in Veszprem (Hungary),
the Krems conference was an open congress which attracted many par-
ticipants from the immediately preceding Sixth International Phonology
Meeting. 1 In effect, the last day of the Phonology conference overlapped
with the first day of the Morphology meeting.
Accordingly our volume opens with the topic "Interface" (mainly
between morphology and phonology) where Spencer's morpholexical
approach to morphophonemics contrasts with Shapiro's semiotic ap-
proach to isomorphism of rule types. Carstairs' paper on suppletion
focuses on phonological triggers, Bayer and Lahiri's on morphosyntactic
constraints on Bengali clitics.
Within the second topic "Word formation" a wide range of subjects is
covered: Corbin presents her own rule-based model of complex words,
whereas Mötsch discusses conflicting proposals for word-structure theory
(e.g., analogy vs. constraints vs. rules). Principles of headedness in com-
pounds are in Di Sciullo's contribution on argument inheritance and in
Vogel's account of shortening of English loans in Italian. Beard argues
against morphemes as lexical items and for a strict separation of meaning
and form in derivational morphology, Dressier and Kiefer deal with the
morphopragmatics of Austrian and Hungarian diminutives, Warren with
types of phonologically modified English compounds. Zwanenburg dis-
cusses the order of compounding and inflection in French, Scalise argues
why Italian adverbs are derivational, not inflectional.
Although belonging to the third topic "Inflectional morphology and
clitics", Zwicky's presentation of his modular approach to both sub-
topics includes their relations to other modules as well; Wurzel's approach
to inflection arrives at comparable conclusions although couched in a
model of markedness. Morin adduces arguments for lexicon-internal
1. The selected papers of that conference will be published under the title Phonologica 1988
by the Cambridge University Press.
VI Preface
inflection from the history of French, P. K. Andersen against a universal

morphological category "passive". Joseph applies Zwicky's classificatory
criteria on Modern Greek clitics reinterpreted as affixes, whereas Man-
oliu-Manea analyses pronominal and pragmatic functions of Rumanian
clitics.
From the conference topic "Computer morphology" only the paper by
Meijs (on his ASCOT model) has been included in this volume; from the
topic "The psycholinguistic study of morphology" there are Derwing's
experimental evidence for a full-listing hypothesis about the representa-
tion of morphology in the mental lexicon and van Marie's reappraisal of
analogy with evidence from Dutch derivational morphology.
The last topic "Typology and non-Indo-European morphology" is
represented by Hagege's substitution of classical morphological types
with intralingual polytypical complexity due to phonetic evolution, and
S. Anderson's comparison of Sapir's morphological typology with recent
theoretical conceptions.
In this way our volume offers a representative cross-section of contem-
porary developments in theoretical morphology. Papers given at work-
shops are published separately: on Aphasia by J.-L. Nespoulous and P.
Villiard,2 on Natural Morphology by J. Mendez Dosuna and C. Pensado
Ruiz.3 The discussion papers have already been published in two separate
volumes.4
Wolfgang U. Dressier Hans C. Luschützky
Oskar Ε. Pfeiffer John R. Rennison
2. Morphology, Phonology and Aphasia (New York: Springer).

3. Naturalists at Krems (Universidad de Salamanca).
4. Wiener linguistische Gazette, supplement 7 Interface, supplement 8 Morphology.
Contents
Topic 1: Interface 1
Josef Bayer and Aditi Lahiri
Bengali emphatic clitics in the lexicon-syntax interface 3
Andrew Carstairs
Phonologically conditioned suppletion 17
Michael Shapiro
On a universal criterion of rule coherence 25
Andrew Spencer
The advantages of morpholexical phonology 35
Topic 2: Word formation 41

Danielle Corbin
Associativite et stratification dans la representation des mots
construits 43
Anna-Maria Di Sciullo
Formal relations and argument structure 61
Wolfgang U. Dressier and Ferenc Kiefer
Austro-Hungarian morphopragmatics 69
Wolfgang Mötsch
Problems of word structure theories 79
Sergio Scalise
Constraints on the Italian suffix -mente 87
Irene Vogel
English compounds in Italian: the question of the head 99
Beatrice Warren
The importance of combining forms Ill
Wiecher Zwanenburg
Compounding and inflection 133
VIII Contents
Topic 3: Inflectional morphology and clitics 139

Paul Kent Andersen
Arguments against the passive as a universal morphological
category 141
Robert Beard
The empty morpheme entailment 159
Brian D. Joseph
The benefits of morphological classification: on some apparently
problematic clitics in Modern Greek 171
Maria Manoliu-Manea
Case markers and pragmatic strategies: Romanian clitics 183
Yves-Charles Morin
Parasitic formation in inflectional morphology 197
Wolfgang Ullrich Wurzel
The mechanism of infleciton: lexicon representations, rules, and
irregularities 203
Arnold M. Zwicky
Inflectional morphology as a (sub)component of grammar .... 217
Topic 4: Computer morphology 237

Willem Meijs
Morphology in LDOCE and in the ASCOT system 239
Topic 5: The psycholinguistic study of morphology 247

Bruce L. Derwing
Morphology and the mental lexicon: psycholinguistic evidence . . 249
Jaap van Marie
Rule-creating creativity: analogy as a synchronic morphological
process 267
Topic 6: Typology and non-Indo-European morphologies 275

Stephen R. Anderson
Sapir's approach to typology and current issues in morphology . . 277
Contents IX
Claude Hagege
Do the classical morphological types have clear-cut limits? . . . . 297
Index of languages 309
Subject index 311
List of contributors 315

Topic 1: Interface
Bengali emphatic clitics in the lexicon-syntax
interface*
Josef Bayer and Aditi Lahiri
1. Introduction
In this paper, we advocate the view that to provide an adequate account

of certain clitic constructions, one must refer to more than one component
of the grammatical system. We will argue that the emphatic clitics in
Bengali must be licensed by both the morphophonemics of the lexicon
and the syntax of logical form, where the latter largely obeys the con-
straints on overt syntactic movements. After presenting some of the core
facts, we will first discuss the lexical account and then turn to the
complementary logical-form account. In the last section, we will present
our solution to a paradox that appears to arise when both the morpho-
phonemic and semantic aspects of these clitics are considered together.
2. The core facts
joj and j\j are the so-called emphatic clitics in Bengali which mean
something like 'too' and [ + emphatic] respectively. Since only joj intro-
duces a new lexical meaning, we will mostly use joj for the examples, but
/i/ behaves alike in all important respects. At a first glance, /o/ and /i/
seem to adjoin as enclitics to an element of type X° which is then the
focus of the clitic, e.g.,
* We wish to thank Probal Dasgupta for his advice and Jogamaya Bayer for discussing
the Bengali data presented here. We are also grateful to Wim van der Wurff and an
anonymous reviewer for their suggestions. Realizing that proposals independently made
in Dasgupta (1984, 1987, in press) turned out to be similar in spirit to ours gave us
encouragement to pursue this work.
4 Josef Bayer and Aditi Lahiri
(la) babul- ο kha- be

Babul- too eat- [fut, 3rd pers]
'Also BABUL will eat'
b) babul kha- be- ο
Babul eat- [fut, 3rd pers]- too
'Babul will also EAT'
In (la) we see that /o/ can adjoin to a syntactic word and (lb) shows
that it can attach to an inflected verb. The clitic can also attach to an
overtly inflected N.
(2) babul cheleke- ο mereche
Babul boy- [obj]- too beaten-has
'Babul has beaten also his SON'
The facts concerning the attachment of the clitic to the verb are more
complicated. As we have seen in (1), jo/ can come after the inflected
verb, and in most instances, it cannot be added between the stem and
the ending. Compare the following pairs of sentences where the second
member is ruled out. 1
(3a) mar- i- ο
beat- [lpers]- too
'(I/we) also BEAT'
b) *mar-o-i
(4a) march- i- ο

beat- [prog]- [lpers]- too
'(I/we) am/are also BEATING'
b) *mar-o-ch-i
Now consider instances where the clitic can be added before the inflec-
tional ending is attached. Examples are given in (5b) and (5d).2
(5a) babul cheleke mereche-o
Babul boy[obj] beaten- has-too
'Babul has also BEATEN the boy'
b) babul cheleke mere-o-che
c) mer- e- ch- i- I- am- ο
beat- [prt]- [prog]- [link]- [past]- [lpers]- too
'(I) have also BEATEN'
d) mer-e-o-ch-i-l-am
Although the clitic can be attached between the stem and the inflection,
it cannot be inserted between affixes. The choice is binary — either the
clitic comes right after the stem, or it must come after all the affixes are
added. Thus the forms in (6) are ungrammatical in Bengali.
(6a) *mer-e-ch-i-o-l-am
b) *mer-e-ch-i-l-o-am
It must be noted that there are no instances of categories other than
verbs in which the emphatic clitic could be "infixed". For instance, (7) is
ungrammatical. joj appears here between a noun stem and the case-
marker -ke, which is arguably an inflectional ending.
(7) *babul chele-o-ke mereche
'Babul has beaten also the BOY'
This restriction also holds for compound-like word formations. Bengali
has a verbal noun, which is derived by attaching the suffix -a to a V-
stem, e.g., por 'read' + a —• pora 'reading'. Similar to German infinitives
such as rad+fahrert 'bike riding', the verbal noun can incorporate an N°-
object into the verb stem involved, e.g., golpo por-a 'story reading'. As
(8a) below shows, joj can adjoin to the verbal noun and select its focus
inside, but as shown in (8b), it cannot adjoin to the focused Ν incorpo-
rated. For these examples, imagine a preceding discourse in which some-
one states that (s)he liked somebody's reading of stories very much.
(8a) tader [kobita por- a]- ο bhalo laglo
their poetry read-ing- too pleased-has
'(I) was pleased by also their reading of POETRY'
b) * tader [kobita-o por-a] bhalo laglo
This also holds true for the so-called dvandva constructions such as in
(9) below. The contrasting sentences with different clitic placements are
given in (10).
(9a) bap ma
father mother
'parents'
b) uttor dokkhin
north south
(10a) [bap ma]-o

b) *[bap-o ma]
c) [uttor dokkhin]-o
d) *[uttor-o dokkhin]
The interpretation of dvandvas such as (10a) suggests that /o/ attaches

to the whole dvandva as indicated by the bracketing, not to its second
constituent. Thus (10a) can only mean 'the parents too', but not 'father
and also mother'.
Given these observations there seem to be two obvious questions that
come to mind. First, what is the domain to which the clitics can attach?
Second, if this domain is not identical to the stem to which all affixes
can attach, how do the clitics differ from regular affixes?
3. Emphatic clitics in the lexicon
The fact that (3b) and (4b) are bad but not (5b) can be accounted for by
examining the minimal phonological unit that /o/ can attach to. Observe
that joj can attach to mere but not to mar. We will argue that the host
of the clitics must minimally be a phonological word. Under this view,
mere constitutes a phonological word while mar is merely a stem. In the
following discussion, we will focus on the facts which constitute evidence
for differentiating phonological words from stems, indicating also how
the clitics themselves are different from regular affixes.
3.1. Bengali has a rule which degeminates syllable-initial geminates. The

underlying form of the progressive affix is /cch/, a geminate affricate.
After a vowel-final stem the geminate is retained (since it can close the
preceding syllable), but is degeminated when preceded by a consonant-
final stem.
(11a) kha- cchi —• khac.chi (closure of preceding syllable)

eat- [prog, lpers]
'(I) am eating'
b) mar- cchi —*• mar.chi (C deleted)
The same holds true after the causative affix /a/ is added to the stem;
the geminate is retained as after a stem final vowel.
(12) mar- a- cchi —• ma.rac.chi
beat- [caus]- [prog, lpers]
'(I) am having (him) beaten'
Resyllabification is, however, blocked outside the domain of a phonolog-

ical word; after mere, the geminate remains syllable initial and is dege-
minated. 3
(13) mar- e- cchi —• me.re.chi (not: *me.rec.chi)
beat- [past prt]- [prog, lpers]
Implicit in this view is the claim that affixes can be added to stems as
well as to larger units like words. The compound-like constructions (cf.
9 — 10) also have affixes added at the end ([bap ma]r 'of parents', [*bap-
er ma]). Moreover, the prosodic unit after the addition of a clitic to a
phonological word, is still a word to which an affix can be added.
3.2. Evidence that mere is indeed a phonological word and that there
is less cohesion between word + affix than stem + affix comes from re-
duplicating echo words. In Bengali, an echo word can be formed by
reduplicating the entire word except for the initial consonant which is
usually replaced by a coronal. The echo word could be interpreted as 'X
and so forth' with perhaps a slight pejorative tinge. There is a constraint,
however, in what can be reduplicated. All stem + affix constructions can
be reduplicated, but no stems alone.
(14a) Nominal forms
pa — ta 'leg'
bari — tari 'house'
chele — tele 'boy'
kobita pora — lobita pora 'poetry reading'
b) Verb forms:
mare — tare '[3rd pers] beat'
khae — tae '[3rd pers] eat'
mere — tere '[past part] beat'
c) Inadmissible verb stems:

*kha — ta 'eat'
*mar — tar 'beat'
The affixation pattern also shows the distinction between the stem and
the word. The past-participial form mere can have the affix -che added
to it and then become reduplicated; but the -che can also be added after
reduplication has taken place. Compare the forms under (15).
8 Josef Bayer and A did Lahiri
(15a) mereche tereche '[past part + 3rd pers] beat'

b) [mere tere]che
c) marche tarche '[3rd pers pres] beat'
d) *[mar tar]che
This gives more evidence that mere is a unit which can stand on its own.
3.3. Evidence that the clitics differ from superficially similar affixes can
be obtained from phonological rules of deletion and shortening. Deriv-
ative vocalic suffixes can trigger vowel shortening in stem vowels and
deletion of vowels; clitics, however, do not trigger such processes. The
examples in (16) are taken from Dasgupta (1984).
(16a) na:k 'nose'
b) naki 'nasal'
c) ra:g 'anger'
d) ragi 'angry'
e) na:k-o 'the nose too' (not: *nak-o)
0 ra:g-o 'anger too' (not: *rag-o)
g) no:t '(male) dancer'

h) noti '(female) dancer'
j) no:t-i '(male) dancer/[+emphatic]'
k) pagol 'idiot'
1) pagli 'mad woman'
m) pogol-i 'idiot/[ + emphatic]'
To summarize, we have shown that the clitic jo/ and jij are different from
derivational suffixes and have as their minimal host category the pho-
nological word. The latter fact, however, still does not account for the
fact that sentences such as (7a) are ungrammatical. It seems unreasonable
to suggest that chele does not constitute a minimal phonological word. 4
In the next section, we will therefore explore a completely independent
line of reasoning.
4. Emphatic clitics in syntax and logical form
Let us make the assumption that the clitics under consideration impose
quantificational properties on their morphological/syntactic domain, sim-
ilar to only and even in English. As Rooth (1985) and others before him
have argued, a phrase narrowly focused by only, even, etc., must be

interpreted with respect to a quantificational domain. Such a domain is
naturally provided by the verb, although not necessarily by the verb. Let
us assume that (17a) below is an S-structure, (17b) is the logical form
derived from it, and (17c) is a rough semantic representation which
transduces the logical form into a proposition with a universal quantifier
having scope over it.
(17a) we saw only John

b) [only John], [we saw Xj]
c) For all χ [we saw χ —> λ: = John]
Only John like John-o is a quantifier which must be assigned scope over
the clause (proposition) at the level of logical form. Scope assignment,
however, is constrained in language-specific ways. For instance, even in
English cannot appear in an unconstrained fashion, although there is no
prima-facie semantic reason which could prevent this.
(18a) They have killed [NP even [NP my dog]]
b) *They have killed [NPmy [N' even [N- dog]*\\
As Bayer (1988, 1990) has shown, only, even, etc., and their respective
correspondents exhibit different island effects in Dutch, English, and
German. (See also Longobardi (in press) for Italian.) Once they are
adjoined to an XP which does not correspond to the predicate of a root
sentence, this XP must be canonically governed by a verb, or it must
connect to a "dynasty" of uniformly oriented governors in the sense of
Köster (1986). According to Köster, with the exception of VP, all maximal
projections XP of lexical categories are virtual bounding nodes. A bound-
ing domain can, however, be extended when XP is governed by an element
which conforms to the basic orientation of government in the language.
The direction that counts as basic in a VO-language such as English is
— w h i l e it is <— in OV-languages such as German and Bengali. Some
examples contrasting English and German will show what is at stake.
(19a) John would [even [talk to MARY]] (adjoined to VP)

b) John would talk [to [even [MARY]^\
(adjoined to N P inside PP)
(20a) weil Hans [sogar [mit MARIAN sprechen würde

(adjoined to PP)
10 Josef Bayer and Adit ι Lahiri
b) *weil Hans [mit [sogar [MARIAN sprechen würde
(adjoined to NP inside PP)
Adopting the rule of quantifier raising (QR) as suggested in May (1977),

English allows for even Mary to undergo QR because Ρ governs in the
same direction as V: the PP ceases to be a bounding node, and (19b) is
well-formed. English employs uniformly rightward-looking governors,
but German employs a mixed system: Ps that govern to the right, but Vs
that govern to the left. Since sogar Maria is governed by a rightward-
looking governor, the PP counts as a bounding node, and QR is impos-
sible as (20b) shows. The derivation of well-formed logical forms seems
to be constrained by the following principles.
(21a) Focus-sensitive quantifiers {only, even; nur, sogar; -ο etc.) must

have access to a domain of quantification.
b) Raising to S (or at least to predicate-level) provides a domain
for quantification.
c) Governed quantifiers must be (canonically) governed in a dynasty
(gi, ..., g n ) of uniformly oriented governors, up to a tree height
where a quantification domain is found.
4.1. Bengali does not have a mixed system as German does. It has
exclusively postpositions. Since it is an OV-language, PPs should not lead
to island effects. At least partially, this expectation is borne out.
(22a) ami fipra- r- ο fange kotha bolbo

I Sipra- [gen]- too with talk say [fut]
Ί will talk with also SIPRA'
b) ram- er- i fommondhe alocona hocchilo
Ram- [gen]- [emph] about discussion was-taking-place
'Discussion took place about RAM'
There are other postpositions, however, which do not allow similar

constructions. Diye 'with', derived from the past participle of deowa
'give', for example, does not. It is possible that by careful classification
an independent difference between Bengali postpositions could be derived
which could account for the asymmetry. 5
4.2. Like in German, Bengali adjectives take NP-complements to the

left. As expected, island effects are absent here, too.
(23) weil das Gericht []\dem Patienten] sogar] bekömmlich] ist

since the meal the patient even suitable is
'since the meal is even suitable for the PATIENT'
(24a) ram- er- ο prio

Ram- [gen]- too dear
'dear to Ram, too'
b) babul- er- ο bhokto
Babul- [gen]- too fond
'fond of Babul, too'
4.3. Quantified NPs which are adjoined to X max can freely undergo
quantifier raising because they are ungoverned. The X max to which they
belong does not count as a bounding node. The Bengali possessor-NP
appears to be adjoined, i.e., not in [SPEC, NP]-position as is the case in
English or Standard German. For lack of space we have to simplify
somewhat and propose the following structures. 6
(25a) [NP. [NP. amar baba- r] [NP. oi bari ]]
my father [gen] this house
'this house of my father's'
b) [npj [npj my father 's] [N< house]]
c) *my father's this house
Both the grammar of syntactic movement and the grammar of scope
assignment behave accordingly, as we will show next.7
(26) NP-split in the syntax
a) tumi [kon lok- ta- r bari] dekhecho?
you which man- [defj- [gen] house seen-have
'Which man's house did you see?'
b) [kon lok- ta- r] tumi [bari] dekhecho?
This is not possible in English. It would violate what has since Ross
(1967) become known as the "left branch condition". 8
(27a) [ Whose house] did you see?
b) *[ Whose] did you see [house]?
Exactly the same constraint seems to be at work in logical form, as

indicated by the grammaticality difference between (28) and (29):9
(28) NP-split in logical form
[amar ΒAB A- r- o] bari bikri hoe gseche
my[gen] father- [gen]- too house sold become has-gone
'Also my FATHER'S house was sold'
(29a) *[my FATHER even's] house was sold
b) *[my father TOO's] house was sold
Let us assume that Ν does not lexically govern a structural position such
as [Spec, NP]. Despite the canonical direction of government that holds
in a language, this guarantees that NP is an island for the specifier, while
it is not an island for an adjoined position such as the possessor-NP in
(25a).
We can now make the following generalization: 10 only those quantified
elements X can undergo quantifier raising which are either adjoined to
some YP or linked in a dynasty. A dynasty is built by a chain of successive
governors when they govern in the same canonical direction. [ x X particle]
or [χ particle X] is c(anonically)-governed, if the minimal maximal cate-
gory dominating X contains a governor G which precedes X in a right-
branching language and follows X in a left-branching language. In the
following examples, these requirements are not met:
(30a) *ami [NP nil- ο bari] dekhechilam
I blue- too house seen-have
Ί have seen a house which is also BLUE'
b) * They have killed [NP my [N- even [N dog]]\
c) *fipra [NP chele- o- ke] mereche]
Sipra boy- too- [acc] beaten-has
'Sipra has beaten also the BOY'
d) *tader ο] [v ρ or]]- a]] bhalo laglo
their poetry- too reading pleased-has
(30a —c) are ruled out because the NP is the minimal maximal category
dominating the quantified X, and X fails to be c-governed in NP. The
category dominating the quantified nil-o in (30a) is the NP nil-o bari;
analogously, the relevant category in (30b) and (30c) is the NP my even
dog and chele-o-ke respectively. Notice that /o/ in (30c) c-commands only
the uninfected noun. After the attachment of the case-suffix -ke, which
we do not consider to be a governor, the quantified element can only be
the bare noun. The dominating minimal maximal category is the NP

across which the quantificational domain cannot be extended. Under the
assumption that kobita-o por-a in (30d) ( = 8b) is a VP which undergoes
nominalization due to the affixation of -a, por c-governs kobita-o, but
the dominating and intervening N P blocks the formation of a dynasty
with the verb (bhalo)laglo. Obviously, the nominalized VP itself does not
count as an appropriate quantification domain. Thus, all the examples
in (30) arguably constitute "interpretive islands" in the sense of Bayer
(1990).
5. An apparent paradox
The results achieved in Section 3 left us with an apparent paradox: when

the minimal lexical category for the affixation of an emphatic clitic is the
phonological word, why is the clitic only allowed inside inflected verbs,
but not — as we have just seen — inside inflected nouns? In the light of
our discussion in Section 4, the answer to the problem is quite obvious.
Following Chomsky (1986), we assume that S is a maximal projection of
the inflectional element I(NFL), i.e., S = IP. Apart from this, however,
we do not assume that in Bengali I heads a separate node in the syntax.
Rather, I being an inflectional feature seems to merge with V such that
it becomes the formal head of IP, while V becomes the semantic head. 11
Wherever the emphatic clitic attaches to the inflected verb — observing
the morphophonemic constraint — it will quantify over an element that
is the head of the clause. Irrespective of the availability of a c-governor,
the cliticized/quantified past participle or the cliticized/quantified [V + I]
will automatically percolate to a domain where quantification can take
place. (31), which means 'Sipra has also BEATEN the boy', symbolizes
this percolation process with a dotted line.
14 Josef Bayer and Adit ι Lahiri
(31) IP
V CLITIC Γ + tense]
L+AGRJ
mer-e- ο -che
beat-en too (past, 3pers)
Notes
1. mar is the V-stem.

2. mer-e is the form of the past participle.
3. The stem vowel of verbs alternates in specific morphological environments.
4. One piece of evidence is that the above discussed reduplication rule produces better
results when the uninfected noun chele is affected: [chele tele]-ke versus V.[chele-ke
tele-ke].
5. See Kayne (1981) where the question is raised why English but not French or other
Romance languages allow for preposition stranding. Kayne argues that prepositions in
English are structural governors, but not in French. Bengali could be a language with
a "mixed system" in the sense of Kayne's proposal. The details of such an account,
however, would have to be worked out.
6. A more refined analysis could probably be given with the hypothesis that our NP is in
fact a determiner phrase (DP). See Abney (1987).
7. See also the similar case of Hungarian (Szabolcsi, 1984). Notice that Bengali keeps
WH-words in situ like Chinese and Japanese. What looks like WH-movement in (26b)
is rather a case of "scrambling".
8. In the framework of Chomsky (1986) one could account for the left-branch condition
because adjunction to arguments (here NP) for purposes of movement is generally not
allowed in this theory.
9. Notice that post-NP even as well as too is generally acceptable, as in my FATHER even
would agree on that and my father TOO would agree on that. Under the assumption
that the English genitive -s is a clitic which is unselective with repect to its host (cf.
Zwicky — Pullum 1983), the examples in (27) cannot easily be excluded otherwise.
10. The predicate "quantified" here pertains only to the cases under consideration, namely
focusing particles and emphatic clitics.
11. On V as the head of S, see Jackendoff (1977). On the distinction of syntactic and
semantic heads, see Abney (1986). Notice that in Bengali, I cannot take VP as its
complement because this would predict that in babulke mere-o-chilam (Babulfobj] beat-
too-have [I]) Ί/we have also beaten Babul' the clitic can focus on babul. This, however,
is never possible. The clitic can only focus on the verb in such cases.
References
Abney, Steven
1986 "Functional elements and licensing" [Paper presented at the GLOW con-
ference Girona, Spain],
1987 The English noun phrase in its sentential aspect [Dissertation, MIT, Cam-
bridge, MA],
Bashir, Ε. —M. Deshpande —P. Hook (eds.)
1987 Selected papers from SALA 7. (Bloomington: Indiana University Linguistics
Club).
Bayer, Josef
1988 "Directionality of government as a locality constraint for scope assignment"
[Paper presented at the GLOW conference Budapest, Hungary],
1990 "Interpretive islands: evidence for connectedness and global harmony in
logical form", in: G. Grewendorf — W. Sternefeld (eds.).
Chomsky, Noam
1986 Barriers. (Cambridge, MA: MIT Press).
Dasgupta, Probal
1984 "Bangla emphasizers and anchors", Indian Linguistics 45: 102—117.
1987 "Sentence particles in Bangla", in: E. Bashir —M. Deshpande —P. Hook
(eds.), 4 9 - 7 5 .
[ms.]. "The word in Bangla", unpublished paper, Pune, Deccan College.
Grewendorf, G. —W. Sternefeld (eds.)
1990 Scrambling and barriers. (Amsterdam: Benjamins).
Huang, J. —R. May (eds.)
in press Logical structure and linguistic structure. (Dordrecht: Reidel).
Jackendoff, Ray_
1977 X-Syntax. (Cambridge, MA: MIT Press).
Kayne, Richard S.
1981 "On certain differences between French and English", Linguistic Inquiry 12:
349-371.
Koster, Jan
1986 Domains and dynasties. (Dordrecht: Foris).
Longobardi, Giuseppe
in press "In defense of the 'correspondence hypothesis': island effects and parasitic
constructions in logical form", in: J. H u a n g - R . May (eds.).
May, Robert
1977 The grammar of quantification [Dissertation, MIT, Cambridge, MA],
Rooth, Mats
1985 Association with focus [Dissertation, University of Massachusetts, Amherst].
Ross, John R.
1967 Constraints on variables in syntax [Dissertation, MIT, Cambridge, MA],
Szabolcsi, Anna
1984 "The possessor that ran away from home", The Linguistic Review 3:
89-102.
Zwicky, Arnold — Geoffrey Pullum
1983 "Cliticization vs. inflection: English n't", Language 59: 502 — 513.
Phonologically conditioned suppletion*
Andrew Carstairs
In some morphological alternations, the alternants are dissimilar pho-

nologically (sometimes as dissimilar as the prototypically suppletive pair
go and went), yet the distribution of the alternants depends on phono-
logical characteristics of the context. We may call this phenomenon
"phonologically conditioned suppletion". Quite a few instances of it have
been noticed, and the logical independence of phonological conditioning
and phonological similarity has been pointed out by Mugdan (1986: 34);
but no one seems to have examined the general bearing of phonologically
conditioned suppletion on morphological or phonological theory. Some
implications of this suppletion for lexical semantics and the inflection-
derivation distinction have been discussed elsewhere (Carstairs 1988); the
present paper looks at its phonological and morphophonological impli-
cations, at least in a preliminary fashion.
At (1) —(11) is a selection of examples:
(1) Hungarian: 2nd singular indefinite present indicative: -ol after
sibilants and affricates, -(a)sz elsewhere (Sauvageot 1951: 72;
1971: 85, 318).
(2) Turkana: deverbal abstract nouns: suffixes -isi, -u and -f^C,

(subject to vowel harmony) after roots with structure CVC,
qViQViC, and CVCVjQ, respectively (Dimmendaal 1987:
205-206; cf. 1983: 270-274).
(3) Seri: Passive prefix on verbs: p- before vowel-initial roots, a:9-

before consonant-initial ones (Marlett — Stemberger 1983).
* Acknowledgements are as for Carstairs (1988) with the addition of Brent de Chene,
Wolfgang Dressier, Brian Joseph, Paul Kiparsky, and Steve Marlett. Faults remain my
responsibility. The work was supported by University of Canterbury English Department
research grant 87/2.
18 Andrew Carstairs
(4) Kalkatungu: Dative suffix on nouns: -ku after consonant-final

stems, -V,· after Vj-final stems (Blake 1979: 44, quoted by Mugdan
1986).
(5) Fulfulde (Southern Zaria district): General future active tense:

suffixes -Vit- before Vj-initial suffixes, -ay- before consonant-
initial ones (Mary Mcintosh, personal communication; cf. Arnott
1970: 53, 213, 224).
(6) Turkish: 3rd singular possessive marker on nouns: suffix (subject

to vowel harmony) -i after consonants, -si after vowels (Lewis
1967: 39).
(7) Italian: finire 'finish' (and similarly for many other verbs of the
-ire conjugation): stem fin- when unstressed, finisc- (i.e., [fi'nisk]
or [fi'niJJ], according to context) when stressed (Matthews 1981).
(8) Italian irregular preterit stems: romp- 'break', mov- 'move', prend-
'take', etc., when unstressed; rupp-, moss-, pres-, etc., when
stressed.
(9) Sanskrit: asthi 'bone' (and similarly for three other nouns): stem
asthi word-finally and before consonant-initial case-number suf-
fixes; asthn- or asthan- or astha.n- (according to context) before
vowel-initial suffixes (Whitney 1889: 122, 160).
(10) French oeufs 'eggs': [0] after [z], [oef] elesewhere, e.g., les oeufs,
des oeufs, trois oeufs with [0] versus quatre oeufs, cinq oeufs, huit
oeufs with [cef] (Swiggers 1985).
(11) Italian masculine definite articles: lo or /' (singular) and gli

(plural) before vowels and any cluster of one or more consonants
containing a sibilant (except [ # s V ] ) or a palatal sonorant; il
(singular) and i (plural) elsewhere (Plank 1984: 332 — 335; Dress-
ier 1985).
For some of these, one can imagine an analysis in which the surface
alternants are all derived from a single underlying phonological represen-
tation by brute force, invoking otherwise unmotivated phonological rules.
This is particularly so where the suppletion is only partial — that is,
where the alternants share an invariant portion, as in (6) —(9) and (11).
But grossly suppletive examples like (1) —(5) show that this approach will
not always work. There is no escaping the conclusion that morphological
alternants may have phonologically complementary distributions without
sharing phonological representations at any level. The consequence for
phonological theory is that the existence of a phonologically conditioned
alternation does not by itself prove the existence of some synchronic
phonological process giving rise to it. This consequence is serious. Few
linguists now hold that the main function of phonological rules is to
account for morphological alternations. Even so, many phonologists still
implicitly assume that suppletive alternants can be distributed only on a
grammatical or lexical basis (like English past went versus non-past go,
or regular -es in foxes versus lexically marked -en in oxen), not on a
phonological basis, and consequently that a phonologically predictable
distribution always implies a shared phonological underlier and phono-
logical rules to generate the surface shapes. If the mistakenness of this
assumption had been recognized earlier, phonological theory might have
been spared some of its morphological digressions of the late 1960s and
early 1970s.
Despite that sobering conclusion, there still appear to be interesting
phonological restrictions on where and how phonologically conditioned
suppletion occurs. In all of (1) —(9) the phonological determinants of the
alternation are within the same word as the alternants, and we can
probably say the same of (11) if we interpret "word" as "phonological
word" or perhaps "clitic group" (Nespor — Vogel 1986). We do not find
phonologically conditioned suppletions in which the conditioning factors
reside in other words within the same phrase or clause. One apparent
exception is (10). But there is evidence that French speakers are uncom-
fortable with the distribution of alternants for oeuf described by Swiggers
(1985). In some varieties of colloquial French the [(z) 0] alternant is
maneuvered into all plural contexts, with (for example) cinq oeufs being
replaced by cinq [z] oeufs or cinq beaux oeufs (i.e. [... boz0]). It does not
matter for our present purposes whether this [z] is analyzed as part of
the preceding word, as a plural prefix, or as part of a new stem alternant
[z0]. The important point is that the distribution of the alternants is no
longer inescapably dependent on phonological factors; it can now be
analyzed instead as an 'ordinary' suppletion, based on number ([cef]
singular, [0] or [zo] plural) — a development which may have been
motivated by pressure to eliminate a phonologically conditioned supple-
tion of an unusual and highly marked kind. If so, the examples at (1) —(11)
fortify Plank's (1984) suggestion that "the locality requirement on the
triggering of morphophonemic rules be extended to rules of exponence",
20 Andrew Carstairs
as Zwicky (1985: 435) puts it; for phonologically conditioned suppletion,

"locality" seems to involve not only linear adjacency but also membership
of the same constituent, in some sense.
A second restriction relates to the relative positions of the suppletive
alternants and their conditioning environment. Both in "lexical phonol-
ogy" (Kiparsky 1982; Mohanan 1986) and in the "extended word-and-
paradigm" model of inflection (Anderson 1982), complex words are built
up from the root outwards, by a process like the accretion of layers on
a hailstone. In these models, it is unsurprising to find the shape of affixes
being conditioned by phonological characteristics of stems, as in (1) —(4);
it is startling, however, to find stem shape apparently conditioned by
characteristics of affixes, as in (7) —(9), or one affix conditioned by a
more peripheral one, as in (5) — what we may call "inward" as opposed
to "outward" conditioning. How can the choice between alternants be
effected by the phonological shape of material which has yet to be added?
But closer examination of these instances suggests that the conditioning
that they involve is of a kind that hailstone models ought to be able to
accommodate without compromising their spirit, in that they all seem to
involve alternants which are either phonetically uninterpretable by them-
selves or else analyzable as "empty morphs", similar to the "intermorphs"
which have long been recognized in descriptions of Slavic word formation
(Szymanek 1985).
Consider the Italian examples at (7) and (8), involving stress. In most
current treatments of stress, whether metrical or autosegmental, it is
regarded as inherently relational, in some sense; few phonologists are
happy any longer with [+ stress] as a vowel feature on a par with
[ ± back]. So, although the lexical entry for (say) the Italian verb rompere
must specify in some fashion that its preterit stem rupp- is intrinsically
stressed, this specification is uninterpretable until rupp- has been incor-
porated into some unit (say, a phonological word) within which stress
relationships are statable. The right way to look at the alternation between
rupp- and romp- in the preterit, I suggest, is not in terms of inward
conditioning of the stem by affixes but in terms of suprasegmental
characteristics of the word as a whole; given that only one primary stress
per word is permitted, the choice of a stressed suffix (whether its stress
is lexically determined or not) precludes the choice of an intrinsically
stressed alternant of the stem, and so enforces the stress-indifferent
'default' alternant romp-. (I call romp- "stress-indifferent" because outside
the preterit it may be stressed, e.g., present römp-o Ί break'.) The
conditioning of the alternants fin- and finisc- will operate similarly, except
that -isc- can plausibly be detached from fin- as a kind of semantically
and morphosyntactically empty intermorph whose sole function is to

protect fin- from ever bearing stress; Matthews (1981: 62) calls it "an
augmentation of the root in cases where the accent remains on it". The
only thing which distinguishes -isc- from the majority of the intermorphs
which Szymanek (1985) discusses is the fact that its distribution is de-
pendent on phonological factors.
The Fulfulde alternation at (5) likewise involves one alternant which
is uninterpretable on its own. The -Vt- alternant copies an immediately
following vowel, so it is phonetically incomplete until some following
suffix has provided a vowel to copy. Once a following suffix has been
added, however, not only its first vowel but its whole phonological
composition is available for inspection, and in particular its initial con-
sonant, if any; so the fact that a consonant-initial suffix triggers the -ay-
alternant rather than the -Vt- alternant does not involve any inward
conditioning beyond what was required anyway to fill in the unspecified
vowel. 1
It is not an accident, I suggest, that the Italian and Fulfulde instances
of "inwardly conditioned" suppletion both involve phonetic uninterpret-
ability. What would be really serious for hailstone models of morpholog-
ical organization would be instances of inward phonological conditioning
in which the rival alternants are just as fully specified phonologically as
those in the Hungarian and Seri examples at (1) and (3). For the time
being, it seems reasonable to propose that a universal restriction on
phonological conditioning excludes inward conditioning of that kind.
Inward conditioning is allowed, in other words, only where it is essential
in order to make one of the relevant alternants pronounceable.
At first sight, the Sanskrit alternation at (9) breaches the universal
restriction just proposed. The stem alternants asthi and asth(a(:))n- are
all interpretable phonetically as they stand, yet are seemingly subject to
inward conditioning on the basis of the segmental composition of the
suffixes. But this presupposes that the variable portions of these "stems",
namely -/' and - ( a ( : ) ) n - , really are to be analyzed synchronically as
belonging to the stem. At least as plausible is an analysis under which
these are seen as morphosyntactically empty elements which accompany
the case-number suffixes of asth- under certain conditions. In forms such
as the instrumental plural asth-i-bhih or the genitive singular asth-n-ah,
we are not dealing with sequences of two affixes proper (that is, two
affixes each with its own morphosyntactic or semantic content), such that
that shape of the inner affix is paradoxically dependent on an affix which
is added "later". Rather, we are dealing in each instance with only one
22 Andrew Carstairs
affix proper {-bhih or -ah), which can plausibly be analyzed as added "at
the same time as", rather than "after", the empty intermorph which
precedes it. The conditioning involved here is quite complex, in that the
need for an intermorph and the range of alternants for it must be specified
in the lexical entry for the stem, while the choice of the individual alternant
is determined by phonological and grammatical properties of the case-
number suffix; but none of this conditioning seems to violate the spirit
of hailstone models.
If we accept that phonological factors may affect the incidence and
shape of intermorphs, we will expect to find instances where these factors
reside in the stem rather than in an accompanying affix; and we do find
them. The most plausible analysis of the Turkish example (6) involves
taking -s- as an intermorph which is inserted between vowel-final stems
and the possessive affix proper. An analysis under which -si as a whole
is simply an independent alternant of third singular possessive, alongside
-i, is equally compatible with hailstone models, because the conditioning
involved is outward rather than inward; but the fact that an intermorph
analysis is also available provides some reassurance that the resort to an
intermorph analysis in an instance like (9) is not a resort to an otherwise
unmotivated ad hoc device.
There are a variety of reasons why phonological theory has moved
away from its earlier preoccupation with morphological alternations.
Although much remains to be discovered about the incidence and impli-
cations of phonologically conditioned suppletion, I hope to have shown
that it constitutes one more reason why that move is wise.
Notes
1. The terms "phonetically uninterpretable" and "phonetically incomplete" here are not to
be confused with "underspecified", as used by, e.g., Kiparsky (1982) and Pulleyblank
(1988). Underspecification in underlying representations is intended to avoid duplication
between "morpheme structure conditions" and phonological rules. That issue is not
directly connected with the sort of phonetic incompleteness discussed here.
References
Anderson, Stephen R.
1982 "Where's morphology?", Linguistic Inquiry 13: 5 7 1 - 6 1 2 .
Arnott, D. W.
1970 The nominal and verbal system of Fula. (Oxford: Clarendon Press).
Blake, Barry J.
1979 A Kalkatungu grammar (Canberra: Australian National University)
(= Pacific Linguistics Series B, No. 57).
Carstairs, Andrew
1988 "Some implications of phonologically conditioned suppletion", Yearbook
of Morphology 1: 67 — 94.
Dimmendaal, Gerrit J.
1983 The Turkana language. (Dordrecht: Foris).
1987 "Drift and selective mechanisms in morphological change: the Eastern
Nilotic case", in: A. G. Ramat et al. (eds.) Papers from the 7th International
Conference on Historical Linguistics, (Amsterdam: Benjamins), 193 — 210.
Dressier, Wolfgang U.
1985 "On the definite Austrian and Italian articles", in: E. Gussmann (ed.) Phono-
morphology: Studies in the interaction of phonology and morphology, (Lublin:
Wydawnictwo Katolickiego Uniwersytetu Lubelskiego), 35—47.
Kiparsky, Paul
1982 "From cyclic phonology to lexical phonology", in: H. van der Hulst —Ν.
Smith (eds.) The structure of phonological representations (I), (Dordrecht:
Foris), 1 3 1 - 1 7 5 .
Lewis, Geoffrey L.
1967 Turkish grammar. (Oxford: Oxford University Press).
Marlett, Stephen Α. —Joseph P. Stemberger
1983 "Empty consonants in Seri", Linguistic Inquiry 14: 617 — 639.
Matthews, Peter
1981 "Present stem alternations in Italian", in: H. Geckeier et al. (eds.) Logos
semantikos: Studia linguistica in honorem Eugenio Coseriu, (Berlin/Madrid:
de Gruyter/Gredos), vol. 4 (Gramätica), 57 — 64.
Mohanan, Karuvannur P.
1986 The theory of lexical phonology. (Dordrecht: Reidel).
Mugdan, Joachim
1986 "Was ist eigentlich ein Morphem?", Zeitschrift für Phonetik, Sprachwissen-
schaft und Kommunikationsforschung 39: 29 — 43.
Nespor, Marina —Irene Vogel
1986 Prosodic phonology. (Dordrecht: Foris).
Plank, Frans
1984 "Romance disagreements: phonology interfering with syntax", Journal of
Linguistics 20: 3 2 9 - 3 4 9 .
Pulleyblank, Douglas
1988 "Vocalic underspecification in Yoruba", Linguistic Inquiry 19: 233 — 270.
Sauvageot, Aurelien
1951 Esquisse de la langue hongroise. (Paris: Klincksieck).
1971 L'edification de la langue hongroise. (Paris: Klincksieck).
Swiggers, Pierre
1985 "How to order eggs in French", Folia Linguistica 19: 63 — 66.
Szymanek, Bogdan
1985 "On intermorphic extensions in English and Polish", in: E. Gussmann (ed.)
Phono-morphology: Studies in the interaction of phonology and morphology,
(Lublin: Wydawnictwo Katolickiego Uniwersytetu Lubelskiego), 177 — 191.
Whitney, William D.
1889 Sanskrit grammar. (Cambridge, Mass.: Harvard University Press).
Zwicky, Arnold M.
1985 "Rules of allomorphy and phonology-syntax interactions", Journal of Lin-
guistics 21: 431—436.
On a universal criterion of rule coherence
Michael Shapiro
One of the factors commonly retarding progress in linguistic theory is a

kind of amnesia affecting its exponents, whereby well-grounded, highly
productive principles of language structure discovered in the past are
forgotten in contemporary discussions. The principle of isomorphism has,
alas, suffered from just such neglect — with predictable consequences. 1
In its simplest form this principle states that different levels of language
structure embody identical rules of organization. The first explicit appli-
cation of this principle was made in Jakobson (1932), and the first
significant recognition of the pervasiveness of isomorphism between the
different levels of language was achieved by Hjelmslev (1938).
One of the consequences of the discovery of the isomorphism principle
was an important shift in the understanding of linguistic arbitrariness
(associated chiefly with de Saussure). Due largely to Jakobson's studies
after the war, it became increasingly clear that the core of language is
constituted by extensive patterns of similarity and difference among the
shapes of grammatical morphemes which correspond to relations of
similarity and difference among their meanings. In semiotic terms such
correspondences between relations on the expression level and the content
level of languages are to be understood as diagrams, relations represented
by relations. In recent discussions (to the limited extent that this idea is
explored in depth) 2 mapping relations of this sort have come under the
designation of "iconicity" in grammar (e.g., Haiman 1985).
There is another, equally fundamental sense in which isomorphism
can be said to pervade the structure of language, namely the sense in
which rules at the core of grammar are not merely statements of regu-
larities but are coherent. The notions associated with the terms "rule"
and "coherence" need to be examined separately. Although the concept
of rule was not prominent among the theoretical advances of the early
European structuralists, it is nonetheless clear that its ubiquitousness
today owes much to an understanding of grammatical relations as pat-
terning and regularity that goes back to prewar discussions (principally
26 Michael Shapiro
in Prague and Copenhagen) of the foundations of linguistic theory. What

is missing from both pre- and postwar theorizing, however, is the notion
of the coherence of linguistic relations, and as a corollary, the precise
means whereby coherence is to be expressed in the practice of linguistic
description.
All along, the potential for making coherence an explicit principle in
the understanding of language structure existed unexploited among the
many overt achievements of early structuralism, specifically in the idea
of markedness. Coherence obtains when rule relations signify the mirror-
ing of markedness values, either parallelistically or chiastically (more on
this later), across content and expression levels, or between different
aspects of expression (as in the case of some morphophonemic congru-
ences). In the latter case — which will be the focus of this paper — the
term automorphism has been proposed (Haiman 1985: 4, cf. Shapiro 1987:
165), imported into linguistics from mathematics. 3 Since patterning is
present at all levels of grammar, to the extent that the rules of language
structure expressing this patterning reflect congruences of markedness
values we can attribute their coherence (their raison d'etre) to such
cohesions. What is more, we can do this uniformly by virtue of the
isomorphism of grammar. Nothing proves the validity of this universal
notion of coherence better than the evidence of linguistic change. The
drift of a language involves the actualization of patterns that are coherent
in just this sense and the rejection of those that are not (Shapiro 1985;
cf. Plank 1979, 1981; Andersen 1980b: 203).
Rules are more than mere generalized formulas of patterns when they
embody specifications of coherence between linguistic elements, namely
cohesions between units and contexts. This criterion of rule coherence
remains true and valid but practically vague without the necessary in-
volvement of markedness because it is markedness that provides the
explicit means of expressing coherence. In order to illustrate how this
criterion works to make linguistic analysis aspire to explanatory adequacy
I have chosen a set of morphophonemic alternations in English that affect
the distribution of tenuis and media obstruents as between nouns (also a
few adjectives) and verbs:
Nominals Verbals
[-f] proof [-v] prove
safe save
wife wive
[-Θ] mouth [-Ö] mouth

sheath sheathe
loath loathe
[-s] diffuse [-z] diffuse
close (adj) close
refuse refuse
advice advise
(ab)use (ab)use
excuse excuse
house house
This alternation between /f θ s/ and /ν δ ζ/ also affects (albeit to a
much lesser extent) English plural formation, e.g., staff/staves [obsolete],
hoof/hooves, wolf'/wolves, oath/oaths (cf. cloth/clothes, identifiable only by
etymological analysis), and house/houses. Since English orthography reg-
ularly reflects the media obstruent in the case of [-v] by writing ν and [-Ö]
by writing e after th (not always; cf. sheathe but mouth above), the
interesting and numerically greater set of examples is the one exhibiting
an alternation between [s] and [z] — which subsumes orthographically
atypical items like advice and device. The alternation is basically unpro-
ductive as it affects noun/verb pairs in contemporary English, but its
persistence requires ultimately that some accounting be reached as to its
raison d'etre, which is to attempt an answer to the question "Why does
it persist?" It is certainly not unusual for languages to rationalize unpro-
ductive pockets of the grammar by investing morphophonemic alterna-
tions with just the sort of motivation we are seeking here (cf. Shapiro
1969). Where no motivation arises, an alternation is liable to be elimi-
nated, as it is in some samples of the English case under consideration
(see below).
To proceed, then, to answer the question in terms of grammatical
coherence, one must examine the markedness values of the sounds and
the meanings of the grammatical categories in which they are imple-
mented. Taking the markedness values of the meanings first, we know
that verbs are marked vis-ä-vis nouns in that verbs necessarily make
reference to the time axis while nouns do not. Given the marked value
of verbs and the correspondingly unmarked value of nouns we would
expect the distribution of tenuis and media obstruents to mirror the
markedness values of the grammatical categories.
At this point we have to be careful about exactly what "mirror" means.
In most discussions of iconicity or isomorphism, it is taken for granted
28 Michael Shapiro
that the mapping of sound/meaning relations is one of simple replication;

indeed, where the values are at variance with each other, such instances
are termed "counter-iconic" (Wurzel 1984: 204).4 My own extensive
investigations of this issue (with particular reference to the structure of
Russian; cf. n. 3) have led me to the conclusion that replication of
markedness values between expression and content is typically confined
to morphology, defined as the domain of relations between basic signs
and contrasted to morphophonemics, defined in turn as the domain of
relations between contextual variants of the same linguistic sign(s) (An-
dersen 1969: 807; cf. 1980b: 89 — 90). In privileging chiastic semiosis —
complementation of markedness values — morphophonemics is only
being true to its nature, i.e., as the domain of grammar defined by
contextual variation of signs ( = alternation). Because this is an area of
the theory of grammar that has not been sufficiently investigated, any
assertions must necessarily be tentative. Thus, chiasmus as a typical form
of semiosis in morphophonemics may ultimately turn out to pertain to
stem alternation but not to the contextual variants of desinences (cf.
Andersen 1980a: 4 3 - 4 4 ) . 5
To a significant extent, of course, this whole way of looking at linguistic
coherence turns on the determination of the markedness values themselves
— not always an easy or straight-forward task for learners as well as
analysts. But there is a heuristic element to which the notion of coherence
gives learner and linguist alike direct access, and it is this element that
becomes particularly helpful in resolving problematic cases (Shapiro
1974b; cf. Anttila 1980: 276 for an assessment). In the English data
utilized for purposes of illustration above, for instance, it is clear that
tenuis obstruents co-occur with the unmarked category of nouns, while
media obstruents co-occur with the marked category of verbs. Assuming
this to be an instance of morphophonemic iconicity affecting stem alter-
nants and invoking the principle of chiastic semiosis, we ought to expect
the markedness relations between sounds and meanings to be oppositely
valued ( = markedness complementarity). We have already stipulated the
values of the two categories involved, noun and verb, but what about
the stem-final obstruents?
Trubetzkoy claimed (1962: 142) that it is impossible to say whether
English has distinctive voicing or distinctive protensity (tense vs. lax) in
its obstruent system, but the concept of rule coherence with its reliance
on markedness considerations now makes this agnosticism seem ground-
less. Positions of neutralization are diagnostic in this respect because
neutralization rules provide contexts in which variation rules tend typi-
cally to produce diagrams of the markedness values of the terms of

phonological oppositions (Andersen 1979: 381). The relation between
syllable peaks and contiguous obstruents in English is such that syllable
peaks are [ — long] before tense obstruents but are [ + long] before lax
obstruents, sonorants, and in final position. Hence beet is [bit], but bead,
beam, and bee are [bi:d], [bi:m], and [bi:], respectively. Beyond the fact
that the complementary distribution here is semiotically significant — as
a sign of the non-distinctiveness of quantity in English — it is the
stipulation of the tenuis and media obstruents as distinctively tense vs.
lax that allows the variation rule to be coherent. Since tense obstruents
are marked relative to the unmarked lax obstruents, and shorter reali-
zations understood as abridgements of syllable peaks are marked relative
to unabridged peaks (which are unmarked), the markedness values of the
vowels replicate those of the contiguous obstruents. If the tenuis and
media obstruents were assumed to implement the opposition voiced vs.
voiceless (as they often erroneously are),6 the variation would lack co-
herence because the markedness values of the obstruents would not match
those of the vowels (voiced obstruents are marked, voiceless unmarked).
This little digression into the phonological problematics of English is
necessary, apart from its practical bearing on the argument about rule
coherence, because it illustrates the methodological status of what is now
commonly referred to as "independent motivation" in linguistic expla-
nation. Trubetzkoy's perception of irresolvability of the English tenuis/
media problem in its obstruent system can be seen as justified only as
long as circularity is barred from explanations of language structure (as
if language were not a hermeneutic object). The mutual dependency of
the elements of the solution proposed above — the shorter realizations
of the syllable peaks seen as abridgments rather than the longer ones as
prolongations, the stipulation of protensity as the relevant phonological
category rather than voicing, and the invocation of markedness consid-
erations as the vehicle of grammatical coherence — testifies to the fact
that these elements cohere as an ensemble of conditions informing the
data. This is the structural coherence that emanates from an evaluation
(intrinsically, in the grammar) of the units and the contexts in tandem,
in a mutually dependent manner, so that randomness and arbitrariness
are reduced to a minimum (if not always to nil).
In an unavoidably circuitous way we now come back to the earlier
examples of alternation between English tenuis and media obstruents in
stem-final position. The alternation becomes coherent when we under-
stand it to subsume two conditions: (1) the alternation is morphopho-
30 Michael Shapiro
nemic, hence the markedness values as between sounds and meanings

will be chiastic (complementary); (2) the obstruents involved are distinc-
tively tense vs. lax. The recognition of these two conditions enables us
to assert a coherence based on markedness values. Otherwise the alter-
nation would be strictly arbitrary, non-iconic, and non-coherent.
When units and contexts do not cohere, the typical outcome in the
long run is a heightened tendency toward the reduction of such instances,
to the limit of their wholesale elimination from the language. With
reference to our English examples there are attested historical changes
that confirm the correctness of the analysis, specifically by showing
morphophonemic coherence in just the sense advanced to be the telos of
the changes. Where coherence has already been reached, no further
changes occur. Thus in the history of English there is evidence (Horn
1954: 799 — 802) of generalization of either the tense or the lax obstruent
in words which now regularly have orthographic s, e.g., enterprise, com-
promise, purpose, promise, practise ( = American practice). In Middle
English texts one can observe the testing of the contemporary rule in the
occasional writing of ζ instead of s, particularly in verbs (but not only).
In a Milton manuscript, for example, one finds the spelling practiz'd; and
as late as 1836, the pronunciation of the infinitive with a ζ is proscribed
as vulgar by normative grammarians (Horn 1954: 800). The same differ-
ence in pronunciation as between the nominal and the verbal forms of
the word evidently obtained for enterprise and compromise in Middle
English, the difference here being in the particular obstruent that was
generalized. In the case of practise and promise, it was the tenuis obstruent
that was generalized; in enterprise and compromise it was the media
obstruent. Exactly why it was s in the first pair and ζ in the second
constitutes a separate problem that might be treated in the spirit of rule
coherence and markedness, but I hesitate to offer an explanation. Perhaps,
in the presence of a primarily or secondarily stressed vowel in the verb
form there is a discernible tendency to generalize the unmarked media ζ
(compromise, but also close, as in at the close of...). Similarly, one might
want to explain the gradual elimination of the pronunciation of greasy
with [-ζ-] as a case of an unmotivated alternation being dropped from
the (standard) language: there is no chiastic distribution of markedness
values within the category of nominals. The semantic split between louse
and lousy would tend to confirm such an analysis albeit obliquely.
Although the focus of this exposition has been on phonology and
morphophonemics, there is no doubt that the principle of isomorphism
in grammar would predict the equal and uniform applicability of the very
same notion of coherence to an understanding of lexis and syntax. Some

efforts in that direction are available here and there in Haiman (1985),
but Battistella (1985) is the first study known to me that utilizes my
methodological strategy exactly in explaining a set of related syntactic
changes. The contemporary English use of the semantically empty or
periphrastic do that serves as a carrier of tense in interrogative, negative,
and emphatic sentences lacking an auxiliary verb {Did anyone tell you the
news?; I didn't get the notice; I did read the assignment!) is absent from
Old English. It can be explained as the result of a development wherein
the telos of change is determined by an alignment of markedness values
(gradually, over the span from Old English to Modern English) that is
generally analogous to the morphophonemic example, even though the
specific form of semiosis (parallelism rather than chiasmus) is different.
The markedness-based isomorphism in these syntactic examples accounts
not only for the spread of do from its originally non-periphrastic locus
to semantically marked paradigms or contexts but also explains its loss
in the unmarked simple declarative forms. 7
There may be ways of defining linguistic coherence other than by
appealing to markedness isomorphism. But coherence thus construed is
clearly a theoretical advance in the understanding of grammatical struc-
ture and language change as unitary phenomenological domains, and
where structure at any given synchronic point is always the cumulative
teleological result of change. The passage of some five decades since the
heyday of our Prague and Copenhagen predecessors need not be an
insuperable obstacle to present-day efforts at achieving a potent theory
of grammar. More cooperation unfettered by ideological or socio-aca-
demic biases is called for (cf. Itkonen 1986). In this enterprise the work
of naturalist grammarians in the German-speaking countries of Europe,
of North American exponents of "natural" or functional syntax, and of
linguists with a semiotic outlook on both sides of the Atlantic will need
to emphasize common structuralist foundations and points of conceptual
convergence in order that progress be made on all methodological fronts. 8
Notes
1. I am using the term "isomorphism" in the traditional sense to refer to grammatical

relations and not in the sense explored by writers like Givon (1985) or Haiman (1985),
Haiman (ed., 1985), namely as a way of characterizing (grosso modo) the diagrammatic
relation between language and world or grammar and reality. Givon and Haiman refer
to such relations as "meta-principles" and talk about the way in which a linguistic code
32 Michael Shapiro
encapsulates experience. As Itkonen (1986: 40) points out, this "discourse-pragmatic"

sense of isomorphism (and, for that matter, of iconicity) goes back at least as far as the
modistic grammarians. Note that Haiman (esp. in the edited volume (1985), but also
before and since) wants to distinguish between isomorphism and diagrammatization, on
the one hand, and isomorphism and economy, on the other, for which he has been
appropriately criticized by Givon (1985: 188) and Itkonen (1986: 39), respectively.
2. See especially Dressier (1985: Chap. 10) and the references therein; also Wurzel (1984,
1985); Mayerthaler (1981; now available in English translation), and Plank (1979, 1981).
A compact English source for work on morphology along naturalist lines is now
represented by Dressler (et al. 1987). All of these studies are by scholars who have been
influenced to one extent or an other by Jakobson (esp. 1965). Their narrow under-
standing of the concept of icon and iconicity has been justly criticized by Andersen
(1980b: 202).
3. In a rare concession of priority, Andersen (1980a: 34) testifies that "patterns... where
one morphophonemic alternation mirrors another so that they together form what may
be called an automorphic structure in the paradigm, were first described explicitly by
Shapiro, who gives several examples of existing or emerging patterns of this kind from
contemporary Russian (1969)." Since Plank, Wurzel, and Mayerthaler do not cite my
work and habitually take their exploitation of the idea of iconicity back to Jakobson
through Anttila, this is perhaps the appropriate time to set matters straight. Anttila
(1969) cites my paper "Toward the recognition of iconicity in language" which was given
in December 1968 at the Winter Meeting of the Linguistic Society of America in New
York and then became part of Shapiro (1969); (he also cites the manuscript version of
Shapiro 1971). Unfortunately, Anttila (1969) appeared in Finnish and was thus not
digested outside of Finland; however, Anttila (1972) refers to Shapiro (1969) and the
semiotic ideas promulgated therein. In subsequent publications, Anttila is scrupulously
consistent in giving credit to his former U C L A colleague, as well as to Andersen. In a
series of publications following on my 1969 book, I continue to credit Andersen (1966)
and his succeeding publications (see esp. the long lists in Shapiro 1976 and 1983). Anttila
(1977: 4) gives a brief historical view of the same period that overlaps to some extent
with the one I have just given. With the prominent exception of Dressier (who has
exchanged offprints with me over many years), the members of the Natural Morphology
school who most often recur to iconicity (Wurzel and Mayerthaler) never cite Shapiro
(1969) or subsequent publications (like Shapiro 1974a, an important milestone in my
thinking). I trust that the contacts initiated immediately before and during the meetings
in Krems will help overcome these lacunae in future.
4. Andersen (1980b: 202) points out that Wurzel (as Greenberg before him, on whom he
relies) goes wrong here because he fails to take the special relationship between genitive
plural and nominative singular desinences in Russian into consideration. I could add
that Wurzel seems totally to ignore the idea of markedness complementarity (as devel-
oped, for instance, in Shapiro 1974a, and recapitulated in Shapiro 1980 and 1983).
5. Battistella (1990: Chap. 3) explores this matter in some depth, but the question still
remains open.
6. One is reminded here of the persistence, in American linguistic literature at least, of the
mistaken notion that standard German has distinctive voicing in its obstruent system
(cf. Andersen 1972: 45, n. 22). The careful analysis in Vachek (1976) supports the claim
that English has distinctive protensity and not voicing.
7. Battistella (1990) recapitulates his earlier analysis of these constructions and embeds it
in a much wider framework with many further examples.
8. For an extended attempt to cross-pollinate naturalist ideas about grammar with Peircean
semiotic, see Dressier (1985: Chap. 10); cf. also Dressler (et al. 1987).
References
Andersen, Henning
1966 Tenues and mediae in the Slavic languages: A historical investigation [Un-
published Ph. D. Thesis, Harvard University.]
1969 "A study in diachronic morphophonemics: the Ukrainian prefixes", Lan-
guage 45: 8 0 7 - 8 3 0 .
1972 "Diphthongization", Language 48: 1 1 - 5 0 .
1979 "Phonology as semiotic", in: S. Chatman et al. (eds.) A semiotic landscape
(The Hague: Mouton), 3 7 7 - 3 8 1 .
1980a "Morphological change: towards a typology", in: J. Fisiak (ed.), Recent
developments in historical morphology (The Hague: Mouton), 1 —50.
1980b "Summarizing discussion: Introduction", in: T. Thrane et al. (ed.) Typology
and genetics of language ( = Travaux du Cercle Linguistique de Copenhague
20) (Copenhagen: Linguistic Circle of Copenhagen), 197 — 210.
Anttila, Raimo
1969 Uusimman äännehistorian suunnasta ja luonteesta [About the trends and
character of the newest historical phonology] ( = Publications of the Pho-
netics Department of the University of Turku 5).
1972 An introduction to historical and comparative linguistics. (New York: Mac-
millan).
1977 Analogy. (The Hague: Mouton).
1980 "Language and the semiotics of perception", in: I. Rauch —G. F. Carr (eds.)
The signifying animal: The grammar of language and experience (Blooming-
ton: Indiana University Press), 263 — 283.
Battistella, Edwin
1985 "Markedness isomorphism as a goal of language change: the spread of
periphrastic do in English", Lingua 65: 307 — 322.
1990 Markedness: The evaluative superstructure of language. Albany: S U N Y
Press.
1985 Morphonology: the dynamics of derivation (Ann Arbor, Mich.: Karoma
Press).
1987 "Semiotische Grundlagen einer Theorie der Natürlichen Phonologie und
Morphologie", in: Semiotica Austriaca, herausgeg. v. J. Bernard (Wien:
Österreichische Gesellschaft für Semiotik), 165 — 172.
Dressier, Wolfgang U. — Mayerthaler, Willy — Panagl, Oswald — Wurzel, Wolfgang U.
1987 Leitmotifs in natural morphology. (Amsterdam: Benjamins).
Givön, Talmy
1985 "Iconicity, isomorphism, and non-arbitrary coding in syntax", in: J. Haiman
(ed.), 1 8 7 - 2 1 9 .
Haiman, John
1985 Natural syntax. (Cambridge: Cambridge University Press).
Haiman, John (ed.)
1985 Iconicity in syntax. (Amsterdam: Benjamins).
Hjelmslev, Louis
1938 [1970] "Essai d'une theorie des morphemes", Essais linguistiques? (Kobenhavn:
Nordisk Sprog- og Kulturforlag), 152 — 164.
Horn, Wilhelm
1954 Laut und Leben: Englische Lautgeschichte der neueren Zeit (1400—1950),
II, edited by Μ. Lehnert (Berlin: Verlag der Wissenschaften).
34 Michael Shapiro
Itkonen, Esa
1986 "Form-meaning isomorphism, or iconicity, in diachronic linguistics (and
elsewhere)", in: M. Remmel (ed.) Symposium on formalization in historical
linguistics: Summaries, (Tallinn: Academy of Sciences of the Estonian SSR),
38-46.
Jakobson, Roman
1932 "Zur Struktur des russischen Verbums", in: Charisteria Guilelmo Mathesio
oblata (Praha: Prazsky Lingvisticky Krouzek), 74 — 84 [reprinted in Jakob-
son 1971: 3 - 1 5 ] ,
1965 "Quest for the essence of language", Diogene 51: 21—37 [reprinted in
Jakobson 1971: 345-359],
1971 Selected writings, II: Word and language. (The Hague: Mouton).
Mayerthaler, Willi
1981 Morphologische Natürlichkeit. (Wiesbaden: Athenaion).
1988 Morphological naturalness [translated by Janice Seidler]. (Ann Arbor, Mich.:
Karoma Press).
Plank, Frans
1979 "Ikonisierung und De-Ikonisierung als Prinzip des Sprachwandels", Sprach-
wissenschaft 4: 121 — 158.
1981 Morphologische (Ir-) Regularitäten. (Tübingen: Gunter Narr).
Shapiro, Michael
1969 Aspects of Russian morphology: A semiotic investigation. (Cambridge, Mass.:
Slavica).
1971 "Markedness and Russian stress", Linguistics 72: 61 —77.
1974a "Morphophonemics as semiotic", Acta Linguistica Hafniensia 15: 29—49.
1974b "Tenues and mediae in Japanese: A reinterpretation", Lingua 33: 101 —114.
1976 Asymmetry: An inquiry into the linguistic structure of poetry. (Amsterdam:
North-Holland).
1980 "Russian conjugation: theory and hermeneutic", Language 56: 67 — 93.
1983 The sense of grammar: Language as semeiotic (Bloomington: Indiana Uni-
versity Press).
1985 "Teleology, semeiosis, and linguistic change", Diachronica 2: 1—34.
1987 "Sapir's concept of drift in semiotic perspective", Semiotica 67: 159 — 171.
Trubetzkoy, Nikolai S.
1962 Grundzüge der Phonologie 3 (Göttingen: Vandenhoeck & Ruprecht).
Vachek, Josef
1976 "The opposition of voice and tension in modern English paired consonants",
in: J. Vachek, Selected writings in English and general linguistics, (Praha:
Academia), 364 — 372.
Wurzel, Wolfgang U.
1984 Flexionsmorphologie und Natürlichkeit: Ein Beitrag zur morphologischen
Theoriebildung (= Studia Grammatica 21) (Berlin: Akademie).
1985 "Zur Determiniertheit morphologischer Erscheinungen: Ein Zwischenbe-
richt", Acta Linguistica Academiae Scientiarum Hungaricae 35: 151 — 168.
The advantages of morpholexical phonology
Andrew Spencer
In the SPE approach to morphophonemics, all phonological processes

follow all morphological processes. This means that it is impossible for
a morphological process to be sensitive to derived phonological structure
without the intervention of essentially ad hoc readjustment rules. Lieber
(1982) has argued against all readjustment theories of allomorphy, arguing
that there are cases in which we have to say that derived allomorphs are
available in the lexicon prior to the morphology. Her alternative is to list
allomorphs and relate them by means of redundancy relations, mor-
pholexical rules.
In lexical phonology (Kiparsky 1985), phonological rules are inter-
mingled with morphological rules. This gives a new perspective on the
problem of allomorphy, by allowing, for instance, phonological processes
to feed or condition morphological processes. However, even in lexical
phonology the first (structure-changing) phonological rule of the cycle
has to be preceded by a morphological rule. In Spencer (1988a), I have
argued that Czech presents a counterexample in that a morphological
process selects a derived allomorph on the first cycle, before any pho-
nology has had a chance to apply. Somewhat modifying Lieber's original
proposals, I argue, following Booij —Rubach (1987), that there are three
types of phonological processes: cyclic lexical, postcyclic lexical, and
postlexical (phrase level). I claim that the cyclic rules are actually mor-
phological relationships and not rules at all.1
It is not difficult to recode a rewriting rule of the form A —> Β / X
as a set of allomorphy statements plus selectional statements. Essentially,
we provide for the existence of the A and Β allomorphs (either by direct
listing or by generating one allomorph from a basic form by means of a
directionally triggered, but otherwise context-free lexical redundancy rule)
and specify that the Β allomorph is selected in the X context. Concrete
proposals for achieving this without losing phonological generalizations
are presented in Spencer (1988a, b) with additional details in Spencer
(ms). In a sense the approach can be seen as a radical extension of
Kiparsky's (1988) proposals on suppletive allomorphy. The result of this
36 Andrew Spencer
notational recoding is that phonologically sensitive morphological proc-

esses do not have to precede the phonology proper, for in a sense, all
cyclic phonology now precedes all morphology. However, there are a
number of other advantages to the morpholexical framework, some
descriptive, some conceptual.
The descriptive gains center on the possibility of stating morphological
generalizations about "derived" allomorphs in the lexicon (say, of the
form "the present participle stem is always identical to the third-plural
stem"). One obvious spin-off is that certain forms of analogical leveling,
which are notoriously difficult to handle satisfactorily in a derivational
theory, can be seen as straightforward morpholexical rule simplification.
Thus, the leveling of the paradigm of Early Latin honos, honorem, ... to
Classical honor, honorem, ... is no more than generalization of the honor-
stem allomorph for the relevant (sub)paradigm, leaving the adjectival
honestus 'honest' as an irregular remnant. (Cf. also Spencer 1988 a on
Czech conjugation.)
The morphophonemics of palatalizations in Slavic illustrate a related
advantage. The Slavic languages share a core of distinct palatalizations.
The surface inventory of consonant and, crucially, vowel phonemes shows
important differences from language to language, even in the case of
closely related ones (e.g., Czech and Slovak). Particularly interesting are
the interactions between palatalizations and vowel-zero alternations.
Some alternating vowels trigger palatalization, while others do not. In
some languages the two vowels surface as different, in others only one
vowel surfaces. In some cases, this is a front vowel (e.g., Polish, Czech /
e/), in others a back vowel (e.g., Serbo-Croatian /a/).
Despite the differences, the morphologically motivated palatalizations
and vowel-zero alternations are virtually identical across all the languages,
applying in roughly the same morphological contexts (modulo leveling)
and to the same lexical items (modulo morphological class shifts). The
problem for a derivational (SPE-type) account is to provide a set of
grammars for Slavic in which the palatalizations result from essentially
the same set of rules for all the languages. For if the rules generating the
palatalizations are substantially different, we have a curious translingual
conspiracy on our hands. Why should Slavic language learners so con-
struct their grammars as to preserve a roughly constant morphophonemic
surface pattern?
If we assume that the palatalizations and a large part of the vowel-
zero alternations are morpholexical alternations, then there is no con-
spiracy. For the acquisition of morphophonemics is the acquisition of
allomorphic variants and the extraction of redundancy rules relating
them. It is thus lexical learning rather than (phonological) rule learning.

Variation amongst related languages can therefore be seen as the result
of morphological and lexical drift plus the effects of categorial phono-
logical changes, such as restructuring of the phoneme inventory, or
underlying distinctive feature set. On this perspective we would expect a
basic core to be shared, namely common vocabulary elements and com-
mon morphosyntactic and morpholexical classes. Since there are no
derivational rules, there is no need to manipulate them to ensure a
particular type of output.
The conceptual arguments are all ultimately connected with learnabil-
ity. The first two concern Kiparsky's (1985) strict cyclicity condition and
structure preservation. The strict cyclicity condition prevents a cyclic rule,
e.g., English velar softening, from applying to a single, underived, mor-
pheme such as king to produce sing. If all cyclic rules are actually
morpholexical relations then this is an automatic consequence of universal
grammar. For a morpholexical relation is a type of allomorphy statement,
not a rewrite rule. Hence, by definition it only applies to items which
alternate. Since king fails to alternate, the language learner will not have
reason to put king into the velar-softening class. This means that a
morpheme such as -cept (as in concept, accept) is also non-alternating
(underlyingly it is /sept/). The prefix allomorphy found in accept (vs.
assess) and voicing differences between re[z]ignjcon[s]ign and re[s]eive/
con[s\eive must be regarded as facts about allomorph selection, with
redundancy rules stating that the morphemes which select the /ak/ prefix
allomorph are also the ones which do not have a voiced initial alternant.
The trade-off here is between arbitrary, largely non-phonetic, redundancy
statements of this sort for small portions of the vocabulary, as against
extrinsic rule ordering and opacity.2
In a structure-preserving phonology, lexical contrasts which are not
present underlyingly are not created in a derivation. To some extent the
explanatory force of this notion hinges on the constraints put on under-
lying phoneme inventories and absolute neutralization. In any case, it
follows automatically from the present proposals since the allomorphic
variation is stated at a level which precedes phonology proper. Thus, the
only segment types over which it can be defined are phonemes and
archiphonemes (cf. Sproat 1985).
Much effort has been expended on constraining the degree of "ab-
stractness" a phonology can exhibit. The rationale is that abstract and
opaque grammars are harder to learn than concrete and transparent ones.
It is worth contrasting the learning of morphophonemics with the learning
of syntax here. Current syntactic theory (e.g., Chomsky 1986) includes a
38 Andrew Spencer
good many abstract elements (e.g., empty categories, restructurings, and

so on) but has reduced opacity by eliminating the rule component of
grammar in favor of the representational component. The abstract ele-
ments are dictated by the form of universal principles and subcomponents
of grammar. Their ultimate conceptual justification lies in the "poverty
of the stimulus" problem: learners seem to hit on certain classes of
grammars, excluding other, apparently equally accessible, classes, even
though the type of data that would normally motivate such a decision is
either very rarely found or is in principle unavailable (for example, due
to the unbounded nature of syntax).
Morphophonemics is exactly the opposite of syntax in crucial respects.
First, the domain is essentially finite. Therefore in principle learners could
simply memorize all alternations. Thus, there can be no genuine "poverty
of the stimulus" problem. This deprives morphophonemics of the need
for a principles-and-parameters approach. However, an SPE-based ap-
proach to morphophonemics introduces opacity (through extrinsic rule
ordering), and abstract elements (through absolute neutralizations). This
poses a learnability problem which has never been seriously tackled.
Consider the concrete case of Slavic. To figure out the palatalizations,
the learner has to have worked out the abstract underlying vowel system
and the complex pattern of vowel-zero alternations. In some cases, this
means correctly analyzing portions of the vocabulary which is of very
low type or token frequency. But the vowel-zero alternations themselves
cannot be correctly deduced without at least partially figuring out the
palatalizations. This situation is completely typical of rich morphopho-
nemic systems. The kind of bootstrapping deduction it requires is typical
of the hypothesis-testing mode of scientific inquiry, of the kind which
Fodor (1983) associates with central cognitive systems. It is not the kind
of operation we associate with an input module. In other words, SPE-
based morphophonemics takes a finite domain which should thus pose a
trivial learnability problem, and introduces its own, completely artificial
"poverty of the stimulus" problems, without providing satisfactory uni-
versal principles for their resolution.
In the morpholexical approach, these problems do not arise. Most
learning is lexical learning, supplemented with very simple universal
principles of selection. Phonologically describable relationships are a
consequence of the representational aspects of phonological theory and
not the result of derivational devices. This means that the only artificial
"poverty of the stimulus" problems which can arise concern abstract
underlying segments. Since the domain is finite, and since lexical listing
is therefore always open as a possibility, the extent to which learners

project such abstractions is an empirical question, possibly more the
concern of the psycholinguist than the phonologist.
The final aspect of the learnability problem concerns the nature of the
data to which the child is exposed. Morphophonemic systems are typically
riddled with exceptionality and complex patterns of subregularity. A
learning theory for a derivational morphophonemics would have to show
how it is possible for a child to handle such patterns of exceptionality by
means of ordered rules. The main question concerns repair of errors:
how does the child restructure his grammar when he finds he has made
a mistake in the rule formulation, the underlying format, or the ordering
of rules, given the immense choice of hypotheses? Lexical learning in
morpholexical phonology, as in syntax, encounters no such problem since
lexical learning is essentially a question of establishing a taxonomy. If an
incorrect taxonomy is projected (or if the child is confused by errors in
the input itself), then this can be corrected by simply changing morpho-
lexical class membership (and perhaps redefining the criteria for mem-
bership of a class). Thus, the adult grammar can be approached (more
or less) asymptotically. It is hard to see how a derivational morphopho-
nemics could be so learned. Certainly, the onus of proof must lie on
advocates of derivational approaches.
Notes
1. I am inclined to say that the postcyclic and postlexical components take the form of
output constraints governed by universal and language-particular principles such as
syllable structure, metrical structure, autosegmental structure, and so on. This would
make phonology look rather like syntax on the "principles-and-parameters" approach.
2. The strict-cyclicity condition also prevents a rule A which counterfeeds a rule Β from
being fed by Β on a subsequent cycle. In a morpholexical framework, this effect can
only be reconstructed using complex diacritics. It seems to me undesirable that a principle
of universal grammar should have the effect of maintaining rules in a marked order, so
I am happy with this situation.
References
Booij, Geert —Rubach, Jerzy

1987 "Postcyclic versus postlexical rules in lexical phonology", Linguistic Inquiry
18: 1 - 4 4 .
Chomsky, Noam
1986 Barriers. (Cambridge, MA: MIT Press).
40 Andrew Spencer
Fodor, Jerrold
1983 The modularity of mind. (Cambridge, MA: MIT Press).
Kiparsky, Paul
1985 "Some consequences of lexical phonology", Phonology Yearbook 2: 83 — 136.
1988 "Two approaches to suppletive allomorphy" [Paper presented at the 3rd
International Morphology Meeting, Krems, Austria].
Lieber, Rochelle
1982 "Allomorphy", Linguistic Analysis 10: 2 7 - 5 2 .
Spencer, Andrew J.
1988a "Arguments for morpholexical rules", Journal of Linguistics 24: 1 —30.
1988b "Morpholexical rules and lexical representations", Linguistics 26: 619 — 640.
ms "Morpholexical phonology".
Sproat, Richard
1985 On deriving the lexicon [Dissertation, MIT, Cambridge, MA].
Topic 2: Word formation
Associativite et stratification dans la representation
des mots construits
Danielle Corbin
Au risque de paraitre aller a contre-courant de bon nombre de recherches

actuelles en grammaire generative, qui dissocient et traitent dans des
modules differents la representation de la «syntaxe interne» des mots
construits et celle de leur interpretation semantique 1 , je defendrai ici le
bien-fonde d'un composant derivationnel associatif ä l'interieur d'un
composant lexical autonome et stratifie. « Associatif» signifie que le role
assigne au composant derivationnel est d'engendrer conjointement et de
fagon associee la structure morphologique et Interpretation semantique
des mots construits. «Stratifie» signifie que l'organisation interne du
composant lexical et la hierarchie des operations touchant les mots
construits qu'il comporte refletent la complexite des associations forme/
sens qui caracterise les mots construits, et qui ne saurait se reduire ä la
dichotomie regulier/irregulier.
Pour des raisons de place, ce texte ne fait qu'esquisser le cadre general
de cette theorie associative, en se concentrant sur la fagon dont celle-ci
peut traiter les diverses distorsions forme/sens; il ne presente pas de fagon
detaillee les arguments justifiant ses analyses2. Tous les exemples traites
sont empruntes au frangais.
Apres une mise au point sur la notion meme d'interpretation seman-
tique des mots construits (§ 1), je passerai en revue les reanalyses que le
modele presente permet d'effectuer face aux divers types de distorsions
apparentes entre la structure morphologique et l'interpretation seman-
tique des mots construits (§ 2). Dans chaque cas, je montrerai que la
distorsion n'est qu'apparente, c'est-a-dire qu'il est possible d'apporter des
reponses modeliques ä ces distorsions sans remettre en cause le principe
de l'associativite, comme cela a souvent ete fait dans la litterature 3 .
J'expliquerai pour terminer la stratification correspondante du modele
de composant lexical que je propose (§ 3).
44 Danielle Corbin
1. Mise au point sur la notion înterpretation semantique

(Tun mot construit
Je propose d'introduire deux distinctions fondamentales portant sur I n -

terpretation semantique des mots construits.
D'une part entre la notion de «sens atteste» (SA) et celle de «sens
predictible» (SP): j'appellerai sens atteste le sens d'un mot construit tel
qu'il figure dans un ou plusieurs dictionnaire(s) de langue contempo-
rain(s)4. Au sens atteste directement observable, j'opposerai le sens pre-
dictible, c'est-a-dire le sens d'un mot construit tel qu'il est formulable ä
partir de l'operation semantique propre ä la regle qui a construit le mot.
Comme cette definition le fait apparaitre, le sens predictible est le resultat
d'une construction hypothetique du linguiste, en fonction de l'idee qu'il
se fait de l'economie generale de la grammaire derivationnelle de la langue,
en fonction de sa theorie du lexique. Dans ma theorie, le sens predictible
a deux proprietes fondamentales: sa compositionnalite obligatoire par
rapport a la structure morphologique du mot construit, et son indepen-
dance relative par rapport au sens atteste.
D'autre part, ä l'interieur du sens predictible, entre le sens predictible
construit par la regle (SPcr) et le sens predictible herite de la base (SPhb).
Cette distinction vise ä clarifier les divers composants entrant dans l'in-
terpretation semantique predictible d'un mot construit. Le sens predictible
construit par la regie est le sens commun a tous les produits d'une meme
regle de construction de mots, independamment des types de bases et des
procedes morphologiques utilises. Le sens predictible herite de la base est
le sens commun ä tous les produits d'une regle de construction de mots
construits sur des bases ayant le meme type semantique pertinent; il
represente une specification du sens predictible construit par la regie
particuliere ä un type semantique de base donne 5 .
J'illustrerai cette stratification du sens predictible par un exemple
simple, dont je ne developperai pas les justifications 6 : la regle de construc-
tion de mots ä laquelle est associe (entre autres procedes) le suffixe -eux
en frangais attribue ä ses produits un sens predictible construit par la
regle de type relationnel, que l'on peut representer par la paraphrase « en
relation avec [nom de base]» (simple transposition en termes semantiques
du changement categoriel d'un nom en adjectif)· Quand la regie s'applique
ä des bases designant des maladies, le sens predictible herite de la base
des adjectifs construits peut prendre la forme « atteint de [nom de base]».
L'exemple de l'adjectif coquelucheux illustre ci-dessous ces diverses strates:
Associativite et stratification 45
(1) SPcr «en relation avec la coqueluche»

ex. une toux coquelucheuse = une toux de coqueluche
le bacille coquelucheux — le bacille de la coqueluche
SA « de la coqueluche »
SPhb « atteint de coqueluche »
ex. un enfant coquelucheux — un enfant atteint de coque-
luche
SA « atteint de coqueluche »
Cet exemple est simple, car il ne fait guere apparaitre de distorsion entre
les sens predictibles et les sens attestes. Comme on va le voir ci-dessous,
ce n'est pas toujours le cas.
2. Reponses aux distortions apparentes entre l'interpretation

semantique et la structure formelle (Tun mot construit
On partira de l'idee, generalement admise, que la situation la plus «na-
turelle» est celle ού la structure morphologique d'un mot construit et
son interpretation semantique se correspondent biunivoquement 7 . Mais
il n'en est pas toujours ainsi, apparemment du moins. J'envisagerai
successivement quatre types de distorsions possibles entre l'interpretation
semantique et la structure ou la forme du mot construit qui ont pu ou
pourraient servir d'arguments empiriques contre l'associativite, et je ten-
terai de leur proposer des reponses. Deux principes methodologiques
guideront mon expose:
(i) un meme phenomene apparent peut recevoir plusieurs interpretations
differentes;
(ii) il est necessaire, pour atteindre quelque generalite, de depasser le
niveau de l'evidence observationnelle du lexique atteste.
Je m'en tiendrai ici aux arguments empiriques, ayant developpe ailleurs
des arguments theoriques contre les modeles dissociatifs 8 .
2.1. Premier type de distorsion: non-conformite apparente de

l'interpretation semantique ä la structure morphologique
Le type le plus evident de distorsion apparente est celui ou l'interpretation
semantique parait ne pas etre conforme ä la structure morphologique du
mot construit. Je prendrai deux exemples, qui illustrent deux fapons de
resoudre le probleme.
46 Danielle Corbin
2.1.1 Four che tie a pour structure morphologique (2a), done pour sens
predictible (2b)9, mais pour sens atteste (2c):
(2a) [[fourche]N(ette)af]N
b) SP « petite fourche » (cf. flechette, hachette, serpette, etc.)
c) SA «1° ustensile de table (d'abord ä deux, puis a trois, quatre
dents), dont on se sert pour piquer les aliments »
Le fait que le sens predictible ait ete applique a un domaine particulier

de l'experience extra-linguistique ne peut etre ni predit ni explique lin-
guistiquement, et Ton doit considerer ce mot comme superficiellement
idiosyncratique, sans que cette idiosyncrasie, posterieure a la construction
du mot, puisse constituer un argument contre l'associativite10.
2.1.2 Le sens atteste actuel (3d) de publiciste ne correspond pas ä sa

structure morphologique apparente (3a). Le sens predictible construit par
la regle correspondant ä cette structure est (3b), et le sens predictible
specifie par le processus morphologique (SPsp) quelque chose comme
(3c)11:
(3a) [[[[public]A]N(iste)af]A]N
b) Spcr « N en relation avec le public»
c) SPsp « Ν [ +hum] specialiste du public »
d) SA «3° [...] agent de publicite»
Ici, contrairement au cas precedent, le sens atteste est irreductible au sens

predictible, meme au prix d'un ajustement superficiel. Je poserai qu'il y
a en fait en frangais deux noms (provenant par conversion de deux
adjectifs) homonymes qui ont la forme publiciste: l'un a pour structure
(3a) et pour sens predictible (3b) et (3c), mais n'est pas atteste, l'autre,
qui correspond au sens atteste (3d), a pour structure (4a) et pour sens
predictible (4b) et (4c):
(4a) [[[[public] A (ite) a f < + T>]N(iste) a f < T + > ] A ]N

b) SPcr « Ν en relation avec la publicite »
c) SPsp « Ν [ + hum] specialiste de la publicite»
Dans la structure (3a), le nom °publiciste/2 est converti ä partir de

l'adjectif °publiciste,, lui-meme construit sur le nom public, converti ä
partir de l'adjectif correspondant. Dans la structure (4a), le nom
publiciste2 est converti ä partir de l'adjectif °publiciste2, lui-meme construit
sur le nom publicite, dont le suffixe -ite a subi une troncation devant
-isten.
Dans cet exemple, c'est l'homonymisation de la forme publiciste qui
permet de conserver Γ associativite.
2.2. Deuxieme type de distorsion: ä une seule structure morphologique

paraissent correspondre plusieurs interpretations semantiques
Le deuxieme type de distorsion apparente est celui ou ä une seule structure

morphologique paraissent correspondre plusieurs interpretations seman-
tiques. Si tel est le cas, une theorie associative repond, soit que l'inter-
pretation semantique est unique, malgre les apparences, soit, comme on
l'a vu egalement a propos de publiciste, qu'a chaque interpretation se-
mantique correspond une structure morphologique differente.
2.2.1. Du premier cas de figure (interpretation semantique unique sous
plusieurs sens apparemment differents), je ne donnerai qu'un exemple.
En frangais, les noms suffixes par -oir construits sur des bases verbales
(structure (5a)) sont susceptibles de recevoir des sens attestes de type
instrumental (5b) et de type locatif (5c). Un certain nombre d'arguments,
que la place m'interdit de developper ici14, permettent de montrer que,
dans cet exemple, la distinction instrumental/locatif n'est qu'apparente,
et qu'en realite, la ou eile apparait irreductible dans le lexique atteste
(mouchoir par exemple peut difficilement etre interprete comme un locatif
dans l'univers de reference), eile provient de la projection sur le derive
de proprietes semantiques et argumentales du verbe de base. Les sens
instrumental et locatif peuvent done etre consideres comme les sens
predictibles herites de la base des mots construits ä l'aide de ce suffixe
(5d), et le sens predictible construit par la regle associe ä la regle de
construction de mots en question etre formule par la paraphrase (5e):
(5a) [[X]v(oir)af]N
b) SA mouchoir: «1° petite piece de linge, generalement de

forme carree, qui sert ä se moucher, a s'essuyer le visage »
c) SA dortoir: «1° grande salle commune ού dorment les
membres d'une communaute »
d) SPhb « Ν qui sert ä [verbe de base]»
«lieu ού Ton [verbe de base]»
e) SPcr « Ν pour [verbe de base]»
48 Danielle Corbin
2.2.2. Le deuxieme cas de figure (c'est-a-dire en fait l'homonymie struc-

turelle) peut s'expliquer de diverses fapons:
— Dans l'exemple dejä cite de publiciste, l'homonymie de surface etait
due ä la troncation d'un affixe.
— II peut s'agir aussi de l'application de deux affixes homonymes ä deux
bases homonymes de categories differentes: par exemple, la forme plumage
a deux sens attestes (6b) et (6d), irreductibles ä un meme sens predictible,
ä chacun desquels correspond une structure morphologique differente
(respectivement (6a) et (6c)):
(6a) [[plume]N(age,)af]N
b) SA «1 0 l'ensemble des plumes recouvrant le corps d'un
oiseau [...]»
c) [[plume] v (age 2 )af]N
d) SA «2° [...] action de plumer»
Ou bien encore il peut s'agir de l'application des memes regies de
construction de mots dans un ordre different. Par exemple la forme
invalidable15 peut recevoir deux interpretations, (7b) et (7d), dont chacune
est compositionnelle par rapport ä une structure morphologique diffe-
rente: ä (7b) (a laquelle repond un sens atteste) correspond (7a); ä (7d)
(ä laquelle ne repond pas de sens atteste) correspond (7c):
(7a) [[[(in)af{valide]A]A]v(able)af]A
(valide —> invalide —> invalider —•» invalidable)
b) SPcr « Qui peut etre invalide » ( = qu'on peut rendre invalide
(l'etat initial du Ν predique est suppose valide))
c) [(in)af{[[valide]A]v(able)af]A]A
(valide —• valider —> validable —> °invalidable)
d) SPcr « N o n validable» ( = qu'on ne peut pas rendre valide
(l'etat initial du Ν predique est suppose non valide))
Dans tous ces cas d'homonymie, on fait done bien correspondre un
seul sens ä une seule structure morphologique.
2.3. Troisieme type de distorsion: a une seule interpretation semantique

paraissent correspondre plusieurs structures morphologiques
Lorsque, ä l'inverse de ce qui vient d'etre expose, ä une seule interpretation
semantique paraissent correspondre plusieurs structures morphologiques,
la reponse d'une theorie associative est, soit que, dans certaines condi-
tions, des structures differentes formellement peuvent etre considerees
Associativiti et stratification 49
comme equivalentes semantiquement, soit que l'analyse fondee sur la

perception apparente est fausse, c'est-a-dire que sous un seul sens atteste
se cachent plusieurs sens predictibles differents.
2.3.1. On peut illustrer le premier cas par la construction des verbes de
changement d'etat en frangais: sur des bases adjectivales de sens resultatif,
on peut construire des verbes soit ä l'aide de prefixes (les principaux sont
α-, έ-, en-), soit ä l'aide de suffixes (-ifi(er), -is(er)), soit par conversion
(structures (8a)). Tous ces procedes donnent des verbes au meme sens
predictible construit par la regle (8b), souvent attestes avec des definitions
comparables (8c):
(8a) [(Y)af{X]A]v (ex. appauwir, elargir, enrichir)
[[X]A(Y)af]v (ex. humidifier, immobiliser)
[[X]A]v (ex. blanchir)
b) SPcr « rendre (plus) [adjectif de base]»
c) SA appauvrir: «1° Rendre pauvre»
elargir. «1 ° Rendre plus large »
enrichir: « 1 0 Rendre riche »
humidifier. « Rendre humide »
immobiliser: «2° [...] rendre immobile [...]»
blanchir. «1 ° rendre blanc »
Je propose dans ce cas de considerer que les trois structures (8a) sont
equivalentes, c'est-a-dire que tous les procedes morphologiques auxquels
elles sont associees font partie du meme «paradigme morphologique»
associe ä la meme regle de construction de mots 16 , meme si certains
different entre eux par des proprietes specifiques (disponibilite,
contraintes particulieres sur la base, etc.)17.
Les deux implications — importantes — de cette proposition sont
d'une part que ce ne sont pas les affixes, dans ma theorie, qui sont
fondamentalement responsables du sens des mots construits, mais les
regies de construction de mots 18 , d'autre part qu'une telle regie est iden-
tifiee avant tout par l'association d'un rapport categoriel et d'une ope-
ration semantique, et non prioritairement par une forme affixale.
2.3.2 II y a des cas ού il ne faut pas se fier ä la similitude des sens
compositionnels attestes de derives, meme s'ils paraissent correspondre a
un meme sens predictible. L'exemple de chenaie et sapiniere illustrera
l'apparence d'une meme interpretation semantique correspondant ä des
procedes morphologiques differents (les suffixes -aie et -ierj-ieref9, sans
que la situation soit celle, exposee au paragraphe precedent, de procedes
appartenant au meme paradigme. Les sens attestes de ces mots sont en
50 Danielle Corbin
effet comparables (9a), et on pourrait penser que les suffixes -aie et -ierj
-iere appartiennent tous deux au paradigme morphologique associe ä la
regle de construction de mots qui construit, sur des bases nominales
designant des vegetaux, des noms dont le sens predictible peut se para-
phraser par «lieu plante de [nom de base]». Mais, si ces proprietes
paraissent bien etre celles de -aie, elles ne conviennent pas ä -ierj-iere,
dont on peut dire tres brievement qu'il construit fondamentalement des
adjectifs ä sens relationnel sur des bases nominales 20 . La structure mor-
phologique (9b) et les sens predictibles de sapiniere (9c) ne sont done pas
directement deductibles de l'apparence:
(9a) SA chenaie « plantation de chenes »

sapiniere « bois, foret, plantation de sapins »
b) [[[sapin]N(ier/iere)af]A]N
c) SPcr °sapinierA « en relation avec [nom de base]»
sapiniereN « N en relation avec [nom de base]»
SPhb « Ν contenant des, producteur de sapins »
Pour repondre a l'argument, souvent avance contre l'associativite, de
la «classification croisee des affixes et des sens» 21 , j'ai propose dans ce
paragraphe que les procedes morphologiques et Interpretation seman-
tique soient associes de fagon asymetrique: ä un procede morphologique
donne ne peut correspondre qu'un sens predictible construit par la regie,
mais ä un tel sens peuvent correspondre plusieurs procedes morpho-
logiques appartenant au meme paradigme.
2.4. Quatrieme type de distorsion: les accidents formels

Pour en terminer avec la faôn dont une theorie associative peut repondre
aux divers types de distorsions forme/sens, il faut parier rapidement des
accidents formels qui, etablissant une distorsion entre la forme predictible
du mot construit et sa forme attestee, provoquent indirectement des
distorsions entre la forme du mot construit et son sens. Ces accidents
formels, allomorphies et troncations, seront analyses comme des modi-
fications superficielles apportees ä une structure morphologique reguliere.
Je me borne ici ä donner le principe d'une theorie exposee dans D. Corbin
(1987: 283 — 370). Je distinguerai deux cas, en citant a chaque fois un
exemple d'allomorphie et un exemple de troncation.
2.4.1. Si la modification de la forme attestee par rapport ä la forme
predictible du mot construit est recurrente dans le lexique, je suppose que
cette derniere, reguliere mais bloquee ä la sortie du composant deriva-

tionnel, est modifiee par une regie mineure, posterieure aux regies de
construction de mots, declenchee par des traits diacritiques dont sont
porteurs les constituants du mot construit.
Par exemple, le suffixe -ite provoque toujours la posteriorisation vo-
calique de la syllabe finale -eux de sa base, que cette finale soit affixale
(verbe —• verbeux —• verbosite) ou non (genereux —• generosite). Comme
le montrent les schemas (10), les formes *verbeusite, *genereusite sont
regulierement produites par les regies de construction de mots, bloquees
parce que deux constituants successifs de leur structure interne sont
marques de traits allomorphiques complementaires, et corrigees en ver-
bosite, g0n0rostä par une regie mineure, posterieure aux regies de
construction de mots:
(10a) [[[verbe]N(eux)af< + PV>]A(it£)af<pv + )]n

(RCM —> * verbeusiti; regie mineure —• verbosite)
b) [[genereux]A< +Pv>(ite)af<pv+>]N
(RCM —•» *genereusiti; regie mineure —»· generosite)
Le schema (11) rappelle la troncation du suffixe -ite par le suffixe -
iste dans l'exemple deja rencontre de publiciste22: la regle de construction
de mots engendre *publiciteiste, bloque parce que deux constituants
successifs de sa structure interne sont marques de traits de troncation
complementaires, et la regie mineure de troncation corrige cette forme
en publiciste:
(11) [[[public] A (ite) af< + T>]N(iSte)af<T+ >]a
(RCM —• *publiciteiste; regie mineure —* publiciste)
2.4.2. Si la modification n'est pas recurrente, le cas est different pour
rallomorphie et la troncation.
Si la modification est de type allomorphique, je suppose que le mot
construit a pour base un item non atteste suppletif du mot atteste
correspondant: par exemple, la variation consonantique qui affecte la
paire nager/natation est unique, ä ma connaissance (comparer obliger —*
obligation, propager —> propagation, etc.). Natation n'est done pas derive
de nager, mais de °nater, synonyme de nager23.
Si la modification est de type troncatoire, le comportement de l'affixe
tronque est exceptionnel par rapport ä son comportement habituel. Je
propose alors de marquer idiosyncratiquement < + T> le mot construit,
regulier mais non bloque ä la sortie des regies de construction de mots.
On peut ainsi comparer verbosite, dans lequel le suffixe -eux a ete
52 Danielle Corbin
normalement allomorphise devant -ite (cf. §2.4.1), et αηχϊέίέ24, dans

lequel le meme sufflxe a ete tronque dans le meme contexte:
(12) [[[anxie]N(eux)af<+PV>]A(ite)af<pv+>]n
(RCM —*• *anxieusite; regie mineure d'allomorphie:
°anxiosite\ troncation idiosyncratique —• anxiete)
2.5. Conclusion
Pour conclure cette partie, je rappellerai les principes essentiels qui y ont
ete defendus:
— A toute structure morphologique est associee une interpretation se-
mantique compositionnelle et un seul sens predictible construit par la
regie, qui ne correspond pas necessairement au sens atteste.
— S'il apparait une distorsion entre la structure morphologique et l'in-
terpretation semantique, ou bien la distorsion est reelle (exemples (8),
(10) ä (12)), et des dispositifs modeliques (paradigme morphologique,
regies mineures d'allomorphie et de troncation) permettent de ne pas
remettre en cause l'associativite, ou bien la distorsion n'est qu'apparente
(exemples (2) ä (7), (9)), et il faut deplacer l'analyse.
— II n'y a pas de bijection entre l'ensemble des affixes et celui des regies
de construction de mots.
— Ce n'est done pas l'unite de forme affixale qui delimite une operation
derivationnelle, mais l'association entre un rapport categoriel, une ope-
ration semantique et un Operateur morphologique (affixe, conversion)
appartenant ä un paradigme dont le nombre d'unites est variable.
— II en decoule que la morphologie associative que je preconise est
fondamentalement homonymique, dans la mesure ou le critere d'identi-
fication d'un element (affixe ou mot construit) est l'association de pro-
prietes formelles et de proprietes semantiques. Au contraire, les morpho-
logies dissociatives sont polysemiques.
3. La stratification du composant lexical
Je me bornerai ici ä citer l'ensemble des proprietes qui font du modele

de composant lexical que je propose, et dont une representation figure
ci-apres25, un modele stratifie conformement aux principes de l'associa-
tivite enumeres ci-dessus.
REGLES
D'INSERTION
LEXICALE
Figure 1. Organisation du composant lexical

54 Danielle Corbin
L'hypothese fondamentale est que les operations s'appliquent de fagon

hierarchisee en fonction de leur caractere plus ou moins predictible et
plus ou moins regulier, les moins regulieres etant toujours subordonnees
aux plus regulieres.
Du point de vue derivationnel, la stratification du composant lexical
s'illustre de trois fagons.
3.1. La stratification des composants

La stratification la plus visible est le fait que le composant lexical est
divise en trois sous-composants:
— Le composant de base, dans les «entrees lexicales de base», liste les
morphemes (entrees majeures non construites et affixes) et toutes leurs
proprietes, ainsi que les regularites semblables ä Celles des mots construits
qu'ils sont susceptibles de presenter («regies de base»).
— Le composant derivationnel engendre les mots construits et toutes
leurs proprietes regulieres. II definit ce qui est possible linguistiquement.
— Le composant conventionnel rend compte des sous-regularites et ir-
regularites touchant les mots construits. II definit, a partir du possible,
ce qui est atteste. Ainsi, le caractere bien forme d'un mot n'est pas indexe
sur l'attestation de celui-ci.
3.2. La stratification des operations

Une deuxieme manifestation de la stratification du modele est la hierar-
chie des types d'operations qu'il comporte.
— Les regies de base (niveau 2) sont congues comme des regies de
redondance a valeur evaluative, ne s'appliquant qu'aux entrees lexicales
de base.
— Les regies de construction des mots (niveau 3), sont des regies gene-
rales s'appliquant, si les conditions de leur application sont reunies, et
chaque fois qu'elles le sont, aux entrees lexicales de base et a leurs propres
produits (recursivite). On peut les concevoir comme un ensemble d'ope-
rations modulaires associees, dont je ne developperai pas ici le detail26.
Elles construisent tout ä la fois la structure morphologique des mots
construits, leur interpretation semantique (sens predictible construit par
la regle et herite de la base27), une partie de leurs proprietes syntaxiques
(celles qui sont predictibles a partir de leur caractere construit), en
respectant d'une part les contraintes qui leur sont imposees, d'autre part
Celles, plus speciflques, qui sont imposees a l'application de chacun des

procedes appartenant ä leur paradigme morphologique.
— Les regies mineures, representees aux niveaux 5, 7 et 8 du composant
conventionnel, ont un domaine d'application defini ä l'avance, et rendent
compte des sous-regularites que presentent les mots construits et dont ne
sont pas responsables les regies de construction de mots.
— Les operations de l'applicateur d'idiosyncrasies (niveau 6) sont des
operations ad hoc touchant les mots construits. C'est ä ce niveau, par
exemple, que se decide la troncation de °anxiosite en anxiete (cf. (12)).
— Les operations du selectionneur (niveau 9), sont des operations pour
la plupart ad hoc, touchant ä la fois les entrees lexicales de base et les
mots construits. Elles ont pour täche de selectionner les formes et les
proprietes attestees (lexique de fait) parmi les formes et les proprietes
possibles (lexique de droit). Ainsi, seul nager, et non °nater, et, pour
fourchette, seul le sens idiosyncratique « ustensile de table », et non le sens
« petite fourche », seront selectionnes pour faire partie du lexique atteste.
3.3. La stratification des sorties

Une troisieme illustration de la stratification du modele est la possibilite
donnee aux regies d'insertion lexicale (ou ä leur equivalent) d'aller cher-
cher des mots a trois niveaux de sortie differents:
— Au niveau des entrees lexicales de base: ä part les affixes, que leur
categorisation specifique empeche d'etre inseres sous des nceuds
syntaxiques28, toutes les entrees, attestees ou non (nager et °nater), ont
ainsi la possibilite linguistique d'etre inserees dans des phrases.
— Au niveau des mots construits possibles, attestes ou non. Rappelons
que l'une des restrictions qui jouent ä ce niveau est le blocage de mots
comme *verbeusite (cf. (10)) ou *publicitiiste (cf. (11)), dont les consti-
tuants internes sont marques pour subir des regies mineures.
— Au niveau du lexique conventionnel, pour ce qui concerne le lexique
atteste.
Ne sont done pas confondues dans ce modele une propriete de la
langue, qui est la possibilite linguistique qu'ont les mots d'apparaitre
dans des phrases, et une propriete des locuteurs, qui est celle d'exploiter,
de faôn variable selon les individus, les potentialites de la langue.
56 Danielle Corbin
4. Pour conclure
Les principes gouvernant le composant lexical stratifie dont j'ai esquisse

les grandes lignes sont ceux d'une morphologie derivationnelle associative.
Le modele propose reste generatif, dans la mesure ού la description a un
caractere explicite, et ou l'objectif qu'elle se fixe est de rendre compte de
la competence lexicale des locuteurs. Quelles que puissent etre ses imper-
fections actuelles, son originalite consiste a deplacer les ordres de priorite
habituels, en affmant la typologie des associations forme/sens qui carac-
terisent les mots construits. C'est pourquoi il ne pourra veritablement
etre confronte aux propositions dissociatives faites dans un cadre generatif
que pour autant que celles-ci accepteront de se preoccuper de la repre-
sentation de Interpretation semantique des mots construits, ce qu'elles
ne font pas encore, a ma connaissance 29 .
Notes
1. Voir par exemple Selkirk (1982), Toman (1987).

2. Pour un expose detaille, voir Corbin (1987).
3. Voir par exemple Jackendoff (1975), Lieber (1980).
4. Ici, le dictionnaire de reference sera le PR86, sauf indication contraire.
5. Dans les cas de concurrence entre plusieurs procedes morphologiques associes ä la
meme regle de construction de mots (voir infra §2.3.1.), il peut s'ajouter a ces deux
strates fondamentales du sens predictible un troisieme niveau, celui du sens predictible
specifie par le procede morphologique, qui represente le sens attribue par un procede
morphologique (affixe, conversion, composition) ä tous les mots qu'il sert a construire,
sous la forme d'une specification du sens predictible construit par la regie. Une
illustration en sera donnee ci-dessous avec les exemples (3) et (4).
6. Voir, sur ce point, Puchulu (1987, 1988) et Corbin —Corbin, ä paraitre.
7. Cf. Dressier (1987) par exemple.
8. Voir Corbin (ä paraitre).
9. Si Ton admet que le sens predictible construit par cette regie est diminutif, le sens (2b)
represente ce sens; si Ton pense que le sens est hyponymique («Nb d'une certaine
sorte»), (2b) represente le sens predictible specifie par le processus morphologique. Je
ne discuterai pas ici cette question.
10. Je suis d'accord sur ce point avec Booij (1979).
11. On admettra ici (cf. Corbin 1988) que le suffixe -iste sert a construire fondamentalement
des adjectifs de relation sur des bases nominales, et que les noms en -iste sont des
conversions des adjectifs correspondants. Contrairement ä d'autres procedes morpho-
logiques associes ä la meme regie, le suffixe -iste, applique ä des bases designant des
personnes ou des doctrines, a pour particularite de favoriser la specification du sens
predictible construit par la regie sous les formes « specialiste de [nom de base]» (dans
un contexte nominal [ + hum] et « partisan de [nom de base]» (dans un contexte nominal
[±hum]). Comparer par exemple un parti anarchiste ( = «partisan de l'anarchisme (ou
de l'anarchie)») et un parti anarchique ( = «qui manifeste de l'anarchie»).
12. Le signe 0 precede des formes ou des sens grammaticaux quoique non attestes, par
opposition au signe * qui precede des formes ou des sens agrammaticaux.
13. Cette troncation est representee dans le schema (4) par les traits < + T>, signifiant que
-ite est predispose ä subir une troncation dans un contexte approprie, et <T + >,
signifiant que -iste est predispose a provoquer la troncation d'un suffixe predispose a
la subir. Voir sur ce point Corbin (1987: 341 —370).
14. Certains sont developpes dans Corbin (1987: 247 — 248).
15. Encore attestee dans le LXXe et son successeur le GLE avec le sens «Qui peut ou doit
etre invalide», mais plus dans le GDEL, ni aucun dictionnaire frangais strictement
contemporain. Validable est atteste dans le GRLF.
16. Ma proposition differe sur plusieurs points de ce que Zwanenburg (1984) appelle «types
de derivation »:
— Peuvent etre inclus dans le meme paradigme morphologique des prefixes, des
suffixes, la conversion, alors que Zwanenburg n'inclut dans ses «types » que des suffixes.
— Zwanenburg propose qu'ä l'interieur d'un type donne, les suffixes soient exclusifs
Fun de l'autre, c'est-ä-dire que son dispositif lui sert ä expliquer certains blocages, alors
que je ne pense pas que ce principe de blocage soit adequat: si les procedes appartenant
ä un meme paradigme obeissent aux memes contraintes, ils peuvent s'appliquer ä une
meme base remplissant ces contraintes (en temoignent par exemple, dans le lexique
atteste: abonnir / bonifier; clarifier / eclaircir, etc., et dans le lexique possible °laidifier
face ä enlaidir, °sauvagiser face ä ensauvager, etc.).
— Zwanenburg raisonne dans un cadre dissociatif, moi dans un cadre associatif.
17. Certaines de ces contraintes sont explicitees dans Corbin (1987: 450).
18. Ce qui s'oppose ä l'hypothese «One affix, one rule» formulee notamment par Aronoff
(1976).
19. Le genre feminin de sapiniere n'est pas plus predictible que celui de saliere compare a
poivrier.
20. Par exemple, un argument contre l'assimilation du suffixe -aie et du suffixe -ier/iere est
que, contrairement ä ce qui se passe pour -aie, on trouve des noms de lieu en -ierj-iere
sur des noms de base ne designant pas des vegetaux {sablonniere («lieu d'ou Ton extrait
le sable»), heronniere («endroit amenage pour l'elevage des herons»), etc.). On trouvera
une argumentation plus detaillee concernant le suffixe -ierj-iere dans Corbin — Corbin
(ä paraitre).
21. Voir par exemple Jackendoff (1975), Zwanenburg (1982, 1984, 1987: 64).
22. Cette troncation est recurrente, comme le montre l'exemple finalite —> finaliste,, au sens
« Qui croit ä Taction des causes finales et, en general, ä la finalite comme explication
de l'univers » (sens predictible specific: « Partisan de la finalite »).
23. Ce verbe, non atteste aujourd'hui, etait atteste au XIVC siecle (Greimas, 1968).
24. Si ce nom est derive d'anxieux. Je simplifie ici le traitement de ce mot. Pour des details,
voir Corbin (1987: 505-508).
25. Extrait de Corbin (1987: 417).
26. Voir sur ce point Corbin (1987: 476-504).
27. Et, si necessaire, sens predictible specifie.
28. Je m'oppose sur ce point ä des propositions recentes comme celle de Roeper (1988) qui
propose d'inserer des affixes, qui selon lui sont des «tetes», sous des nceuds syntag-
matiques.
29. A l'exception, remarquable, de Botha (1988).
58 Danielle Corbin
References bibliographiques
Aronoff, Mark
1976 Word formation in generative grammar (Linguistic Inquiry Monograph 1).
(Cambridge, Mass.: MIT Press).
Booij, Geert
1979 "Semantic regularities in word formation", Linguistics 17: 985—1001.
Botha, Rudolf P.
1988 Form and meaning in word formation. A study of Afrikaans reduplication.
(Cambridge: Cambridge University Press).
Corbin, Danielle
1987 Morphologie derivationnelle et structuration du lexique, 1—2. (Tübingen:
Niemeyer).
1988 «Une hypothese a propos des suffixes -isme, -ique, -iste du frangais: la
troncation reciproque », in: R. Landheer (ed.) Aspects de linguistique fran-
Qaise. Melanges Q. I. M. Mok (Amsterdam: Rodopi), 63 — 75.
a paraitre «Contre une transposition de la theorie X a la morphologie derivation-
nelle», Acta Linguistica Hungarica 37.
Corbin, Danielle —Corbin, Pierre
ä paraitre «Pour un traitement unifie du suffixe -ier/-iere», Lexique 10.
1987 "Word formation (WF) as part of natural morphology", in: W. U. Dressier
et al., Leitmotifs in natural morphology (Amsterdam: Benjamins), 99 — 126.
<GDEL>
1982—1985 Grand dictionnaire encyclopedique Larousse. (Paris: Larousse), 10 vol.
<GLE>
1960 — 1964 Grand Larousse encyclopedique en dix volumes. (Paris: Larousse), Supple-
ments 1968, 1975.
Greimas, Algirdas Julien
1968 Dictionnaire de l'Ancien Frangais. (Paris: Larousse).
<GRLF>
1985 Le Grand Robert de la langue Frangaise. Dictionnaire alphabetique et ana-
logique de la langue frangaise de P. Robert, 2e ed. entierement revue et
enrichie par A. Rey. (Paris: Dictionnaires Le Robert), 9 vol.
Jackendoff, Ray
1975/1977 « Regularites morphologiques et semantiques dans le lexique », in: M. Ronat
(ed.) Langue. Theorie generative etendue (Paris: Hermann), 65 — 108.
<LXXe>
1928 — 1933 Larousse du XX" siede en six volumes publie sous la direction de P. Auge
(Paris: Larousse).
Lieber, Rochelle
1980/1981 On the organization of the lexicon [Dissertation, MIT] (Bloomington: In-
diana University Linguistics Club).
<PR86>
1986 Le Petit Robert. Dictionnaire alphabetique et analogique de la langue fran-
gaise. (Paris: Le Robert) [1(1967), 2(1977)].
Puchulu, Agnes
1987 Etude des suffixes -aire dans le cadre de la morphologie derivationnelle
[Memoire de maitrise <dactylographie>, Universite de Lille III].
1988 Propositions pour une interpretation semantique des adjectifs denominaux
[memoire de DEA <dactylographie>, Universite de Lille III].
Associative et stratification 59
Roeper, Thomas
1988 «Arguments implicites et la relation Tete-Complement», Lexique 7:
121-141.
Selkirk, Elizabeth
1982 The syntax of words (Linguistic Inquiry Monograph 7). (Cambridge, Mass.:
M I T Press).
Toman, Jindrich
1987 Wortsyntax. Eine Diskussion ausgewählter Probleme deutscher Wortbildung.
(Tübingen: Niemeyer).
Zwanenburg, Wiecher
1982 « Types de derivation comme universaux », Recherches de linguistique fran-
ςaise d'Utrecht (Utrecht: Institut d'Etudes frangaises), 57 — 66.
1984 "Word formation and meaning", Quaderni di Semantica 5: 130 — 142.
1987 «Structure et forme des adjectifs denominaux», Recherches linguistiques
d'Utrecht·. 5 5 - 8 0 .
Formal relations and argument structure*
0. Introduction
This paper is a refinement of the argument structure calculi of Di

Sciullo —Williams (1987), where the argument structure of a morpholog-
ical object 1 is calculated in terms of the arguments of its parts. 2
We will identify relations between the elements of argument structures,
preserving the idea that there is a basic difference between the external
argument (the italicized variable in an argument structure) and the inter-
nal arguments.
We will first discuss the notion of head, then consider different ways
in which a head may be related to a non-head, and propose well-
formedness conditions for argument structures.
1. The head relation
We will assume that the external argument is the head of the argument
structure. But how can we define the head with respect to the argument
structure in morphological objects which contain more than one argument
structure? A possible definition is found in (1), if we define feature F as
being the argument structure:
(1) Head F (read: head with respect to the feature F)
The head F of a word is the rightmost element of the word marked
for feature F. (Di Sciullo-Williams 1987: 26)
In English, (1) holds for a large class of morphological objects. There-
fore, in (2), for instance, the suffix -ed has no argument feature, and thus,
* This work was conducted as part of the Argument Structure Project at the Linguistics
Department of the Universite du Quebec ä Montreal. Support for the project was
provided in part by the SSHRCC grant no. 4 1 0 - 8 6 - 0 7 6 0 .
62 Anna-Maria Di Sciullo
even if it is the rightmost element of the word, it is not the head with
respect to that feature. Of the two constituents with the relevant features,
it is the rightmost which is the head with respect to the argument structure.
Thus, the external argument of formalize is the external argument of the
verbal suffix -ize and not the external argument of the adjective formal?
(2) formal\A ize]v ed\w (x, y)
ω (*, y)
There are languages where (1) is empirically false. Lieber (1988), for
instance, reports cases of Vietnamese and Breton compounds, as well as
affixed forms in Tagalog, where the head, with respect to the categorial
features, is on the left. Moreover, (1) may fail within the same language.
Consider the fact that, in Italian, the head with respect to the category,
and for our purpose the head with respect to the argument structure, is
on the right in affixed forms and generally on the left in compounds.
(3a) dimostrazione/ mangiabiL·/ giocare

'demonstration' 'eatable' 'to play'
b) «ave-tragetto/ blu-notte/ «ma-forte
'ferry-boat' 'night-blue' 'safe'
We propose a revision of (1) in terms of types of words (henceforth
word T ) as in (4), where word T includes suffixed forms and compounds.
(4) The head F of a word T is the X-most element of a word marked
for feature F.
Parameter: X-most: initial/final position.
It has been suggested in the literature, in Lieber (1988) for instance,
that the head of a word could be defined only in terms of inherent
features, and not in terms of positions. The definition in (5) expresses
this view:
(5) The head of a word is the element whose features F project to

the word.
However, (5), as opposed to (4), does not account for the cases where
two elements of a word are marked for a given feature F, as in (2) for
instance. It is not clear how a proposal which does not define the head
in terms of positions can account for these cases. Furthermore, (4)
accounts for the language variation noted above. This is done by fixing
the variable X to initial or final position, according to the word T .
Basically, the definition in (4) develops the idea of relativized head,

and for our purposes, it defines the head with respect to the argument
structure. This notion of head is specific to the morphology. There can
be no relativization of the head within the syntax, because there is only
one potential head in a phrase.
2. Control relation
In Di Sciullo — Williams (1987), a head with respect to argument structure

can be related to a non-head by controlling an argument of the non-
head, as in (6). We will assume here, as in Levin — Rapaport (1987), that
control applies to the predicate argument structure variables. Thus in
(6a), the head controls the external predicate-argument structure variable,
and in (6c), the head controls the internal predicate-argument structure
variable. Similar cases are found in French, as in (7), as well as in other
languages.
(6a) employ er (r, y)
( * i , Y ) fa)
b) the employer of John

c) employ ee (r, x)
(*. yd fa)
d) the employee of John
(7) agresseur / agressi,

'aggressor' 'aggressed person'
employeur / employe
'employer' 'employee'
hacheur / hachis
'chopper' 'chop'
Let us define the control relation as stated in (8):
(8) In a morphological object, A controls Β iff,
A is an argument of the headcontroi
A controls an argument Β of the non-head control
A and Β are co-indexed.
The control relation is determined by the lexical property of a listeme,
which is the headcontroi. Control in morphological objects is distinct from
syntactic control, since in the latter case it is only the external argument
that is controlled. The -e (-ee) nominals show that the internal argument
can also be controlled, which is not the case in syntax.
3. Binding relation
There is another type of relation which holds between the elements of

argument structures, which is not discussed in Di Sciullo — Williams
(1987), and which involves two predicate-argument structure variables of
the same type, 4 such as the [Χ Χ] χ compounds in (9).
(9a) telephone book (r), producer-distributor / plumber-contractor
(r i) (rO
b) Italian English (y), deaf-mute j bitter-sweet
(yd (yd
c
) up on (y, w), onto / into
Oj, w i) (Vj, Wi)
d) blow dry (x, y), stir-fry j slam-dunk
(xj, yj) (xj, yO
In (9a) and (9b), two external predicate argument structure variables

are bound. The examples in (9c) and (9d) include the binding of two
internal variables. These variables must be bound within the word. If it
were otherwise, their derived predicate argument structures would be ill-
formed, as in (10). Furthermore, according to the projection principle
(Chomsky 1986), lexical structures must be represented categorially at
every syntactic level. Thus, all the arguments of the predicate argument
structures in (10) must be projected in the syntax. However, the syntactic
realization of these structures is excluded since there is only one external
argument position for every argument taking lexical head.
(10a) telephone-book: *(r, r)
b) Italian-English: *(y, y)
c) into: *(y, w, y, w)
d) blow-dry: *(x, y, x, y)
We will define the binding relation which applies in morphological
objects as in (11):
(11) In a morphological object, A binds Β iff,

A and Β are predicate-argument structure variables of the same
type,
A and Β are co-indexed.
A possible explanation for the existence of (11) is that it results from
the following well-formedness condition of argument structures:
(12) Distinctivity
All the variables of a predicate-argument structure are distinct.
(12) states that there is no word whose predicate-argument structure
includes more than one instance of the same type of variable. This is the
case for external arguments, since predicators may have only one external
argument. One consequence of (12) is that the predicate-argument struc-
ture of verbs does not include two direct internal arguments. This is the
case if we assume binary branching structures for the syntactic realization
of arguments. Furthermore, the predicate argument structure of verbs
does not include two indirect internal arguments associated with the same
preposition, which is the case for verbs such as walk and fly, for instance.
Distinctivity forces the binding of two instances of the same type of
predicate-argument structure variable. When two variables are bound,
they constitute a unique variable. Thus, the facts in (9) are a consequence
of the interaction of (11) and (12).
Morphological binding, as defined in (11), is formally distinct from
syntactic binding. The syntactic binding of lexical anaphors, for instance,
typically involves arguments that are not of the same type: an external
argument and a direct internal argument.
4. Function-composition relation
In suffixed forms, the head can functionally combine with the non-head,
as in (13). As defined in Di Sciullo —Williams (1987), when function
composition occurs, the argument of the head becomes the external
argument of the whole, and the argument of the non-head becomes the
internal argument of the whole.
lazi ness
(13a) y)
construct ion (r, x, y)
(*, y) W
the construction of the city by the enemy
In Di Sciullo —Williams (1987), function composition is dependent on

specific suffixes, which we have called "functors". For instance, -ness,
-ude, -ion, and -ment are functors, but not suffixes such as -ee or -er,
which do not combine with the non-head, but control an argument of
the non-head.
However, note that function composition does not hold for examples
such as those given in (14), since the argument structures in (15), which
are obtained by function composition, are not proper argument structures
for (14).
(14a) univers al (y), rudimentary, courageous, alpine
(r) (y)
b) box er (χ, y), monopoliser, coudoyer
(r) (*, y) 'to monopolize' 'to elbow'
'to box'
(15a) * 0 , r)
b) *(x, y, r)
We want to explain this fact and reduce the reference to specific
functors in the system, by proposing that function composition is the
general case for the combination of arguments in suffixed forms, and
that its non-occurrence is attributed to (16):
(16) Specificity
The semantic nature of the external predicate-argument structure
variable of a word is specific to its category.
The semantic nature of an external predicate-argument structure var-
iable is fixed at the lexical-conceptual structure level. For a nominal such
as construction, for example, the r predicate argument structure variable
is projected from a semantic concept which is "the process r" or "the
result r"; for a derived adjective such as fearful, the y variable is projected
from a semantic concept which is "the affected thing y", and so on. 5
According to (16), there are semantic types of arguments which cannot
be the external arguments for certain categories. This is the case for r,
which cannot be the external argument for words other than nouns.
Moreover, r cannot be the internal argument for any other category, as
it is specific to the external argument of nouns. Thus, specificity blocks
function composition in (14), and prevents r from being the internal
predicate argument structure variable of adjectives and verbs.
5. Summary
In this paper, we developed the theory of morphology proposed in Di

Sciullo —Williams (1987), which calculates the categorial and argument
structure properties of derived words from the properties of their parts.
Formal relations hold between the sub-parts of derived words; refine-
ments to these relations were suggested, in particular with respect to the
head and binding relations. For all of these morphological relations
between arguments, we indicated ways in which they differed from equiv-
alent syntactic relations. Moreover, the well-formedness conditions of
argument structures that we have proposed affect function composition
and binding, as well as account for different cases of affixation and
compounding which the former argument-structure calculi had left unex-
plained.
Notes
1. In Di Sciullo —Williams's (1987) theory, a morphological object is a word whose

properties are derived from the morphological laws. Regular suffixed forms and com-
pounds are morphological objects. They are not part of the lexicon which includes
listemes and syntactic atoms.
2. In Di Sciullo —Williams (1987), an argument structure was defined as a set of theta
roles (agent, theme, etc.). We will assume here that a predicate argument structure is a
set of variables (r, ..., x, y, z) selected by a predicator, that is an argument taking a
lexical head, as in Levin — Rappaport (1987). The variables in the predicate-argument
structure serve as placeholders for arguments, which must be saturated at some level of
grammatical representation. They are distinguished by the way in which the NPs
corresponding to the arguments they stand for are realized in D-structure positions. The
predicate-argument structure contains variables which stand for the external argument,
the direct internal argument and the indirect internal argument of a predicator. These
variables are projected from concepts which can be defined in terms of deeper semantic
predicates at the level of lexical conceptual structure.
3. The following interpretations can be used for the predicate-argument structure variables:
" r " is realized at D-structure when a nominal head meets the specifer (as in Higginbotham
1985); "x" is agent; "y" is the theme; and " w ' " is the location. I will assume here that
theta roles are not primitives in the grammar.
4. We will assume that there are only three types of such variables: external, direct internal,
and indirect internal. When two variables are of the same type, they are both either
external, direct internal, or indirect internal. Furthermore, two indirect internal variables
are of the same type if they are associated to the same preposition.
5. It would be possible to suppose that "event" would be the external argument for verbal
predicators. This would make the external argument of nouns and verbs very similar,
to the extent that they would not be realized as NPs, but via functional categories:
Determiner and Inflection respectively. We will not investigate this hypothesis here,
given space limitations.
References
Chomsky, Noam
1981 Lectures on government and binding. (Dordrecht: Foris).
1986 Knowledge of language. (New York: Praeger).
Di Sciullo, Anna-Maria
1988 "Argument satisfaction, lexical and syntactic". [Paper presented at the
GLOW 1988 workshop on the interaction of the lexicon with other modules
of the grammar, Budapest],
Di Sciullo, Anna-Maria—Williams, Edwin
1987 On the definition of word (Linguistic Inquiry Monograph no. 14). (Cam-
bridge, Mass.: MIT Press).
Grimshaw, Jane
1987 "Psych verbs and the structure of argument structure". [Ms. Brandeis
University],
1988 "Adjuncts and argument structure", Lexicon Project Working Papers 21.
(Cambridge, Mass.: MIT Occasional Paper no. 36).
Hale, Ken —Keyser, Samuel Jay
1987 "Explaining and constraining the English Middle". [Ms. MIT],
Higginbotham, James
1985 "On semantics", Linguistic Inquiry 16: 547 — 593.
Jackendoff, Ray
1983 Semantics and cognition. (Cambridge, Mass.: MIT Press).
Levin, Beth — Rappaport, Malka
1987 "Non-agentive -er nominals: probe into argument structure". [Ms. MIT],
Lieber, Rochelle
1980 On the organization of the lexicon. [Ph. D. dissertation, MIT].
1988 "Phrasal compounds in English and the morphology-syntax interface",
Papers from the Parasession on agreement in grammatical theory, Chicago
Linguistic Society, PCLS 24-11.
Rappaport, Malka —Levin, Beth
1986 "What to do with theta roles", Lexicon Project Working Papers 11, MIT.
Williams, Edwin
1981 "Argument structure and morphology", The Linguistic Review 1: 81 — 114.
Zubizarreta, Maria-Louisa
1987 Levels of representation in the lexicon and in the syntax. (Dordrecht: Foris).
Austro-Hungarian morphopragmatics
Wolfgang U. Dressier and Ferenc Kiefer
When studying meaning in morphology, we claim that the following four

areas must be distinguished:
1. Morphosemantics = semantic meaning of morphological rules, i.e.,

regular semantic change between inputs and outputs of a morpholog-
ical rule; here we can differentiate word formation rules and inflec-
tional rules
2. Lexical semantics of morphology = semantics of morphologically
complex words, especially of the semantic idiosyncrasies of each output
of a morphological rule
3. Lexical pragmatics of morphology = idiosyncratic pragmatic
meanings/effects of individual, morphologically complex words
4. Morphopragmatics = general pragmatic meanings/effects of morpho-
logical rules, i.e., regular pragmatic change between input and output
of such rules.
From a semiotic point of view, morphopragmatics (cf. Dressier —

Merlini-Barbaresi 1987) comprises (a) the universal pragmatic founda-
tions of morphology (e.g., for the semantic relations among personal
pronouns), (b) the relations between morphological rules and their inter-
preters as well as the interpretant of a potential (or actual) output of
such a rule (particularly a word-formation rule).
In this dualistic investigation of morphopragmatic phenomena of
Viennese German and the Hungarian of Budapest we will deal with the
second aspect (b).
Hungarian EXCESSIVE - German aller-
In Hungarian, the excessive is formed by means of the prefix legesleg-
(the superlative has the prefix leg-) which is added to the comparative:
e.g., nagy 'big' — nagy + obb 'bigger' — leg + nagy + obb 'biggest' —
legesleg +nagy + obb 'the very biggest', and csiinya 'ugly' — csünyä + bb
'uglier' — leg + csünyä + bb 'ugliest' — legesleg +csünyä + bb 'the very
70 Wolfgang U. Dressler and Ferenc Kiefer
ugliest'. In contrast to the derived forms of nagy 'big', the derived forms
of csünya 'ugly' are not used relatively, the things compared are presup-
posed to be ugly.
What is the difference between the excessive and the superlative? Let
us have a look at the pairs legeslegnagyobb — legnagyobb and leges-
legcsünyäbb — legcsünyäbb. The excessive legeslegnagyobb can only be
used of things which are big, i.e., it introduces a presupposition in contrast
to the corresponding superlative which does not. On the other hand, the
difference between legcsünyäbb and legeslegcsünyäbb cannot be purely
presuppositional since the superlative already introduces the relevant
presupposition.
One important function of the excessive is, then, to introduce a se-
mantic presupposition in cases where it is absent in the superlative. In
order to find out what the pragmatic difference, if any, is between the
superlative and the excessive, let us then concentrate on the pair leges-
legcsünyäbb — legcsünyäbb. Compare the following sentences
(1) Εζ α legcsünyäbb rajz, amit valaha lättam.
'This is the ugliest drawing I have ever seen'.
(2) Ez α legeslegcsünyäbb rajz, amit valaha lättam.

'This is the very ugliest drawing I have ever seen'.
(1) is a factual statement about the quality of the drawing, (2), on the
other hand, is an emotionally colored statement, it is more naturally
interpreted as an exclamation. (1) can easily be denied, it is not the last
word to be said about the matter. On the other hand, (3) sounds odd
(possible only as a denial)
(3) ?Ez nem α legeslegcsünyäbb rajz, amit valaha lättam.
'This is not the very ugliest drawing I have ever seen'.
Apparently, the speaker attaches special importance to his qualifica-
tion; he does mean it to be the last word on the matter. No further
discussion is expected. The meaning of the excessive can thus be char-
acterized in the following manner:
(4) The speaker believes that the thing considered possesses the
absolutely highest possible degree of the property in question
(and he wants the listener to share his belief).
(4) also explains the presence of the semantic presupposition. The
highest possible degree of a property must, in fact, be qualifiable by
means of the property in question in an absolute sense. That is, the very
biggest house cannot refer to a small house since the very biggest expresses
the biggest possible. In other words, whenever the semantic presupposi-
tion is not present in the case of the superlative, it has to be introduced
via the excessive. (4) formulates the discourse function of the excessive
which is in principle independent of semantic presuppositions. One may
thus claim that the excessive introduces a discourse function which clearly
belongs to morphopragmatics, it is neither lexically conditioned nor a
property of morphosemantics.
The German correspondence of the Hungarian agglutinating prefix
legesleg- is aller- (originally a genitive plural of all- 'all') preposed to a
superlative, e.g., in Wissen Sie schon das Aller #neu + este? 'Do you know
the most recent news of all?' (neu 'new' superlative neu + est-). Although
less grammaticalized than its Hungarian equivalent (as is typical for
compounding, vs. derivation, cf. Lehmann 1982), it seems to share the
same pragmatic and semantic effects: it expresses the absolutely highest
possible degree of a property and it is only compared with items which
do have this property to a high degree; it is used for emphasis and for
impressing the hearer. Therefore it is often found at the end of a text
chunk, as in the "Auszählreim" (nursery rhyme):
(5) Du, bist schön und duj bist schön und duk die aller schönste.
'YoUi are beautiful and youj are beautiful and youk the most
beautiful of all'.
If aller- is prefixed recursively (e.g., die alleraller allerschön + sie), then
the respective form must be used at the end of the respective text chunk.
An exception is — for diagrammatic reasons — aller φ erst 'first and
foremost' which usually introduces the enumeration of things to be done
first, cf. the Ancient Greek equivalent pän#prötos in Homer (Ilias VII
324, IX 93, XVII 568, Odyssey IV 577, IV 780).
Similarly the excessive can be used as a corrective device in discourse:
(6a) Das ist sehr schlimm.
'That is very bad',
b) Aber das aller φ schlimmste ist, dass X ...
'But the very worst of all is that X ...'.
Emphasis and interactive orientation of expressing the absolutely high-
est possible degree also explain the use of the German excessive as
equivalent to Neo-Latin (and Hungarian!) superlatives in the following
feudal terms:
(7a) Der aller # durchlauchtig + ste Herr = Lat. serenissimus

'His (Ever) Most Serene Dignity'
b) aller # untertänig + st
'most devote of all'
c) Der aller # christlich + sie König = rex christianissimus
'His Most Christian Majesty' = le roi tres chretien (de France)
In the text worlds where these and similar expressions were used, they
underlined the absolutely highest degree of the property present. Cf.
allerhöchst 'the very highest' qualifying the absolute monarch or God.
Typical for Viennese (but not only Viennese) irony are pragmatic
reversals emphasizing the highest degree as in lexicalized Der Allerwerteste
'the dearest of all = buttocks' or in expressions such as Das wär' ja doch
das allerschönste 'This would be the most beautiful ( = indecent, imper-
tinent, scandalous) of all'.
Thus the morphopragmatic properties of the Hungarian and German
excessives are all-important; the introduction of a semantic presupposition
is secondary.
Diminutives
Let us now turn to the Hungarian "diminutive" suffix -i. Nouns are
formed from nouns by means of the suffix -i in two different ways:
(a) -i is added to the truncated form of the initial noun, cf. csok-i
(from csokoläde 'chocolate'), dir-i (from direktor 'director'), ov-i (from
ovoda 'kindergarten'), fagy-i (from fagylalt 'icecream'), zong-i (from zon-
gora 'piano'), szak-i (from szaktärs 'fellow-worker'), nyug-i (from nyu-
galom 'quiet').
(b) the suffix -i is added to the stem of the noun: läb-i (from läb 'foot'),
comb-i (from comb 'thigh'), has-i (from has 'belly'), hus-i (from hüs 'meat'),
ägy-i (from ägy 'bed'). As can be gathered from the examples cited, the
class of nouns in (a) is quite heterogeneous. In contrast, the nouns (b)
seem to belong to the semantic classes of body parts, baby food, and
objects which are important in the small child's life.
The suffix -i has no semantic meaning (it does not denote smallness):
both (a) and (b) are nouns with the same denotations as the corresponding
simplex nouns. If csoki and csokoläde 'chocolate' and läbi and läb 'foot'
are semantically identical, is there any pragmatic difference between the
corresponding forms? The forms in (a) presuppose a certain degree of
intimacy and have often a jocular touch. They are used in school slang
as well as in adult communication. The forms in (b), on the other hand,

belong to nursery talk, they are typically used in a discourse in which at
least one small child participates (even if only as passive hearer), and
they can also occur when addressing animals. It can also be transferred
to the purely adult world in certain specific speech situations: lovers in
the language of love, in attempts to please. Love and attempts to please
are also elements of nursery talk. One may thus generalize the cases of
transference by saying that the forms of nursery talk can be transferred
to the adult world just in case one of the ingredients of nursery talk
becomes the defining property of the speech situation. A detailed inves-
tigation of this problem would go beyond the scope of the present paper,
however. To summarize, nouns in -i have a clear pragmatic function: they
determine the social setting of the speech situation (in the case of (a)), or
they determine the relationship between speaker and hearer (in the case
of (b)). The suffix -/' has no other contribution to the meaning of the
noun. That is, the suffix -i is a morpheme with exclusively pragmatic
meaning. Notice that Hungarian -i (e.g., hus-i) does not denote smallness
in talking to animals either.
Similarly, the equivalent Austrian suffix -i can be extended to speech
addressed to animals, e.g.:
(9) Hund-i/ Katz-i, da ist das Fleisch-i, dein Papp-i,
'dogg-ie/ catt-ie, here is the meat-ie your papp-ie,
dein Tschapp-i
your food-ie'
But only Katz-i may be used in the language of love, otherwise
Fleisch + erl, Papp + erl must be substituted. (Tschapp+erl has a different
meaning (and etymology) 'awkward person', but Tschapp + erl#salat 'the
first leaves of lettuce' is semantically related).
The primary locus of -i suffixation is child language and baby talk,
cf. the nursery rhyme
(10) Kommt ein Maus-i aus dem Haus-i
'Comes a mous-ie out of its hous-ie'
Both Maus-i and Haus-i can be used when addressing pets, but only
Maus-i would seem appropriate for language of love (cf. also dialect
names for animals in Weber 1960: 120, 130, 136, 144, 146).
In a children's book (Nöstlinger 1987) which captures Viennese lan-
guage use quite well, diminutives in -i are only used by a four-year old
child, by his granny when speaking to him or about him, and — rarely
— by the rest of the family when sarcastically referring to them (i.e.,

there is a pragmatic sanction against the transfer of Austrian diminutives
in -i into a purely adult text world).
This pragmatic sanction is even stronger for the (nominal) diminutive
suffix -erl when attached to auxiliaries, principally hat 'has' and is(t) 'is'
— this is only possible when addressing small children or pets in a jocular
way, e.g.,
(11) Ja, was is-erl denn?
'Oh, what is it?'
(Nöstlinger 1987: 30)
All this contrasts strikingly with -erl affixation to nouns which is
extremely productive in Vienna. Of course, we must exclude here lexi-
calized forms such as Mäd-erl (= Mädchen) 'small girl' vs. Mäd-l 'girl'
which is typical for /-diminutives being more lexicalized than
er/-diminutives. Lexicalized e/7-diminutives are, e.g., Buss-erl 'kiss',
Flank-erl 'flake', Nock-erl 'dumpling', Pock-erl 'turkey, fir-cone'. There
is no *Buss or *Pock (Buss-i occurs in child language and language of
love), Flanke and Nocke have very different meanings (only Nock-i occurs
in child language). Cf. the qualification by Willi Schlamm (1938/1987:
150): "Die Sprache der Wiener neigte zum Diminutivum" ('The speech
of the Viennese tended towards the diminutive'), which is still valid today.
Cf. for Southern German in general Brandstetter (1964), Schirmunski
(1958).
The er/-diminutives either denote smallness, i.e., a small size or a
relatively small degree of the prototypical properties ascribed to the
respective nouns, they may connote endearment, intimacy, irony, sarcasm,
etc. (cf. Dressier — Merlini-Barbaresi 1987: 9ff.). Due to both meanings
they are used more frequently in discourses in which at least one small
child participates (even if only as a passive hearer or in his/her absence
as a person referred to) and therefore er/-diminutives can be also used
metaphorically for evoking childhood or connotations associated with it.
The precise connotative effect of an er/-diminutive is text-pragmatically
determined. Let us just mention a few generalizations. Whereas in socially
convergent verbal interactions er/-diminutives tend to receive positive
connotations, they are liable to receive negative ones in socially divergent
interactions. When the referent of a diminutive clashes with the denota-
tional meaning of the suffix, there is an effect of strong understatement,
e.g., mein Wag-erl 'my car' (Wag-en 'car') when referring to a Rolls
Royce.
This clash occurs only if the referent has a more than normal size,
whereas a normal-sized referent may be easily expressed with a diminu-
tive. This also explains the difficulty of accepting potential diminutives
such as Ries-erl 'giant-ie'. This example illustrates also that the presence
of the possessive 'my' in the microcontext conveys the connotation of
endearment. However, the macrocontext may add negative connotations
as in the oral narrative of a Jew who was forced to leave Vienna in 1938:
(12) So hab ich mein Koffer-l gepackt und bin weg.
'Thus I packed my suitcas-ie and went off.'
{Koffer-l is rather a haplology for *Koffer-erl than an instance of an
/-diminutive.)
Consider next the diminutive suffixes -kaj-ke and -cskaj-cske in Hun-
garian. The former occurs mainly after noun stems ending in I, r, n, m,
ny, s or ό, ö, and i. That is, -kaj-ke also occurs after the suffix i: läbi-ka
'small foot', combi-ka 'small thigh', husi-ka 'little meat', etc. Its primary
meaning is diminutive: level-ke 'small letter', asztal-ka 'small table',
leäny-ka 'little girl', tänyer-ka 'small plate', kanal-ka 'small spoon',
väros-ka 'small town', szekreny-ke 'small wardrobe' (some of these are
lexicalized forms). The same with -cskaj-cske, e.g., felhö-cske 'small
cloud', bokr-ocska 'small bush', vödr-öcske 'small bucket', läb-acska 'little
small foot', vaj-acska 'little butter', terd-ecske 'small/little knee'. The
diminutive meaning is exclusive except for the following semantic classes:
body parts, food, and familiar objects (this latter class is yet to be defined
more precisely). In the case of the above-mentioned semantic classes the
diminutive may acquire a purely pragmatic function when it is transferred
to the adult world (in the language of love, or if used jokingly in certain
speech situations). In such cases denotative diminutive meaning disap-
pears. That is, szäjacska 'small mouth', for example, will no longer mean
small mouth but, say, a mouth which I love. Or, läbika may even be a
big foot, in the language of love it means something like a foot to be
caressed. In nursery talk, it may still keep the diminutive meaning while
at the same time expressing affection. That is, depending on the speech
situation, the diminutive suffixes may acquire either a purely pragmatic
meaning or, in addition to the denotative meaning, certain pragmatic
overtones.
To sum up, it would seem that both Austrian and Hungarian dimin-
utives prove essentially the same thing, i.e., that diminutives may have
purely pragmatic functions. All morphological rules which contain a
pragmatic variable in the description of their meaning are morphoprag-
matically relevant.
On separation of form and meaning

If our assumption is correct that morphopragmatics must be separated
both from morphosemantics and from lexical pragmatics, then one can
find a new argument against the strict separation of morphosemantic
derivation rules and morphotactic affixation rules (as postulated, e.g., by
R. Beard (this volume) and A. Pounder (1988)). According to Beard's
(e.g., 1986, 1987) model there are rather few, general, and possibly
universal, semantic derivation rules, but many diverse affixation rules. It
would follow from these assumptions for our diminutives that we should
try to posit one morphosemantic derivation rule for all diminutives of a
language and then several morphotactic affixation rules (e.g., for Hun-
garian -i, -cska, -ka or Viennese -i, -I, -erl, -lein). But this implies that the
meaning effects of these word formation rules should be synonymous.
However we have found various (and regular) differences (a) in denotative
semantics (e.g., Hung. -/ not denoting smallness), (b) in connotations, (c)
in their morphopragmatic potential. Thus, e.g., for Hungarian diminu-
tives, we would have to postulate at least three different rules: two for -i
and a third one for -cskaj-ka.
In this way an apparent advantage of Beard's model, the simplicity
and generality of morphosemantic derivation rules, would vanish. And
this situation is not unique for diminutives, because most "synonymous"
word-formation rules are not really synonymous (pace Pounder 1987),
i.e., their putative synonymy resides in their classificatory labels such as
diminutives, agent nouns, action nouns (cf. Benveniste's (1948) classical
study). As far as morphopragmatics is concerned, it always depends on
the particular combination of form and meaning as embodied in the
various usual formats of morphological rules.
References
Beard, Robert
1986 On the separation of derivation from morphology: toward a lexeme I
morpheme-based morphology. (Bloomington: Indiana University Linguistics
Club).
1987 "Morpheme order in a lexeme/morpheme-based morphology", Lingua 72:
1-44.
Beneveniste, Emile
1948 Noms d'agent et noms d'action en indo-europeen. (Paris: Adrien-
Maisonneuve).
Brandstctter, Alois
1964 "Semantische Studien zum Diminutiv im Mittelbairischen", Zeitschrift für
Mundartforschung 30: 335 — 351.
Dressier, Wolfgang U. — Lavinia Merlini-Barbaresi

1987 Elements of morphopragmatics (Linguistic Agency, University of Duisburg,
Series A, Paper 194). [also to appear in the 1987 International Pragmatics
Conference proceedings]
Lehmann, Christian
1982 Thoughts on grammaticalization. (Köln: Arbeiten des Kölner Universalien-
Projekts 48).
Nöstlinger, Christine
1987 Wetti und Babs. (Wien: Jugend und Volk).
Pounder, Amanda
1987 Systemangemessenheit in der Wortbildung am Beispiel desubstantivischer Ad-
jektivableitung im Deutschen. [Phil. Dissertation, Universität Wien.]
1988 "On the status of conversion as a word-formation process", in: W. Dressier
et al. (eds.); Discussion papers, Third International Morphology Meeting,
Krems 1988 (Wiener Ling. Gazette, Suppl. 8), 1 9 - 2 1 .
Schirmunski, Viktor
1950 "Verstärkte Wortformen in den deutschen Mundarten", Zeitschrift fur
Mundartforschung 26: 225 — 238.
Schlamm, Willi
1938 [1987] "Das war Wien", Das Neue Tagebuch 6,18 (30. 04. 1938): 4 2 2 - 4 2 5 [Öster-
reichs Fall, herausgeg. v. U. Weinzierl (Wien: Jugend und Volk), 147 — 157)].
Weber, Otto
1960 Haustierbezeichnungen in den steirischen Mundarten. [Phil. Dissertation,
Universität Graz.]
Problems of word structure theories
Wolfgang Mötsch
1. In recent theoretical approaches to word formation a particular level

of word syntax has often been adopted. There are two major arguments
in favor of this assumption:
(1) Particular rules defining well-formed word structures seem to offer a
proper basis for semantic interpretation; and
(2) a representation of word structure seems to be necessary to demon-
strate the operation of some general grammatical principles like
thematic-structure satisfaction and feature percolation within word
structures.
However, if we look seriously at the theories proposed so far, some
fundamental doubts will arise as to whether these arguments really hold.
In this paper, I want to point out some crucial problems concerning this
approach. I am going to argue in favor of a revised version of redundancy
rules.
2. In my argumentation, I am going to touch three implications involved

in the assumption of a separate level of word-syntactic structure:
(i) There are rules determining well-formed syntactic words. These rules
are either based on separate principles or turn out to be determined
by general principles of phrase structure.
(ii) Representations on the level of word structure are the basis to which
processes of semantic interpretation apply.
(iii) Word-structure representations allow for the application of general
grammatical principles.
3. As to the first implication mentioned above, that is, the type of rules,
adopted to account for properties of compounds and derived words, the
most general device are rewriting rules of the type
X° Y° x°
80 Wolfgang Mötsch
There are more refined versions of this rule scheme, but they may be
neglected for the purpose of my argumentation.
Rewriting rules of this form have some general properties in common
with phrase-structure rules of the X type. They are, however, restricted
in some fundamental respects:
(i) Only a limited subset of lexical categories may enter into these rules.
(ii) Only zero-bar categories may appear in word structures, that is, no
phrasal structures are allowed within word structures.
(iii) Word-structure rules generate binary structures.
(iv) One of the constituents of a word structure is identical with the
dominating category. In our version of the rule scheme it is the
right-hand constituent, which may be defined as the head of the
construction.
All predictions of this rule scheme are empirically more or less proble-
matic. There are many counterexamples to each of them. This, however,
is not necessarily a sufficient argument against the theory, because it may
turn out to be an interesting generalization which represents the unmarked
properties of word structure. This presupposes that deviating cases might
be interpreted as marked exceptions. Unfortunately, a proper theory of
markedness which excludes purely intuitive judgements on markedness is
not available.
Despite the lack of a theory of markedness, I think, there are some
problematic predictions which — as far as I can see — may not be
resolved by markedness assumptions. One general problem of word-
structure theories is the fact that some of the presumably marked cases
cannot be described other than in an ad hoc way. Difficulties arise with:
(i) The prominent role of affixation among other kinds of devices in-
volved in morphological processes.
(ii) Synthetic compounds and other types of words containing phrases,
cf. German Arbeit-nehm-er, Saure-Gurken-Zeit
English car driv-er, peace lov-ing,
generative-grammar approach.
(iii) The restriction to binary branching.
Word-structure rules of the above-mentioned type generate hierarchical
structures with lexical categories as final elements and as dominating
nodes. It is assumed that the lexical insertion rule applies to these
structures. This, in turn, presupposes that affixes — like roots — are to
be analyzed as lexical entries. Affixes, of course, are considered to be
lexical entries requiring a sister constituent which belongs to a certain

class of words. The assumption that affixes have a separate status as
lexical entries, to make it clear, follows from the theory. In my view, the
treatment of affixes as separate dictionary entries has two serious dis-
advantages:
(i) It rules out morphological processes like conversion, discontinuous
affixes (cf. German Gt-schrei-€), modification of the base string,
reduplication, and subtractive techniques.
(ii) It blurs the difference between two types of restrictions which affixes
impose on their base words. There are restrictions which ultimately
depend on the semantics of a derivational process and which are
represented in terms of lexical categories and subcategories of the
base word, and there are restrictions on purely morphological and
perhaps phonological grounds. In German, to give some examples,
the distribution of -heit und -keit depends on properties of the latter
kind. The derivational process
(entpref last N ) v en inf
excludes complex words. Cf. entflecken but *entfettflecken. Further-
more, derivational affixes in German may be divided into two types.
One of them attaches the affix directly to the stem, the other one
requires an extended form of the base.
Cf. frau-lich, weib-isch, Bieg-ung
frauen-haft, weiber-mäßig, biegt.-fähig
The distribution of class I and class II affixes in English seems to
belong to this type of restrictions too. If this is correct, many of the
so-called bracketing paradoxes may be simply paradoxical descrip-
tions of the facts, because a description of these morphological
regularities does not necessarily presuppose hierarchical structures.
All attempts to extend the lexical entry approach to morphological
processes other than affixation involves exceptions to general properties
of lexical entries. We may, for example, admit phonologically empty
affixes in order to account for conversion. Attempts to represent the
regular nature of processes like modification of the base string by affixes
may only lead to purely technical solutions. In other words, it seems to
be impossible to reduce all types of morphological processes involved in
derivational morphology to separate affixes, even if we adopt a very
abstract concept of affix.
A second exception to the standard conditions on lexical entries arises
with prefixes. Prefixes are morphemes which typically lack an indication
82 Wolfgang Mötsch
of lexical category. This is simply a consequence of word structure

properties. In word structures, only that constituent must be lexically
categorized which bears the properties of the entire newly coined word.
The category membership of the non-head constituent only plays a role
in the formulation of conditions on the base to which a certain affix is
attachable. It is redundant, as far as I can see, in the representation of
the derived word. There are, apparently, no rules referring to the lexical
category or subcategory of the non-head constituent of a complex word.
One strong argument in favour of the lexical-entry approach to affixes
seems to be the possibility to apply the concept "head" to word structures.
We have to take into account, however, that the notion "head of a word"
may not be defined without exceptions in many languages. In German
and English, for example, there are prefixes which should be analyzed as
heads. This, of course, would be an exception to the general definition
of the head as the right-most constituent of a word.
In German ent-laus-en, ent-wanz-en (which correspond to English de-
bug) and be-stiick-en, be-last-en are examples of prefixes indicating the
categorial status of the entire derived word.
Thus, a dilemma remains in any case. Prefixes without category mark-
ing are an exception to the general property of lexical entries to be a
member of a certain class of words. Prefixes with category marking are
an exception to the assumed principle of the head position.
One consequence of the assumption that prefixes must not be marked
for lexical categories is the rejection of rules involving lexical categories
in order to derive prefixed words. Lieber's (1980) modification of the
word structure approach avoids this problem with prefixes. It also avoids
a further obscurity of rules containing lexical categories. These rules
predict subclassifications like
[A - Ν AffixA ]
parallel to subcategorization in phrasal structures, cf.
[VP - V NP]J
However, whereas verbs requiring NPs as direct objects are an important
syntactic generalization, no class of "adjective suffixes requiring nouns
as co-constituents" is relevant in word structures. Restrictions concerning
the base word in derivational processes are a purely individual property
of affixes, not of classes of affixes.
A crucial problem for all sorts of word-structure theories evolves from
synthetic compounds which may be observed in many languages. In
English and German, there are many highly productive types of synthetic
compounds, that is, processes which attach an affix to a phrasal base.
However, this phrasal base is restricted to a concatenation of words, that
is, no normal phrase is allowed. As far as I can see, no word-structure
approach has a sufficient solution to the problem of how restricted
phrases may enter into word-structure trees. Even if we agree with an
analysis like
gle
it will remain a secret how phrasal substructures of this restricted sort

may enter into word trees on the basis of solutions that are other than
ad hoc.
4. Let me now turn to questions of semantic interpretation. The first

observation concerning word structures is that they do not offer any
direct basis to semantically necessary distinctions between subordinated,
coordinated, and exocentric compounds. The interpretation depends on
information available from the lexical entries which are part of a com-
pound, in particular on information about the thematic structure of
relational elements. If a compound lacks a relational element, a relation
will have to be added by semantic processes. I think it is quite clear that
the concatenation of lexical categories in words neither depends on
genuine word-syntactic principles nor does it offer interesting restrictions
to semantic interpretation. Syntactic information involved in rules of the
above-mentioned type appears to be only a reflection of underlying
semantic processes.
5. It should furthermore be realized that the concept "head of a word"

is of little help if we want to try to extend devices of feature percolation
to the description of argument-inheritance phenomena. In these cases the
84 Wolfgang Mötsch
verb, that is, the non-head constituent of a nominalization, transmits its

thematic structure to the whole word. In any case, argument-structure
inheritance in nominalizations is only one case of the reconstruction of
theta-grids involved in derivational processes.
6. To conclude, some of the implications of word-structure theories

seem to be in a serious conflict with facts we want to be covered by a
theory of word formation. Some of the attempts to account for these
facts force us to accept deviations from standard assumptions. Facts like
purely morphological devices and restrictions cannot be sufficiently ac-
counted for in this theoretical framework. A level of word-syntactic
structure seems neither to be justified by processes of semantic interpre-
tation nor by the description of thematic structure reconstruction involved
in derivational processes. One major disadvantage of the word-structure
approach is the blurring of the relatively independent purely morpholog-
ical processes in word formation on the one hand and the
semantically-based processes on the other.
In my view, redundancy rules offer a descriptive device which avoids
most, perhaps all, of the problems mentioned. Redundancy rules of the
type sketched by Jackendoff (1975) following Chomsky's (1970) proposal
enable us to make explicit all aspects of relatedness between base words
and derived words, as well as between pairs of words and compounds.
Redundancy rules may be divided into two parts as Jackendoff has already
observed. One part accounts for morphological conditions on the con-
catenation of stems and affixes. The other one is organized by properties
of the semantic, and in part of the thematic, structure involved in
word-formation processes. Redundancy rules describe types of word
formation as global processes of sign creation. In derivational processes,
there is no need to designate a head, because the category marking of
the derived word is a property of the entire word. Each rule application
results in a representation of the lexical-entry properties of its products.
There is no need for special rules generating hierarchies. The existence
of more complex words is simply a consequence of the fact that some
affixes (and compounds in general) allow complex bases. Synthetic com-
pounds may be analyzed as a combination of compounding and deriva-
tion, quite in the spirit of the traditional analysis, that is, there is no
syntax involved. The concatenation can be described by special types of
redundancy rules.
Of course, a theoretical approach has to offer restrictions to the very

powerful device sketched by Chomsky and Jackendoff. In general, these
restrictions must account for principles underlying the morphological and
semantic aspects of word-formation processes. Only if this has been done
can the two approaches be compared more thoroughly. What we also
want to be added to a theory of word formation is an evaluation of
different realizations of possible structural types or processes. A scale of
markedness seems appropriate to do that. The proposals by Dressler et
al. (1987) concerning markedness involved in derivational processes is an
illustration of what I have in mind.
There is still one further argument in favor of redundancy rules which
I want to point out. Rules of this sort enable us to account for lexical
relatedness of all kinds in lexicalized complex words. The prediction of
new words and the application of redundancy rules to coin new words
may be considered to be a special function of these rules which presup-
poses some particular properties of redundancy rules, cf. Mötsch (1988).
Thus, we may in principle use the same theory to account for highly
productive processes and for restricted relations between lexicalized words
which, in turn, may be models for analogical word formation.
References
Chomsky, N o a m A.
1970 "Remarks on nominalization", in: R. A. Jacobs — P. S. Rosenbaum (eds.)
Readings in English transformational grammar (Waltham, MA: Ginn), 184 —
221.
Dressler, Wolfgang U.— Willi Mayerthaler —Oswald Panagl — Wolfgang U. Wurzel
1987 Leitmotifs in natural morphology. (Amsterdam: Benjamins).
Jackendoff, Ray
1975 "Morphological and semantic regularities in the lexicon", Language 51:
639-671.
Lieber, Rochelle
1980 On the organisation of the lexicon (Cambridge, M A ) [MIT dissertation],
Mötsch, Wolfgang
1988 "On inactivity, productivity, and analogy in derivational processes", Lin-
guistische Studien 179: 1—30.
Constraints on the Italian suffix -mente*
Sergio Scalise
It has been maintained that the suffix -ly in English, being fully produc-
tive, could be considered "inflectional". 1 In this paper I intend to dem-
onstrate that the Italian suffix -mente, which may be viewed as equivalent
to English -ly, although very productive, is not "fully" productive. On
the contrary, it is subject to a number of subtle semantic and morpho-
logical constraints which cannot be interpreted as "inflectional". It is
therefore claimed that -mente (and consequently -ly) are derivational
affixes.
Constraints on simple adjectives and complex adjectives (compounded
and derived) will be examined in that order.
1. Negative Constraints
1.1 Simple adjectives

-mente cannot be attached to possessive, demonstrative, indefinite, or
numeral adjectives:
(la) mio —• *miamente 'my'
b) quest ο —*• *questamente 'this'
c) qualche —• * qualchemente 'some'
d) due —• *duemente 'two'
As far as qualifying adjectives are concerned, in general, the most signif-
icant constraints are semantic. That is, -mente cannot be attached to
adjectives denoting physical properties:
(2) brutto —• *bruttamente 'ugly'
bello —• *bellamente 'goodlooking'
calvo —> *calvamente 'bald'
* This paper is based on Scalise et al. (in press).

88 Sergio Scalise
Nor can the suffix ever be attached to adjectives denoting color:

(3) giallo —> *giallamente 'yellow'
viola —> *violamente 'purple'
When the base adjective has more than one meaning — a "literal" one
and a more abstract "metaphoric" one — m e n t e always selects the latter
(cf. also 1.3.3.1):
(4) arido 'arid' 1: 'free from water' —> *-mente
2: 'lacking feeling' —» aridamente
Finally, there appears to be a constraint relating to the "spatio-temporal"
meaning of the base. If an adjective has two readings, one referring to
time and the other to space, -mente selects the temporal meaning:
(5) lungo 'long' 1: (rel. to space) —* *-mente
2: (rel. to time) —> lungamente
1.2 Compounded adjectives

-mente cannot be attached to compounded adjectives even if it can be
attached to the second element of the compound. This is true both in the
case of "strict" compounds (6a) and in the case of "loose" compounds
(6b):
(6a) [dolce] [amaro] + *mente 'sweet and sour'
but cf. amaramente 'bitterly'
dolcemente 'sweetly'
b) [storico] [critico] + *mente 'historical-critical'
but cf. criticamente 'critically'
storicamente 'historically'
This appears to be a "morphological" constraint. However, another
explanation may be offered: in a model of ordered morphology such as
"lexical morphology", composition follows derivation and therefore the
derivation of compounds is excluded a priori.
1.3 Complex adjectives

Three subclasses of complex adjectives will be considered here separately:
(a) superlative/comparative adjectives, (b) adjectives modified by an
evaluative suffix, and (c) adjectives deriving from a lexical category X by
adding an adjective-forming affix.
1.3.1 Comparative/superlative adjectives

Only the "synthetic" comparatives from this class will be tested since
analytic comparatives cannot receive a suffix (except inflection for gender
and number). The suffix -mente can be attached to a superlative but not
to a comparative:
(7) ottimo ottimamente 'best'

migliore *migliormente 'better'
pessimo pessimamente 'very bad'
peggiore *peggiormente 'worse'
As regards the superlative form of the adjective, on the other hand,
the adverb formation is fully productive:
(8) leggero leggermente 'light'
leggerissimo leggerissimamente 'very light'
tranquillo tranquillamente 'quiet'
tranquillissimo tranquillissimamente 'very quiet'
This pattern is very regular and can be summarized as follows: if
A + mente, then A + issima +mente will always be possible whereas if
*A + mente, then it will hold true that *A + issima + mente.
1.3.2 Adjectives with evaluative suffixes

-mente cannot be attached to adjectives modified by an evaluative suffix:
(9) grande —> grandemente 'big'
grandicello —> *grandicellamente 'quite big'
leggero —> leggermente 'light'
leggerino —> *leggerinamente 'rather light'
This case as well may be explained in terms of order. In Scalise (1988),
an attempt was made to demonstrate that the evaluative suffixes occupy
a particular position in the system, that is, they apply "after" the normal
word-formation rules. The output "adjective + evaluative suffix" there-
fore is not available for the rules adding -mente.
1.3.3 "Derived" adjectives

What follows is a list of the suffixes forming Italian adjectives and the
results of -mente attachment:
90 Sergio Scalise
(10)
*
1. -accio golosaccio — • *golosacciamente 'greedy'
*
2. -acco polacco — • *polaccamente 'Polish'
*
3. -ace seguace — > *seguacemente 'follower'
*
4. -aceo cartaceo — • *cartaceamente 'papery'
Yj* 5. -ale postale — v *postalmente 'postal'
naturale — • naturalmente 'natural'
Y/* 6. -aneo cutaneo — • *cutaneamente 'cutaneous'
temporaneo — V temporaneamen te 'temporary'
*
7. -ano isolano — • *isolanamente 'insular'
Yj* 8. -ante ignorante —• *ignorantemente 'ignorant'
brillante — • brillantemente 'brilliant'
*
9. -ardo savoiardo *savoiardamente 'Savoyard'
YJ* 10. -are immobiliare — • *immobiliarmente 'immovable'
militare — • militarmente 'military'
Yj* 11. -ario funerario — • *funerariamen te 'funerary'
arbitrario — • arbitrariamente 'arbitrary'
*
12. -asco comasco — > *comascamente 'from Como'
*
13. -astro biancastro — • *biancastramente 'dirty white'
*
14. -ate arpinate — • *arpinatemente 'from Arpino'
Yj* 15. -atico asmatico — • *asmaticamente 'asthmatic'
dogmatico — • dogmaticamente 'dogmatic'
Yj* 16. -ato stellato — • *stellatamente 'starry'
dettagliato — • dettagliatamente 'detailed'
Yj* 17. -bile utilizzabile — • * u tilizzabilmen te 'utilizable'
amabile — • amabilmente 'lovable'
*
18. -eccio mangereccio — • *mangerecciamen te 'eatable'
*
19. -ello miserello — • * miserellamente 'quite poor'
*
20. -eno madrileno — • *madrilenamente 'Madrilenian'
Yj* 21. -ente adiacente *adiacentemente 'adjacent'
precedente — • precedentemente 'previous'
*
22. -eo europeo *europeamente 'European'
*
23. -eo argenteo *argenteamente 'silvery'
*
24. -escente fosforescente *fosforescen temen te 'phosphorescent'
Y/* 25. -esco cardinalesco —> * cardinalescamente 'cardinal'
principesco — • principescamen te 'princelike'
*
26. -ese bolognese — » *bolognesemente 'Bolognese'
*
27. -etto furbetto —>· *furbettamente 'cunning'
Y/* 28. -evole pieghevole *pieghevolmente 'pliable'
amichevole — • amichevolmente 'friendly'
Y/* 29. -(i)ano transiberiano —». * transiberianamen te 'trans-Siberian'
cristiano —> cristianamente 'Christian'
*
30. -iccio torbidiccio —> *torbidicciamente 'rather turbid'
*
31. -icello grandicello —>· *grandicellamente 'quite big'
*
32. -icino grandicino *grandicinamen te 'rather big'
*
33. -ico 1 libico —»· *libicamente 'Libyan'
Y/* 34. -ico 2 desertico — • *deserticamente 'desert-'
barbarico barbaricamente 'barbarian'
Y 35. -ido lucido — • lucidamente 'polished'

*
36. -iero alberghiero — • *alberghieramente 'hotel-'
Y/* 37. -ifico sudorifico — • * sudor ificamen te 'sudorific'
pacifico — * pacificamente 'pacific'
*
38. -igno asprigno — > *asprignamente 'rather sour'
Y/* 39. Hie versatile • — • *versatilmente 'versatile'
duttile — > duttilmente 'ductile'
Y/* 40. -He vescovile — • *vescovilmente 'episcopal'
signorile — • signorilmente 'gentlemanlike'
*
41. -ineo sanguineo — • *sanguineamente 'bloody'
*
42. -ingo ramingo — • *ramingamente 'wandering'
* 'salty'
43. -ino 1 salino — > *salinamente
*
44. -ino 2 tunisino — • *tunisinamente 'Tunisian'
*
45. -ino 3 beiI ino — • *bellinamente 'quite nice'
*
46. -(i)ota cipriota — • *cipriotamente 'Cyprian'
Y 47. -issimo grande — • grandissimamen te 'very big'
YJ* 48. -istico violinistico — • * violinisticamente 'rel. to violin'
idealistico — » idealisticamente 'idealistic'
*
49. -ita vietnamita — • *vietnamitamente 'Vietnamese'
Y/* 50. -ivo purgativo — • *purgativamente 'purgative'
impulsivo — > impulsivamente 'impulsive'
*
51. -lento sonnolento — • *sonnolentamente 'sleepy'
*
52. -occio belloccio — • *bellocciamente 'rather nice'
*
53. -occo bustocco — • *bustoccamente 'from Busto'
*
54. -ognolo amarognolo — > * amarognolamen te 'bitterish'
*
55. -olino marzolino *marzolinamente 'of March'
*
56. -olo romagnolo —v * romagnolamen te 'from Romagna'
*
57. -one furbone — » *furbonamente 'cunning'
Y/* 58. -oso erboso — > *erbosamente 'grassy'
coraggioso — • coraggiosamen te 'bold'
*
59. -otto 1 chioggiotto — • * ch ioggio t tame η te 'from Chioggia'
*
60. -otto 2 stupidotto — > * stupidottamente 'quite stupid'
γ 1*
61. -(t)orio venatorio — > * venator iamente 'venatorial'
obbligatorio — • obbligatoriamente 'obligatory'
*
62. -trice ricama trice — • * ricamatricemen te 'embroideress'
*
63. -uccio caruccio — • * carucciamente 'pretty'
*
64. -uto barbuto —>. *barbutamente 'bearded'
*
65. -uzzo avaruzzo * avaruzzamente 'rather stingy'
These data are very clear: 45 out of 65 suffixes never allow the affixation
of -mente·, 20 suffixes sometimes allow and sometimes do not allow the
affixation and only two "always" allow the affixation (including the
superelative suffix -issimo, for which cf. 1.3.1). The quantitative data
alone would suffice to show that the claims regarding the "full" produc-
tivity of the suffix under analysis prove to be empirically unfounded.
-mente "avoids" entire semantic sets: in fact it is not added to adjectives
92 Sergio Scalise
denoting provenance (2,7,9,14,20,22,26,33,44,49,53,56,59) or to adjectives

formed by evaluative suffixes (1,13,19,27,30,31,32,38,45,52,54,55,57,60,
63,65), or to material adjectives (4,23,43).
The remaining suffixes include 18,24,36,39,41,51,62, and 64. Of the
suffixes belonging to this group, -eccio, -escente, -ineo, -ingo, -lento, and
-trice are marginal "adjectival" suffixes in that -ineo is a variant of -eo,
-escente is a variant of -ente, -lento and -eccio are no longer productive
and only occasionally does -trice form adjectives. As far as the two
suffixes left, -uto and -iero, are concerned, see the reasons given in the
next section.
1.3.3.1 Noun + adjectival suffix + -mente

Among the suffixes listed in (10), there are some such as -oso, -ale, -ario,
and -ico to which -mente may or may not be attached. In all these cases
the affixation of -mente seems to be possible when the noun of the base
is [ +abstract] but is not possible when the noun of the base is [—abstract].
The fact that a feature such as [ + abstract] can be relevant for a deriva-
tional process is not surprising: for example, the Italian suffix -aio can
be added to [ — abstract] nouns but not to [ +abstract] nouns:
(11) libro —> libraio 'book — book seller'
pazienza —> *pazienzaio 'patience — *patience seller'
What is surprising, though, is the fact that the suffix seems to be sensitive
to non-"adjacent" information. This would obviously be a violation of
the adjacency condition because -mente would be able to "see" infor-
mation contained in a non-adjacent cycle2:
]N + Suf]A + mente]
[ — abstr]
However, if the noun of the base is [—abstract] and the derived adjective
has two meanings, then the affixation of -mente is possible (but only with
the "metaphoric" meaning):
(12a) vorticoso 'whirling' 1 'full of whirls' —> * +mente
2 'very quick' —* + mente
b) teatrale 'theatrical' 1 'related to the theater' —• * -[-mente
2 'dramatic, exaggerated' —> + mente
As can be seen, this constraint is the same as the one applying to simple
adjectives (cf. (4) in 1.1).
1.3.3.2 Verb 4- adjectival suffix + mente

The suffixes -bile, -ante, -ente, -evole, -ile, and -ivo are added to verbs.
After such suffixes, the derivation in -mente may or may not occur:
(13) asfissiante asfissian temen te 'asphyxiating'
galleggiante *galleggian temen te 'floating'
prudent e prudentemente 'prudent'
contenente * con tenen temen te 'containing'
lodevole lodevolmente 'laudable'
sdrucciolevole *sdrucciolevolmente 'slippery'
impulsive impulsiv amente 'impulsive'
purificativo *purificativamente 'purifying'
All of these cases seem to follow the same strategy: if the derived adjective
is very transparent with respect to the verb, then affixation is blocked.
If, on the other hand, the adjective has undergone a process of semantic
drift, then the affixation is possible.
2. Positive constraints
The suffix in point is also subject to positive constraints. One of the
constraints affecting the various suffixal categories has to do with the
notion of "time", -mente may or may not be added to a certain set of
suffixes. If these suffixes form adjectives denoting "time", then -mente
may be added freely:
(14) acciden t+ale 4- men te 'accidental'
episod + ica + mente 'episodically'
quotid + iana + mente 'daily'
regol + are + mente 'regular'
mens + ile + mente 'monthly'
This constraint must be linked to the one seen above for simple adjectives
(cf. 1.1).
3. -mente and prefixation

Prefixation may have effects on the attachment of -mente. Consider the
following possibilities
94 Sergio Scalise
(15a) if [\X\ +mente] then [pre + [X]] + men te

b) *[[X] + mente] then * [pre + [X]] + men te
c) [\X\ +mente] but * [pre + [X]] + men te
d) *[[X] + mente] but [pre + [X]] + men te
My data provide evidence for all of the four cases in the following order
of frequency: (b), (a), (c), (d) (only for the prefix in-).
Examples of (15a) and (15b) are the following:
(16a) critico criticamente 'critical'
acritico acriticamente 'not critical'
estetico esteticamente 'esthetic'
antiestetico an tiesteticamen te 'antiesthetic'
b) diluviano *diluvianamente 'diluvian'
antidiluviano * antidilu vianamen te 'antidiluvian'
navigabile *navigabilmente 'navigable'
circumnavigabile * circumna vigabilmente 'circumnavigable'
As regards group (d), the following examples will suffice:

(17) consolabile * consolabilmente 'consolable'
inconsolabile inconsolabilmente 'inconsolable'
qualificabile * qualificabilmente 'qualifiable'
inqualificabile inqualificabilmente 'not qualifiable'
As can be seen, the suffixation by -mente of an adjective ending in -bile
is made possible by the presence of the negative prefix in-. The fact that
a prefix can favor the attachment of a suffix is a counterexample to the
"atom condition" (cf. Williams 1981), but this possibility has already
been observed in connection with the suffix -mento and the prefix a- in
Italian (cf. Scalise - Zannier 1982-1983: 195).
The situation in (c) is more complex as a greater number of prefixes
are involved. Let us consider, however, what happens when the base
adjective is semantically complex, that is when it has both a literal and
a metaphoric meaning as in the following examples:
(18) spasmodico 'spasmodic' 1 'affected by spasm'
2 'frenetic'
rigido 'rigid' 1 'hard'
2 'severe'
The consequences of prefixation for the attachment of -mente can be seen
below:
(19) rigido —> rigidamente 'rigid'

semirigido —> *semirigidamente 'semi-rigid'
spasmodico —• spasmodicamente 'spasmodic'
antispasmodico —> *antispasmodicamente 'antispasmodic'
In general, a prefix selects one (and only one) meaning, actually the more
"literal" one. 3 Also in the examples just given the prefix chooses the more
literal meaning (antispasmodico 'antispasmodic', for example, refers to
products to prevent spasms, not to something to avoid rushing, semirigido
'semirigid' does not mean 'severe but not too much') but doing so it
automatically disfavors the attachment of -mente which, as we have seen,
prefers metaphoric meanings.
In connection with these facts (as well with others as observed in
Scalise 1988), my claim is that prefixation can favor or disfavor only the
attachment of derivational material and not of inflectional material.
4. Conclusions
The suffix -mente seems to draw morphological distinctions between

simple, compounded, and derived adjectives. It distinguishes between
indicative and qualifying adjectives in the first case; it never applies in
the second case, and it triggers a number of subtle distinctions in the
third case.
Among the derived suffixes, -mente clearly distinguishes between su-
perlative (to which it always attaches), evaluative (to which it never
attaches), and derived suffixes proper. In the latter case, -mente makes
global semantic choices: it does not attach to adjectives denoting "prov-
enance" or "material". As far as the other classes of adjectives are
concerned, -mente is added only if the base is an abstract noun. Further-
more, when the base is an abstract noun and the derived adjective has
two meanings, a literal and a metaphoric one, -mente selects the meta-
phoric meaning.
Finally, this suffix is never added to adjectival bases deriving from
verbs where the derivation V —> A is still clearly identifiable. Thus a
present or past participle does not, as a rule, constitute the base for an
adverbial derivation unless it underwent some sort of semantic drift. The
positive constraints which have been singled out, on the other hand, are
generally related to the notion of "time", including frequency.
96 Sergio Scalise
As for the relation between -mente and prefixation, in some cases

(especially with the prefix in- and the suffix -abile) prefixation favors
suffixation which is typical of the relationship between prefixation and
derivational suffixation, thus confirming the initial hypothesis that -mente
behaves more as a derivational suffix than as an inflectional one.
5. Post-scriptum
A reader of the present paper (whom I want to thank for his observations)
suggested to me that the restrictions on the attachment of -mente can be
explained very easily if one assumes that "A + mente" is possible when
the morphological sequence corresponds to 'in an A way' (e.g., semplice-
mente 'simply' = 'in a simple way'). The reader maintains, furthermore,
that in a language with a case "ablativus modi", the constraints on the
adjectives are the same and therefore there is no point in arguing in favor
of the derivational status of -mente.
Even though such a proposal has some advantages, it cannot be
accepted here for the following reasons. In the first place, the theoretical
framework adopted here, being strictly "lexicalist", rules out the possi-
bility of "matching", so to speak, words with phrases. In the second
place, the offered explanation needs, in my opinion, to incorporate some
diachronic information which is not within the reach of a synchronic
model such as the one adopted here.4 In the third place, not every problem
would be explained anyway. In general, if the offered explanation were
correct, we would expect both the following implicational patterns:
(20a) if 'in an A way' then A+mente
b) if *'in an A way' then *A + mente
This pattern holds true for many cases, but not for all. In fact we also
find the following possibilities:
(21a) 'in an A way' but *A + mente
b) *'in an A way' but A + mente
Examples of (21a) are the following:
(22) in modo rivelatore *rivelatormente
in modo calzante *calzantemente
in modo invidiabile *invidiabilmente
in modo distorto * dist or lament e
Examples of (21b) are the following:

(23) *in modo bimestrale bimestralmente
*in modo estetico esteticamente
*in modo caldo caldamente
*in modo grafico graficamente
The hypothesis in (20) is thus falsified in both directions. (20a) is falsified
when the adjective is deverbal (cf. the examples in (22), and (20b)) is
falsified in various cases: (i) when the adverb express "periodicity" (cf.
(24a)), (ii) when the resulting adverb modifies adjectives (cf. (24b)), (iii)
when the adverb can only be used in metalinguistic contexts (cf. (24c)):
(24a) *in modo settimanale settimanalmente

'weekly'
b) * in modo alto altamente (prevedibile)

'highly (foreseeable)'
c) * in modo batter iologico batteriologicamente (parlando)
'bacteriologically (speaking)'
I will not try to account for these data here. Clearly some aspects of the
problem are open to further investigation, but I consider the evidence
discussed in this paper as a little step in favour of the "derivational"
status of the suffix -mente.
Notes
1. Cf. Bauer (1983: 8 4 - 8 5 ) and Bybee (1985: 225).

2. Cf. Siegel (1977). The reported observation would also be a violation of the "atom
condition" of Williams (1981).
3. Cf. Mutarello (1987).
4. In Scalise (1983) the difference of status between -mente in Latin and in Italian is
discussed.
References
Bauer, Laurie
1983 English word-formation. (Cambridge: Cambridge University Press).
Bybee, Joan
1985 Morphology. (Amsterdam: Benjamins).
98 Sergio Scalise
Mutarello, R.
1987 Alcuni aspetti delta prefissazione in italiano. [Unpublished thesis, Universita
di Venezia.]
Scalise, Sergio
1983 Morfologia lessicale. (Padova: CLESP).
1984 Generative morphology. (Dordrecht: Foris).
1988 "Inflection and derivation", Linguistics 26: 561 — 581.
Scalise, Sergio —Irene Zannier
1982/83 "Restrizioni sulle regole di formazione di parola: la condizione di adiacenza
e la condizione atomo", Quaderni Patavini di Linguistica 3: 159 — 210.
Scalise, Sergio —F. Bevilacqua —A. Buoso —G. Piantini
in press "II suffisso -mente", in: Raccolta di saggi in memoria di A. Limentani
(Venezia).
Siegel, Dorothy
1977 "The adjacency condition and the theory of morphology", ΝELS 7:
189-197.
Williams, Edwin
1981 "On the notions 'lexically related' and 'head of a word'", Linguistic Inquiry
12: 145-274.
English compounds in Italian: the question of the head*
Irene Vogel
1. Introduction
In this paper, I will examine, and propose an account of, the fact that
when Italian borrows a particular type of compound from English,
typically only one of its members is retained (e.g., night club —• night). I
will first describe the relevant data and then consider several hypotheses
as to how to account for them. While the phenomenon in question
represents a relatively small area of Italian morphology, it will be shown
that it nevertheless involves a more general issue, the relationship between
compound structure and basic sentential word order.
2. English compounds in Italian
Italian began borrowing compounds from English regularly around the

middle of the last century. Zolli (1976) cites, for example, the items in
(1), where it will be noticed that the form used in Italian consists of only
one member of the English compound.
(la) waterproof coat —* (It.) waterproof (1868)
b) smoking jacket —> (It.) smoking (1888)
While (1 a) is no longer in use, (1 b) is very much alive. Other forms heard
commonly include the following:
(2a) night club —• night
b) scotch tape —• scotch
c) water closet —> water
d) plaid blanket —• plaid
* I would like to thank Wolfgang Dressier, the anonymous reviewers, and the participants
in the Krems meeting for their helpful comments.
100 Irene Vogel
A number of the items in (1) and (2) also appear in other European
languages (e. g., smoking in French, Dutch, German). In such cases, it is
not always clear whether each language borrowed the word from English
independently or whether one language borrowed it first and other
languages borrowed it from this one rather than from English. As Haugen
(1988: 8) points out, there are serious difficulties "in teasing out the
English models, their origins in speech or writing, their tortuous ways of
reaching the European public, and the current results in the form of local
replicas". While there may be some doubt as to the origin of a number
of the earlier borrowings in Italian which also appear in shortened form
in other languages, Italian differs from a number of other languages
which have borrowed many of the same items. Of the fifteen languages
and their treatment of English loanwords reported in the studies in
Filipovic (1982), none shows as consistent a pattern of shortening of
compounds as Italian. Instead, the full form is usually retained, though
sporadic instances of shortening are also mentioned (e.g., Albanian strip
'comic strip' (Mehmeti 1982); Dutch keeper 'goal keeper' (Gerritsen 1982);
Finnish klosetti 'water closet' (Oresnik 1982)). In Italian, on the other
hand, the reducing of English compounds is quite productive and con-
tinues to contribute new forms to Italian such as those in (3) (T. Cravens,
personal communication).
(3 a) Pink Floyd —• Pink

b) Rolling Stones —» Rolling
It should be noted, furthermore, that the shortening is not restrictecd to
"popular culture" items among speakers who may not have a good
command of English. In fact, (il) British is often heard as the shortened
form of British Council among those who have occasion to speak of this
organization (W. Dressier, personal communication).
Finally, still further support for the productivity of the reduction rule
comes from my own observations of the same process being applied by
an Italian child who spontaneously produced the shortened forms in (4),
having heard the full forms.
(4 a) self service —• self

b) pop corn —> pop
c) snack bar —*• snack
d) beauty case —» beauty
What is interesting about the way in which Italian borrows and shortens
the type of compounds in question is not only that it reduces the com-
English compounds in Italian 101
pound to a single word, but that the word that is retained is precisely
the one native speakers of English would not choose if they were to
shorten the same compounds. Compare the English examples and their
Italian equivalents in (5).
(5a) night club: Let's go to the club.
Andiamo al night.
b) smoking jacket: Give me the jacket, please.
Dammi lo smoking, per piacere.
c) scotch tape: I need the tape.
Ho bisogno dello scotch.
The question that must be addressed now is why Italian reduces the
compounds in the opposite way from the way in which they are reduced
in English. This is not to say, of course, that all English compounds
borrowed by Italian are shortened. Some are retained in their full form
(e.g., T-shirt; pickup 'record player'). Others are used in either their full
form or in the shortened form (e.g., blue jeansjjeans, basketball/basket).
Still others have undergone translation, either retaining the English word
order, particularly in earlier borrowings (e.g., banconota 'banknote', gen-
tiluomo 'gentleman' (originally from French). While these patterns are
rather sporadic in contemporary Italian, the one illustrated in (1) —(4) is
productive and it is this one that will be the focus of the rest of this
investigation.
3. Compounds in English and Italian
The structure of the compounds under investigation here is either

Noun + Noun (N + N) or Adjective 4- Noun (A + Ν). Ν + Ν and
A + Ν compounds are common in English and in both cases the entire
compound is always a noun, as illustrated in (6), in conformity with the
"Is a" Condition stated in (7).
(6a) N + N: [[water]» [lily] Ν ]n
b) A + N: [ [blue]^ [cheese] N ]N
(7) "Is a" Condition (cf. Allen 1978: 105)

[[X] [Y] ] z : Ζ is a Y
102 Irene Vogel
Thus, water lily "is a" type of lily and blue cheese "is a" type of cheese.
This follows from the fact that the head of complex words in English is,
with few exceptions, on the right (cf. Williams 1981). That is, the category
label and certain other features of the right member of a compound
percolate up to the node that dominates it and its sister, as shown below.
(8 a)
watery
b)
bluet, cheeseN
Italian, like English, has both Ν + Ν and A + Ν compounds, as

illustrated in (9).
(9a) Ν + N: croce via 'cross road'
capo stazione 'station master'
b) A + N: alto piano 'plateau' (lit.: 'high level')
giallo limone 'lemon yellow'
Unlike English compounds, however, Italian compounds do not always
have the head on the right, as shown below (cf. Scalise 1984).
(10a)
croceN via^
b) N.
altoi piano N
(11a)
capoN stazioneN
b)
giallof limoneN
The compounds in (10) have the same structure as English compounds;

those in (11) have the opposite structure. That is, while the features for
syntactic category, as well as number and gender, percolate from the right
member of the compound in (10), they percolate from the left member
in (11). As it turns out, however, only (11) represents a productive pattern,
a point we will return to below.
4. Three hypotheses about Italian compound borrowing
In analyzing the Italian data, we will consider the following hypotheses

regarding the shortening phenomenon:
HI: It is due to some basic linguistic strategy.
H2: It is due to some aspect of English compounds.
H3: It is due to some aspect of Italian compounds.
To maintain HI, we would have to argue that there is something
particularly salient about the first member of compounds, regardless of
the characteristics of either the source language or the borrowing lan-
guage. A basic strategy in this case might be something like "delete
everything but the first word". This, however, would run counter to
findings in other areas of linguistic behavior which indicate that it is the
rightmost, not the leftmost, part of a string that is most salient. For
example, it has been demonstrated in free-recall experiments that there
is a strong recency effect in which it is the last part of the material
presented that is the best retained (cf. Klatzky 1975). In addition, it has
been observed that the ends of words tend to be more salient to children
learning language than the beginnings. Slobin's (1973: 191) Operating
Principle A for language acquisition, "Pay attention to the ends of words",
expresses this strategy and accounts, for example, for the fact that children
tend to learn suffixes before preposed elements with similar linguistic
functions. A strategy that retains the first element of a compound solely
on the basis of its position would therefore seem to lack independent
motivation and, at best, be ad hoc.
According to the second hypothesis, it is the nature of the English
compounds themselves that is responsible for the Italian borrowing pat-
tern. Both syntactically and semantically the first element, whether it is
a noun or an adjective, in some way modifies the head of the compound.
A more detailed examination of the semantic and syntactic relations
between the two words, however, is not particularly revealing. Despite
various attempts to categorize the relations between the members of
104 Irene Vogel
compounds (e.g., Levi 1978; Warren 1984a, 1984b), according to Selkirk

(1982: 25), for the type of compounds under consideration "the range of
possible semantic relations between the head and nonhead is so broad
and ill defined as to defy any attempt to characterize all or even a
majority of the cases". More generally, it is unlikely that the syntax and/
or semantics of one language would influence the borrowing patterns of
another one since this would mean that the speakers of the borrowing
language know the relevant structures of the source language. In fact,
while we can assume that the first person or group to introduce an
English compound into Italian knows the meanings of its component
words and the relation between them, it would be unrealistic to make
the same assumption about the majority of Italian speakers who subse-
quently incorporate the item into their vocabulary.
The possibility that the phonology of the source language influences
the borrowing pattern, however, is not as easy to dismiss since speakers
of the borrowing language can be sensitive to certain aspects of the
sounds they hear without knowing anything about the structure of the
source language. Of relevance to the phenomenon under investigation is
the often made observation that the main stress of noun compounds in
English is typically on the leftmost member (e.g., Chomsky —Halle 1968).
We might thus hypothesize that Italian retains the first word of a com-
pound because this is the one bearing the main stress. Of the examples
given above in (1) —(4), however, it turns out that only half follow this
pattern, as can be seen below.
(12 a) smoking jacket
night club
wäter closet
pop corn
snäck bar
beäuty case
b) waterproof coat
scotch täpe
plaid blanket
Pink Floyd
Rolling Stones
self service
Thus, the phonology of the English compounds does not seem to be

crucial either. At this point, then, we must also reject our second hy-
pothesis, (H2), since there does not seem to be anything in the phono-
logical or in the syntactic and semantic structure of English compounds

that could account for the Italian data.
Finally, let us consider the third hypothesis. While it is true that Italian
has noun compounds with the head on the right like English, it was
shown above that there are also compounds with the head on the left.
In fact, it is the latter type that is more common and represents the
productive pattern in contemporary Italian. Of the almost 3,000 items
listed as compounds by Tollemache (1945), approximately 560 are of the
form NA, NN or AN. Approximately 68% of these have their head on
the left, while only 32% have their head on the right.1 New ("loose")
compounds when they are created in contemporary Italian typically have
the form shown below, where Ή ' = head and 'M' = modifier.
(13 a)
'ferry boat'
valigiaH armadioM
'wardrobe-valise'
In contrast to these "loose compounds", the ones in which the head is

on the right are generally older, lexicalized forms (e.g., gran cassa 'bass
drum' (lit.: large box), mezzanotte 'midnight') or forms that have been
translated from another language, maintaining the original word order
(e.g., croce via 'cross roads') (cf. Tollemache 1945). From these obser-
vations, we see that what underlies the retention of the leftmost element
of English compounds is the structure of Italian itself. That is, in its
borrowing pattern, Italian is, in fact, respecting its own contemporary
compound structure where the head is on the left. Hypothesis 3 is thus
confirmed.
If we now look beyond the structure of compounds to the basic word
order of Italian, we find further support for H3. That is, the compound
patterns are actually part of a more general phenomenon. From a ty-
pological point of view, it has been observed (cf. Greenberg 1963) that
the unmarked pattern in SVO languages is to have modifiers follow the
head. For example, in noun phrases, adjectives typically follow the noun,
as do genitive constructions, as illustrated in (14) and (15), respectively.
(14) Ν + A: un uomo alto 'a tall man' (lit.: 'a man tall')
(15) Ν + Gen: il libro di Clelia 'the book of Clelia'
106 Irene Vogel
Since Italian is an SVO language, compounds with the head on the left
are precisely what we would expect. The other cases are mostly transla-
tions from other languages with different word order patterns or from
earlier stages in the development of Italian when the basic word order
was not strictly SVO.2
Thus, we can conclude that in retaining the leftmost member of English
compounds, Italian is behaving in a way consistent with other aspects of
the language. That is, it is retaining the element that is in the crucial
position according to its own structure. The fact that the crucial element
(i.e., head) is on the right in English is irrelevant for Italian, as are other
semantic and phonological characteristics.
5. Further considerations
If the above analysis is correct, this should allow us to predict how a

language will borrow compounds in those cases in which only a part is
consistently retained through the operation of a rule. What is needed is
information about (a) the internal structure of compounds in the bor-
rowing language, and (b) the basic word order in that language. Specif-
ically, what is predicted is that the position of the head in the borrowing
language will determined that the member of the borrowed compound
that occupies this position will be the one that is retained. Data from
other languages, and in particular ones with different word orders than
Italian are needed, of course, to evaluate the validity of the analysis
beyond Italian.
As we have already seen, however, there may also be data which at
first glance tend to obscure the patterns of compound formation and
borrowing. We must, therefore, take several additional factors into con-
sideration, the most important of which is productivity. The proposal
advanced here only deals with cases where there is a distinct rule in
operation; it does not apply to sporadic instances of the reduction of
borrowed compounds. In the case of Italian, it was seen that there are
compounds with the head on the right as well as the left, so we might
expect these to have an effect on the borrowing pattern too. On further
examination, however, it was found that those with the head on the right
typically had sources other than an Italian word formation rule (e.g.,
croce via 'cross roads', gentiluomo 'gentleman' translated from the anal-
ogous English or French compounds).
At first glance several other forms also seem to complicate the situation
in Italian:
(16 a) blue jeans —• jeans
b) plum cake —• cake
While these items appear to contradict our generalization that it is the
leftmost element of the English compounds that is retained, further
investigation shows that they do not, in fact, constitute counterexamples
to the above analysis. Jeans was borrowed from English already in its
shortened form, and we can hypothesize the same for (16b), though I
have never encountered an Italian who actually uses the short form cake
(cited in Zolli 1976) rather than plum cake. Thus, it is often essential to
know something about the way in which specific words enter a language.
Other sources of borrowings may also provide relevant data. For
example, in Italian, we find additional compounds of German origin
which are shortened in the same way English compounds are:
(17 a) Volkswagen —• ν ο Iks
b) Blitzkrieg —• blitz
While blitz most likely entered Italian through English, already in its
shortened form, volks appears to be a form Italian developed on its own.
We thus have some confirmation of our prediction since the shortening
phenomenon also appears to work the same way on compounds from at
least one other language, though in general there are far fewer borrowings
from other languages in Italian.
Finally, the extent to which borrowed forms are integrated into the
borrowing language must be taken into consideration. In the case of
Italian, let us consider the small set of recently formed items illustrated
in (18).
(18 a) scuola bus 'school bus'
b) fiera district 'fair district'
In these items, the head is on the right, contrary to the claim made above
that the productive compound formation rule in Italian places the head
on the left. These forms, however, are not felt to be Italian, and their
low degree of integration appears to be determined not only by the fact
that they contain elements that are, indeed, foreign, but also precisely
because of the order of the words (i. e., the head is on the right). Related
to this, we find some speakers who refer to gli Stones and i Floyd (rather
than using the more common i Rolling and i Pink), and we also occa-
108 Irene Vogel
sionally encounter the shortened form club, instead of the more regular
form night. In the case of club, the "non-Italianness" is also usually
marked phonetically. That is, we often hear [kleb], which seems to indicate
that the speaker knows the English"u" is not pronounced [u], as it would
be in Italian. That [ε] is not the correct English vowel either is not relevant
here. What all these cases have in common is the fact that they reflect
an attempt on the part of speakers to exhibit their knowledge of English
by using forms that do not conform to the patterns of Italian. As long
as such items are felt to be "exotic" and external to the system of Italian,
they do not constitute counterexamples to the proposal advanced here.
That is, they are not Italian forms and therefore do not need to conform
to the linguistic patterns of Italian. To the contrary, they are intentionally
used to signal knowledge of a different language.
6. Conclusions
In this paper, I have investigated the way in which English compounds

are borrowed into Italian. Three hypotheses were considered in account-
ing for the observed pattern whereby compounds, when borrowed, are
reduced to only their first element. It was argued that neither a general
strategy such as "keep the leftmost element" nor a strategy based on the
syntactic, semantic, or phonological properties of the English compounds
themselves could account for the data. Instead, it was proposed that the
pattern observed in Italian borrowing depends on the structure of native
Italian compounds and this, in turn, is related to more general charac-
teristics of word order in Italian sentences. On the basis of the analysis
of Italian, it was suggested, furthermore, that it might be possible to
predict borrowing patterns involving shortening of compounds in other
languages as well. Finally, several additional factors and apparent coun-
terexamples to the proposed analysis were examined and the importance
of such issues as productivity and degree of integration of borrowings
were considered.
Notes
1. This pattern may, in fact, be fairly common in other Romance languages as well. See
Bauer (1978: 41), where it is reported that French, too, has a strong preference for (at
least NN) compounds to have their head on the left, in contrast with Danish and English.
2. This is, in fact, not an isolated phenomenon. Changes in compound structure related to
changes in basic word order have also been noted, for example, in Irish (cf. Ahlqvist
1985).
References
Ahlqvist, Anders
1985 "The ordering of nominal compounds in Irish", in: J. Fisiak (ed.) Historical
semantics — historical word formation (Berlin: de Gruyter), 1—9.
Allen, Margaret
1978 Morphological investigations. [Ph. D. dissertation. University of Connecti-
cut],
Bauer, Laurie
1978 The grammar of nominal compounding. (Odense: Odense University Press).
Chomsky, Noam —Morris Halle
1968 The sound pattern of English. (New York: Harper and Row).
Filipovic, Rudolf (ed.)
1982 The English element in European languages. (Zagreb: Zagreb University
Press).
Gerritsen, Johan
1982 "English influence on Dutch", in: R. Filipovic (ed.), 1 5 4 - 1 7 9 .
Greenberg, Joseph H.
1963 "Some universals of grammar with particular reference to the order of
meaningful elements", in: J. H. Greenberg (ed.) Universals of language,
(Cambridge, Mass.: M I T Press), 6 1 - 1 1 3 .
Haugen, Einar
1988 "The influence of English: a transatlantic perspective", Folia Linguistica 22:
3-9.
Klatzky, Roberta L.
1975/1980 Human memory. Structures and processes. (San Francisco: Freeman).
Levi, Judith N.
1978 The syntax and semantics of complex nominals. (New York: Academic Press).
Mehmeti, Ismail
1982 "A morphological and semantic analysis of the adaptation of anglicisms in
Albanian", in: R. Filipovic (ed.) 28 — 56.
Oresnik, Brigitta
1982 "On the adaptation of English loanwords into Finnish", in: R. Filipovic
(ed.), 1 8 0 - 2 1 2 .
Scalise, Sergio
Selkirk, Elizabeth O.
1982 The syntax of words. (Cambridge, Mass.: M I T Press).
Slobin, Dan I.
1973 "Cognitive prerequisites for the acquisition of grammar", in: Ch. A. Fer-
guson—D. I. Slobin (eds.) Studies of child language development. (New
York: Holt, Rinehart & Winston), 1 7 5 - 2 0 8 .
110 Irene Vogel
Tollemache, Federico
1945 Le parole composte nella lingua italiana. (Roma: Edizioni Rores di Nicola
Ruffolo).
Warren, Beatrice
1984a "Covert connectors", Quaderni di Semantica 5.2: 331—349.
1984b "The functions of modifiers of nouns", Quaderni di Semantica 5.1: 111 —123.
Williams, Edwin
1981 "On the notions 'lexically related' and 'head of a word'", Linguistic Inquiry
12: 245-274.
Zolli, Paolo
1976 Le parole straniere. (Bologna: Zanichelli).
The importance of combining forms*
Beatrice Warren
1. Introduction
Consider the morphemic and semantic analyses of the noun alcoholic. A

morphemic analysis yields the two morphemes alcohol and ic; the semantic
analysis could be said to yield three elements: "person"-"addicted to"-
"alcohol". We can match the semantic element "alcohol" with the mor-
pheme-form alcohol·, -ic, I would argue, is a morpheme-form devoid of
lexical meaning; it does, however, serve to indicate a change from non-
personal to personal reference in this case, and so indirectly it can be
coupled with the element "person". The element "addicted to" is not
formally expressed. It is a semantic element which, I suggest, was origi-
nally constructed and not retrieved from the mental lexicon. I have
elsewhere (Warren 1988) outlined how the construction of unexpressed
relational meanings within words and phrases can be accomplished. I will
therefore not dwell on this aspect of the semantic analysis here. Let me
only point out that meaning construction of this kind involves linguistic
and extra-linguistic knowledge as well as the application of pragmatic
rules of interpretation and also that the outcome of this meaning con-
struction is only predictable provided we have the particular extra-
linguistic knowledge required. It need not have been "addicted to", but
could have been, say, "selling", "producing", "appointed to prevent the
imbibing of", etc. For some reason, however, "addicted to" is the relation
which established itself and which is now a part of the meaning of the
noun alcoholic and as such retrievable from the mental lexicon, although
it would be difficult to assign it any special morpheme-form.
* As the reader will detect, the ambiguity of the title is intended: either interpretation being
apposite. I would here also like to thank most heartily Gunilla Malmborn and Gabriele
Stein — the former for pointing out inadvertencies in the original manuscript and above
all for helping me excerpt examples; the latter for generously sending me some relevant
articles.
112 Beatrice Warren
Consider next the morphemic and semantic analyses of workaholic.

According to word-formationists, its morphemic analysis yields: work
plus a, which represents a linking phoneme, plus holic\ the semantic
analysis yields "person"-"addicted to"-"work". The semantic element
"work" can be matched with the morpheme-form work, which is a
straightforward content morpheme; the two semantic elements "person"
and "addicted to" can be linked with the rest of the phonemes of the
word, i.e., (a)holic, thus making this sequence of phonemes into another
straightforward content morpheme. Coming across the word workaholic
for the first time, we have to associate with -(a)holic the semantic elements
"person"-"addicted to". To do so, we crucially depend on knowing the
meaning of alcoholic.
The above claim, i.e., that -aholic is a straightforward morpheme is as
a matter of fact infelicitous, since -aholic is a combining form and
combining forms are morphemes of a rather special kind. Being neither
proper roots, nor proper affixes, they upset the morphologists' neat
subdivision of morphemes into roots and affixes. How then do word-
formationists deal with combining forms? As we will see in the next
section, standard works on word formation yield somewhat heterogene-
ous descriptions and definitions.
2. Survey of definitions of combining forms
Adams (1973) discusses combining forms in a chapter entitled "Neo-

classical compounds". She points out that the OED reserves the term
"combining form" for first elements in neo-classical compounds and calls
second elements endings. Thus, micro- would be an example of a com-
bining form, whereas -scope would exemplify an ending. However, she
herself does not make this terminological distinction. (See Adams 1973:
129, where she refers to -graph, -cide, -meter, and -scope as combining
forms.) Also, in the supplements of the 1970s, the editors of the OED
seem to have abandoned this distinction themselves.
Quirk et al. (1985) claim that combining forms are almost obligatorily
initial, giving psycho- in psychotherapy as a model example of a combining
form. Semantically, combining forms are said to be like the first constit-
uents of compounds, but in other respects they are like prefixes. That is
to say, they tend to be bound morphemes and they tend to be unstressed.
The importance of combining forms 113
Typically they are neo-classical. A vowel (usually -o- but often -/-) is
inserted as a link between the combining form and the base. They point
out that there are certain bases which are particularly common, viz.
-meter, -graphy, -gram, and -logy.
Bauer (1983) does not hesitate to apply the term to first as well as last
elements, distinguishing simply between initial and final combining forms.
Unlike prefixes, initial combining forms almost invariably end in a vowel.
If they do not, the linking vowel -o- may be added. The crucial difference
between initial combining forms and prefixes, according to Bauer, is that
the former may combine with final combining forms whereas prefixes
may not. Consequently, hyper- in hyperactive could be considered a prefix,
but must be considered a combining form in hypertrophy. Similarly, the
crucial difference between suffixes and final combining forms is that —
unlike suffixes — final combining forms may co-occur with initial ones.
In fact, Bauer (1983: 272) suggests that they invariably do.
Marchand (1969) claims that the term "combining form" is usually
employed for prepositive elements which derive from Latin or Greek
stems of full words, whereas prepositive elements which go back to Latin
or Greek prefixes, prepositions, or particles are termed prefixes. He does
not think that this distinction is justified, however (Marchand 1969:
3.1.5). Nevertheless, he himself distinguishes between prefixes (bound
morphemes that are prefixed to full English words) and prepositive
elements which occur in compounding on a Neo-Latin basis, such as
astro-, electro-, galato-, hepato-, osteo-, which he thinks are of "a purely
dictionary interest". As far as final elements are concerned, Marchand
seems to make the following distinctions: suffixes, semi-suffixes, and
terminal elements. As examples of semi-suffixes he gives -like, -worthy,
-monger, -way/ways, -wise, -wort, -wright. These are like full words in that
their word character is still recognizable. They are like suffixes in that
some of them are used only as second words of compounds (Marchand
1969: 4.80). Suffixes Marchand defines as elements which are not inde-
pendent words and which are tacked on to full English words or to
allomorphs of English words. Last elements which are not combined with
full words or allomorphs of full words, such as, for instance, -scope in
galvanoscope, he considers not to be proper suffixes; they are simply
terminal elements and excluded from consideration in his book. Never-
theless, Marchand (1969: 4.1.8.1—4.1.8.2) does discuss the creation of
last elements such as -athon, -burger, -furter, -rama and calls them suffixes,
which seems inconsistent since they often do not have English words or
allomorphs of English words as bases.
114 Beatrice Warren
Hansen et al. (1985) make a simple distinction between proper suffixes

on the one hand and semi-suffixes alias suffixoids on the other. Conse-
quently, no distinction is made between, for example, -teria and -monger
— both are semi-suffixes alias suffixoids. In parallel fashion, they distin-
guish between proper prefixes on the one hand and semi-prefixes or
prefixoids, alias combining forms, on the other, keeping to the tradition
that the term "combining form" should be restricted to the first element.
As a conclusion of this survey, let me quote the entry under "combining
form" in the Second Barnhart Dictionary of New English. Since this is
already in condensed form, I quote it in full:
"Combining Forms. This term is usually restricted to forms that
occur in COMPOUNDS 1 and DERIVATIVES, such as semi- and
-naut; sometimes, of course, the form coincides with a free-standing
word, such as graph as contrasted with telegraph·, and even a free-
standing word may be used as a combining form, such as -PERSON
in place of -man.
Many of our combining forms are borrowings from Latin or Greek
and are so well established in English that they are freely used to form
new words. Even the simplest list of words beginning with AGRI-,
bio-, INDUSTRIO-, micro-, astro-, PETRO-, SEXO-, illustrates the
range and productivity of such forms. Combining forms are particu-
larly frequent and important in the creation of new TECHNICAL
TERMS.
Like other linguistic forms, these word components are subject to
change and extension of meaning. For instance, the development of
astronautics in the 1960's expanded the meanings of the old combining
forms of astro-, and cosmo- to include outer space and space travel.
More recently, the energy crisis has caused the form PETRO- to take
on an extended meaning related to the oil industry, while -ATHON
was applied to any prolonged activity resembling a marathon. Similar
changes are noted under the entries SYN-, FLEXI-, -ORIENTED,
and -WATCHER. Other examples of the use of combining forms
taken out of older words, such as -GATE in WATERGATE, are cited
in the separate notes on ABSTRACTED FORMS and NONCE
WORDS."
3. The corpus of combining forms
It is obvious from this survey that there are certain word components
which linguists intuitively feel are neither affixes nor roots. It is also
obvious that there is as yet no uniform terminology for these elements,
nor any generally valid description of them. 2 For example, some linguists
insist on their classical origin; others include in the category native
morphemes too; some emphasize the ability of these word components
to occur in words without proper roots; others ignore this interesting
aspect of their character. The present paper is an attempt to improve on
the descriptive as well as explanatory adequacy of the account of these
affixlike morphemes. To do so, I have excerpted from Mort (1987) words
containing elements which on intuitive grounds I considered good can-
didates for being formatives of this kind. The excerption yielded 81
different types of likely combining forms. These were of course first
divided into first and last elements. However, both in the case of initial
and final combining forms, my examples naturally divided themselves
into two further groups: one consisting of elements which are allomorphic
variants of some other word, e.g., astro-, which is a variant of Lat.
astrum, and -drome, which is a variant of Gk. dromos; and one consisting
of elements which represent parts of other words, e.g., cyber-, which we
recognize as a part of the word cybernetics and -aholic, which is not a
variant of alcoholic, but a part of it. Among the final combining forms,
there was also a third group of forms, viz., elements which from a purely
formal point of view are not new morphemes, but which have novel
meanings. Gate, for example, is no new morpheme-form, but as a com-
bining form it has a novel meaning, i.e., "political scandal involving a
cover-up". Altogether, there are then five subgroups as illustrated in
Table 1. For additional examples, see lists in the Appendix.
Glancing through the Appendix, we will notice, among other things,
that my examples support the claim that initial combining forms end in
a vowel, more precisely /i/, /o/, or /a/. (Of these, /o/ and probably /i/ —
although my examples happen not to confirm this — can often be looked
as linking phonemes.) In other words, initial combining forms differ from
final ones in that the former have a characteristic phonetic shape, whereas
the latter are amorphous in this respect.
We will also notice differences between the groups, in particular be-
tween Group I, on the one hand, and Groups II and III, on the other.
With a few exceptions, notably debtno- and smello-, both the first and
116 Beatrice Warren
Table 1. The five subgroups of combining forms.
Initial Final
combining forms
Group I: astro- in astrodome -drome in alpinodrome

Allomorphs of model "stadium with "place for climbing
words (in this case (Lat. translucent domed contests"
astrum and Gr. dromos) roof'
Group II: Cyber- in -aholic in spendaholic
Truncated forms of model cyberphobia "fear of "compulsive
words (in this case computers" spendthrift"
cybernetics and alcoholic)
Group III: -gate in Yuppiegate
Parts of model words, "scandal involving five
which happen to be Yuppies"
established morpheme-
forms
last elements in Group I are of classical origin and are established

combining forms. The majority of the examples in Group I can in fact
be looked upon as classical loans which have been phonetically adapted
so as to form smooth components of English words. Among the examples
of the two Groups II and Group III, however, we find non-classical
elements and also novel morphemes, i.e., novel either from a purely
semantic point of view, as exemplified by -gate, or from a formal as well
as from a semantic point of view, as exemplified by -zak in newzak,
"repeated news coverage of the same event" modelled on muzak. It follows
that the formation of a certain kind of combining forms is a source of
truly novel morphemes, which is something morphologists seem to have
overlooked. My main interest in combining forms is the way in which
these novel morphemes can be created. My suggestion is that they have
been formed by a process which Jespersen (1950: 384 ff) called secretion. 3
What this process involves will be the topic of the next section.
4. Secretion
Secretion could be said to involve a rearrangement of semantic elements:

from having been associated with no particular part of a word-form,
certain elements become connected with a certain section of a word. I
tried to demonstrate the nature of this process by discussing the combining

form -(a)holic in the introductory section, -aholic, I claimed, is a content
morpheme meaning "person addicted to", which is a meaning we derive
from alcoholic, although we do not associate "person addicted to" with
the phoneme sequence -holic in alcoholic.
Linguists look upon secretion either as the result of some abbreviation
(i.e., applied to my example, workaholic would be the result of a con-
traction of work alcoholic) or as the result of some folk-etymological
misdivision of morphemes (i.e., alcoholic would mistakenly be believed
to consist of alco + holic).
The suggestion that secreted combining forms are the results of con-
tractions of two syntagmatically related words is clearly implausible in
some cases. To think of prequel as a contraction of *presequel or newzak
as a contraction of *news muzak is simply unreasonable, for instance. In
other cases, "the contraction theory" appears more plausible, at least at
first sight. Wamography "literature or films glorifying war and violence",
for example, could well be a contraction of war pornography. On second
thought, however, we will be reminded that -nography in wamography
does not mean "films, literature designed to cause undesirable excitement
about sex" and we will discover that wamography and war pornography
are in fact not semantically equivalent, the literal meaning of war por-
nography being "pornography issued during wartime". We must conclude
that in forming wamography, we have not only meddled with the form
of the word pornography, we have also meddled with its meaning. There-
fore, to consider -nography as simply an abbreviation of pornography is
an oversimplification which is not innocuous.
The suggestion that secreted combining forms may be the outcome of
uninformed morphemic analyses is an explanation that I accept in some
cases, but again not in all. For instance, it is easy to accept the suggestion
by Marchand (1969: 4.1.7) that lemonade was mistakenly analysed as
lemon meaning "lemon" and -ade meaning "drink", although -ade in fact
was no English morpheme at all. I find it considerably less plausible that
people could have mistaken alco- for a morpheme meaning "alcohol"
and holic for a morpheme meaning "person" or "person addicted to" or
that pornography should be taken to consist of the morphemes por-
meaning "sex" and -nography meaning "film, literature".
Instead I suggest that the coiner of, say, wamography does not mistake
-nography for a morpheme, he makes it into one. That is to say, in
secreting new morphemes we save so much of a word as we feel is
necessary for it (i) to provoke the right associations and (ii) to have the
118 Beatrice Warren
phonological characteristics of a morpheme suitable to be a word com-

ponent. We simply discard the rest. Note, however, that the discarded
part requires replacement by some established content morpheme which
adds some specifying feature of meaning. Warnography, for example,
could then be said to have been formed by secreting -nography from
pornography provoking the following association "films, literature, etc.,
designed to cause undesirable excitement about ...". By adding the word
war, we get the gloss "films, literature, etc., designed to cause undesirable
excitement about warfare".
As I have already pointed out, the elements in Group III (i. e., -boom,
-fare, -gate, etc., see p. 130 in the Appendix and Table 2) are alike in that
from a formal point of view, they are not new morphemes. However, I
claimed, as combining forms they express novel meanings. This is a claim
that needs to be qualified. It is true of -fare "social benefit", -gate
"scandal", -wagon "trend", -wave "programme", but it is not true of
-boom "sudden great increase". The correct claim is instead that as a
combining form -boom is restricted to a particular sense, viz., the sense
it has in the combination baby boom. Similarly, -speak and -meter, for
example, have the senses they have in newspeak and hydrometer (see
Table 2). In other words, the meanings of these combining forms are not
novel but particularized. However, this particularization is derived from,
and depends on, a model word. Also, it is my contention that from a
semantic point of view, the morphemes in this group are all bound.
Table 2. The meanings of the combining forms in Group III.
"Suffix" Meaning-particularization
-boom "sudden great increase (in) "

-fare "social benefits (involving) "
-gate "the scandal involving some cover-up (which occurred at or had
to do with) "
-mare "past or imagined future disaster (involving) "
-meter "instrument (for) measuring "
-smith "a person (who achieves) "
-speak "jargon (occurring in/having to do with) "
-wagon "sudden popular trend (consisting in) "
-ware "computer programme (stored in) "
(The dotted line signifies that some complementing semantic feature is obligatory.
Brackets surround relational meanings, which, as is shown in Section 6, are in
principle variable.)
5. Secretion and abbreviation
In Section 3, I mentioned that it is in the two Groups II and in Group

III that we find examples of secreted combining forms. It should be
pointed out, however, that I do not think of all the combining forms in
these groups as secreted. More precisely, I suggest that all the combining
forms in Group II of last elements are secreted, whereas — with one
possible exception — the combining forms of Group II of first elements
are not secreted but simply abbreviated. (The possible exception is cy-
ber-, which expresses the meaning "machine/computer" and may do so
through transferance of a semantic element originally associated with no
particular section of cybernetics.) Among the elements in Group III, there
are two combining forms that I consider to be abbreviated, viz., -wagon
and -mare. These will be discussed presently.
What then is the difference between secretion and abbreviation? The
combining form -burger in cheeseburger and fishburger nicely illustrates
the difference. In cheeseburger, -burger is an abbreviated combining form;
in fishburger, it is a secreted form. Cheeseburger means "hamburger with
cheese"; fishburger does not mean "hamburger with fish", but "fried patty
made of fish served in a bun". Secretion is a process in which certain
semantic elements in a linguistic unit are kept and others discarded.
Abbreviation is a process in which all the semantic elements are kept,
although the form of the unit is made shorter. Secretion is like meta-
phorization in that the interpreter is asked to disregard certain semantic
elements of a word or a phrase. It differs from metaphorization in that
the form of the word or the phrase is not kept intact. This can be
summarized (and, admittedly, oversimplified) as below:
FORM CONTENT RESULT

Secretion shortened shortened incomplete unit
Abbreviation shortened intact complete unit
Metaphorization intact shortened 4- complete unit
augmented
Note that linguistic units which have undergone an abbreviation or a
metaphorization process remain complete units; linguistic units which
have undergone a secretion process become, I maintain, semantically
incomplete units. 4 The reason for this is that the discarded semantic
elements require overt replacement. Compare in this respect the meta-
120 Beatrice Warren
phorization process. When we use a word in a metaphorical sense, we

expect the interpreter to work out — without overt manifestation — not
only which features are inapplicable, but also which new features of
meaning we wish to express or highlight.
Now consider -wagon and -mare. Branwagon is glossed by Mort as
"the increasing trend towards eating foods considered healthy (e.g. bran
and fresh vegetables)". Wagon in branwagon obviously derives from
bandwagon in the phrase jump on the bandwagon "to join the most popular
movement without due consideration". It is included among the examples
in Group III, since it has a particularized sense, which can only be worked
out provided one is familiar with the model word and its occurrence in
the idiom cited above. However, since the removal of band from band-
wagon does not involve any semantic modification, it must be seen merely
as an abbreviated, not as a secreted form. This is confirmed by the fact
that bran bandwagon is semantically equivalent to and semantically as
felicitous as branwagon. For similar reasons, -mare in nukemare must be
considered an abbreviated rather than a secreted form.
6. Interpreting combining-form composites
I have suggested above that secretion can be a fairly straightforward

exploitation of shared lexical knowledge. That is to say, finding that we
need a word for the notion "person addicted to work", which is like the
notion "person addicted to alcohol", except for the semantic element
"alcohol", we make use of the word for this latter notion by keeping so
much of it that the notion "person addicted to" will be provoked and
indicate the semantic modification by replacing the discarded element
with the morpheme-form for "work".
Constructing the meanings of words containing secreted combining
forms may, however, require more mental processing than the above
suggests. First of all, relational meanings are not necessarily carried over
as in workaholic but may have to be constructed afresh. Consider, for
example, Radfemspeak "jargon of radical feminists", computerspeak "jar-
gon for computers", and cataloguespeak "jargon occurring in catalogues".
Moreover, just as in the case of so many compounds and derivatives, we
will also find that the meaning of the whole often depends on some
particular extra-linguistic knowledge assumed to be shared and assumed
to be prompted by the morphemes in question. Spendaholic, for example,

suggests more than simply "person addicted to spending". Mort, realizing,
as it were, the close connection between reference and meaning, feels
obliged to add to his definition of this word the following explanatory
comment: "The behaviour of a spendaholic — is an important social
phenomenon. It often results from lack of security. It sometimes afflicts
young people and often those who have been deserted by their spouse.
It is not restricted to those two typical groups, of course."
In order to understand words with secreted combining forms for the
first time, we need to (i) recognize and be familiar with the model word
— its form, meaning and referents — and (ii) be able to adjust its
meaning, which then requires — as I have frequently pointed out —
discarding some features of meaning and adding others. We are guided
in this by the added content morpheme and usually by our knowledge
of the referents of the model word and the new composite. That is to
say, of all the facts that we know about the referent(s) of the model
word, we choose some as features of meaning. Which ones depends on
the nature of the referent of the new composite. In this way, we can
explain how the concrete noun gate comes to mean "scandal". We can
also explain why, in specifying the meaning of a combining form as part
of langue, we often find a set of possible features rather than a stable set
of necessary features. Bauer (1983: 272), for example, specifies the mean-
ing of -naut as (adventurer, explorer) (travelling) (especially in space) (in
an enclosed space), adding — not quite correctly — that any combination
of the parenthesized elements is possible. The implication is that when
-naut is part of parole — that is, when it occurs in combination with
another morpheme in an actual context — then one particular combi-
nation of features will suggest itself. I accept this with some modifications.
Firstly, obviously not all contexts are equally effective in singling out one
and only one combination. Also, certain types of features invite greater
uncertainty as to their appropriateness than others, viz., typically features
which are not empirically confirmable. For example, in selecting features
of meaning for -zak in newzak, we will, I believe, be more certain that
"something recorded" and "transmitted with great frequency" are in-
tended features of meaning than the feature "regrettably". Semanticists
tend to give these empirically unconfirmable features of meaning special
status by referring to them as connotations. It is my impression that
words containing combining forms tend to be particularly rich in con-
notations. Consider, for example, -speak, which I believe owes its popu-
larity to its negative connotation. Finally, fuzziness of meaning may also
122 Beatrice Warren
be caused by the fact that often there is more than one model word
possible. In principle, as soon as a composite containing a combining
form has been established, it may in turn serve as a model word. As an
example of this, let us consider -naut again. Originally, it was secreted in
the sense "explorer of unknown territories" from argonaut, referring to
a sailor in the legendary Argo (see Adams 1973: 185), to form aeronaut,
which refers to people travelling in balloons. Aeronaut in turn served as
a model word for astronaut, cosmonaut among others, which explains the
existence of the features "especially in space" and "in an enclosed space"
in Bauer's definition.
7. Root, affix, or neither?
The question which will be pursued in this section is: are combining
forms a special kind of root or affix or are they in fact neither?
Let us first consider similarities and differences between roots and
combining forms. Combining forms and roots are both morphemes with
lexical content. Roots may be free or bound. Combining forms, I have
argued, are invariably bound. It follows that a free morpheme must be
a root, since — again by definition — it cannot be an affix either. The
difficulty is therefore to distinguish between bound roots and combining
forms. Granted that we know which elements are suffixes, this is possible.
It follows from the following: a bound root must combine with a suffix
to make a word; a single combining form combined with a suffix does
not make a word. Consequently, a bound morpheme in word-final po-
sition is either a suffix or a combining form, but not a root. A single
bound morpheme followed by a suffix is a root. If it is followed by
anything else, it is either a prefix or a combining form.
In this connection, let us deviate from the topic of the section and
compare ordinary compounds and composites containing combining
forms. An initial combining form may co-occur with a final one (e.g.,
neurology) or with a free morpheme (neurophilosophy). A final combining
form may consequently be found together with an initial combining form,
but also with a free morpheme (feature-itis)5 and very exceptionally with
a prefix [prequel). Whatever type of combination, the first element is the
modifier and the last the head. 6 This is also true of regular compounds.
It may appear then that the only difference between ordinary compounds
and combining-form compounds is that in the former the components

are free morphemes, whereas in the latter, first or last or both elements
are bound. There is, however, another important difference. Compound-
ing is normally a hyponymy-creating device, i.e., steamship is a hyponym
of ship, tablespoon a hyponym of spoon, etc. Composites containing a
secreted final combining form are co-hyponyms rather than hyponyms
of their heads. Warnography is not a hyponym but a co-hyponym of
pornography. Similarly, spendaholic is not a type of alcoholic but a type
of addict. Here we witness one of the consequences of secretion.
Having established that roots and combining forms are different, let
us return to the subject of the section and consider whether it is possible
to distinguish between affixes and combining forms.
Distinguishing between suffixes and combining forms is normally not
problematic. Suffixes may have derivational or inflectional functions or
serve to modify the content of the head {kitchenette = "small kitchen").
Final combining forms do not serve any of these functions. They are
lexical elements, invariably functioning as heads. Distinguishing between
prefixes and initial combining forms, however, may sometimes be prob-
lematic, which is evidenced by the fact that not infrequently we find
inconsistencies among word-formationists and lexicographers as to which
individual morphemes should be considered affixes and which combining
forms. For example, Marchand (1969: 3.36) classes neo- as a prefix,
whereas Quirk et al. (1985: 1.29) identify it as a combining form or,
conversely, Marchand (1969: 3.50.1) classes pseudo- as a combining form,
whereas Quirk et al. (1985: 1.23) call it a prefix.
Prefixes, like initial combining forms, tend to be modifiers. Both tend
to have lexical content. These are important similarities. There are,
however, some differences: prefixes do not have to end in /i/, /o/, or /a/,
although they may and quite often do. More importantly, prefixes, like
all affixes, have, or have had, productive force. Initial combining forms,
like all combining forms, need not have productive force. Indeed, they
may be nonce formations. Expressed differently, prefixes are modifiers
of a well-established kind, belonging to a more or less closed set. Com-
bining forms are open-set items. They are lexical in nature and may have
rather specific meanings. Finally, initial combining forms, like all com-
bining forms and unlike affixes, can be connected with a "model word".
These last two differences between affixes and combining forms inspire
the following hypothesis:
As a rule, affixes were originally independent words — hence no
allomorphic status — with some auxiliary function in a syntagm. This
124 Beatrice Warren
auxiliary function, if it turned out to be repeatedly useful, triggered off

a grammaticalization process, involving phonetic as well as semantic
erosion to different degrees — hence the once free morphemes become
bound morphemes with productive force and often with little lexical
content. The auxiliary functions that these elements served would be
derivational, inflectional, or to modify the lexical content of some head.
In the former two cases, semantic erosion would follow naturally; in the
latter case complete semantic erosion would be prevented by the fact that
an element which serves to modify the meaning of some other element
must carry some meaning; it cannot be purely functional.
Combining forms, on the other hand, as a rule, started off as incom-
plete variants or bits of words — hence their allomorphic status. These
elements had to be combined with some other unit to make an inde-
pendent unit. In other words, combining forms were originally bound,
lexical elements, better equipped to resist grammaticalization. However,
they would not be immune to this process. Hence we find combining
forms, particularly among initial combining forms, which have developed
features characteristic of affixes. Pseudo- and neo- may serve as examples.
These are productive, bound, initial modifiers with prefix-type meanings.
Therefore, from a synchronic point of view, it will seem unjustified to
class, say, semi- and pre- as prefixes, but pseudo- and neo- as combining
forms.
My position then is that combining forms are different from affixes
in that they are open-set dictionary elements, whereas affixes are closed-
set elements that have undergone a grammaticalization process. Since,
however, combining forms are not precluded from undergoing this proc-
ess, we may come across morphemes, which from a diachronic point of
view are combining forms, but from a synchronic point of view are
affixes. How such morphemes should be classed is necessarily arbitrary.
Summing up this section, combining forms are neither roots, nor
affixes. Nevertheless, in individual cases, it may not be possible to decide
unequivocally whether a particular morpheme should be considered an
affix or a combining form.
8. Summary
The present study reveals that there are two main ways in which so-
called combining forms may be formed: by phonetic modification of some
existing morpheme or by secretion. It is possible to distinguish between
two types of phonetic modification: minor ones such as we witness in

alpino- (from alpinus), circo- (from circum), smello- (from smelt) and
-scope (from scopium) and proper clippings as exemplified by eco- (from
ecology) and -wagon (from bandwagon). Secretion is not simply an ab-
breviation process but is seen as a process which makes it possible to
create new morphemes (see examples among the last elements in Group
II in the Appendix), or new meanings for established morpheme-forms
or to transfer certain meaning-particularizations (see examples in Group
III: Table 2). Note that secreted forms tend to be final elements. That is
to say, it appears that secretion serves above all as a co-hyponym-creating
word-formation device.
The examples also show that it is not justified to apply the term
"combining form" to morphemes of classical origin only, although among
non-secreted elements, classical morphemes predominate. Finally, to re-
strict the term "combining form" only to first elements seems unwar-
ranted, although, admittedly, there are definite differences between first
and final combining forms. Initial combining forms are modifiers, final
combining forms are heads. Initial combining forms have a characteristic
phonetic "shape", ending in a vowel; as just noted, they tend not to be
secreted. Final combining forms are phonetically amorphous and are
either abbreviated or secreted.
In most cases, it is easy to identify combining forms. There are,
however, morphemes which may qualify as combining forms and affixes
equally well. This happens when a morpheme formed in the manner
characteristic of combining forms develops grammatical features. The
most important difference between combining forms and affixes is the
fact that combining forms are open-set items which may be formed on
the spur of the moment. The products of compounding and affixation
are novel combinations of existing morphemes; the products of conversion
are semantic modifications of existing morphemes; the products of clip-
ping, acronym-formation, and blending are abbreviations, i.e., phonetic
modifications, of existing morphemes. In forming composites containing
combining forms, we are not restricted to existing morpheme-forms. The
accounts of word formation I am familiar with neglect to make clear this
crucial difference between the process of forming composites involving
combining forms and other word formation processes. The real reason
for my interest in combining forms is, however, that once again we find
support for the idea steadily gaining ground among semanticists, viz.,
knowing what other people know and knowing how they are likely to
process this knowledge are essential parts of meaning formation, which
of course is an essential part of word formation.
126 Beatrice Warren
Appendix
List A: First elements
"Prefix" Containing noun Model word
Group I
alpino- alpinodrome Lat. alpinus
"place for climbing contests"
angio- (balloon) angioplasty Gk. angeion
"technique for treating angina"
antenno- elect roan tennogram ?E antenna
"chart of electricity in antennae"
laqua- aquatube Lat. aqua
"water slide"
astro- astrodome Gk./Lat. astrum
"stadium with translucent domed
roof'
bio- biocomputer, biohazardous, Gk. bios
bioholonics, biomotor et al.
chrono- chronobiology et al. Gk. chronos
"the study of biological rhythm in the
human body"
circo- circotherm( oven ) Lat. circum
"oven with fan circulating heat"
debtno- debtnocrat Ε debt
"banker dealing with international
debts"
electro- electroantennogram, et al. Lat. electricus
"tracing of electricity in antennae"
ethno- palaeoe thnobo tany Gk. ethnos
Igiga- gigadisc, gigaflops ?Gk. Gigas
"disc capable of storing large
amounts of data"
hetero- heterosexism, (heterosexist) Gk. heteros
"discrimination agabinst homosexu-
al«"
alS
iatro- iatroblast Gk. iatros
"embryonic doctor"
irido- iridology, (iridologist) Gk. iris
"technique of diagnosis involving iris"
litho- lithotripter Gk. lithos
"device for pulverizing kidney stones"
mega- "great" mega-blockbuster, megabrand, Gk. megas
megabuck, mega-city et al.
'Prefix' Containing noun Model word
mega- "million' megaflops Gk. megas

"a million flops"
micro- microburst, microfloppy, microgravity Gk. mikros
mult i- multitask, multi-user, multi-processor Lat. multus
neuro- neurophilosophy et al. Gk. neuron
"philosophy of the mind-brain"
opto- optoelectronic(s) Gk. optikos
"electronics involving interaction of
light and electronic signals"
palaeo- palaeoethnobotany, Gk. palaios
palaeotopomorphologist
petro- petrocurrency E/Lat./Gk. petroleum
"currency of nation depending on oil
production"
phyto- phytolith, phytotherapist Gk. phyton
"minerals in plants",
"therapist using plant extracts"
poly- polymyositis Gk. polys
"inflammation of striated muscle"
psycho- psychographic (s), Gk. psyche
psychoneuroimmunology
radio- radioglaciology Lat. radius
"the study of glaciers by radar"
smello- smellometer Ε smell
"instrument for measuring body
odor"
Isomato- videosomatography Gk. soma
"technique involving video pictures
for improving man-machine interface'
sono- sonochemistry Lat. sonus
"chemistry using ultrasonics"
spiro- spiroplasma Lat. spira
"spirally shaped micro-organisms"
sulph(o)- sulphoxidizer E/Lat. sulphur
"person who can break down sul-
phur"
techno- technobattle, technofear, technofreak, Gk. techne
technophobia et al.
tele- telebanking, telecommute, Gk. telos
teleshopping
topo- palaeotopomorphology Gk. topos
"study of land to discover traces of
primitive cultures"
uretero- ureteroscope Gk./Lat. ureter
"instrument for viewing urinary
tracts"
128 Beatrice Warren
"Prefix" Containing noun Model word
Iviru- virustasis ?Lat. virus

"stopping of growth of viruses"
Group II
cyber- cyberphobia cybernetics
"fear of computers"
eco- ecosocialism ecology
"socialism concerned with ecology"
econo- econospeak economics
"the jargon of economists"
Euro- Eurofighter, Eurο feebleness, Europe(o)
Eurosclerosis, Eurowimp et al.
Ipluro- pluro-communism pluralism
"communism and pluralism com-
bined"
vege- vegeboom vegetarianism
"sudden popularity of vegetarianism"
List B: Last elements
"Suffix" Containing noun Model word
Group I
-blast iatroblast Gk. blastos
"embryonic doctor"
-cracy yobbocracy ?aristocracy
"rule by louts and thugs"
-erat debtnocrat ?bureaucrat
See debtno-
-drome alpinodrome Gk. dromos
"place for climbing contests"
-ectomy lumpectomy Gk. -ektomia
"surgical removal of cancer"
-gram electroantennogram et al. Gk. gramma
"chart of electricity in antennae"
-graphic psychographic (s) Gk. graphos
"market research of attitudes"
-graphy videosomatography Gk. graphia
See somato-
"Suffix' Containing noun Model word
-lith phytolith Gk. lithos

"mineral particle from plants"
-logue biologue Gk. logos
"biographical radio programme"
-logy chronobiology et al. Gk. -logia
"study of biological rhythm in the hu-
man body"
-mania mergermania Gk./Lat. mania
"craze for mergers"
-phobia cyber phobia et al. Gk./Lat. phobia
"fear of computers"
-plasty angioplasty Gk. -plastia
See angio-
-scope ureteroscope Lat. scopium
"instrument for viewing urinary
tracts"
-stasis virustasis Gk. stasis
See viru-
Group II
-(a)holic spendaholic alcoholic
"compulsive spendthrift"
-(a) thon bikeathon, duckathon, pedalathon, marathon
readathon, swimathon
-erati slopperati, glitterati lliterati
"the deliberately untidy rich";
"jetsetters", respectively
-gram Tarzangram, potatogram telegram
"greeting delivered by costumed
person"; "message on a potato"
-itis feature-itis Ίarthritis, etc.
"excessive occurrence of feature arti-
cles"
-nik Wappnik Irefusenik
"journalist refusing to cross picket
line in Wapping to go to work"
-nography warnography pornography
"literature or films glorifying war and
violence"
-tro outro intro
"ending of song or musical number"
-rrhoea processorrhoea diarrhoea
"excessive flow of words induced by
word processor"
130 Beatrice Warren
"Suffix" Containing noun Model word
-quel prequel sequel

"the story of the events leading up to
those in an existing work"
-zak newzak muzak
"repeated news coverage of the same "soft music played
events" continuously in
stores, etc."
Group III
-boom vegeboom baby boom

"sudden popularity of vegetarianism"
workfare welfare
-fare
"social benefits in return for unpaid
work"
Yuppiegate, Westlandgate Watergate
-gate
"scandal involving five Yuppies, West-
land helicopters", resp.
nukemare nightmare
-mare
"fear of disaster induced by nuclear
materials"
magnometer, smellometer barometer, etc.
-meter
"instrument measuring yang and body
odour", resp.
-smith leaksmith ?blacksmith, etc.
"person leaking information"
-speak cataloguespeak, computer speak, newspeak
econospeak etc.
Ί-wagon branwagon band wagon
"trend to eat health foods, such as
bran"
-ware firmware software
"computer programme stored in a
read only memory"
Comments
(i) In excerpting examples, problems of inclusion always arise. Admittedly, some

of the examples in the Appendix are not indisputable examples of combining
forms. These have been marked with question marks. Similarly, some composites
which rightly should have been included may have been excluded. Only combining
forms part of novel composites have been considered; that is to say, from, for
example, palaeotopomorphologists, which consists of four combining forms, mor-

pho- and -logist have not been extracted, being parts of an established composite.
(ii) For reasons of space, the glosses are brief. The interested reader is referred
to Mort (1987), who supplies full explanations.
(iii) Linguists normally think of -o- and -i- as linking vowels added to the initial
combining form, whereas -a- in -(a)holic, for instance, is considered a linking
phoneme added to the final combining form. Although I cannot quite justify this
practice, I have adhered to it here.
Notes
1. Capitalization signifies that the particular word or form has an entry of its own in the
dictionary.
2. I am not the first to have noticed inconsistencies in the treatment of combining forms.
See Stein (1977 and 1984).
3. Barnhart et al. (1980), however, use the phrase "abstracted form" for secreted elements.
4. The fact that combining forms are bound morphemes does not prevent them from
occasionally being converted into proper nouns. Consider, for example, itis, which can
be used as a noun in the sense "bodily condition or disease".
5. Bauer's claim that final combining forms are restricted to the company of initial
combining forms is not correct.
6. There is an exception to this rule, viz., the so-called dvandva compounds exemplified
by Franco-German, Anglo-Polish, Sino-Italian. These compounds consist of two heads
rather than a modifier and a head.
References
Adams, Valerie
1973 An introduction to Modern English word-formation. (London: Longman).
Bauer, Laurie
1983 English word-formation. (Cambridge: Cambridge University Press).
Barnhart, Clarence L., et al.
1980 The second Barnhart dictionary of New English. (Bronxville, Ν. Y.: Barnhart
Books).
Hansen, Barbara — Klaus Hansen —Albrecht Neubert — Manfred Schentke
1985 Englische Lexikologie. (Leipzig: Enzyklopädie).
Jespersen, Otto
1950 Language — its nature, development and origin. (London: Allen & Unwin).
Marchand, Hans
1969 The categories and types of present-day English word-formation. (München:
Beck).
Mort, S. (ed.)
1987 Longman Guardian new words. (Harlow: Longman).
Quirk, Randolph, et al.
1985 A comprehensive grammar of the English language. (London/New York:
Longman).
132 Beatrice Warren
Stein, Gabriele
1977 "English combining forms", Linguistica 9 (= Tartu riikliku ülikooli toime-
tised 437), 140-147.
1984 "Word-formation in Dr. Johnson's dictionary of the English language",
Journal of the Dictionary Society of North America 6: 35 — 112.
Warren, Beatrice
1988 "Ambiguity and vagueness in adjectives", Studia Linguistica 42: 122 — 177.
Compounding and inflection
Wiecher Zwanenburg
0. The general idea behind this paper is the hypothesis that in the
unmarked case languages organize their word structure hierarchically in
such a way that inflection is peripheral to compounding.
I want to discuss here the apparently exceptional relationship between
compounding and inflection in the Romance languages, and more par-
ticularly in French. In French, we find cases where inflection seems to
be inside compounding, as in (la), next to the expected cases where
inflection is peripheral, as in (lb):
(la) des secretaires-generaux 'secretaries-general'
des basses-cours 'poultry yards'
b) des en-tetes 'letter headings'
des apres-midi{s) 'afternoons'
I will argue that the cases of (la) are only apparent exceptions to the
above-mentioned generalization.
Before discussing this point in Section 2, I will consider in Section 1
the possibility to distinguish in a satisfactory way between derivational
and inflectional affixation. This possibility, questioned by quite a number
of linguists, is of course a necessary condition for the above-mentioned
generalization to make sense.
1. As to derivational and inflectional affixation, one of those who have

recently argued that it is doubtful whether one can distinguish the two
in a principled way is Selkirk (1982: 69 — 77). She argues convincingly
that the two types of affixation must be accounted for by the same
morphological (sub)module. They obey quite similar or maybe identical
principles of morphological X-bar structure and show quite similar and
in many cases identical phonological behavior.
But this does not mean that there is not a principled distinction between
the two types of affixation with respect to syntax. Anderson (1982)
proposes the following criterion, which seems to me to be adequate:
134 Wiecher Zwanenburg
(2) Inflectional morphology is what is relevant to the syntax, with

the exclusion of properties of lexical insertion per se and con-
comitant principles of subcategorization etc.
Anderson concludes from this that derivational and inflectional affix-
ation must be treated in different modules, derivation in the lexicon
before lexical insertion at S-structure, and inflection in an extended
phonological module after lexical insertion. But this runs counter to
Selkirk's important insight that there must be some (sub)module where
it is possible to generalize over derivational and inflectional affixation as
to morphological X-bar structure and phonological behavior.
In my view, one can do justice to Anderson's insights within the
general framework of Selkirk's model. That is, both derivational and
inflectional affixation are accounted for in one and the same morpholog-
ical module within the lexicon by comparable and in many cases identical
rules. At S-structure, the necessary syntactic information is available for
lexical insertion of the desired inflectional form of a word, so there is no
need to have an extended phonological module take care of inflection.
But within the morphology, the affixation rules operate at least at two
different levels, first for derivation at a syntax-independent level, and
second for inflection at a level dependent on syntactic information. The
distinction between the two levels must be stipulated within morphology.
According to Di Sciullo — Williams (1987: 25) this distinction need not
be stipulated, but follows from the position of heads in words. For an
inflectional affix to communicate by percolation its syntactically relevant
information to the word, it must be in head position, that is in the
rightmost position, and thus outside derivational affixes. This excludes a
form like *nounshood instead of nounhoods.
But then Di Sciullo-Williams (1987: 2 5 - 2 7 ) relativize the notion
head, defining per feature F the head of a word with respect to F as the
rightmost element of that word marked for F. This accounts for cases
like the French diminutive suffix -ot, forming nouns from nouns, adjec-
tives from adjectives, and verbs from verbs respectively (frer-ot, pal-ot,
and touss-ot-er): -ot is unmarked for the feature category, and thus the
respective bases are heads with respect to category. But now *nounshood
is not excluded any more.
In order to save their analysis, Di Sciullo —Williams (1987: 27 — 28)
assume a principle saying that if a feature is defined for a category, then
all members of that category are marked for that feature. Thus noun in
nounhoods will be marked for (singular) number and the plural affix must
be in rightmost position, as in (3a). The same holds for a compound like

choir boys in (3b). But Di Sciullo — Williams' principle is unable to explain
the cases where inflection is outside a derivational affix unmarked for
category. This is what we find for example in French touss-ot-ait in (3c).
We have seen that -ot is unmarked for category. Thus nothing would
predict the agrammaticality of *touss-ait-ot.
(3a) nounhoods j *nounshood

b) choir boys / *choirs boy
c) touss-ot-ait 'coughed dryly', literally 'cough-little-ed' /
*touss-&\\-ot
So I feel that the hierarchical relation between derivation and inflection

should be stipulated. And this should be done along the lines suggested
by Anderson, but within Selkirk's general framework.
2. Having thus established the possibility and the necessity to distinguish

between derivation and inflection, let us look more closely into the
apparently problematic relationship between compounding and inflection
in French. My examples will all concern nominal number inflection. They
are from written French, including where possible not so frequent cases
where orthographic differences between singular and plural are parallelled
by differences in pronunciation.
Consider first the following cases, all problematic like those of (la):
(4a) Ν —*• NN: oiseaux-mouches 'humming-birds'
amiraux-poetes 'admirals who are poets'
b) Ν —• N N : timbres-poste 'postage stamps'
(Eufs-mayonnaise 'eggs with mayonnaise'
c) Ν —• N A : coffres-forts 'safes'
secritaires-generaux 'secretaries-general'
d) Ν —• A N : basses-cours 'poultry yards'
beaux-arts 'fine arts'
e) A —* A A: sourds-muets 'deaf-mute (pi)'
verbaux-nominaux 'verbal-nominal (pi)'
In all these cases, the plural affix is on the two words forming the
compound or on the first one only. This is contrary to the hypothesized
generalization, which would predict it to be on the entire compound only.
Now, Di Sciullo-Williams (1987: 7 8 - 8 8 ) argue that examples like
the ones in (4) are syntactic expressions. Insofar as they are idiomatic,
136 Wie eher Zwanenburg
they will be listed in the lexicon and as such be "listemes". Better-known

examples of such expressions are the much-discussed idiomatic verbal
expressions of the type to take to task or French mettre ä profit 'to
exploit'.
If this interpretation is correct, the cases of (4) are not problematic at
all for the generalization mentioned in the beginning. We are dealing
with phrases instead of compounds, and it is only to be expected that
inflection is marked on the words of these phrases. This means that not
only does Di Sciullo — Williams' analysis help to explain our data con-
cerning the position of inflectional affixes, but at the same time our data
provide independent evidence for their analysis.
Now consider the other kind of French compounds, illustrated in (lb).
In fact we can distinguish two types, as in (5a) and (5b). We have seen
that they are in principle unproblematic for our generalization, because
inflection, if present at all, is on the entire word and not on one of its
parts, and thus it is outside compounding:
(5a) Ν —> V N: des essuie-glace 'screen wipers', literally 'wipe-ice'
des couvre-chefs 'head-dresses'
b) Ν —*• PP: des en-tetes 'letter headings', literally 'at-heads'
des apres-midi(s) 'afternoons'
This would conclude my argument, but for the fact that compounds
like those in (5) confront us with problems which are indirectly linked to
the topic of this paper. So let us examine the cases of (5a) and (5b) in
that order before concluding.
Di Sciullo —Williams argue that the type of (5a) allows for other
complements than nouns, as illustrated in (6). On the basis of these
examples they argue that this type must in fact be analyzed as in (7).
(6a) N - -> V N : essuie-glace 'screen wiper'
b) N - ^VA: gagne-petit 'poorly paid person'
c) N - -> V Adv: couche-tard 'night-reveller'
d) N - -•VP: frappe-devant 'sledge hammer'
(7) N - -> VP
This leads them to admit more generally, for the periphery of the
grammar, nonmorphological word-creating rules reanalyzing phrases
such as VP as words. Unlike the cases of (4) and like ordinary words
created by morphological rules, syntax may only refer to these words as
a whole and cannot "look into them".
Di Sciullo —Williams then argue that in fact there are more possibilities
for this type of words, as shown in (8). And they conclude that the right
generalization is as in (9).
(8a) des trompe-l'oeil 'trompe l'ceil paintings'
b) des boit-sans-soif 'boozers'
c) des bons-ä-rien 'good-for-nothings'
d) des hommes-de-paille 'straw-men'
e) des hors-la-loi 'outlaws'
(9) Ν —• XP
Now, this would constitute a problem for our generalization: in (8c)
and (8d), unlike the other cases, inflection is not on the entire compound-
like word but on a part of it only, contrary to our generalization. But it
is easy to see that precisely this fact allows us to interpret the cases of
(8c) and (8d) as syntactic expressions with idiosyncratic meaning or
listemes, like those in (4). Thus, such expressions turn out to be more
widespread than Di Sciullo — Williams suggest.
In this way we are led to maintain rule (7) instead of (9), but generalized
over nouns and adjectives to account for adjectival cases like (10). The
rule will then take the form of (11).
(10) un enfant brise-tout 'a child that breaks everything'
(11) A/N —> VP

This will take care of all of Di Sciullo — Williams' examples except
(8e). Now, it turns out that (8e) represents the second case of (5) to be
discussed, that of Ν —> PP in (5b). Consequently, we need a second
nonmorphological word-creating rule of the form Ν —> PP, but this rule
can be generalized over nouns and verbs, as the examples in (12) show.
I will discuss elsewhere in more detail the generality of the cases of (12)
and their prepositional character, as well as the occurrence of PP in
adjectives. The rule will have provisionally the form of (13).
(12a) N: en-tete
b) V: en-chain-(er) 'to enchain'
(13) N/V —> PP

This leaves us with a question for future research concerning the status
of nonmorphological word-structure created by rules such as (11) and
138 Wiecher Zwanenburg
(13). The examples given so far suggest that its hierarchical relationship
with inflection is the same as that of compound rules: inflection is outside.
But then Di Sciullo — Williams do not discuss the not infrequent case of
VP nouns where VP contains a plural object, like casse-noisettes 'nut-
cracker', which have the same form in the plural. This fact does not
impair the reasoning above about the ungrammaticality of *nounshood
and * choirs boy as plural forms of nounhood and choir boy. The question
to be examined is to what extent syntax-like nonmorphological word
structure admits inflected elements. And more generally, what are the
restrictions on such a structure other than nonreferentiality discussed by
Di Sciullo —Williams.
3. I conclude that the French facts do not force us to abandon the

hypothesis that in the unmarked case languages organize their word
structure hierarchically in such a way that inflection is peripheral to
compounding.
In order to reach this conclusion I have followed Di Sciullo — Williams
(1987) in analyzing French compounds partly as listed phrases, partly as
cases of marked nonmorphological word structure. My treatment has
given independent evidence for this analysis. I have argued against
Di Sciullo — Williams that the distinction between derivational and
inflectional affixation must be stipulated. Moreover I have argued
that their rule (9) Ν —• XP must be replaced by (11) N/A —• VP and
(13) N/V —• PP.
References
1982 "Where's morphology?", Linguistic Inquiry 13: 571 —612.
Di Sciullo, Anna-Maria —Edwin Williams
1987 On the definition of word. (Cambridge, MA: MIT Press).
Selkirk, Elizabeth O.
1982 The syntax of words. (Cambridge, MA: MIT Press).
Topic 3: Inflectional morphology and clitics
Arguments against the passive as a universal
morphological category
Paul Kent Andersen
The purpose of this paper is to argue that what is generally regarded as

a "passive" in the literature does not in fact constitute a single,
universal/prototypical grammatical category, but on the contrary, that
there is evidence for the existence of several, universal/prototypical cat-
egories that should replace the single category "passive". Due to the
limitations imposed on this paper, the following discussion will center
around only one of the many aspects of the "passive", namely "passive"
morphology. Accordingly, the basic thesis that will be defended here is
that the varieties of verbal morphemes employed in constructions gen-
erally regarded as "passives" in the literature are in fact instances of a
number of distinct, prototypical, morphological categories. Consequently,
there is preliminary evidence that "passive" constructions employing these
respective morphological categories are themselves instances of distinct,
prototypical constructions.
1. Introduction
For quite some time now it has been recognized that constructions
generally regarded as "passives" differ in morphological, syntactic, se-
mantic, and even discourse features; in fact, there is very little that is
shared by all "passive" constructions, cf., among others, the conclusions
reached by Siewierska (1984: 259). Thus we are faced with the problem
that there is no single, generally accepted definition of the "passive" that
does justice to all instances of "passives" attested in the literature. This
is particularly evident in the most recent collection of papers concerning
the "passive" and Voice edited by Shibatani (1988): here we find that the
definitional properties of the "passive" are regarded by some authors,
e.g., Rude and Cooreman, as discourse features, by others as syntactic
142 Paul Kent Andersen
properties, e.g., Siewierska and Kimenyi, and by others still, e.g., Dezsö,
as the "passive" morpheme. Of course, one of the easier ways out of this
problem is to define the "passive" in a rigid manner and then to claim
that instances not covered by the definition are not "passives". Since,
moreover, "theoretical consideration of passive constructions has natu-
rally focused on English" (Langacker — Munro 1975: 789), it therefore
comes as no surprise to find that the English "passive" generally ends up
either as the expression par excellence of "the passive prototype" — cf.,
e.g.,
We shall refer to passives like (lb), John was slapped, as 'basic pas-
sives' Our justification for calling such passives 'basic' is that they
are the most widespread across the world's languages (Keenan 1985:
247)1
and
The characterization in 40 applies to prototypical passives like English
Many soldiers were killed, or its Japanese equivalent Takusan no heitai
ga koros-are-ta (Shibatani 1985: 837)2
or positioned at one extreme along "the passive continuum", cf.
If the correlations (6), (7), and (8) above indeed hold, then one excepts
one major continuum in the typology of passivization, with the English
type representing one extreme of the continuum... . The Ute extreme
of the scale would have just the opposite properties (Givon 1981: 171).
Had theoretical considerations been focused on a language other than
English, we may have found that an entirely different prototypical defi-
nition of the "passive" would have been proposed and that the English
"passive" would have been regarded as an "alternative",
3
"pseudo-passive", "peculiar passive", or the like. In the present paper
an alternative approach to the problems of the "passive" will be taken,
i.e., instead of looking for one single "passive", the possibility of the
existence of a number of distinct "passives" will be investigated.
2. Passive morphology: preliminary remarks
The point of departure for the following morphological discussion will

be the questioning of the following fundamental axiom underlying most
discussions of passive syntax and morphology. "Passive constructions"
Arguments against the passive 143
are defined in terms of specific characteristics (varying from author to

author) including the stipulation that the verb/predicate contain a "pas-
sive morpheme", 4 whereas when it comes to concrete examples of "passive
constructions" the relevant verbal morphology — i.e., morpheme — is
isolated and invariably glossed as PASSIVE. In other words, a construc-
tion is a "passive" because it contains (among others) a "passive" mor-
pheme and a morpheme is a "passive" because it is employed in a
"passive" construction. That this axiom is blatantly circular appears to
pose no serious problems for the majority of authors. 5
The question that we will now pursue is whether there is any empirical
evidence for recognizing a specific morpheme as a "passive" other than
the fact that it is employed in a "passive" construction. Even if we would
assume that specific verbal morphology is a definitional characteristic of
"passive" constructions, does this automatically mean then that all in-
stances of these specific verbal morphemes are necessarily representatives
of one and the same prototypical morphological category? It is now my
contention that the recent typological studies of verbal morphology by
Bybee (1985) and Dahl (1985) will offer us a working framework in which
to answer these questions.
3. Verbal categories
Bybee (1985) and Dahl (1985) have each devoted an entire monograph
to the investigation of verbal categories based upon two independent
samples of languages especially suited for typological research. Although
the results that were reached do differ in various aspects, they both
demonstrate the existence of a number of prototypical verbal categories.
But because they both concentrate on Tense, Aspect and Mood, they
unfortunately have very little to say about "passive" Voice; both do,
nevertheless, agree that such a morphological category exists. So, for
example, according to her principles of relevance6 and generality,7 Bybee
establishes the following hierarchical order of verbal morphemes relevant
to the verb root: VERB-Valence-Voice-Aspect-Tense-Mood-number/
person/gender Agreement. 8 Since the "passive" is regarded as an instance
of the category Voice, we see that within this framework the "passive"
not only has a specific place in this hierarchy of verbal categories, it also
has a specific function (in terms of relevance and generality) that distin-
guishes it from the other categories. In this respect it is worth noting the
definitions of the categories for Valence, Voice, and Aspect employed by
Bybee in her investigation:
Valence refers to differences in the number or role of arguments that
the verb stem can take. Voice indicates the perspective from which the
situation described by the verb stem is viewed. Aspect refers to the
way the internal temporal constituency of the situation is viewed
(Bybee 1985: 28).9
Thus we have reason to believe that should there be a single, universal
morphological category for "passive voice", it will exhibit not only a
specific place along the hierarchy of verbal categories, but also a specific
function with regards to Bybee's principles of relevance and generality.
Dahl, on the other hand, gives the following reason why he has chosen
not to investigate the "passive" with his method based upon a compre-
hensive questionnaire:
In addition to the combinatorial explosion, there are additional reasons
why it is hard to study e.g. passive constructions with a methodology
of this type. In some languages, passives may be very marginal or even
not occur at all; other languages may have several constructions that
could be labelled passive, with more or less subtle differences in
conditions of use. This means that by simply giving an English sentence
in the passive you cannot guarantee that what comes out in the
translation is the passive construction you are looking for, if it is a
passive at all (Dahl 1985: 47).
In the following investigation we will assume that it is indeed possible
to classify the distinctive verbal morphology employed in constructions
generally regarded as "passives" according to the principles established
by Bybee and Dahl without recourse to the circular axiom mentioned
above. Our first indication that the morphemes employed in "passive"
constructions are not all instances of one and the same verbal category
can be seen from the fact that morphemes isolated and glossed as
"passives" generally perform functions other than that of appearing in
"passive constructions", i.e., "passive" morphemes are generally em-
ployed in constructions other than the "passive". 10 A closer look at these
other constructions will reveal the fact that they are themselves not
identical, but rather that they do define certain independent sets of
construction types. Thus it is not the case that all of the other functions
of the "passive" are the result of the mere extension of a single, basic
"passive" meaning in all of these instances.
4. Agreement
The first class of morphemes to be discussed is the set of "middle"

inflectional endings of Classical Greek and other Indo-European lan-
guages. The reason for starting with the "middle" should be rather
obvious: our modern term "passive" derives via Latin from the Greek
grammatical tradition, in which the term pathos (which was translated
into Latin as passivum) designates the morphological set of personal
endings generally referred to now as the "middle" in contrast to the
corresponding "active" ones.11 So, for example in (la) below from Clas-
sical Greek the "middle" form of the verb is employed (among others)
in "passive" constructions (TV represents the thematic vowel) and con-
trasts with the corresponding "active" form in (lb):
(la) lou-o-metha
wash-TV-1 PL/MIDDLE
'We are being w a s h e d 2
b) Ιού-ο-men
wash-TV-1 PL/ACTIVE
'We are washing.'
This now brings us to the actual morphological category employed in
(la). First of all, the categories generally referred to as "active" and
"middle" cannot be separated from the categories for person and number,
thus we have evidence that the categories for "active" and "middle" are
expressions of the category Agreement. These morphemes for Agreement
(along with the TV) are furthermore placed after the morphemes for the
categories Tense, Aspect, and Mood, 13 thus giving evidence that the
morphological categories expressed by the "active" and "middle" Agree-
ment markers are not instances of the category "Voice" within Bybee's
framework. Second of all, we can determine that a number of other
construction types also employ the same "middle" inflectional endings,
cf., e.g., Barber (1975) and Haspelmath (1987). These include the reflexive,
reciprocal, anticausative, numerous so-called "middle constructions" (see
the passage from Klaiman quoted below) and the "passive". 14 From this
list of construction types we can furthermore determine that the common
functional denominator of all of these constructions is the following:15
"The implications of the middle (when it is in opposition with the active)
are that the 'action' or 'state' affects the subject of the verb or his interests"
(Lyons 1968: 373). So, for example, Klaiman (1982: 408) gives the fol-
lowing functions of the category "middle":
a) Plain Middle: results of action accrue to Subject

b) Reciprocal Middle: referents of plural Subject do action to one
another
c) Reflexive Middle: Subject performs action to self
d) Deponent Middle: action denotes physical/mental disposition of
Subject
e) Nucleonic Middle: object of action belongs to, moves into, or
moves from sphere of Subject
f) Plain Passive: Subject does nothing, is affected in consequence of
action
By characterizing the function of the "middle" inflectional category in
this way we can now see that this morphological category is not as
relevant to the verb as the categories Valence, Aspect, Tense, and Mood
are: the difference between the "active" and "middle" Agreement cate-
gories here does not directly affect or modify the semantic content of the
verb. Moreover, the categories for "active" and/or "middle" are neces-
sarily applicable to all verb stems (there are no finite verb forms not
conjugated for one or the other category) and thus — according to Bybee
(1985: 16 f.) — the semantic function of these categories must be rather
general and have only minimal semantic content. Thus it comes as no
surprise to find that they are placed farther away from the verb root
than the other categories.
This now brings us to the semantic characterization of the "passive"
constructions that employ the "middle" inflectional endings as the dis-
tinctive, morphological feature. Since the "middle" inflectional endings
are expressions for the category Agreement, we can determine that the
fundamental function is referential in nature, cf.:
My basic thesis is that agreement is referential in nature. It helps
identify or reidentify referents. It does this by giving information on
grammatical properties of its referent and, thus, of the NP representing
it if one is around. The functions of agreement in the marking of
syntactic relations derive from this primary function (Lehmann 1988:
55).
Recall, furthermore, that according to Jakobson (1971: 134) "PERSON
characterizes the participants of the narrated event with reference to the
participants of the speech event". If, however, we assume that there is
not only a speech event and narrated event, but also a narrative discourse
— as distinguished, e.g., in Dahl (1985: 112). "I define a narrative
discourse as one where the speaker relates a series of real or fictive events
in the order they are supposed to have taken place" — we will be in the
position to better characterize the actual function of the "middle" mor-
pheme in this type of "passive" construction: one participant of the
narrated event — i.e., the "Actor" — does not participate in the narrative
discourse (/speech act) within its respective narrated context; in the
corresponding "active" the "Actor" participates not only in the narrated
event, but also in the narrated discourse (/speech act). Thus we see that
the narrated event is not altered at all in this type of "passive" construc-
tion.
To sum up, in Classical Greek and other languages the specific mor-
phological means employed in the "passive" can be identified as an
inflectional category for person/number "Agreement". The basic/
prototypical semantic function of the "middle" is furthermore to indicate
that the "subject" is "affected" by the verbal action. The specific function
of this morphological category when employed in the "passive" can be
characterized as indicating that the "Actor" participates in the narrative
event without, however, participating in the narrative discourse (/speech
act). Since reflexive affixes in numerous languages throughout the world
exhibit almost 16 the very same functions as the "middle", there is some
indication that they may also represent the same prototypical verbal
category. 17
5. Perfect
Let us now turn to the type of "passive" found in languages like English,
cf., e.g.:
(2) John was seen (by Mary).
Our very first problem will be to isolate the specific morpheme employed
in this construction that distinguishes it from a corresponding "active"
construction. Unfortunately, it is a general consensus among various
linguists (especially those working within formal frameworks) that the
distinctive morphology in "passives" such as (2) above is the so-called
"passive participle" — or more accurately, the suffix -en which forms the
"passive participle". 18 Nevertheless, it should be rather obvious that this
form of the verb is not in any way restricted to "passive" constructions,
cf.:
(3) John has seen Mary.
(4) The glass is unfortunately broken. 19

What is morphologically distinctive of this type of "passive" construction
is the combination of a copula (or other auxiliary) and the "past parti-
ciple", i.e., a periphrastic construction.
Having isolated the specific morphological form, we can now attempt
to specify the morphological category expressed by this periphrastic
construction. In a recent investigation of this type of "passive" in English,
German, and Russian, Beedham (1982) has come to the conclusion that
the "passive" can be characterized as a construction describing "an event
leading to a state", cf.:
It is postulated ... that passives mean the portrayal of a state as the
result of a preceding action... . It is the contention of the present work
that the passive is an aspect, like in English the perfect and the
progressive... . The very similar restrictions of co-occurrence that the
passive and the resultative perfect display suggest that they are se-
mantically similar, thus confirming that the passive describes an event
leading to a state (Beedham 1982: 147).
With respect to verbal categories it therefore appears that this type of
"passive" represents the category "Perfect" or some closely related cate-
gory (or variant thereof)· 20 Although Bybee (1985: 159 ff.) considers the
"Perfect" — she prefers the term "anterior" in this context — to be a
special instance of "Tense", Dahl's characterization of the "Perfect"
(PFCT) is more relevant to our discussion, cf.: "PFCT is rather consis-
tently marked periphrastically... . Typically, constructions involving a
copula or some auxiliary together with some past participle or similar
form of the verb are used" (Dahl 1985: 129). Dahl (1985: 133 ff.) fur-
thermore makes the distinction between the Perfect of result and Resul-
tative constructions and remarks that "there also seems to be a high
correlation between passive voice and resultative constructions: indeed,
in many languages, resultative constructions are only found in the passive
voice or some functional equivalent of it..." (1985: 135). The difference
between the Perfect of result and Resultative constructions — according
to Dahl (1985: 134 f.) — is that in the Perfect of result there is a focus
on the event, whereas in the Resultative construction the focus is on the
(resulting) state. Although Dahl apparently assumes that "passives" are
instances of Resultative constructions, there is reason to believe that at
least in some languages the "passive" could in fact represent a Perfect of

result, cf. the following examples from Swedish:
(5) När hon kom ut pä gatan, upptäckte hon, att bilen var stulen.
'When she came out onto the street, she discovered that the car
was stolen.'
(6) När hon kom ut pä gatan, upptäckte hon, att bilen blev stulen.
'When she came out onto the street, she discovered that the car
was being stolen.'
(7) Bilen blev stulen, medan hon var inne pä banken.

'The caf was stolen while she was in the bank.'
(8) * Bilen var stulen, medan hon var inne pä banken.

In (5) the focus is on the resulting state and hence represents a prime
example of a Resultative construction. In (6) and (7), on the other hand,
the focus is on the event itself, thus representing examples of the Perfect
of result. Notice that in Swedish two different constructions are employed
in these two instances: in (5) the auxiliary is vara "to be" and in (6) and
(7) it is bli "to become". 21
We can therefore conclude that the specific morphological category
employed in these types of "passive" constructions is a variant of the
"Perfect", i.e., Perfect of result and/or Resultative; hence it is placed in
an entirely different position along Bybee's hierarchy of verbal categories
and exhibits an entirely different function than the morphological means
employed in the previous type of "passive" constructions. Needless to
say, the consequence of this is that the semantic structure of this type of
"passive" differs considerably from that of the previous type: here the
narrated event itself is altered to present a state resulting from a preceding
action, whereas in the "middle"/"reflexive" type of "passive", the narrated
event is not changed in any such manner. Moreover, the specific construc-
tion types in addition to the "passive" that employ this type of verbal
morphology likewise differ from the range of construction types that
employ "middle" or "reflexive" morphology: certainly, this periphrastic
means cannot be employed in and of itself to express reciprocals, anti-
causatives, "plain middles", "deponent middles", "nucleonic middles",
etc., just as the "middle" and "reflexive" cannot be employed to express
resultati ves/states.
6. Valence
The next types of "passive" constructions that we will discuss employ

morphemes representative of the category Valence. Here there are basi-
cally two different types of morphemes that can be distinguished: (a)
causatives in which the valence is increased and (b) "detransitives" in
which the valence is reduced. Let us therefore start with an investigation
of "passive" constructions that employ causative morphology or mor-
phemes that have developed from causatives. Although the use of caus-
ative morphology in the "passive" is not as well-known nor as widely
discussed in the literature as other types of "passives", in recent years
there has been a growing literature on the subject. So, for example, in
his discussion of the development of "passives" from causatives Haspel-
math (1988: 34 ff.) makes reference to the following languages that exhibit
this development: Hungarian (see also Dezso (1988: 306, 316 f.)), Green-
landic (Inuit), Kafa, Arawak, Wolof, Chinese (see also Hashimoto (1988)),
Gujarati, and various Altaic languages. To this list we can add Korean
(Keenan 1985: 262), Finnish (Lehtinen 1984: 14 ff.), and perhaps many
more. It should be perfectly clear now that this type of morpheme
represents a completely different verbal category than those discussed
previously (i.e., the "middles/reflexives" and "Perfects"): not only are
these morphemes placed in positions different from the others discussed
above with respect to Bybee's hierarchical order of verbal categories, thus
giving evidence of a basic difference in relevance and generality, they also
exhibit completely different functions as can be seen from the fact that
the construction types that employ this type of morpheme are not the
same construction types that employ the other types of "passive" mor-
phemes. With regard to the "get passive" in English and similar construc-
tions, for example, we can determine the specific semantic function of
the causative morpheme when employed in the "passive", cf.:
The choice of a form like The window got broken over The window
was broken seems to imply that the window somehow brought the
catastrophe onto itself — if only by being in the wrong place at the
wrong time. Its presence, as it were, catalyses the action performed by
a quite separate agent (Barber 1977: 22).
Whether this is true of all "passives" with causative morphology is not
at all clear. Notice, though, that there are — semantically speaking — a
number of different types of causatives; each will presumably entail a
slightly different semantic analysis.
Since the "passive" is traditionally characterized as entailing detran-

sitivization it should also come as no surprise to find that morphemes
that decrease Valence are employed in "passive" constructions. The fol-
lowing example from Basque may in fact exhibit one such instance,
should the morpheme -a actually perform the function of decreasing the
Valence of the verb resulting in the auxiliary agreeing with one core
constituent less than the corresponding active.
(9a) Gizon-a-k txakurr-a-0 maluskatu zuan
man-the-ERG dog-the-ABS beat AUX(3SgSubj3SgObj)
'The man beat the dog.'
b) Gizon-a-k txakurr-a-0 maluskatua zan
man-the-INSTR dog-the-ABS beat(PASS) AUX(3SgSubj)
'The dog was beaten by the man.'
(cf. Keenan 1985: 248 f.)
Note also the discussion in Givon (1988) in which it is argued that the
"passive" suffix -ka in Ute was also associated with detransitivization.
Again, we can determine that such morphemes that decrease Valence
represent yet another morphological category distinct from all of those
discussed above not only with respect to the actual function expressed
but also by the specific types of construction types that also employ this
same morpheme.
7. Aspect
When it comes to Aspect we can recall the generalization drawn by

Keenan (1985: 267): "If a language has any passives it has ones which
can be used to cover the perfective range of meaning." Accordingly, all
languages that have at least one "passive" construction at their disposal
have a "passive" with perfective Aspect. We should note, though, that
there is a great deal of confusion in the literature with respect to the
categories perfective Aspect and Perfect: certainly, the "passive" in Eng-
lish is to be regarded as an instance of the Perfect and not of the perfective
Aspect. It therefore appears that this confusion is incorporated into
Keenan's generalization here. This is not to say that perfective should be
replaced by Perfect in Keenan's generalization: surely there may be
"passives" that do indeed employ verbal morphology representative of
the category perfective Aspect. In fact DeLancey (1981), who does dif-
ferentiate between perfective Aspect and Perfect, states that "Perfective
views an event from its terminal point" (1981: 647) and proposes that in
the "passive" there is a "terminal view point with respect to the transitive
vector" (1981: 647). Thus we see that according to DeLancey the "passive"
is an expression of terminal viewpoint and hence perfective Aspect.22
Should the specific verbal morphology that is employed in such "pas-
sive" constructions actually represent the category perfective Aspect, then
we would be able to conclude that this category is again distinct from
the other categories discussed above, cf.:
A PFV verb will typically denote a single event, seen as an unanalysed
whole, with a well-defined result or end-state, located in the past.
More often than not, the event will be punctual, or at least, it will be
seen as a single transition from one state to its opposite, the duration
of which can be disregarded (Dahl 1985: 78).
We would furthermore expect to find that if these morphemes are em-
ployed in constructions other than the "passive", then these construction
types would differ from those construction types that employ other verbal
categories.
8. Voice
Up to this point in our discussion we have shown that the morphemes

distinctive of many "passive" constructions do not represent instances of
the verbal category Voice within Bybee's (1985) framework. This should
not, however, be taken as evidence that a category "passive" Voice does
not exist; certainly Bybee must have found enough instances of mor-
phemes for the verbal category Voice (that are placed between morphemes
for Valence and Aspect along her hierarchy of verbal categories and that
do furthermore exhibit a distinctive function in terms of relevance and
generality that is situated between those of Valence and Aspect) to warrant
her postulation of this category in this particular hierarchical position.
Unfortunately, Bybee does not give any examples or references to lan-
guages where such a situation occurs. Presumably, there were many
languages in her sample that exhibit a morpheme for "passive" Voice in
addition to morphemes for Valence and Aspect that is indeed placed
between these other morphemes. Since I have no direct access to, or

knowledge of, these languages, I have not yet been able to investigate
these morphemes; thus, we are not yet in the position to give a charac-
terization of them. It should be evident, however, that such morphemes
for "passive" Voice will indeed exhibit a function entirely different from
those discussed above.
9. Summary and conclusions
The basic assumption of this paper has been that it is possible first of all
to isolate and identify the verbal morphology that is distinctive of con-
structions generally claimed to be "passives" in the literature, and second
of all to classify the specific morphemes as instances of prototypical
verbal categories in accordance with the methods proposed by Bybee
(1985) and Dahl (1985). The preliminary results presented here give
sufficient evidence that the morphemes traditionally regarded as "pas-
sives" do not represent instances of a single, prototypical morphological
category, but rather that they represent instances of the following distinct
categories:
(i) Agreement
(ii) Perfect of result
(iii) Resultative
(iv) increase in Valence (causative)
(v) decrease in Valence ("detransitive")
(vi) perfective Aspect
(vii) "passive" Voice
Each of these prototypical verbal categories furthermore has its own
distinct function, thus giving evidence that the constructions employing
these verbal categories are themselves distinct and not merely variants of
a single, prototypical construction type. Semantically, these functions can
be interpreted as various processes by which a narrated event is trans-
formed into an appropriate discourse event — or state — within the
respective discourse context.
Since the basic function of Agreement is referential in nature (cf.
Lehmann 1988), we see that the specific function of Agreement has to
do with identifying participants of the narrated event with respect to the
discourse event within the respective discourse context without changing

the narrated event in any way. In "passive" constructions employing
"middle" or "reflexive" Agreement morphology we furthermore can con-
clude that one participant — i.e., the "Actor" — of the narrated event
does not participate in the discourse event within the respective discourse
context.
In the case of the two types of Perfect we see that the verbal category
has the effect of changing the narrated event into the state that results
from the narrated event. The Perfect of result places the focus on the
event itself, whereas the Resultative places the focus on the resulting
state.
In the case of an increase in Valence we find that the narrated event
is expanded by the addition of a Participant, or more to the point by
embedding the narrated event into another event, thus giving evidence
for event-embedding.
In the case of a decrease in Valence we have evidence that the narrated
event is changed by removing one of the Participants; hence we have an
instance of event-splitting in the sense of Reichenbach (1947: 268).
In the case of the perfective Aspect we see that the narrated event is
changed to another, more compact and perfective event.
Unfortunately, the present investigation has not led to any concrete
conclusions concerning the semantic function of morphemes for "passive"
Voice.
Notes
1. Notice that although this particular construction may be widespread in the world's
languages syntactically speaking, the specific morphology found in this construction is
certainly restricted to a rather small number of languages.
2. Shibatani's characterization 40 of the passive prototype contains the following mor-
phological property: Active = P; Passive = P[ +passive] (837). As I will argue in the
course of this paper, there is no evidence for a single, prototypical passive morpheme,
i.e., P[ +passive],
3. Iconoclastic though it may be, in actual fact the type of "passive" attested in English
should be classified as either a stative or a resultative according to Nedjalkov and
Jaxontov's (1988: 6) definition: "... the stative expresses a state of a thing without any
implication of its origin, while the resultative expresses both a state and the preceding
action it has resulted from". So, for example, in the sentence "When I came at five,
the door was shut, but I do not know when it was shut" the first occurrence of the
"passive" was shut is a stative, whereas the second is a resultative. Note, however, that
not all passives can be classified as states or resultatives (especially those discussed in
Sections 4, 6, 7, and 8 below). For a further discussion see Section 5 below; note,
though, the difference in terminology.
4. Cf., among others, Keenan (1985: 245): "Thus consider how, as field workers, we can
tell if a sentence in a language is passive or not. What is it about passives that makes
them observably distinct in surface from basic actives? ... Thus we cannot recognize a
passive in terms of its NPS being marked or positioned in the sentence in ways different
from those used in basic actives... . In fact the only way we know that (4b) above is
passive is by the presence of a specifically passive suffix, -aki, on the verb. And this
observation turns out to be general across languages. That is, in general in a language,
what is distinctive about the observable form of passives is localized within the predicate
or verb phrase (understood broadly enough to cover auxiliary verbs)." See also the
relevant discussion in Haspelmath (1988: 3 — 11). A notable exception to this is Sie-
wierska (1984: 256).
5. Notice that some authors have even attempted to motivate this axiom, cf., e.g.,
Haspelmath (1988: 1): "My view that the nature of passive morphology is inseparable
from the nature of passive syntax..." and (1988: 51, fn. 3): "And that, more generally,
whenever a syntactic construction is associated with morphological marking, the nature
of the morphology is inseparable from the nature of the syntactic construction." Should
the results of this paper be accepted, Haspelmath would then be forced to accept the
existence not of one single "passive" construction, but rather a number of distinct
"passive" constructions.
6. Cf. "A meaning element is relevant to another meaning element if the semantic content
of the first directly affects or modifies the semantic content of the second" (Bybee 1985:
13).
7. Cf. "By definition, an inflectional category must be applicable to all stems of the
appropriate semantic and syntactic category and must obligatorily occur in the appro-
priate syntactic context. In order for a morphological process to be so general, it must
have only minimal semantic content. If a semantic element has high content, i.e., is
very specific, it simply will not be applicable to a large number of stems" (Bybee 1985:
16 f.).
8. Bybee's hierarchy of verbal categories is reminiscent of, and certainly related to, the
distinction made by Foley —Van Valin (1984) between different, structured layers of
the clause and specifically between different operators of the various layers.
9. It should be pointed out that Bybee's (1985: 20) characterization of Voice is based
upon Barber (1975), which in turn represents the characterization of only one single
type of "passive" construction, i.e., those discussed under the heading "Agreement"
below. Should the results of this study be accepted, the characterization of Voice will
necessarily need modification.
10. Cf., e.g., Shibatani (1985: 825): "Cross-linguistic evidence that passives are often related
to other constructions is overwhelming. In fact, it is far more deep-rooted and wide-
spread than has been indicated by casual observations found in many descriptive
materials". See also Keenan (1985: 253): "Moreover, the same types of formal mor-
phological means used in deriving basic passives are often used to derive VPS which
are not passives".
11. See Andersen (1989) for a detailed discussion.
12. As will be discussed below, the Classical Greek clause louo-metha can also mean 'We
are washing ourselves' as well as 'We are washing each other'.
13. Notice, in this respect, the following generalization already mentioned in Greenberg's
discussion of implicational universals: "Universal 30. If the verb has categories of
person-number or if it has categories of gender, it always has tense-mode categories"
(Greenberg 1966: 93).
14. Since there are numerous references in the literature of other "passive" morphemes
that are also employed in reflexive — possibly even reciprocal and anticausative —
constructions, these construction types may prove to be not indicative of this type of
morphological category. Notice, though, that these other "passive" morphemes have
not been identified as instances of any particular verbal category (other than the
"passive"): a detailed study of these morphemes may reveal the fact that they too are
instances of this prototypical verbal category. At any rate, what is distinctive of the
"middle" is its employment in a number of so-called (intransitive and transitive) "middle
constructions".
15. Although there has been much discussion of this notion of "affectedness" as the primary
function of the "middle" in recent years, it should be noted that this very function has
been recognized in some linguistic circles for more than a century. In fact, the very
term päthos that was employed for this morphological category in the Greek gram-
matical theory has the meaning "affection", cf. Andersen (1989). The characterization
of the function of the "middle" as "affectedness" should also be regarded as a very
useful, but still only a prototypical characterization, because there are some exceptions.
So, for instance, there are examples of the "middle" employed in impersonal "passives"
in which there is no "subject" to be affected. This applies not only to impersonal
"passives" of intransitive verbs such as Latin venitur, curritur, itur (cf. Lehmann 1985:
247 f.), but also to impersonal "passives" of transitive verbs that therefore result in
"non-promotional passive" constructions, cf. the two examples from Plautus discussed
in Wackernagel (1950: 146). Furthermore, there are also other means of expressing the
fact that the "subject" is affected by the verb, cf. the type Latin pudet me and German
Mir ist heiß with "active" morphology.
16. Notice though that in a few languages the reflexive markers are placed closer to the
verb root than categories for Tense and Agreement; they therefore can be regarded as
less prototypical in such instances.
17. Should the semantic characterization of this type of "passive" construction given above
be valid, I see absolutely no reason not to regard constructions with active morphology
that express this very same semantic structure as "passives", cf. Keenan (1985: 247):
"It appears however that languages without passives have somewhat more grammati-
cized means for expressing functional equivalents of basic passives. Perhaps the most
common means is to use an active sentence with an "impersonal" third plural subject.
By impersonal here we mean simply that the third plural element is not understood to
refer to any specific group of individuals." The only reason for not regarding these
constructions as "passives" is by reference to the fundamental — yet totally circular
— axiom discussed above. In fact Keenan (1985: 255) even states that "[o]ne might
expect, then, to find passives VPS which are identical to the transitive verbs they are
derived from." Moreover, there are numerous instances in which such third-person
plural markers do in fact develop into morphemes for the "passive", cf., e.g., Shibatani
(1985: 845 f.) and Haspelmath (1988: 37).
18. Cf., e.g., the discussion in Bresnan (1982: 81 - 8 2 , fn. 5), Hoekstra (1984: 129 ff.), Sells
(1985: 55), Riemsdijk - Williams (1986: 248 f.), Jaeggli (1986: 590 ff.) and McCloskey
(1988: 44 ff.). A notable exception to this rule is Haider (1986).
19. This particular construction contrasts with "passives" such as The glass is regularly
broken by vandals', the morphology of both predicates is, however, identical.
20. According to Nedjalkov — Jaxontov (1988: 6 ff.) the category in question here is the
resultative, or more specifically the objective resultative.
21. Note also the distinction in German between the "Zustandspassiv" with the verb sein
"to be" and the "Vorgangspassiv" with the verb werden "to become".
22. Whether terminal view point does indeed presuppose perfective Aspect is certainly
debatable.
References
Andersen, Paul Kent

1989 "Remarks on the origin of the term 'passive'" Lingua 79: 1 — 16.
Barber, E. J. W.
1975 "Voice — beyond the passive", Berkeley Linguistic Society 1: 16 — 24.
Beedham, Christopher
1982 The passive aspect in English, German and Russian. (Tübingen: Narr).
Bresnan, Joan
1982 "The passive in lexical theory", in: J. Bresnan (ed.) The mental representation
of grammatical relations (Cambridge, MA: MIT Press), 3 — 86.
Bybee, Joan L.
1985 Morphology. A study of the relation between meaning and form. (Amsterdam:
Benjamins).
Dahl, Osten
1985 Tense and aspect systems. (Oxford: Blackwell).
DeLancey, Scott
1981 "An interpretation of split ergativity and related patterns", Language 57:
626-657.
Dezsö, Läszlö
1988 "Passiveness in Hungarian: with reference to Russian passive", in: M.
Shibatani (ed.), 291-328.
Foley, William A. - Robert D. Van Valin, Jr.
1984 Functional syntax and universal grammar. (Cambridge: Cambridge Univer-
sity Press).
Givon, Talmy
1981 "Typology and functional domains", Studies in Language 5: 163 — 193.
1988 "Tale of two passives in Ute", in: M. Shibatani (ed.), 417 - 440.
1966 "Some universale of grammar with particular reference to the order of
meaningful elements", in: J. H. Greenberg (ed.) Universals of language
(Cambridge, MA: MIT Press), 7 3 - 1 1 3 .
Haider, Hubert
1986 "Fehlende Argumente: Vom Passiv zu kohärenten Infinitiven", Linguistische
Berichte 101: 3 - 3 3 .
Hashimoto, Mantaro J.
1988 "The structure and typology of the Chinese passive construction", in: Μ.
Shibatani (ed.), 329-354.
Haspelmath, Martin
1987 "Transitivity alternations of the anticausative type". (Köln: Institut für
Sprachwissenschaft, Universität Köln; Arbeitspapier Nr. 5, Neue Folge.)
1988 "Passive morphology: a crosslinguistic and diachronic study" [ms.].
Hoekstra, Teun
1984 Transitivity. Grammatical relations in Government-Binding Theory. (Dor-
drecht: Foris).
Jaeggli, Oswaldo A.
1986 "Passive", Linguistic Inquiry 17: 587-662.
Jakobson, Roman
1971/1957 "Shifters, verbal categories and the Russian verb", Selected Writings II (The
Hague: Mouton), 130-147.
Keenan, Edward L.
1985 "Passive in the world's languages", in: T. Shopen (ed.) Language typology
and syntactic description (Cambridge: Cambridge University Press), vol. 1:
243-281.
Klaiman, Μ. H.
1982 "Affectiveness and the voice system of Japanese: satisfaction guaranteed or
your money back", Berkeley Linguistic Society 8: 267 — 281.
Langacker, Ronald W. — Pamela Munro
1975 "Passives and their meaning", Language 51: 789 — 830.
Lehmann, Christian
1985 "Ergative and active traits in Latin", in: F. Plank (ed.) Relational typology
(Berlin: Mouton), 243-255.
1988 "On the function of agreement", in: M. Barlow —Ch. A. Ferguson (eds.)
Agreement in natural language. Approaches, theories, descriptions (Stanford:
Center for the study of language and information), 55 — 87.
Lehtinen, Tapani
1984 Itämerensuomen passiivin alkuperasta [On the origin of the Baltic-Finnish
passive] (Helsinki: Suomalaisen Kirjallisuuden Seura).
Lyons, John
1968 Introduction to theoretical linguistics. (Cambridge: Cambridge University
Press).
McCloskey, James
1988 "Syntactic theory", in: F. J. Newmeyer (ed.) Linguistics: The Cambridge
Survey I. Linguistic theory: Foundations (Cambridge: Cambridge University
Press), 1 8 - 5 9 .
Nedjalkov, Vladimir P.— Sergej Ε. Jaxontov
1988 "The typology of resultative constructions", in: V. P. Nedjalkov (ed.) Ty-
pology of resultative constructions (Amsterdam: Benjamins), 3 — 62.
Reichenbach, Hans
1947 Elements of symbolic logic. (New York: Dover) [Reprint],
Riemsdijk, Henk van —Edwin Williams
1986 Introduction to the theory of grammar. (Cambridge, MA: MIT Press).
Sells, Peter
1985 Lectures on contemporary syntactic theories (Stanford: Center for the study
of language and information).
Shibatani, Masayoshi
1985 "Passives and related constructions: a prototype analysis", Language 61:
821-848.
Shibatani, Masayoshi (ed.)
1988 Passive and voice. (Amsterdam: Benjamins).
Siewierska, Anna
1984 The passive: a comparative linguistic analysis. (London: Croom Helm).
Wackernagel, Jacob
1950 Vorlesungen über Syntax mit besonderer Berücksichtigung von Griechisch,
Lateinisch und Deutsch l 2 . (Basel: Birkhäuser).
The empty morpheme entailment*
Robert Beard
1. Separationist morphologies
The view that affixation and other morphological operations are separate
from those of lexical and inflectional derivation has been articulated
recently by Beard (1981, 1987, 1988), M o o r t g a t - v a n der Hulst (1981),
Pounder (in press), Szymanek (1985), Wood (1985) and for inflection
alone by Anderson (1982) and Matthews (1972). On this view, the addition
and adjustment of lexical categories in lexical bases are carried out by
abstract morpholexical (for word formation) and morphosyntactic (for
inflection) derivation rules. Morphology comprises autonomous phono-
logical "spelling-out" operations such as affixation, reduplication, revo-
calization, or autosegmental recombination.
(1) Lexeme Operations on Lexemes Grammatical
Component
Lexical Lexicon
feature Syntax
inventory
I
/p/ Morphology
Phonology
where /P/ = the phonological matrix of the lexical formant. This model,
the "separation hypothesis", implies that language is based on two discrete
types of basic elements, lexemes and (grammatical) morphemes, each with
its own related set of operations located in different parts of the grammar.
* Comments from Andrew Carstairs, Wolfgang Dressier, Yves-Charles Morin, Glanville

Price, and several other participants in the conference have significantly strengthened the
revision of this paper. My deepest gratitude to all of them.
160 Robert Beard
It does not take issue with the claim that lexical morphemes are signs;
lexemes are undisputedly signs, direct associations of sound and meaning.
The separation hypothesis is distinguished by its definition of grammatical
morphemes as semantically empty phonological operations and not lexical
objects.
2. The empty morpheme entailment
If morphology consists of semantically empty operations modifying the

phonological formant (only) of lexemes, such information as gender,
agency, diminutive specification is not brought to a lexical derivation by
an affix from the lexicon as lexical morpheme theories assume. This paper
examines this prediction against the evidence of affix synonymy. It shows
that the evidence of affix synonymy militates against the view that the
gender of new lexical (L-)derivatives in Indo-European languages origi-
nates specifically in suffixes.
Lexical morpheme theories (e.g., Lieber 1981, Selkirk 1982, Di
Sciullo —Williams 1987), assume that affixes are lexemes, direct associa-
tions of sound and meaning stored in the lexicon; hence they predict affix
synonymy as a regular aspect of the synonymy common among lexical
items. These theories predict that affixes combine with stems in essentially
the same fashion that lexical bases compound.
(2a) die [Fleisch [brühe]Ff.m]Fem 'meat broth'
b) die [Schön [heit]Pem]Fem 'beauty'
(2) represents a typical German compound and L-derivation in which the
rightmost constituent seems to transmit its category information to the
newly derived word. If the lexical morpheme assumption is valid and
suffixes project their category features in the same way as lexemes do:
(3a) The affix must be present to project or percolate its features.
b) Affix synonymy should be approximate, not absolute.
c) Synonymous affixes should, ceteris paribus, vary in their sub-
categorization.
(a) "Zero lexemes" are not found among compounds like those of
(4) — (5) such that the rightmost element of a compound must be omitted
without concomitant loss of meaning, e.g. das Fleisch 'meat' —• *die
Fleisch 'meat (beef) broth'. Colloquial forms like eine Weiß (= Weiß-
wurst), eine Brat (= Bratwurst) must be performative wordplay since the
head noun is stylistically optional, not grammatically omitted,
(b) Lexical synonymy is approximate, not exact.
(4) die Fleischbrühe die Fleischbouillon die Bohnensuppe
die Kalbsbrühe (die Kalbsbouillon) die Kalbssuppe
Suppe is distinguished from Brühe and Bouillon by its heavier ingredients.
Brühe also means 'gravy, sauce' besides 'bouillon'.
Finally, (c) while some synonyms of such semantic fields are sporadi-
cally associated with one gender, such associations are only occasional
(5)·1
(5) die Vorspeise das Zwischenessen das Nachtmahl
die Schultasche der Postsack der Sportbeutel
As Bloomfield (1933: 280) put it, "[t]here seems to be no practical criterion
by which the gender of a noun in German, French, or Latin could be
determined".
The separation hypothesis predicts something quite different. Since L-
derivation and affixation are independent, (a) derivation may operate
without affixation, thus generating zero-marked forms. Since separation
allows several affixes to mark an identical L-derivation, (b) it predicts
absolute synonymy among such forms. Finally, in cases of derivational
synonymy, we should (c) find instances of derivationally determined
gender which does not vary with suffix selection. Since derivation deter-
mines meaning and affixes are semantically empty, in most instances
derivation should determine gender, number, and diminution and these
categories should remain consistent across suffix markers.
To test the predictions of the empty morpheme entailment of the
separation hypothesis against the lexical morpheme hypothesis, this paper
examines two Indo-European L-derivations which are marked in each
language by several ostensibly synonymous suffixes. The purpose is to
discover whether suffixes or abstract derivation rules determine gender
among synonymous L-derivations.
162 Robert Beard
3. Two Indo-European derivational classes with

systematic synonymy
3.1. The western Indo-European deadjectival nominalizations

The first evidence that the separation hypothesis makes the correct
predictions is found in the abstract deadjectival nominals throughout the
western Indo-European languages. The recategorization of (6) (there is
no semantics) is not only absolutely the same for each language but
across languages as well.
(6) Italian German Russian French
la fals-itä die Krank-heit mjagk-ost' [F] la moit-eur

la facond-ia die Fähig-keit slep-ot-a [F] la differen-ce
la lucent-ezza die Naiv- ität tis-in-a [F] la modern-ite
la brav-ura die Finster-nis bel-izn-a [F] la faibl-esse
la stupid-aggine die Tief-e sin' [F] la just-ice
For the lexical morpheme position, the examples of (6) should be deri-
vationally identical with, or at least similar to, (4) —(5). But the synonymy
here is absolute, not the approximate synonymy of lexical items. More-
over, all the nouns of (6) bear feminine gender regardless of the suffix
used and regardless of its productivity. Now, if the gender of an L-
derivation is determined by that of the suffix, no reason exists for the
consistent association of femininity in (6). The consistency which we find
suggests that the rule, Adj —> N, itself, not the suffixes marking it, provides
the femininity. 2
One might say that the lexical representations of these suffixes coin-
cidentally associate feminine gender with nominalization and are fortui-
tously synonymous.
(7) -esse -eur -ite -ce -ice

+ Noun + Noun + Noun +Noun +Noun
+ Fem + Fem + Fem -l-Fem + Fem
Adj_ _ Adj_ Adj _ Adj - _ Adj_ _
This exit is blocked two ways: (a) French -eur and German -nis are not
always feminine and (b) gender is also assigned when no affix at all is
added.
German -nis is an unproductive nominalization suffix. French -eur is
highly productive as a masculine agentive marker.
(8a) DAS Erzeug-nis 'product(ion)'

Bekennt-nis 'confession'
Verhält-nis 'relation'
b) LE lis-eur 'reader'
fum-eur 'smoker'
chant-eur 'singer'
gag-eur 'bettor'
Should -eur and -nis be treated as homonyms, at least four problems
arise: (a) an enormous, unjustified proliferation of affixes, (b) contradic-
tion of phonological evidence, (c) circularity in the argument, and (d) no
basis for choosing between homonymy and polysemy. Affix proliferation
occasioned by theoretical homophony like (9) can hardly be exaggerated;
besides the masculine and feminine uses of French -eur, it also marks
adjectives: migrat-eur 'migratory'. If -eur represents a lexical morpheme,
it represents three. Beard (1985) discusses in detail the extent of such
multi-exponence across languages.
Phonologically, almost all such "homonymous" suffixes of (8) are
unitary and historically always have been. Without phonological support,
the argument for homonymy becomes circular. Differences in meaning
may not be used as an argument for postulating two independent lexical
items. An argument designed to explain two unexpected meanings of an
ostensibly single formant cannot postulate two formants from the as-
sumption of the independence of the meanings. Beard (1988: 31—34)
discusses in detail the impossibility of choosing between homophony and
polysemy as a theoretical solution to this quandary.
Theories based on the separation hypothesis posit at most one deri-
vation rule for each of -eur's three functions (denominal adjective, de-
verbal agentive, deadjectival nominal) and one suffix -eur with complex
conditions on insertion which allow it to be inserted in a variety of
conditions.
(9) + Adjective
+ Masculine
+ Noun
0 —* -eur j < + Agent
+ Masculine
+ Noun
4- Feminine
164 Robert Beard
Since the separationist approach operates on a single suffix -eur, it

provides a significant gain in avoiding the otherwise mysterious prolif-
eration of pseudohomophonous lexical entries. Even the exceptions to
the femininization of deadjectival nominals are better explained by empty
morphology. The failure of (9) to apply at all has no effect on the gender
of the derivation within this framework. Hence forms like German die
Tiefe 'depth' and Russian sw' [F] 'blueness' still receive regular gender
features though no morphological markings.
The consistent gender of nominals derived from color adjectives is
exceptionally neuter in German and masculine in French. (The German
exceptions exist alongside regular die Grau-heit, Blau-heit; die Bläue,
Schwärze.)
(10) DAS Gelb, Grün, Rot, Blau, Schwarz, Weiß
LE bleu, vert, rouge, blanc, jaune, noir
Whatever determines the gender of these nouns, it is not any affix. 3
Moreover, the language-specific factor conditioning the anomalous gen-
der of these adjectives is purely semantic, not morphological.
3.2. The Slavic locative nominalizations

The nomina loci of Serbo-Croatian, Polish and Ukrainian present osten-
sible problems for the empty morpheme entailment. The grammatical
gender of locative nominals in these languages is, in complete accord
with the lexical morpheme hypothesis, (a) always marked by phonolog-
ically real affixes which (b) represent all three genders and (c) have only
approximately similar meanings.
(11) knjig-a 'book' knjiz-n-ic-a [F] 'library'
raz 'rye' raz-ist-e [N] 'rye field'
gus-k-a 'goose' gus-in-jak [M] 'goose pen'
Each form in (11) represents a productive locative nominalization in
contemporary Serbo-Croatian. Assuming that each suffix, -nica, -iste,
-in-jak means 'location of N', the lexical morpheme hypothesis is osten-
sibly substantiated. However, closer examination demonstrates that even
where affixation varies widely, patterns of meaning-gender associations
persist which cannot be explained by affixes.
Unlike other Indo-European languages, these Slavic languages have
two nomina loci, an "in'Mocative, meaning 'place in which' and an "on"-
locative, meaning 'place on which'. The two derivations are related to
the division of the locative case into locative case functions with the same
two meanings, i.e., those marked by the prepositions na 'on' and u 'in'
(Beard 1981: 171—202). The verbs below are presented in their past-
tense stem form where this form serves as the base for locative nomin-
alizations; otherwise, the citation form is provided. /I/ is regularly replaced
by /o/ unless followed by a vowel.
(12) igral- 'play, dance' igral-ist-e 'playground'
igrao-n-ic-a 'dance hall, casino'
radil- 'work' radil-iste 'work site'
radio-n-ic-a 'workshop'
vezbal- 'exercise' vezbal-ist-e 'training field'
vezbao-n-ic-a 'gymnasium'
Assuming with Zepic (1970) and others that the "in'Mocative means
'small, enclosed place of N' and the "on'Mocative means '(large, open)
place of N', where the latter is the unmarked locative, the separation
hypothesis also predicts the data quite well. Since two derivations are
necessary, either the derivations or their suffixes, e.g., -n(ic)a and -iste,
respectively, could assign gender appropriately. However, the "in'Moca-
tive is marked by a fairly wide range of affixes — all feminine.
(13a) pepeo 'ash' pepel-jar-a 'ashtray'
pil-a 'saw' pil-an-a 'sawmill'
racun-i- 'calculate' racun-ic-a 'arithmetic notebook'
rafin-ir-aj- 'refine' rafin-er-ij-a 'refinery'
The examples show how derived locative nominals associated with the
sense of 'inside', i.e., referring to an enclosed space, all exhibit feminine
gender. Here, even the idiomatic variations (13a) reveal feminine gram-
matical gender. Most striking, again, are those "in'Mocatives which, like
the French examples in -eur, display a masculine agentive suffix, -ar,
declined in the feminine declension I paradigm, that is, -ara, -are, -ari,
..., instead of the masculine, declension II paradigm -ar, -ara, -aru, etc.4
(13b)
mes-o 'meat' mes-ar 'butcher' mes-ar-a, mes-ar-n(ic)a 'meatshop'
pek- 'bake' pek-ar 'baker' pek-ar-a, pek-ar-n (ic) a 'bakery'
knjig-a 'book' knjiz-ar 'bookseller' knjiz-ar-a, knjizar-n-ic-a 'bookstore'
The femininization cannot percolate from any suffix here for the only
suffix is -ar, which is no more feminine than masculine.
The exceptions to this pattern of feminine markers on "in'Mocative
nominalizations themselves form a subregular pattern. (14) exhibits an
166 Robert Beard
array of suffixes, -ac, -n-jak, -in-jak, -ar-n-ik, -n-ik (the masculine variant
of -n-ic-a) plus an unusual occurrence of the agentive suffix -or, kokos-
ar 'chicken coop', marking those "in'Mocatives based on animate nouns.
The Slavic languages distinguish animacy among masculine (and plural)
nouns at the inflectional level; that locative nominalizations are masculine
is hence perfectly consistent. Only the empty morpheme entailment ex-
plains this interlevel consistency.
(14) svinj-a 'hog' svinj-ac 'pig sty'
zee 'rabbit' zec-ar-nik, zec-in-jak 'rabbit warren'
zab-a 'frog' zab-n-jak, zab-ljak 'frog pond'
Exceptions to this subregularity follow the inanimate pattern in exhib-
iting the default, feminine gender: ovc-ar-a 'sheep pen', koz-ar(ic)a 'goat
pen', konj-us-n-ic-a 'horse stable', krav-ar-a 'cowshed', pset-ar-n(ic)a
'kennel'. Milojevic (1934), however, points out that terms like konj-ar-
nik often occur in speech and other masculine forms with -ar-nik are
acceptable to him: ovc-ar-nik 'sheep pen', koz-ar-nik 'goat pen', krav-ar-
nik 'cow shed'. We can only conclude that masculinity is now productively
associated with "in'Mocatives derived from animate nouns according to
the animacy-masculine association in inflectional usage. The gender con-
flict in these forms no doubt arises from the conflict between marked
and unmarked genders for nomina loci.
4. Other determiners of gender
It does not follow, of course, from the fact that derivation rules can
determine gender that they always do. The diminutive derivation generally
does not determine gender (cf. Italian il bimb-o : il bimb-ett-o, la bimb-
-a : la bimb-ett-a); however, in German it does: der Bart : das Bärt-chen,
die Haut : das Häut-chen; das Mäd-el, Mägd(e)-lein, Mäd-chen. Such
variation is the result of language-specific factors which do not impinge
on the crucial issue raised by separation.
The Pashto deadjectival nominals seem to represent another violation
of the empty morpheme entailment.
(15) äräm 'quiet' äräm-i [F] 'quietness'
sur 'red' sur-tiä [F] 'redness'
arzan 'cheap' arzan-tob [M] 'cheapness'
klak 'firm' klak-välay [Μ] 'firmness'
"Gender" here refers strictly to grammatical gender, and in Pashto

grammatical gender is determined phonologically. Pashto has only two
cases in two nominal declensions. If any word ends on any vowel except
/u/, it receives the "feminine" oblique variant; otherwise, it receives the
"masculine". As long as the morphological component supplies all the
derivational affixes prior to inserting inflectional markers, the proper
oblique desinence can be assigned without reference to derivation. The
important point is that just as assignment of gender varies across lan-
guages in the diminutive, so does it vary with the deadjectival nominal-
ization across languages. In Pashto, the deadjectival nominalization does
not assign grammatical gender; gender is determined on purely phono-
logical grounds.
5. Conclusion
It is a metatheoretical given in science that of two competing theories,

the one which explains the data with the structure of the theory itself,
rather than with ancillary subtheories, is the preferable one. While lexical
morphemes can explain most of the phenomenon of gender-specific lexical
derivations, ancillary theories of semantic analogy, theoretical homo-
phony, conversion must be introduced or extended to do so. This slowly
erodes the lexical morpheme assumption and/or moves it toward the
separation hypothesis. Whether ancillary subtheories accrete to the lexical
morpheme hypothesis or the crucial, definitional distinctions between the
two positions disappear, lexical morphology eventually loses its appeal.
The separation hypothesis with its empty morphemes, however, explains
the limited range of morphological phenomena explained by the lexical
morpheme, plus nonaffixational, derivationally determined gender such
as the animacy transparency of "in'Mocatives, without any theoretical
accretions.
Notes
1. Zubin —Kopeke (1984, 1986) argue that words referring to "unimageable" superordinate
categories tend to be neuter, e.g., das Ding, das Gut, das Stück vs. der Tisch, die Tür.
However, their arguments (a) are necessary but not sufficient, (b) are based on converse
accidence, underestimating synonyms like die Sache, die Ware, and (c) ignore some
massive regularities which are counterexamples (das Fräulein, Mädchen) while depending
168 Robert Beard
heavily on others, e.g., das Kraut, Unkraut, Heilkraut, Würzkraut (p. 160) as examples
of independent unimageable superordinate categories.
2. This is not unexpected within the lexeme-morpheme based morphology of Beard (1988).
That article explains that the derivation rules are much more regular than affixation
rules, probably because they are universal, while affixation rules are locally determined
by specific languages.
3. The same is true of a handful of absolute French exceptions (e.g., le chaud 'warmth', le
froid 'cold') though not of the dozen or so Slavic exceptions (e.g., bogat-stvo 'wealth',
zdorov-'e 'health'). Here the logical gender, neuter, may be accounted for by affixation
but only exceptionally. Since masculine is the default (unmarked) gender in French, and
all zero marked derivations are generally masculine, lexical morphologists might argue
that the lexicon automatically assigns masculine to derivations not marked by gender-
bearing affixes. However, these derivates are still lexically excluded from affixation rules
and whether they are excluded from affixation by their exceptional masculine gender or
their masculine gender results from their exceptional null morphology is presently a
moot point.
4. Some grammars suggest that forms like those of (15b) are derived from the corresponding
agentives, e.g., mes-ar 'butcher' : mes-ar-a 'butcher shop'. This is unlikely, however,
since (a) agentives are not always available for locatives (ucio-nica 'classroom') and (b)
they are often irrelevant when available, e.g., igl-ar 'needle-maker': igl-ar-a 'pin cushion',
ovc-ar 'shepherd' : ovc-ar-a 'sheep pen, barn'. Such a derivational path would make the
lexical morpheme position even more difficult since (a) /-a/ is not a derivational affix
but a nominative singular declension I desinence and (b) -ar is a masculine agentive
suffix.
References
1982 "Where's morphology?", Linguistic Inquiry 13: 571-612.
Beard, Robert
1981 The Indo-European lexicon: a full synchronic theory ( = North-Holland
Linguistics Series 44) (Amsterdam: North-Holland).
1985 "Is separation natural?", Studia gramatyczne 7: 119 — 134.
1987 "Morpheme order in a lexeme/morpheme-based morphology", Lingua 72:
1-44.
1988 "On the separation of derivation from affixation; toward a lexeme/mor-
pheme-based morphology", Quaderni di Semantica 9: 3 — 59.
1989 Review of Di Sciullo - Williams (1987). Lingua 77: 81 - 9 0 .
Bloomfield, Leonard
1933 Language. (New York: Holt).
Boskovic, G.
1934 "Povodom clanka G. A. Paunova" [On the occasion of an article by G. A.
Paunov], Νas jezik 1: 275 — 278.
Chomsky, Noam —Morris Halle
1968 The sound pattern of English. (New York: Harper & Row).
Di Sciullo, Anna Maria —Edwin Williams
1987 On the definition of word. (Cambridge, MA: MIT Press).
Jackendoff, Ray
639-671.
Lieber, Rochelle
1981 On the organization of the lexicon. (Bloomington: Indiana University Lin-
guistics Club).
Matthews, Peter H.
1972 Inflectional morphology. (Cambridge: Cambridge University Press).
Milojevic, V. Z.
1934 "Jedan predlog za nekoliko naziva" [A proposal for some terms], Νas jezik
2: 209-210.
Moortgat, Michael —Harry van der Hulst
1981 "Geinterpreteerde morfologie", Glot 4: 179-214.
Moskovljevic, Milos S.
1934 " Opkladionica' i tim povodom ο imenicama na -ionica, -aonica" [Opkla-
dionica, and on this occasion on substantives in -ionica, -aonica], Nas jezik
2: 7 6 - 7 7 .
Pounder, Amanda
in press "The semantic organization of word-formation paradigms and diachrony",
in: Proceedings of the International Morphology Conference at Veszprem ( =
Acta Linguistica Academiae Scientiarum Hungaricae 36).
Roeper, Thomas
1987 "The syntax of compound reference" [XIV International Congress of Lin-
guists, Berlin, August 1 0 - 1 5 , 1987],
Selkirk, Elizabeth
1982 The syntax of words ( = Linguistic Inquiry Monograph 7). (Cambridge,
MA: MIT Press).
Szymanek, Bogdan
1985 English and Polish adjectives: a study in lexical word-formation. (Lublin:
Wydawnictwo Katolickiego Uniwersytetu Lubelskiego).
Wood, Mary
1985 "The semantics and dynamics of derivation", in: G. A. J. Hoppenbrouwers
et al. (eds.) Meaning and the lexicon (Dordrecht: Foris), 49 — 54.
Zepic, Stanko
1970 "Izvedenice sa sufiksima za tvorbu mjesnih imenica (nomina loci)" [Inves-
tigation of suffixes forming place-names], Jezik 18: 83—90, 195 — 109.
Zubin, David —Klaus-Michael Kopeke
1984 "Affect classification in the German gender system", Lingua 63: 41 —96.
1986 "Gender and folk taxonomy: the indexical relation between grammatical
and lexical categorization", in: C. Craig (ed.) Categorization and noun
classification (Philadelphia: Benjamins), 139 — 180.
The benefits of morphological classification: on
some apparently problematic clitics in Modern
Greek*
Brian D. Joseph
1. Introduction and theoretical background
The act of classification is essential to linguistic investigation at all levels

of analysis. With regard to morphological classification, a variety of
taxonomic parameters are possible, such as, for words, content versus
function, or for affixes, inflectional versus derivational. Without denying
the importance of such taxonomies, another more basic one for mor-
phology can be adduced, namely a classification into different types along
a scale encompassing words, as the typical free or complex morpheme,
affixes, as the typical bound morpheme, and clitics, somewhere in-between
as a quasi-word/quasi-affix.
This three-way division of morphological elements has a special sig-
nificance, for precise classification is necessary if cross-linguistic gener-
alizations and putative universals concerning these elements are to have
any empirical content, as Zwicky (1985) has emphasized. Particularly
troublesome here are claims concerning the behavior of clitics, for their
ambiguous status between words and affixes represents an implicit chal-
lenge to linguistic theory. Nonetheless, several generalizations about clitics
have been proposed in recent years, and these claims need the benefit of
precise classification in order to be tested adequately.
* This work was supported by a Grant-in-Aid from the College of Humanities of the Ohio
State University, and by a Fulbright Research award that enabled me to spend time in
Greece collecting data and consulting with colleagues there. I would like to thank T.
Christides and the other linguists at the University of Thessaloniki and my colleague A.
Zwicky for stimulating discussion on these topics; A. Aikhenvald, V. Bubenik, B. de
Chene, C. Hagege, R. Janda, E. Petrounias, and E. Ternes all contributed useful comments
at the presentation of this paper at the Third International Morphology Meeting in
Krems, Austria, in July 1988.
172 Brian D. Joseph
Two examples deserve mention here, for the facts to be considered

below bear on their viability as linguistic universals. First, Kaisse (1982)
has proposed, as a modified version of the generalization widely known
as Wackernagel's Law, that S' ( = "S-bar") clitics must occur in second
position within their clause. Second, Zwicky (1987), following Klavans
(1983) and others, has claimed that there are no endoclitics, i.e., clitics
that are positioned within the morphological unit defined by a word (as
if *Bo-'ll-b come were an acceptable variant of Bob'U come, with the clitic
variant 7/ of the future auxiliary will)}
In order to carry out the classification necessary for testing such
hypotheses, a set of criteria for determining categorial status is essential.
One especially cogent set are those in Zwicky (1985, 1987) and
Zwicky — Pullum (1983), for they are both internally consistent and de-
rivative from the overall architecture of the highly modular theory of
grammar, sometimes referred to as the "interface program", assumed in
these works. 2
The most relevant of the "Zwicky criteria" for the present discussion
are given in (1); (la—c) distinguish affixes from nonaffixes (i.e., clitics
and words), while (Id —e) distinguish words from nonwords (i.e., clitics
and affixes):
(la) Selectivity in combination (affix: high selectivity; nonaffix: low
selectivity)
b) Idiosyncratic behavior (affix: shows idiosyncrasies; nonaffix:
shows few or none)
c) Parallel to morphophonological process (affix: such parallels
exist; nonaffix: none)
d) Ordering (nonword: strictly ordered; word: some degree of free
ordering)
e) Phonological dependence (nonword: dependent; word: independ-
ent)
The overall thrust of these criteria is that affixes are characterized by
a high degree of idiosyncrasy in their realization and behavior, while
nonaffixes, i.e., clitics and words, show a high degree of regularity and
predictability in realization and behavior. Zwicky's criteria are relative
ones, so that a collective positive indication for affixal status need not
be taken to indicate an absolute classification, especially if the criteria
define a continuum between words and affixes. On the other hand, an
absolute interpretation is possible if one holds that affix, clitic, and word
represent not a continuum but three discrete classificatory assignments.
The benefits of morphological classification 173
The nature of this classificatory scale is beyond the scope of this paper,
but it should be clear that although the absolutist view of the criteria is
adopted here, the relativist interpretation is available for analysts who
are so inclined.
With these criteria in place, the two hypotheses about clitic behavior
mentioned above — second position for S-bar clitics and the ban on
endoclitics — can be explicitly tested. More particularly, apparent coun-
terexamples to these claims provided by some facts from Modern Greek
can be subjected to rigorous testing via Zwicky's criteria.
2. The Greek facts
The problems posed by the Greek facts both center on the finite verbal
complex, a unit consisting of the inflected verb plus various elements
traditionally called "clitics", in the case of the weak object pronouns
(1SG ACC me, 1SG GEN mu, 3SG ACC NTR to, etc.), and "particles",
in the case of the modal markers na and as, the future marker θα, and
the negation markers den and min. These elements modify the inflected
verb for tense, mood, negation, and argument structure (see Joseph 1985
for some discussion). The general schema for the verbal complex is
sketched in (2), with some examples of possible expansions given in (3):
(2) (na)- (min)- (θα)- (WEAK PRONOUNS)- VERB

as Sen [me, mu, etc.]
MOOD NEGATION TENSE ARGUMENT.MARKER HEAD
(3a) Sen θα ta fäo

NEG FUT 3PL.ACC.NTR eat/1 SG
Ί won't eat them'
b) den tus to dosame
NEG 3PL.GEN 3SG.ACC.NTR gave/IPL
'We didn't give it to them'
The classification of these elements is crucial for the claims concerning
clitics noted above, for if any of the interior ones are true clitics (note,
e.g., the traditional label of "clitic" for the weak pronouns) while the
exterior ones are affixes, there would be a clear case of endoclitic posi-
tioning within the bounds of a stem-plus-affix word unit. Similarly, the
negation markers, whose distribution correlates with verbal mood — den
174 Brian D. Joseph
with indicative mood, min with subjunctive — pose a problem for the
modified Wackernagel's Law second-position generalization. To focus
just on the indicative negator, discussed further below,3 it need not occur
in second position (note (4a) with sentence-initial Sen) even though its
scope is demonstrably sentential, 4 in that it determines the selection of
the negative polarity indefinite pronoun kanenas as subject, as opposed
to the nonnegative käpjos:
(4a) den irOe kanenas / * kanenas irOe
NEG came/3SG no-one/NOM
'No one came'
b) käpjos irOe / *δen irOe käpjos
someone/NOM came/3SG
'Someone came'
It turns out, however, that there is a solution to these problems. A
close examination of the properties of the morphemes in question — the
weak pronominals and the indicative negator — with regard to the Zwicky
criteria allows for an analysis of these elements as more affixal in nature
and thus not clitics in the strict sense that these criteria now permit.
Crucial to this solution is the ability to classify the problematic elements
on a principled basis as being outside the domain of these generalizations.
In that way, both the negator den and the weak pronominals become
irrelevant for these hypotheses, and so do not constitute counterexamples
to them.
3. Evidence for the affixal analysis
The affixal analysis that provides the key to preserving the above-men-
tioned generalizations is well-supported by facts from both Standard
Modern Greek and various dialects. The dialect evidence cannot of course
bear on the analysis in Standard Modern Greek per se. However, it does
provide "pan-Hellenic" support for the affixal analysis by lending typo-
logical credence; the dialects, as varieties of Greek, are closer to Standard
Greek in all respects — lexically, structurally, grammatically — than any
other language.
With regard to the indicative negator Sen, it is clear first that it must
be a nonword. It is phonologically dependent, unable to stand alone, for
instance as a negative response word ('No!') or as a prohibitive utterance

('Don't!'). Also, as is characteristic of nonwords, it shows strict ordering
with respect to elements it combines with, and must be leftmost in the
verbal complex:
(5a) den θα vlepo I * θα den vlepo j * θα νΐέρο den
N E G F U T see/1 SG
Ί will not be seeing'
b) den ton vlepo j *ton den vlepo / *ton vlepo den
N E G him/ACC see/1 SG
Ί don't see him'
Other evidence suggests that of the nonword possibilities, den shows
some idiosyncratic behavior characteristic of affixes. For instance, den is
highly selective in its combinatory possibilities, occurring with only in-
dicative finite verbs (6a), but not subjunctive finite verbs (6b) or nonfinite
verbs, (6c):5
(6a) den vlepo Ί don't see (PRES)'
den evlepa Ί wasn't seeing (IMPRF)'
den ida Ί didn't see (AOR)'
den θα vlepo Ί won't see (FUT)'
b) *den na νΐέρο j *na den vlepo 'that I not see (SUBJUNC)'
c) *den vlepondas 'not seeing/ACT.PPL'
*den des 'Don't see/IMPV.SG'
*den deste 'Don't see/IMPV.PL'
Also, den shows a semantic idiosyncrasy in its use in the expression den
mu les 'tell me ...', which introduces an inquiry without a trace of negative
meaning, even though its literal meaning is "you don't tell me". While
the non-negative meaning of den mu lis may lie partly in its pragmatics,
nonetheless, in this expression, den idiosyncratically does not have its
usual negative value, and so prevents a fully compositional meaning for
this phrase. The other criteria — parallelism with a morphophonological
process and morpho(phono)logical idiosyncrasies — provide no relevant
evidence bearing on the classification of den. Thus, all the available
evidence from Standard Modern Greek points in the direction of an
affixal realization of indicative negation.
Evidence from the Tsakonian dialect of the southwest Peloponnesos
(see Pernot 1934) leads to a similar conclusion. The Tsakonian indicative
negator is 6, but it is parallel to Standard Greek den, and derives from
the first part (ou) of the Ancient Greek form oudin 'in no wise; not at
176 Brian D. Joseph
all' (composed of ou 'not' + άέ 'but' + hin 'one/NTR') which yielded

δ en.
Two pieces of data are relevant here. As (7) shows, Tsakonian has
innovated a virtual negative auxiliary through the fusion of ό with the
verb 'to be', which is used in the formation of all present and imperfect
tenses periphrastically in this dialect:
(7)
1SG oni 'I'm not' (< ό + eni) 1PL ome 'we're not' (< ό + eme)
2 osi 'you're not' (< ό + esi) 2 othe 'you're not' (< ό + ethe)
3 oni 'he's not' (< ό + eni) 3 Mi 'they're not' (< ό + ini)
The fusion gives the appearance of an inflected negative verb, and if
synchronically analyzable as ό plus 'be', would involve some morpho-
phonological idiosyncrasies. At least the 3PL uni, for instance, would
have to have [ύ] as the outcome of /ό/ + /ί—/ rather than the synchronically
regular [o]. Thus, if not inflectionally affixal because of the fusion,
Tsakonian ό would nonetheless present a synchronic idiosyncrasy, i.e., a
characteristic of affixes.
Also, for certain forms in (7), irregular truncated variants are possible,
specifically: 1SG öi, 3SG ό, and 3PL ύΐ, and at some point in the history
of Tsakonian, the old 3SG oni was separated off from the paradigm and
lexicalized as the emphatic 'surely not!' (3SG oni is a reformation). Such
specialization of individual forms is characteristically found with affixal
formations, according to Zwicky — Pullum (1983), and not with those
involving clitics, which are syntactically and semantically transparent.
Thus, from this evidence, Tsakonian, and by extension, Greek in general,
has affixal indicative negation.
With regard to the weak pronominal elements, a similar case can be
made for analyzing them as affixal elements. The relevant facts, discussed
more fully in Joseph (1988, in press), are sketched here.
As with den, it is clear that the weak pronominal forms cannot be
independent words. They are phonologically dependent, unable to occur,
for instance, as one-word answers; the corresponding strong form must
instead occur:
(8) pjon ίδβ ο jänis ? emena
whom/ACC saw/3SG the-John/NOM me/ACC.STRONG
/ *me
me/ACC.WEAK
'Whom did John see? Me'.
In addition, they are strictly ordered with respect to other elements they
combine with:6
(9) den ton νΐέρο
NEG him/ACC.WEAK see/1 SG
/ *ton Sen vlepo / *ton vlepo den j *Sen vlepo ton
Ί don't see him'.
On other criteria, moreover, the weak pronouns test out as affixes,
and not clitics.7 They show selectivity of combination, in general occurring
only with verbs, though accusatives also occur with the adverb kalos
'well' in the meaning 'welcome' (e.g., kalos ton 'welcome to him!') and
genitives occur with some prepositions (e.g., brostä mu 'near me') and a
few adjectives (e.g., monos mu 'on my own'). The existence of these
occasional nonverb hosts might suggest lesser selectivity, but these com-
binations are actually extremely limited, among adverbs, for instance,
restricted only to kalos. Also, the weak pronouns show various combi-
natory gaps, a type of selectivity recognized by Zwicky — Pullum (1983)
as typical of affixes, e.g., the absence of first person genitives with second
person accusatives (see Warburton 1977):
(10) se dosane
me/GEN.WEAK you/ACC.WEAK gave/3PL
'They gave you to me'.
In addition, the weak pronouns show several idiosyncrasies. The 2SG
genitive /su/ combined with third person weak accusative forms, which
all begin with /t-/, anomalously undergoes a contraction to [s], e.g., /su
+ to/ 'to you itT —* [sto], even though no regular elision process affecting
/u/ occurs in that context in standard Greek. Also, for some speakers,
the initial /t-/ of the weak third person pronouns may be voiced in
combination with the future marker Θα and the modal na, e.g., /0a to
käno/ —> [0a do käno] Ί will do it', even though intervocalic voicing is
not a regular rule of Greek phonology. Semantic and morphosyntactic
idiosyncrasies are found in the occurrence of weak pronouns in idioms
with otherwise intransitive verbs, e.g.:
(11) ρύ θα tin pesume
where FUT her/ACC.WEAK fall/1 PL
'Where will we go?' (literally "*Where will we fall her")
Not only is the weak pronoun semantically empty, with no interpretation
as an argument, but its co-occurrence with an intransitive verb is entirely
irregular too.
178 Brian D. Joseph
Given this behavior, from the perspective provided by the Zwicky

criteria, one has to conclude that the weak pronominals of Standard
Modern Greek, despite their traditional label, are best treated as affixes.
Especially strong supporting evidence for this conclusion comes from
several dialect facts. In many Northern dialects (e.g., of Thessalia, Mac-
edonia, etc., see Thavoris 1977), the first person singular weak pronoun
m(e) occurs inside of an indisputable affix, the plural imperative ending
-ti, as in (12):
(12)
a)pe-m-ti [ = t e l l / I M P V +me + PL] '(You/PL) tell me!'
b) δό-m-ti [ = give/IMPV + me + PL] '(You/PL) give (to) me!'
c)/έη-ηιέ-ίϊ [= bring/IMPV + me + PL] '(You/PL) bring (to) me!'
Such placement is unexpected for clitics, for it amounts to the otherwise
unattested endoclisis. This interior placement, however, is not unusual if
m(e) is an affix.
Also, in Tsakonian, a singular weak pronoun regularly occurs as an
optional but preferred accompaniment to a plural weak pronoun, as in
(13):
(13) tsi m epetsere namu
what me/ACC.WEAK said/2SG us/ACC.WEAK
'What did you say to us?' (literally: "What did you tell me to
us?")
This construction exhibits an anomalous non-agreeing "doubling" of the
plural weak pronoun, with semantically idiosyncratic null interpretation
of the weak pronoun, and a syntactically idiosyncratic agreement pattern
(for person only and not, as with doubling of a strong pronoun by a
weak pronoun, for person and number).
Finally, the dialect of Tirnavo in Thessalia (Tzartzanos 1909) shows
the processual realization of an entire weak pronoun (criterion lc above).
In this dialect, the feminine accusative singular weak pronoun, which is
[tn] before vowels, is realized with verbs having an initial dental stop or
affricate simply as voicing on the verb's initial consonant. For example,
contrasts such as those in (14) occur:
(14a) [täraksi] 'he disturbed' vs. [däraksi] 'he disturbed her'

b) [tsakusi] 'he caught' vs. [dzakusi] 'he caught her'
c) [t'imsi] 'he pinched' vs. [dzimsi] 'he pinched her'
Processual realization — here by the process of voicing — is expected in

Zwicky's system only for affixes, not for clitics, so Tirnavo offers clear-
cut evidence in this framework for the affixal analysis under considera-
tion. 8
4. Conclusion
What emerges from this brief discussion is that these elements from
Modern Greek, which at first glance seem so problematic for claims
regarding clitics universally, turn out to present no problem, once a
stringent set of classificatory criteria is applied to the data. Rather than
being clitics, they are instead well-behaved affixal elements and as such
are irrelevant to the hypotheses discussed at the outset. A ban on endo-
clisis and second-position for S-bar clitics can therefore be maintained as
viable universals, given present knowledge. More examination of the
Modern Greek verbal complex is surely needed, though the preliminary
indications 9 are that it is a word-level unit of a verbal stem plus clitic
modal markers, tense and negation prefixes, affixal object markers, as-
pectual suffixes, and endings for the person and number of the subject.
Thus Zwicky's principles of morphological classification have enor-
mous benefits for research in this area. To the extent, then, that the
Zwicky criteria lead to satisfying results in dealing with these seemingly
difficult facts, the analyses presented here lend considerable credence to
the overall framework and approach to morphological classification em-
bodied in the interface program.
Notes
1. Some apparent cases of endoclitics have been reported for various languages, but closer
inspection in each case yields a better, nonendoclitic, analysis (e.g., Nevis 1984, Klavans
1983).
2. See Zwicky (1990) for an overview of the theory.
3. I am deliberately ignoring the potential counterexample posed by the modal negator
min, largely for reasons of space; although it is a safe assumption that the analysis of
den given below can carry over into the classification of min, the latter presents other
analytic problems that go beyond the scope of this paper (see Joseph—Janda 1987).
4. Similar problems for the modified Wackernagel's Law would arise, of course, if any of
the M O O D and TENSE markers in (2) are sentential in scope and are clitics (so also
with the modal negator min, see note 3).
180 Brian D. Joseph
5. On Greek imperatives as nonfinite, see Joseph (1985).

6. The weak pronominals are regularly enclitic to nonfinite verbs (e.g., vlepondäs ton 'seeing/
ACT.PPL him'). Since this ordering difference is predictable, it does not contravene the
import of the strict ordering criterion for affixhood. Similarly, even in dialects in which
weak pronominals regularly follow even finite verbs, strict ordering is still observed.
7. Note also that, based on vowel contraction, stress assignment, and voicing assimilation,
Malikouti —Drachman and Drachman (1988) conclude that Greek weak pronouns
behave like verbal prefixes not free words, so that, phonologically, they do not require
a separate class label of "clitic".
8. While synchronically this processual realization could derive via /tn + täraksi/ —>
tndaraksi —• tdaraksi —• [däraksi], in Zwicky's system only a morpheme's surface
realization is relevant; thus the contrasting pairs in (14) are significant.
9. This work is part of a larger study of the morphosyntax of the Greek verbal complex,
the full results of which have been made public in various lectures in the United States
and Europe in 1987 — 1988 and will appear in print in the near future.
References
Joseph, Brian D.
1985 "Complementizers, particles, and finiteness in Greek and the Balkans",
Folia Slavica 7.3: 390-411.
1988 "Modern Greek pronominal affixes: the case against clisis", CLS 24:
203-215.
in press "Diachrony and linguistic competence — the evidence from morphological
change", in: University of Chicago Special Publications in Linguistics vol. 2
(Papers from the Conference on the Theory and Practice of Historical
Linguistics).
Joseph, Brian — Richard D. Janda
1987 "On generalizations as linguistic categories: Of rules, redundancy-rules,
meta-rules, rule-conspiracies, and rule-constellations" [Paper presented at
the Conference on Linguistic Categorization, University of Wiscon-
sin—Milwaukee, April 1987].
Kaisse, Ellen
1982 "Sentential clitics and Wackernagel's law", Proceedings of the 1st West
Coast Conference on Formal Linguistics'. 1 — 14.
Klavans, Judith
1983 "The morphology of cliticization", Papers from the Parasession on the
Interplay of Phonology, Morphology, and Syntax (CLS): 103 — 121.
Malikouti-Drachman, Angeliki —Gabereil Drachman
1988 "On Greek clitics and lexical phonology" [Paper presented at the Sixth
International Phonology Meeting (Krems), July 1988],
Nevis, Joel
1984 "A non-endoclitic in Estonian", Lingua 64.2/3: 209-224.
Pernot, Hubert
1934 Introduction a l'etude du dialecte tsakonien. (Paris: Societe d'Editions "Les
Belles Lettres").
Thavoris, Andonis
1977 "Morfolojikä merikön iöiomäton tis öitikis make8onias" [The morphology
of some Western Macedonian dialects], First Symposium for Northern Greek
Linguistics (Thessaloniki: Institute for Balkan Studies), 75 — 95.
Tzartzanos, Ahilefs
1909 Pragmatia peri tis sinxronu Oessalikis Sialektu [A treatise on the contem-
porary Thessalian dialect] (Athinai: P. A. Peträku).
Warburton, Irene P.
1977 "Modern Greek clitic pronouns and the 'surface structure constraints'
hypothesis", Journal of Linguistics 13: 259 — 281.
Zwicky, Arnold M.
1985 "Clitics and particles", Language 61: 2 8 3 - 3 0 5 .
1987 "Suppressing the Zs" Journal of Linguistics 23: 133 — 148.
1990 "Inflectional morphology as a (sub)component of grammar", this volume,
217-236.
Zwicky, Arnold M. —Geoffrey K. Pullum
1983 "Cliticization vs. inflection: English η 'Γ, Language 59: 5 0 2 - 5 1 3 .
Case markers and pragmatic strategies: Romanian
clitics
Maria Manoliu-Manea
It has often been claimed that Romanian is a highly deviant Romance

language, characterized by a great deal of redundancy, especially in the
area of noun-phrases. Alain Guillermou (1953: 51), for example, empha-
sizes that Romanian "entoure le substantif d'un luxe de procedes deter-
minants dont certains peuvent paraitre inutiles" (see also Seidel-Slotty
1940: 142). Among these intriguing phenomena, the agglomeration of
attribute markers and the combination of clitics, prepositions, and suffixes
for marking indirect or direct objects have been a controversial matter
in Romance linguistics in general and in Romanian linguistics in partic-
ular. Let us consider the following examples:
(a) multiple object markers:
(i) direct objects:
(1) l- α väzut pe copil,
him-ACC has-he seen on child-ACC
'he saw the child' 1
(ii) indirect objects:
(2) i -a spus mamei
her-DAT has-he told mother-DAT-the-DAT
'he told (my) mother'
(b) multiple attribute markers:
(3) i -am dat rochia alba
her-DAT have-I given dress-the white
a Mariei
POSS-CL. Mary-GEN-the-GEN
Ί gave her Mary's white dress'
In the present paper I intend to bring evidence in favor of the hypothesis
that: (a) There is at least one context in which the items under discussion
184 Maria Manoliu-Manea
are not redundant, but serve as case markers, (b) Even when they are
copies of NP's in the same sentence, clitics preserve some of their endo-
phoric function, and as such they may become means of discourse
strategies.
1. Noun declension
1.1. The inherited case markers

The typological pattern of Central Europe, rich in nominal inflections,
can account for the preservation of Latin case markers, if they were
different, after the replacement of the genitive markers by the dativus
adnominalis or by prepositional markers (see Manoliu-Manea 1985):
(4) Latin first declension: casa 'little house'
Case Lat. Vulg.Lat. (East) Rom.
NOM casa casa casä
GEN casae case case
ACC casam casa casä
ABL casa casa casä
but
(5) Latin second declension: lupus 'wolf
NOM lupus lupu lup
GEN lupi lupojad lupu- lup
DAT lup ο lupo lup
ACC lupum lupu lup
As (4) and (5) show, only feminine nouns have inherited two case markers,
while the other nouns have not. New case markers have been sought in
order to replace the ones that have been lost.
1.2. The "new declension"

Pronominal noun determiners, which have better preserved their inflec-
tional structure in all Romance areas, were among the first candidates.
Old Romanian has both proclitic and enclitic definite articles and so has
contemporary Romanian. The construction which is often invoked as
Romanian clitics 185
being the key-context in the development of the enclitic definite article is

Noun-Adjective (see Graur 1967), for example:
(6) Lat. homini Uli bono dixit 'he said to the man, the good one',
which, in fact, rests upon a pragmatic device belonging to the area of
indexicality, namely an additional identifying procedure.
1.2.1. The transformation of a demonstrative pronoun into an article,

a general Romance phenomenon, has also offered the means of marking
the syntactic function of the noun, since, in Romanian, noun determiners
agree in gender, number, and case with the head-noun. Although the
definite article is not the only noun determiner which has this function,
in certain cases the definite article is the sole bearer of the case marker:
(7) cas -a copil -ului
house -the child -the-GEN-SG
'the child's house'
The inherited case endings have favored the generalization of postnominal
definite articles playing the role of case markers. With invariable common
nouns the clitic definite article has also become a number marker:
pui -ul
chicken -the-NOM/ACC-SG
pui -ului
chicken -the-GEN/DAT-SG
pui -ule\2
chicken -the-VOC-SG
pui -i
chicken -the-NOM/ACC-PL
pui -lor
chicken -the-GEN/DAT/VOC-PL
In Romanian grammars this inflectional pattern is called declinarea ar-
ticulatä, i.e., 'the articulated declension'.
1.2.2. The dative/genitive forms of the definite article have different

distributional patterns according to the opposition between common and
proper nouns, on the one hand, and between masculine and feminine
gender on the other. The definite article always precedes masculine proper
nouns:
(9) i- am spus lui Ion

him-DAT have-I said the-DAT John
Ί told John'
but follows common nouns and feminine proper nouns:

(10) copilului
child -the-DAT
'to the child'
(11) Mari-e -i
Mary-GEN/DAT -the-gen./dat
'Mary's/to Mary'
This distribution is probably due to the fact that, when co-occurrent with
proper nouns, the definite article loses its anaphoric value and becomes
a mere case marker. As it has been shown in 1.1, masculine nouns have
not preserved case endings which could attract the new ones, i.e., the
cliticized demonstratives. Feminine proper nouns have preserved a post-
posed case ending and consequently the attraction under discussion took
place. Compare:
(11a) Old Roman ei Marie

the-GEN/DAT-FEM-SG Mary-GEN/DAT
b) Marie -i
Mary-GEN/DAT -the-GEN/DAT-FEM-SG
'Mary's'
In contemporary spoken Romanian, the masculine genitive-dative
marker lu(i) occurs also before feminine proper nouns with an atypical
feminine ending (that is -i or a consonant):
(12) i-am spus lui Cici / lui Carmen
I said to Cissy / to Carmen
and even before common nouns (substandard):

(13) in locu lu clopotu ästa
in place-the-ACC the-GEN bell-the this-MASC-SG,
'instead of this bell' (see Graur 1968: 302)3
In this case, the noun takes the nominative/accusative form.
2. Clitics as case markers

In spite of the development of the "articulated declension", there have
always been ambiguous contexts, in which the case system in question is
not sufficient for marking the syntactic function of the noun.
Case 1: The verb accepts a noun characterized by the feature Person,
either as its subject or as its direct object, and the noun in question
belongs to the topic framework. 4
(14) α väzut omul
has-he seen man-the-NOM/ACC
which can mean either (a) 'he saw the man' or (b) 'the man saw'; cf.
(15) α väzut omul, nu-p face grijil
'the man saw (it), don't worry (about it)'
Case 2: The noun may be interpreted either as an attribute or as an
indirect object:
(16) a dat car tea unei fete
has-he given book-the-ACC a-DAT/GEN girl-DAT/GEN
which can mean either (a) 'he gave the book to a girl' or (b) 'he gave the
book of a girl'.
Consequently, new desambiguating means have been developed, and,
as the pan-Romance preference for proclisis would predict, the new
markers occupy pre-nominal positions. In spite of the variety of means
available, two main ways have been followed:
(a) The generalization of certain prepositions from one case to another;
for example, ad 'to' spread from indirect objects to attributes:
(17) tatä a cinci copii
'father of five children'
p(r)e 'on' spread from locatives to direct objects: e.g.
(18) am väzut pe Ion
Ί saw ON John'
(b) The conversion of various means of pragmatic strategies (topical-
ization, focalization of objects, indexical devices) into syntactic markers.
In the following paragraphs I shall present some of the features
characterizing the clitics which have acquired the function of case mark-
ers.
2.1. The possessive article

The so-called possessive or genitive article (al masc.sg., a fem.sg, ai
masc.pl, and ale, fem.pl.) is a cliticized anaphor or pronoun which refers
to a "possessum" and it is determined by an NP referring to the possessor,
expressed either by a possessive pronoun (adjective) or by a genitive:
(19) cartea ta a mea jAnei
book-the yours and poss.-cl. mine /Ann's
'your book and my/Ann's (book)'
In contemporary Romanian, the possessive anaphor may copy a noun
in the following contexts: (i) The governing noun copied by the possessive
clitic is predetermined by a demonstrative or by a quantifier:
(20) oj fiecarej oricare/ aceastä carte a prieten -ului meu.
a/ every/ any/ this book poss.-cl. friend -the-GEN mine
'a/every/any/this book of my friend'
(ii) The governing noun is followed by an adjective or a prepositional
phrase:
(21) rochi-a albä/ de searä
dress-the-FEM-SG white-FEM-SG/ of evening
a Ane- i
poss.-cl. Ann-GEN- the-GEN
Although it might seem redundant, since it precedes the genitive/dative
form of the "possessor", one may easily find contexts where the possessive
clitic has a clear-cut disambiguating function. Compare:
(22a) i- am dat cäma§a albä
him-DAT have-I given shirt-the-ACC white
lui Ion
the-GEN/DAT John
Ί gave the white shirt to John'
and
(22b) i- am dat cäma§a albä
him-DAT have-I given shirt-the-ACC white
a lui Ion
poss.cl. the-GEN/DAT John
Ί gave him John's white shirt'
In (22b), the possessive clitic /a/ is the only means of marking the
following noun as an attribute, while in (22a), the absence of the posses-
sive clitic leads to the interpretation of lui Ion 'the-gen./dat. John' as an
indirect object.
2.2. Clitics as copies of objects

2.2.1. When the direct object precedes the verb and is characterized by
the feature — Person, the clitic can become the only disambiguating case
marker:
(23) ο comisie ο folose§te
a-FEM-SG committee it-FEM-SG-ACC uses
conducerea fabrici -i
the management factory-GEN/DAT -the-GEN/DAT
'a committee is at the service of the factory management'
In substandard spoken Romanian, the accusative of the clitic copy
may be the only case marker, if both the subject and the object follow
the verb:
(24) le- ο vindut oameni -i
them-FEM have sold men -the-NOM/ACC-PL
vite -le
cows -the-FEM-NOM/ACC
'the men sold the cows' (see Niculescu 1965: 68).
2.2.2. The anaphor which copies an indirect object may also play the
role of a proclitic case morpheme as shown by the following utterances:
(25) i- ai cerut cartea
her-DAT have-you asked-for book-the-ACC
Ane -it — Da' cuil
Ann-DAT/GEN -the-DAT/GEN? - but whom-DAT?
'You asked Ann for the book? — Who(m) else?'
where i 'her-DAT.' is a copy of the noun following the direct object and
plays the role of identifying the copied noun as an indirect object and
not as an attribute.
When the clitic copy of the indirect object is absent, the noun following
the direct object is interpreted as an attribute referring to the possessor
rather than as an indirect object:
(26) ai cerut cartea Anei?

have-you asked-for book-the-ACC Ann-GEN-the-GEN?
— Da' a cui?!
- But poss.cl. whose-GEN/DAT?!
'You asked for Ann's book?' — 'But who else's (could I have
asked for)?!'
2.2.3. The difference between the functions of clitics, namely between

(a) pronouns, (b) copies of fronted objects, and (c) copies of postverbal
objects are syntactically relevant.
(i) No more than one postverbal nominal may be copied by a clitic:
(27) il dau pe bäiat lui Bau Bau
him-ACC give-I on boy to Bogeyman
Ί shall give the boy to the B\
where only the postverbal direct object is copied by a clitic, but not:
(28) *i- I dau pe bäiat lui Bau Bau'
him-DAT him-ACC give-I on boy to B.
(same meaning as 27), where both objects are copied.
(ii) When pronominal, two clitics may co-occur:
(29) imi dai bäiatul? — }i- I dau
me-DAT give-you boy-the-ACC? — you-DAT him-ACC give-I
'Are you going to give me the boy? — I will (give him to you)'
(iii) Both objects may be copied if the corresponding nouns are fronted
as topicalized or focalized:
(30) pe bäiat lui Ion i- I dau,
on boy to Ion him-DAT him-ACC give-I,
nu Ane -i
not Ann-DAT the-DAT
'the boy, to John I shall give him, not to Ann'.
(iv) When the direct object is fronted, the clitic can even copy inanimate
NP's:
(31) riscul ästa mi -I asum, altele nu
risk-the this me-DAT it-ACC assume-I, others not
'this risk, I'll take (it), not others'
while postverbal nouns may be copied only if characterized by the feature

Person and introduced by the preposition pe 'on' (see below, 3.2).
3. Object clitics and discourse strategies
3.1. Left-dislocated objects

As in any Romance language, clitics which copy left-dislocated NPs are
paired with discourse-prominent entities.5
(i) They copy overtly marked topical objects: (a) direct objects:
(32) Petru, I- am väzut ieri
Peter, him have-I seen yesterday
'Peter, I saw him yesterday'
(b) indirect objects:
(33) Petru, i- am dat cartea ieri
Peter, him-DAT have-I given book-the yesterday
'Peter, I gave him the book yesterday'
(ii) They copy focalized objects:
(34) Pe Petru I- am väzut, nu pe tine
on Peter him-ACC have-I seen, not on you-acc.
Ί saw Peter, not you'
(35) lui Petru i- am dat cartea,

him-DAT Peter him-DAT have-I given book-the-ACC,
nu fie
not you-DAT
'to Peter, not to you I gave the book'
3.2. Postverbal direct objects

Left-dislocated direct objects constitute prominent discourse entities, first
and foremost by their syntactic order (see Halliday 1967; Brown —Yule
1984: 134—135) and the clitic copies are mainly case markers. But in
Romanian, the clitic may even copy postverbal objects, in which case
they have become markers of discourse prominence. For example, the

clitic may be paired with afterthoughts:
(36) le -am gäsit, cuiele
them-ACC-PL have-I found, nails-the-ACC
Ί found them, the nails'
If not paired with afterthoughts, clitics may copy only postverbal pre-
positional direct objects. Compare:
(37) U voi vedea pe bäiatul Anei
him-ACC shall-I see on boy-the-ACC Ann-GEN-the-GEN
Ί shall see Ann's boy'
but not
(38) *il voi vedea bäiatul ...
him-ACC shall-I see boy-the-ACC
Ί shall see Ann's boy'
In order to account for this constraint, it is necessary to bring into the
picture the pragmatic differences between the prepositional direct object
and the mere accusative construction in contexts in which both construc-
tions are grammatically acceptable, namely when the direct object is
expressed by a common noun referring to persons.
3.2.1. The accusative construction is compatible only with back-

grounded episodes:
(39) Dar nu de nume e vorba! ... Crezul, domnule Dardalat, asta este
esentialul! Crezul! Au cufundat copilul ΐη αρά §i i-au citit crezul!
'but the name is not the matter! ... The creed, Mr. Dardalat, this
is the essential (matter)! The creed! They immersed the child
('child-the') in the water and read the Creed to him!' (Voitin,
Girueta, 306)
'The Creed' is the speaker's topic, and not the 'child', although it belongs
to the topic framework, since it constituted the speaker's topic several
times in the previous co-text. The prepositional construction, pe copil 'on
child-ACC', would foreground the scene of the baptism, which is not the
case. The baptism is semantically connected with the topic of the dialogue
between the director of a college (Dardalat) and a lady from whom he
hopes to obtain a substantial amount of money for remodeling the school.
But the "baptism" itself belongs to the chain of arguments aimed at
explaining her refusal to make any donation — and it is not meant "to
be seen" by the reader. In such cases, the definite article is sufficient for
marking topical and/or known entities.
3.2.2. The prepositional construction with common nouns occurs only

when the events (or at least the referent of the direct object) are fore-
grounded:
(40) Inainte de Pa§te, plecam cu acceleratul la matu§a mea ... Mä
a§ezasem deci confortabilpe locul meu lingä fereasträ ... Exami-
nam curiosä pe tovarä§ii de compartiment, cind ... (Camil Pe-
trescu, Opere 3,38)
'Before Easter, I was leaving by a fast train, to (see) my aunt ...
I thus sat down, comfortably on my seat by the window... I was
gazing at my companions (lit. on companions-the) in the com-
partment, when ...'
3.2.3. With prepositional direct objects, the clitic has become obligatory
in standard Romanian if the common noun is not determined by a definite
description, but is topical and/or known, because pe, as many other
prepositions, does not accept a definite article, unless the noun is accom-
panied by a definite description 6 :
(41) Cind α väzut-o pe fatä, lupul a luat-o la fugä

when has seen her-ACC on girl, wolf-the has run away
'when it saw the girl, the wolf ran away'
When the noun is determined by a definite description, the clitic is a

marker of the paragraph topic:
(42) intr-o bunä zi, gäsit-a marele §i nemuritorul pictor intr-o bisericä
din Romaun cintäref de strand, ..., care ii pläcu. II rugä deci pe
acest june, Petru Bandinelli, ..., sä-i serveascä drept model, (Brad,
Descoperirea familiei, 25)
'one day, the great and immortal painter found in a church in
Rome a lectern singer, ..., whom he liked. Therefore he asked
this young man (lit. 'him-ACC asked-he therefore on this young
(man)'), P.B., ..., to be his model (for the painting)'
Last, but not least, the clitic can copy a postverbal prepositional object,
if under emphasis:
(43) Bine, dar sä-l invejipe copili lui Artimonu(Brad, Descoperirea

familiei, 13)
'(very) well, but SUBJ-marker them teach also on children-the
ACC-PL of A.', i.e.,
'all right, but you have to teach ArtimonuJ's children too'.
4. Conclusions
4.1. In contemporary Romanian, there are three clitics which can serve
as case markers: (a) the definite article, (b) the possessive article, and (c)
the personal anaphor which copies direct or indirect objects. The definite
article can be the sole bearer of a case marker for the following syntactic
functions: (i) either subject or direct object and (ii) either indirect object
or attribute (see 1.2). The possessive article marks the attribute in op-
position to the indirect object (see examples (22a) and (22b)). The cliti-
cized personal anaphor eliminates the possibility of interpreting the noun
in question as subject (see (23) and (24)).
4.2. Even when they lose their pronominal function, clitics preserve
traces of their endophoric values and, as such, they play an important
role in various discourse strategies. For example, when copying left-
dislocated direct objects and playing the role of case markers, clitics are
always paired with prominent discourse entities (topicalized or focalized
NP's — see examples (32) —(35)). They can also copy postverbal prepo-
sitional direct objects if representing paragraph topics (see (42)), topical/
definite NP's — when not followed by a definite description (see (41)) or
focalized entities (see (43)).
Notes
1. In the literal translation, the dash (-) links morphemes belonging to the same word; for
example, 'child-the-MASC-GEN/DAT' is the morphematic translation of the word
copilului 'of/to the child'. The slash (/) means 'either ... or'. The abbreviations are to be
read as follows: ACC: accusative, DAT: dative; FEM: feminine, GEN: genitive; MASC:
masculine, NOM: nominative; PL: plural, POSS.-CL.: possessive clitic; SG: singular;
SUBJ: subjunctive.
2. With masculine common nouns, the definite article has a special form for marking the
vocative: e.g., omule\ 'man-the-MASC-SG-VOC'.
3. In spoken Romanian, the article -ul is reduced to -u.
4. The "topic framework" is a compound of those entities derivable from the physical
context and from the discourse domain of any discourse fragment. The "domain of
discourse" is defined by the entities which have been introduced in the preceding co-
text (Brown-Yule 1984: 135-137).
5. When talking about the hierarchy of discourse, current approaches in discourse analysis
have brought into the picture interesting converging concepts such as "staging", "theme",
"relative prominence", etc. "Staging", for example, is a dimension of prose structure
which identifies the relative prominence given to various segments of prose discourse
(Clements 1979: 287). The most prominent entity is usually referred to by the term
"theme", either as the left-most constituent in the sentence or clause (Halliday 1967), or
as the grammatical subject of a series of sentences (Katz 1980: 26). The topic is 'the
event, character, etc. speakers want to talk about'. The concept of "stage distance" refers
to the difference between "close-ups", i.e., events which are foregrounded in order to be
presented as taking place before the addressee's eyes, who is supposed to become a
spectator (cinematographic effect) and backgrounded events.
6. For more details see Manoliu-Manea (1988).
References
Brown, Gillian —Yule, George

1984 Discourse analysis. (Cambridge: Cambridge University Press).
Clements, P.
1979 "The effect of staging and recall from prose", in: R. O. Freedle (ed.) New
directions in discourse processing (Norwood, N. J.: Ablex), 287 — 330.
Diaconescu, Paula
1970 Structurä §i evolufie in morfologia substantivului romänesc [Structure and
evolution in the morphology of Romanian nouns] (Bucure§ti: Editura Aca-
demiei R.S.R.).
Graur, Alexandru
1965 "Articolul hotärit la numele de persoane romane§ti" [The definite article
with proper names in Romanian], RRL 10: 551 — 557.
1967 "De nouveau sur l'article postpose en roumain", RRL 12: 3 — 18.
1968 Tendinfe actuate ale limbii romäne [Current tendencies in contemporary
Romanian] (Bucure§ti: Editura §tiin[ifica).
Guillermou, Alain
1953 Manuel de langue roumaine. (Paris: Klincksieck).
Halliday, Μ. A. K.
1967 "Notes on transitivity and theme in English", Part 2, Journal of Linguistics
3: 199-244.
Katz, Jerrold
1980 "Chomsky on meaning", Language 56: 1 —42.
Malkiel, Yakov
1985 "Old and new problems in the latinity of the Lower Danube", Journal of
the American Romanian Academy 6/7: 90 — 104.
Manoliu-Manea, Maria
1985 "Genetic type versus areal coherence: Romanian case markers and the
definite article", in: M. Deanovic et al. (ed.) Melanges de linguistique dedies
ά la memoire de Petar Skok (1881 — 1956) (Zagreb: Djela Jugoslavenske
akad. znanosti i umjetnosti), 301 —308.
1988 "Direct-object constructions in Rumanian. Verbal complementation and
discourse strategies". International Journal of Rumanian Studies 6/2: 53 — 68.
Niculescu, Alexandru
1965 Individualitatea limbii romäne intre limbile romanice. Contribufii gramaticale
I [Romanian among the Romance languages. Grammatical contributions]
(Bucure§ti: Editura §tiintificä).
Rosetti, Alexandru
1986 Istoria limbii romäne de la origini ρΐηά in secolul al XVII-lea [The history of
the Romanian language from its origins to the 17th century] (Bucure§ti:
Editura §tiintifica).
Seidel-Slotty, Ingeborg
1940 "'Hypertrophie' der Pronomina im Rumänischen", BL 8: 142-145.
Zwicky, Arnold — Geoffrey Pullum
1983 "Cliticization vs. inflection: English n't", Language 59: 502-513.
Parasitic formation in inflectional morphology
Yves-Charles Morin
One can oppose two radically different conceptions of inflectional mor-

phology, which can be described as "lexicon internal" and "lexicon
external". This paper examines the formation of past subjunctive verbs
in French which involves a process described by Matthews (1972: 86)
as "parasitic", whose analysis favors a model of lexicon-internal inflec-
tion.
1. The place of inflectional morphology in grammar
In a model of lexicon-internal inflection, lexical items appear fully in-

flected in the lexicon (cf. Bresnan 1982, Halle 1973, Kiparsky 1982, Lieber
1980) and are the only forms which "are lexically inserted into phrase
structures" (Bresnan 1982: 307). Under this conception, the lexicon is not
only a repository of the idiosyncratic properties of the language, but also
generates forms which may be only temporarily retained.
In the lexicon-external model of inflection developed by Anderson
(1977, 1982, 1986), the lexicon may also have some generative capacities,
but only for derivation. It "supplies a comprehensive set of well-formed
stems ... which represents complete words, with exception of inflectional
material" (Anderson 1982: 592). Stems may nonetheless contain some
inflectional information, but only those which are idiosyncratic. For
instance, the lexical entry for naitre in French would not only contain
the default stem /nes-/ (i.e., the stem used in all but the otherwise specified
cases), but also the idiosyncratic stem /naki-/ for preterit indicative,
specified as [preterit indicative] in the lexicon.
198 Yves-Charles Morin
2. Defective paradigms
It is sometimes thought that the existence of defective paradigms might

jeopardize a lexicon-external model of inflection: "if only //-regular in-
flections are listed in the lexicon, we have no ready account of the problem
of defective paradigms" notes Anderson (1982: 593n9). Actually, defective
paradigms would only be problematic if one had to postulate a default
stem in the analysis of a defective paradigm whose gaps are neither
semantically nor phonologically motivated (cf. Morin 1987).
The theoretical difficulties traditionally associated with defective par-
adigms in French (Dell 1970: 227-228, Plenat 1981: 155-156), for
instance, do not result from any specific lexical model for inflection, but
from further assumptions about the allomorphic relationship between
stems. The restrictions on the verb frire, which is limited to infinitive,
past participle, future, conditional, and singular present indicative forms
can be shown to result from the fact that it only has one specific stem
/fri-/ — lexically marked for these inflections — and no default stem (cf.
Morin 1987).
In Anderson's model, lexical insertion allows a stem to be inserted
under a syntactic node only when it is consistent with the requirements
of the morpho-lexical representation of that node. None of the stems of
the verb frire, thus can be inserted under, e.g., a syntactic node [ + V
present indicative lpl] if the lexical entry of this verb neither contains a
default stem nor a specific one for [present indicative pi]. The insertion
procedures proposed by Anderson thus automatically account for frire's
defectiveness.
3. Parasitic formation
The past subjunctive is now obsolete in modern French. The fact that
there nonetheless exist some speakers who have internalized this tense
allows us to assume that an account of the verbal morphology of French
which includes the past subjunctive is a reasonable model of a natural
inflectional system and hence relevant to linguistic theory.
Past subjunctive stems are regularly derived from the corresponding
(unmarked) preterit indicative stems by adding the suffix /s/ for all
persons, except for the third person singular for which the past subjunctive
stem is identical to the preterit stem; e.g., from the preterit indicative
stem /dormi-/ of dormir, one derives two past subjunctive stems: (1)
/dormi/ for third person singular and /dormis-/ for all other persons. This
correspondence between preterit indicative and past subjunctive stems is
systematic , whether the preterit stem is regular, as in the case of excuser,
pret. /skskyza-/, past subj. /Ekskyza(s-)/, semi-regular as in the case of
devoir, pret. /dy-/, past subj. /dy(s-)/, or totally idiosyncratic, as in the
case of etre, pret. /fy-/, past subj. /fy(s-)/ or naitre, pret. /naki-/, past
subj. /naki(s-)/. Defective verbs such as extraire which lack a preterit
stem also lack a past subjunctive stem. Defective verbs which lack a
default stem, but have a specific preterit stem have a past subjunctive
stem, e.g., cheoir, pret. /fy-/, past subj. /fy(s-)/. 1
In this account of past subjunctive stems, one "does derive a 'stem'
(i.e., a part of one word-form) from another 'stem' (a part of another
word-form) of identical status", a process Matthews (1972: 86) calls a
parasitic formation. This runs counter the principles for inflectional
morphology proposed by this author, who then develops other tools to
circumvent the problem (Matthews 1972: 175 — 182).
Parasitic formations are equally problematic for a model of lexicon-
external inflection if the basis for the derivation is a stem which sometimes
does, and sometimes does not, belong to the lexicon. To see why, let us
examine how some specific cases could be analyzed in Anderson's model.
The only stem of naitre which can be inserted under a node having
the morpholexical representation [subjunctive past lpl] is the specific
preterit stem /naki-/ to which the regular affixes /-s-/, /-}-/, and /-5/ would
be added by the rules of the "phonological" component to form /nakisjo/.
The same analysis applies to all the verbs which have a specific preterit
stem. In the case of regular verbs, however, the lexicon does not have
access to any such stem — as regularly inflected forms are not generated
within the lexicon. One would thus be required to have two completely
different mechanisms for past subjunctive depending upon the presence
of a specific preterit stem in the lexicon. This is clearly inadequate, as
the formation of past-subjunctive forms is obviously the same in all cases.
One solution to the problems raised by the parasitic formation de-
scribed here for a model of lexicon-external inflection would require that
both preterit-indicative and past-subjunctive forms be derived from a
common intermediate stem, phonologically identical but nonetheless mor-
phologically distinct from the preterit indicative — that is to say, find an
alternate analysis without parasitic formation. For instance, one would
derive from the intermediate stems /*ekskyza-/, /*dormi-/, and /*naki-/
both the preterit indicative stems /ekskyza-/, /dormi/, and /naki-/and the
past subjunctive stems /ekskyza(s)/, /dormi(s-)/, and /naki(s-)/. The idio-
syncratic stem /*naki-/ would be listed in the lexicon, while the regularly
inflected forms /*skskyza-/, /*dormi-/ would be derived from the default
stems /skskyz-/ and /*dorm-/ outside of the lexicon. The complete par-
allelism between preterit indicative and past subjunctive could now be
accounted for directly by the (lexicon-external) rules of the grammar
which could derive them both in similar ways from the intermediate
stems.
However, there does not appear to be any empirical justification for
such intermediate stems. In particular, the intermediate stems cannot be
interpreted as being "past" stems from which one could derive both past
subjunctive and past (traditionally called "preterit") indicative stems.
Although one can reasonably define a "past" morphological category to
include both past subjunctive and preterit indicative, it would also have
to include at least the imperfect indicative. For instance, congruence
patterns which could be used to justify a common past category, do not
distinguish between imperfect and preterit indicative verbs which can
equally govern past subjunctive embedded sentences: (i) il demanda [pret-
erit indicative] alors qu'on l'excusat [past subjunctive] vs. (ii) il demandait
[imperfect indicative] souvent qu'on l'excusat [past subjunctive] (cf. Grev-
isse 1980: section 2743). The imperfect forms, however, cannot be derived
from such intermediate stems, cf. the imperfect indicative stem /nss-/ of
(il) naissait of naitre vs. the preterit indicative stem /naki-/.
One could perhaps try to justify intermediate stems underlying both
preterit indicative and past subjunctive stems as a matter of principle,
because such stems cannot be morphologically derived from one another,
although they are obviously related. This stance would deny any relevance
to the worries about parasitic formation discussed by Matthews (1972),
and would claim that the distinction between lexicon-internal and lexicon-
external inflection is not an empirical issue. If this is the case, then no
argumentation is possible.
But if we assume that the distinction must be empirical, then there is
but one means to express in a linguistically significant manner the com-
plete regularity by which past subjunctive stems are systematically derived
from the corresponding preterit indicative stems — whether or not these
stems are completely predictable or highly idiosyncratic, and that is, by
allowing all such stems to appear in the lexicon. In other words, the
formation of regular inflected forms must be allowed within the lexicon.
4. Conclusion
The formation of past subjunctive forms in French shows the necessity

for at least some of the inflection to be carried within the lexicon. This
immediate conclusion may even seem acceptable to some proponents of
lexicon-external inflection. For instance, Piatt (1981: 53), who otherwise
adopts Anderson's model, proposes stem-formation rules which derive
perfect and imperfect forms in Old Provencal within the lexicon (she does
not argue, however, that this is an empirical necessity). Piatt, in all other
cases however, describes the inflection as a lexicon-external operation.
One certainly cannot accept that the principles governing the distinc-
tion between lexicon-external and lexicon-internal operations should be
simple matters of convenience. If inflections in some well-defined cases
must be carried within the lexicon, would it not be conceptually simpler
if it were always the case? As a matter of fact, no strong empirical
argument appears to have ever been advanced for lexicon-external in-
flection. In Chomsky (1965: 84 — 88), where an autonomous lexical com-
ponent was first developed in generative grammar, lexical insertion pre-
ceded syntactic transformations, and lexicon-external inflection appeared
at first to be a necessity as the morphosyntactic properties of inflected
forms are not specified at this level of the derivation. This choice, however,
is not felicitous, as there is no simple way to insert idiosyncratic stems
at this stage, cf. Halle (1973: 9), who argued in favor of lexicon-internal
inflection and suggested that "lexical insertion transformations insert
partial or entire paradigms" and that the choice of relevant forms be
made later after the syntactic rules have specified the morpho-syntactic
properties. The decision to have lexical insertion apply after the syntactic
rules, as in Anderson (1982), is another way to circumvent that problem,
but does not justify the choice of lexicon-external inflection, which prob-
ably is one of "the many aspects of this model [which] are assumed
without justification" (Anderson 1982: 594). Indeed, the formal mecha-
nism he proposes for inflection can be straightforwardly transposed to
operate within the lexicon, although a reformulation along the lines
proposed by Piatt for Old Provengal perfect may be more appropriate
(cf. also Morin 1987 for another model of lexicon-internal inflection).
As a more general conclusion, therefore, the characteristics of past
subjunctive inflections in French and the absence of strong empirical
evidence for lexicon-external inflection strongly favor a conception of
lexicon-internal inflection in linguistic theory.
Note
1. Bouix-Leeman et al. (1980: 81) only mention the third singular past subjunctive stem
/Jy-/, but my informants also used the general stem /Jys-/, without hesitation.
References
1977 "On the formal description of inflection", CLS 13: 1 5 - 4 4 .
1982 "Where's morphology?", Linguistic Inquiry 13: 571-612.
1986 "Disjunctive ordering in inflectional morphology", Natural Language and
Linguistic Theory 4: 1 — 31.
Bouix-Leeman, Danielle — Colonna-Cesari, Helene — Dubois, Jean — Sobotka-Kannas,
Claude.
1980 Larousse de la conjugaison. (Paris: Larousse).
Bresnan, Joan (ed.)
1982 The mental representation of grammatical relations. (Cambridge, MA: MIT
Press).
Chomsky, Noam
1965 Aspects of the theory of syntax. (Cambridge, MA: MIT Press).
Dell, Frangois
1970 Les regies phonologiques tardives et la morphologie derivationnelle du franâis
[PhD dissertation, MIT],
Grevisse, Maurice
1980 Le bon usage ( l l c ed.) (Paris & Gembloux: Duculot).
Halle, Morris
1973 "Prolegomena to a theory of word-formation", Linguistic Inquiry 4: 3 — 16.
Kiparsky, Paul
1982 "From cyclic phonology to lexical phonology", in: H. van der Hulst —Ν.
Smith (eds.) The structure of phonological representations (Dordrecht: Foris),
vol. 1, 131-175.
Lieber, Rochelle
1980 On the organization of the lexicon [Ph.D. dissertation, MIT].
Matthews, Peter H.
1972 Inflectional morphology. (Cambridge: Cambridge University Press).
Morin, Yves-Charles
1987 "Remarques sur l'organisation de la flexion des verbes frangais", ITL-
Review of Applied Linguistics 77/78: 13 — 91.
1988 "Disjunctive ordering and French morphology", Natural Language and
Linguistic Theory 6: 271 —282.
Piatt, Diana
1981 "Old Provencal verb inflection: the balance between regularity and irregu-
larity in morphology", in: T. Thomas-Flinders (ed.) Inflectional morphology:
introduction to the extended word-and-paradigm theory, (UCLA occasional
papers 4, Working papers in morphology), 41—71.
Plenat, Marc
1981 L' "autre" conjugaison ou de la rigularite des verbes irreguliers (Cahiers de
grammaire 3, Universite de Toulouse-le Mirail).
The mechanism of inflection: lexicon
representations, rules, and irregularities
Wolfgang Ullrich Wurzel
In agreement with most contemporary morphological concepts, let us

start from the assumption that inflectional morphology is a component
of the lexicon, following specific principles.1 Let us further assume that
the mechanism of inflection is characterized by three types of regularities,
namely:
— lexicon representations of words providing, in their phonomorpho-
logical part, the base information for word inflection;
— paradigm-structure conditions specifying in detail the inflectional be-
haviour of words and, thus, the paradigmatic relations between their
inflected forms;
— inflection rules symbolizing, by formal operations, the various mor-
phosyntactic categories in words.
Below, these three types of regularities shall be considered in more detail,
before discussing the problems of irregularities.
1. Lexical representations
The lexical units of a language are stored in the dictionary of the lexicon.
Usually, i.e., in "regular" cases, the lexicon representation of a word
contains only one representative form of the paradigm, its base form.
This base form, roughly corresponding to the so-called citation form, is
represented in modern Indo-European languages generally by the nomi-
native singular of nouns, the infinitive of verbs, and the unspecified form
of adjectives. But, of course, the categories of lexical representation of
individual word classes may differ from language to language. For ex-
ample, in Russian, the base form of adjectives is not their morphologically
unspecified (masculine) short form but their long form; and in Hungarian,
the base form of verbs is not represented by the infinitive but by the
204 Wolfgang Ullrich Wurzel
third person singular present indicative subjective. 2 In the lexicon, the

base form is not specified in its inflectional category or categories. (We
will return to this important point.)
For a word class in a language, base-form inflection or stem inflection
can apply. In base-form inflection, the entire base form represents the
input for inflection rules; in stem inflection, a (shorter) stem different
from it. This difference is reflected accordingly in lexicon representations,
thus compare 3
(1 a) Base form inflection:
English dog //dog/ B F / N , cry //krai/ BF /v, red //red/ BF / A
German Tisch 'table' //IiJ/bf/n, Hase 'hare' //ha:z + e/ BF / N ,
rot 'red' //ro:t/ BF / A
b) Stem inflection:
German leben 'to live' ///le:b/ St en/ BF / v
Latin hortus 'garden' ///hort/ St us/ BF /N
pratum 'meadow' ///pra:t/ s t um/ B F / N
mensa 'table' ///mens/ s t a/ B F / N
parvus 'small' ///paru/ s t us/ B F / A
Since, in the case of stem inflection, the stem as such is specified as part
of the base form, it is available as input for inflection rules although, in
general, lexicon representation is couched in terms of base forms. The
assumption that, in the dictionary of the lexicon, not stems without any
flexives but base forms are stored rests on the fact that the properties of
base forms including their flexives and not those of stems determine the
inflection-class membership of words. From this fact follow correspond-
ing consequences for inflection-class markedness and inflection-class tran-
sition. For example, the inflection-class membership of Latin mensa is
determined not by its stem mens- but by its nominative singular form in
-a, that is, its base form.
Now, what is the inflectional behaviour of words specified in the
lexicon? In English-oriented investigations often a bipartition into "reg-
ular" and "irregular" is employed which, in the sense of inflection-class
markedness, has to be evaluated as "unmarked" vs. "marked". 4 "Irreg-
ular" inflection forms like feet from foot appear (with respective statement
of categories) in the lexicon representations of words; "regular" inflection
forms like hands from hand are produced by general rules and, since there
is only one general plural rule in English, this need not be explicitly
entered in the lexicon representation of hand. With more complex inflec-
tional systems, such an approach soon proves inadequate. The bipartition
into "regular inflection" vs. "irregular inflection" is not sufficient and,
The mechanism of inflection 205
moreover, imparts a wrong impression of the structure of inflectional

systems. For example, in German beside the "regular" part in the sense
of English s-plural and ed-preterit (which, more strictly speaking, can be
called regular-unmarked) and the really "irregular" part embracing, for
example, suppletive cases, there is a third part which can be most ade-
quately called regular-marked. Compare the following gradient: in the
preterits of verbs and in the plurals of nouns:
(2) Verbs Monosyllabic feminines
Regular-unmarked: sagen — sagte 'to say' Bahn — Bahnen 'train'
Regular-marked: reiten — ritt 'to ride' Wand — Wände 'wall'
Irregular: sein — war 'to be' Stadt — Städte 'town'
(long vowel)
It should be only mentioned here that regular-marked and irregular words
may deviate in different degrees from the corresponding regular-un-
marked ones. For example, in the regular-marked range, different degrees
of markedness should be expected. A downright extreme example are the
"non-weak" verbs in German exhibiting, with respect to deviation from
unmarked "weak" verbs, six different degrees.5 Unlike the irregular range,
the regular-marked range is constituted not by individual cases but by,
in part, quite large inflection classes: about 30 words follow the pattern
of reiten, about 35 words that of Wand, and the regular-marked er-plural
class in German (Mann 'man' — Männer, Kind 'child' — Kinder) embraces
even about 90 words. That regular-marked words are not exceptions to
be covered by lexicon specification of forms, but that their inflection
forms are really regular, i.e., produced by rule, follows from the fact that
often the same inflection rules apply to them as to words of the regular-
unmarked range. For example, in German the «-plural rule applies both
to the unmarked monosyllabic feminines of the type Bahn — Bahnen and
to the marked monosyllabic masculines of the type Mensch 'human being'
— Menschen·, and the e-plural rule applies both to the unmarked mon-
osyllabic masculines of the type Berg 'mountain' — Berge and to the
marked monosyllabic feminines of the type Wand — Wände. This also
makes it clear that, in German as in other languages with more complex
inflectional systems (Latin, Icelandic, Russian), there need not be any
general "regular" inflection for all members of a word class. Which
inflection class is regular-unmarked for a word always follows from the
extramorphological properties of the word concerned. For example, as
mentioned above, in German the e-plural class is unmarked for mono-
syllabic masculines and the «-plural class for monosyllabic feminines and
the 5-plural class for nouns of the Opa 'grandpa'/Kino 'cinema'/Mutti
'mamma' type. Therefore, it is the inflection-class membership of a word

that is unmarked and not the inflection class itself!6
Let us return to the question of lexicon specification of the inflectional
behavior of words. It can now be answered as follows: words with regular-
unmarked inflection-class membership do not require any explicit inflec-
tional information. In contrast, words with regular-marked class mem-
bership do require such information. Different inflection classes differ
with regard to which inflection rules apply to the respective words. It is
therefore fully legitimate to assume an inflectional specification in terms
of features for inflection rules, for example, for nouns of the type Mensch
as [«/PI] and for nouns of the type Wand as [e/Pl], Such rule features
have a double function — on the one hand, they prevent the application
of the inflection rule unmarked for the word concerned and, on the other,
they trigger the application of the specified marked rule. Words with
irregular inflection differ from the others in that they have specific
unsystematic inflectional characterization in the lexicon. We will discuss
these problems in more detail below.
There could be the wrong impression that lexical specification of
inflectional behavior in terms of rule features instead of inflection-class
features dispenses with certain possible generalizations, but the opposite
is the case. Here, paradigm-structure conditions come into play.
2. Paradigm-structure conditions
Based on the relevant lexical-base information, paradigm-structure con-

ditions lay down in detail the inflectional behavior of words. They are
of implicative character and derive (further) inflection-rule features from
extramorphological properties and/or from lexically entered inflection-
rule features. Compare the following examples from German:
— Fem
(3 a) [e/Pl]
# Σ #
b) [e/Pl] ^ [n/D.P1]
e/Pl
c) [s/ G.Sg]
— Fem
Words with unmarked inflection class membership are first assigned

features for "Kennformen" by paradigm-structure conditions. For ex-
ample, paradigm-structure condition (3a) means that the e-plural rule
applies to monosyllabic non-feminines (monosyllabism indicated by
' # Σ # ' ) unless specified otherwise in the lexical entry, compare Berg —
plur. Berge. Further paradigm-structure conditions then assign the other
rule features, again, unless lexically specified as deviating. In our example
this is done by paradigm-structure condition (3 b) meaning that to nouns
to which the e-Plural rule applies, the «-dat.plur. rule also applies, com-
pare plur. Berge — dat.plur. den Bergen, and by paradigm structure
condition (3 c) meaning that to non-feminines, to which the e-plural rule
applies, also the s-gen.sing. rule applies, compare plur. Berge — Gen.
Sing, des Berg(e)s. Words with marked inflection-class membership are
assigned the inflection-rule features not stated in the lexical entry by
paradigm-structure conditions, based on the features entered in the dic-
tionary. So, since the «-plural is unmarked in monosyllabic feminines,
nouns with e-plural are lexically characterized by the feature [e/Pl]. Then
here paradigm structure condition (3 b) also assigns the feature to the n-
dat.plur. rule, compare Wand — plur. Wände — dat.plur. den Wänden.
Thus, paradigm-structure conditions specify, on the whole, the predictable
inflectional properties of words, due to the properties of certain "Kenn-
formen". In the unmarked cases, the lexical base form is also the only
"Kennform"; in the marked cases, reference to further "Kennformen" is
necessary. The various inflection systems differ regarding which inflection
forms represent "Kennformen". In German noun declension the (nomi-
native) plural functions as the canonical "Kennform". With reference to
the plural form, the inflection of all regular-marked nouns in German is
predictable. So, for these words, lexicon representation contains only the
rule features of plural, never features of case rules.
From the preceding remarks, it becomes clear that paradigm-structure
conditions operate according to the default principle, i.e., they assign
inflection-rule features only if their "place" is not yet "occupied", if there
are no features incompatible with them. 7
This does not simply mean that a word can have only one inflection-
rule feature each for a category or category bundle, for there exists the
phenomenon of multiple inflection of categories; compare, e.g., plural
inflection by suffix and umlaut in forms like Männer and Wände or
perfect-participle inflection by prefix, suffix, and ablaut in forms like
geritten 'ridden' and gesungen 'sung'. Incompatible inflection-rule features
are those which relate to the same category (the same categories) and the
same position in the word (slot) as in the German noun [e/Pl] and [s/Pl],
both concerning plural symbolization in suffix position. 8 The default
principle allows us to evaluate the unmarked class membership as the
predictable normal case and to dispense with explicit lexical specification
in cases where inflection-class membership does not strictly follow from
the extramorphological properties of words but where one out of several
inflection classes clearly proves unmarked (by quantitative relationships
and productivity). In German, monosyllabic feminines can fall into the
η-plural class like Bahn — Bahnen or into the e-plural class (with umlaut)
like Wand — Wände; but only the marked membership of the e-plural
class needs to be specified.
It is a consequence of the default principle that, for certain marked
cases, negative rule features have to be provided. For example, for
German feminines in -er and -el (as for monosyllabic ones), the mem-
bership of the «-plural class is unmarked as in Schwester 'sister' —
Schwestern and Wurzel 'root' — Wurzeln, but the two words Mutter
'mother' and Tochter 'daughter' form their plural without plural suffix,
but with umlaut. To avoid assigning them the feature [«/PI] by the
respective paradigm structure condition, they have to be specified in their
lexical entry by the feature [0Suff/Pl] 'no suffix plural rule'. Unlike other
inflection features, this of course prevents the assignment of the feature
for the unmarked rule but does not lay down the application of a marked
rule instead. Such a feature, again, can be the starting point for paradigm-
structure conditions. In the example given, the feature [0Suff/Pl] together
with [ + Fem] then implies the rule feature [Uml/Pl], compare Mutter —
Mütter.
Paradigm-structure conditions have the status of redundancy condi-
tions, that is, they are structural and not rule-like, transformational
regularities and thus comparable, e.g., to phonological structure condi-
tions. They reflect the existence and the specific structure of paradigms
and thus show that the paradigm is more than the sum of its inflectional
forms. The paradigm-structure conditions for a word class in a language
as a whole reflect the structure of the entire inflectional system concerned.
3. Inflection rules
The assumption that morphological markers are introduced ("spelled

out") by inflection rules and are not lexical units of their own, follows
simply from the fact that there exist morphological markers that have
no affix character and, accordingly, cannot be represented in the lexicon

as segmental (or also suprasegmental) strings. Here, subtractive and
metathetic markers but also simple modifying markers like German verbal
ablaut can be mentioned. 9
Inflection rules symbolize morphological categories in the word by
inducing a formal change. This change can be of highly different nature.
Let us assume that inflection rules operate in the lexicon; yet they must
have access to syntactic strings at the same time. They operate on pairs
of the form (LU, M).10 Here, LU is a lexical unit, chosen from the
dictionary, with its lexical features, i.e., gender and inflection-rule features
to which paradigm-structure conditions have been applied, so that all
inflection rules applying to the lexical unit are also specified in terms of
features. Depending on the inflection type expressed in the form of rules,
the lexical unit appears as a base form or as a stem, compare the
nominative plural inflection of German Tisch and Latin mensa:
(4) Input:
a) /AIJ/BF/N
[ + P1] — e / [e/Pl]/ BF - /N
b) //mens/st/N
Output:
a) //tiJ/ BF e/ N
b) //mens/Stae/N
Also previous inflection rules can already have been applied to the lexical
unit, for example, in German the dative plural of nouns is formed on the
basis of the plural forms, compare dative plural Tischen from //tiJ/BFe/N.
Μ is the morphosyntactic representation of the syntactic position into
which the word is inserted. It consists of the corresponding morphosyn-
tactic category features that are partly primary (like tense features in the
verb), partly due to structural or lexically determined assignment (case
features in subjects and objects) and to concord (case features in adjec-
tives). An inflection rule is applied if the feature constellation demanded
by it is given. If several rules relate to the same category/category bundle
and the same position in the word (slot) like the e-, n-, er-, and s-plural
rule for German nouns, then there will be no problems of application,
for the choice of the correct rule is strictly determined by the correspond-
ing rule feature. Then the word covered by an inflection rule will be
inserted into the syntactic string in its inflected form. If there is no

inflection rule for that word which operates on the category features
concerned, the word will be inserted into the string in its uninflected
lexical base form. This is possible because the base form of regularly
inflecting words is not specified with respect to its inflectional category
(categories). For example, German Tisch / / U J / B F / N can be inserted not
only if the nominative singular but also if the accusative or dative singular
is required.
4. Irregularities
Above, we distinguished between the regular-marked and the irregular

range in inflection and illustrated this distinction by plausible examples.
But we have not yet said by what criteria these two ranges can be exactly
delimited from each other. Irregular cases are — in general — such cases
as cannot be explained by general regularities. Now, to what regularities
does the irregular status of such words in inflection relate?
At first thought, it could be assumed that irregular inflection forms
simply contradict inflection rules. Suppletive forms qua prototypically
irregular forms seem to confirm this assumption, for suppletive forms
indeed cannot be explained by general inflection rules, compare German
sein — preterit war, English good — comparative better, and Russian
celovek 'human' — plural ljudi, etc. But not all irregular inflection forms
contradict inflection rules. Compare, for example, the German neuter
Herz 'heart' which is almost the only German noun whose case forms
do not follow by implication from its plural form and whose irregular
status is therefore beyond doubt. It ought to inflect like Ohr 'ear' as in
(5 b) but it inflects as in (5 a):
(5) a. das Herz b. das Ohr

Plural die Herzen die Ohren
Gen.sing. des Herzens des Ohr(e)s
Dat.sing. dem Herzen dem Ohr(e)
Accordingly, Herz has irregular forms in the genitive and dative singular.
Nevertheless, these irregular forms correspond to inflection rules, for in
German there is an ws-rule for the genitive singular that applies to nouns
of the type Funke 'spark' — des Funkens, and an «-rule for the dative
singular applying to nouns of the types Bär 'bear' — dem Bären, Mensch
— dem Menschen, and Funke — dem Funken. So, irregular inflection
forms can also definitely correspond to general inflection rules. Therefore,
irregularity in inflection cannot be claimed merely on the basis of non-
applicability of inflection rules.
Of course, for the word Herz the genitive singular in -ens and the
dative singular in -en are marked forms, but this criterion is insufficient
for delimiting the range of irregularity because it already delimits the
range of regular-marked from that of regular-unmarked forms. Never-
theless, it is without doubt crucial for the irregular status of Herz which
inflection rules apply to the word. For the corresponding irregular forms
of the genitive and dative singular to be produced, Herz must have the
rule features [jw/G.Sg] and [«/D.Sg] in its lexicon representation. This
means that, by way of exception, case-rule features have to be referred
to, while otherwise only plural-rule features occur in the lexical represen-
tations of the entire German noun inflection. So, here the canonical
lexicon information is not sufficient to generate the correct inflectional
forms. Thus, we have arrived at the second decisive criterion for delimiting
irregular from regular cases. Now we can define the three ranges of
inflectional morphology that differ in their regularity and markedness as
follows:
(6 a) The inflectional forms of regularly inflecting words can be ex-
plained by inflection rules on the basis of the canonical "Kenn-
formen" of the system.
(i) In regular-unmarked cases exclusively the lexical base form
with its extramorphological (phonological and/or semantic-
syntactic) properties functions as the "Kennform".
(ii) In regular-marked cases the lexical base form and canonical
derived forms specified by rule features together function as
"Kennformen".
b) The inflectional forms of irregularly inflecting words cannot be
explained by inflection rules on the basis of the canonical "Kenn-
formen" of the system.
By means of this definition also regular individual cases and (non-
suppletive) irregular cases can now be distinguished. Irregularities are
always individual cases, i.e., the respective formation occurs only in one
word as in German Stadt — plural Städte or in very few words as in
bringen 'to bring '/denken 'to think' — preterits brachte/dachte. But not
all individual cases are also irregular, compare the German nouns Floß
'raft' and Kloster 'cloister' as well as Käse 'cheese'. Floß and Kloster are
the only neuters without er-plural but with plural umlaut and their
inflectional behavior can be specified simply by the rule feature [Uml/
PI], i.e., with reference to the canonical plural form. Also Käse, the only
masculine with 0-plural ending in -e requires only the lexical specification
[0Suff/Pl] preventing the application of all suffix rules for plural. Due to
this feature, it is then assigned — quite regularly by the respective
paradigm-structure conditions — the features [s/G.Sg] and [«/D.P1], i.e.,
it functions entirely like a normal 0-plural word, compare Käse — plural
die Käse — gen.sing, des Käses — dat.plur. den Käsen with Sommer
'summer' — plural die Sommer — gen.sing, des Sommers — dat.plur. den
Sommern. So, the words Floß, Kloster, and Käse are — unlike Herz —
not irregular but actually form regular-marked "miniclasses".
Since irregular cases cannot be explained by inflection rules based on
canonical "Kennformen" they require special unsystematic information
for their inflection. Such information can consist in the statement of
unsystematic rule features (not relating to canonical "Kennformen"), of
idiosyncratic category markers and/or of suppletive stems.11 Then, from
these different possibilities there actually follows a typology of irregular-
ities in inflectional morphology. Compare the following types:
Type I: Unsystematic inflection rule features ("Kennformen")
Example: German Herz
Irregular forms: G.Sg. (des) Herzens, D.Sg. (dem) Herzen
Lexicon
representation: //herts/BF/N [+Neutr, njPI; ns/G.Sg, w/D.Sg]
The irregular forms can be explained by (independently motivated) in-
flection rules; the plural form is regular-marked.
Type II: Idiosyncratic categorial marker
Example: English ox
Irregular form: Plur. oxen
Lexicon
representation:
The irregular form cannot be explained by an (independently motivated)

inflection rule; beside oxen there exist only the en-plural forms brethren
and children. The categorial specification for oxen constrains this form
to the respective category. The filling of the plural suffix position (slot)
by -en prevents the assignment of the rule feature [s/PI].
Type III: Suppletive stem

Example: German Stadt 'town'
Irregular form: Plur. Städte (long vowel)
Lexicon j//Jtat/BF/N jr+F<
[+Fem, ejPI]
representation: j//Tta:t/ BF / N [ + PI]J
The stem of the irregular plural form cannot be explained by an (inde-

pendently motivated) inflection rule; the plural suffix and plural umlaut,
conditioned by it, however, are regular-marked. The case of Stadt —
Städte is weakly suppletive, a strongly suppletive counterpart is, e.g.,
Russian tfelovek 'human being' — plural ljudi.n
Type IV: Suppletive stem and idiosyncratic marker
Example: Engl, child
Irregular form: Plur. children
Lexicon
representation: / N [+P1]J
Neither the stem nor the plural suffix of the irregular form can be
explained by (independently motivated) inflection rules.
Type V: Suppletive stem with "incorporated marker
Example: Engl, foot
Irregular form: Plur. feet
Lexicon
representation:
The irregular form cannot be explained by (independently motivated)

inflection rules. In contradistinction to type III, there is no regular plural
suffix, which is prevented by the blocking feature [0/P1]. To distinguish
the two types compare also English good — bett-er vs. bad — worse and
German bringen 'to bring' — brach-te vs. gehen 'to go' — ging.
To types II to V, the condition applies that the units specified in the
lexicon, i. e., idiosyncratic markers and/or suppletive stems, can relate to
canonical "Kennformen" of the system or not. For example, in the case
of Stadt (type III), the irregular inflection characterization relates to the
plural which is anyway without "Kennform" for German nouns.
Unlike the case of English house where, due to the consonant alteration
/s/~/z/, the plural stem also has to be lexically specified, compare
//häus/ßp/fj — plur. //hauz/bf/n> for — unlike the German noun — in
the English noun the plural is definitly not a canonical "Kennform" but
is in general regularly assigned by a paradigm structure condition. Thus,
the irregular plural form of house practically represents a combination of

types III (suppletive stem) and I (unsystematic "Kennform"). This shows
that irregularity is a gradual phenomenon.
There are no problems in choosing the correct forms for lexicon
representations of types II to V if the principle of disjunctive ordering is
presupposed. 13 Let us take the example Stadt: If [ — PI] is demanded by
the morphosyntactic representation, only //Jtat/ BF /N can be chosen, since
//Jtait/BF/N is characterized as [+P1] in the lexicon and incompatible with
the morphosyntactic representation. However, if [+P1] is demanded, both
forms are compatible, since the canonical base forms are not specified
for their categories. Here, the principle of disjunction causes the more
specific form to be chosen, and this is the form with the characterization
[ + P1], that is, //Jta:t/ BF /N·
To be certain, the concept of inflectional mechanism presented here
still leaves undecided quite a number of questions and it is equally certain
that other solutions than those proposed here would be possible in certain
points. Nevertheless, this concept on the whole allows a quite plausible
explanation both of the functioning and of the changing of inflectional
systems. It permits practicable, easy-to-follow delimitations between un-
marked and marked as well as between regular and irregular phenomena
of inflectional morphology. Based on these distinctions, a differentiated
coverage ensues of various "ranges of regularity" to be expected in more
complex inflectional systems, including irregularities, which are thus prop-
erly integrated into this concept of morphology. Future research will
show how viable this concept is.
Notes
1. Compare also recent papers by Anderson and Zwicky, especially Anderson (1986),
Zwicky (1985, 1989). The concept of inflectional system advanced here furthermore
shows a number of other characteristics common with Anderson's and Zwicky's
position.
2. In this field, plausibility considerations are largely dominant. There is not yet any
theory of the base form.
3. Compare Wurzel (1984: 51 ff.) where examples of combinations of base form and stem
inflection can be found.
4. For the concept of inflection class markedness see Wurzel (1988).
5. Compare Bittner (1985, 1988).
6. In Wunderlich (1985) and Wiese (1986), it is explicitly or implicitly assumed that s-
plural is generally regular for German nouns and that all other plurals, among them
the «-plural of feminines ending in -e (the only one possible for such nouns), are
irregular. Such an assumption may meet the requirements of the underlying model of
a lexical multilevel morphology but it is incompatible with the facts of German

inflectional morphology!
7. For the default principle in morphology see Wurzel (1984: 125 fT.) and Zwicky (1985).
8. If a word has several forms for a specific category, the corresponding possibilities will
be specified in the lexicon in a disjunctive way without blocking each other. This applies,
e.g., to the German noun Sau 'sow' and its plural forms Säue and Sauen. Since, for
words of this type, the η-plural is unmarked, the inflectional entry has to be as
i0 1
follows: j ^pjjj · f i r s t c a s e the paradigm structure conditions assign the feature
[«/PI], in the second case the features [Uml/Pl] and [n/D.Pl].
9. Compare Anderson (1986/ 11 fT.) and Wurzel (1989).
10. Compare Anderson (1986: 30 fT.).
11. This applies similarly to words having defective (incomplete) paradigms not discussed
here.
12. An interesting relevant example is the German noun Junge 'boy' with the plural form
Jungs (beside Jungen and Jungens). Here, the lexicon representation takes the following
form:
i//jung + e / B F / N I
f / p | 1
V/jung/ BF / N [ + Pl]j l ' / r , J ·

Thus, instead of the base form inflection applying elsewhere (compare Junge—Jungen),
actually stem inflection occurs in plural formation.
13. For the disjunction principle in inflectional morphology see Kiparsky (1982: 9 fT.),
Anderson (1982: 593 ff. and 606) as well as Zwicky (1985: 3 fT.).
References
1982 "Where's morphology?", Linguistic Inquiry 13: 571 —612.
1986 "Morphological theory", Linguistics: The Cambridge Survey (Cambridge:
Cambridge University Press), 146 — 191.
Bittner, Andreas
1985 "Das 'Stark-Schwach-Kontinuum' der neuhochdeutschen Verben", Acta
Linguistica Hungarica 35: 31 —42.
1988 Starke 'schwache' Verben — schwache 'starke' Verben. Überlegungen zur
Struktur des deutschen Verbsystems im Rahmen der natürlichen Morphologie
[Dissertation A, Akademie der Wissenschaften der DDR, Berlin].
Kiparsky, Paul
1982 "Lexical morphology and phonology", Linguistics in the Morning Calm
(Seoul: Linguistic Society of Korea), 3 — 91.
Wiese, Richard
1986 "Schwa and the structure of words in German", Linguistics 24: 697 — 724.
Wunderlich, Dieter
1986 "Probleme der Wortstruktur", Zeitschrift für Sprachwissenschaft 5/2:
209-252.
Wurzel, Wolfgang U.
1984 Flexionsmorphologie und Natürlichkeit. Ein Beitrag zur morphologischen
Theoriebildung (Studia grammatica 21) (Berlin: Akademie Verlag) [In Eng-
lish: Inflectional Morphology and Naturalness (Dordrecht/Boston/Lancaster/
Tokyo: Foris)].
1987 "Paradigmenstrukturbedingungen: Aufbau und Veränderung von Flexions-

paradigmen", in: A. Giacalone-Ramat — O. Carruba —G. Bernini (eds.) Pa-
pers from the 7th International Conference on Historical Linguistics (Am-
sterdam: Benjamins), 629 — 644.
1988 "Gedanken zur Flexionsmarkiertheit", in: M. Bierwisch — W. Mötsch —I.
Zimmermann (eds.) Syntax, Semantik und Lexikon (Studia grammatica 29)
(Berlin: Akademie Verlag), 259-277.
1989 "Weshalb morphologische Kategorienmarker nicht als selbständige Ein-
heiten im Wörterbuch des Lexikons repräsentiert sein sollten, dürften und
können — Ilse zum Sechzigsten gewidmet", to appear in Linguistische
Studien (Z1SW der ADW der DDR, Reihe A), H. 194, 277-298.
Zwicky, Arnold M.
1985 "How to describe inflection", Proceedings of the 11th Annual Meeting of
the Berkeley Linguistic Society, 372 — 386.
1990 "Inflectional morphology as a (sub)component of grammar", this volume,
217-236.
Inflectional morphology as a (sub)component of
grammar
Arnold M. Zwicky
This is a report on a program of modular grammar (which I sometimes

refer to as the Interface Program), incorporating morphology of several
kinds, syntax, their accompanying semantics, and phonology. I pursue
an enterprise of this scope in the belief that an understanding of any one
of these domains depends crucially on appreciating how it fits with all
of the others. The program is substantive rather than notational in
character, and my goal is a framework that encompasses the full range
of attested phenomena, so that at the moment at least I am willing to
sacrifice theoretical parsimony for the sake of adequacy.
Here I touch on a few of the aspects of this framework that concern
inflectional morphology, recasting and expanding material in Zwicky
(1985, 1986, 1987a, 1988). The ultimate intellectual source of these ideas
lies in Sapir's (1921) notion of "grammatical process"; the more recent
inspiration comes from Matthews's explorations of the Greco-Roman
"word and paradigm" tradition (1972); and in consequence my ideas
have an obvious kinship with those of Anderson (1988a, b), especially
insofar as both of us assign no fundamental theoretical significance to
the morpheme.
1. Lexemes and morphological rules
Clarity will be served by making at least the distinction between the

concepts of a lexeme (the Welsh lexeme CATH 'cat'), its (inflectional)
forms (singular cath, plural cathod), and their (phonological) shapes (the
mutation shapes cath, gath, chath, ..., cathod, gathod, chathod, ...). I will
cite lexemes with capitalization throughout, forms and shapes in standard
orthographies or phonemic transcriptions.
218 Arnold Μ. Zwicky
Lexemes are complexes of grammatically relevant information, includ-

ing at least the following: semantics; a syntactic category; a set of syntactic
subcategories, each encoding the ability of the lexeme to occur as the
head word in a syntactic construction; morphological features, including
paradigm classes; a list of forms, each form being a pairing of a set of
grammatical category features with one or more phonological shapes, as
in the example below; and a list of stems, each stem being a complex of
phonological properties, with one stem distinguished as the primary stem.
A fragment of the form list for the english lexeme RHYME:
{<{PRS, 1, SG}, /raym/>, <{PRS, 2, SG}, /raym/>, <{PRS, 3, SG},
/raymz/), ...}
I assume that derivational morphology and inflectional morphology
are separate subcomponents of grammar, limited in their interactions
with one another in a way that rules within either of these components
are not; the framework I have been exploring thus opts for something
like the subcomponent divisions of traditional grammar, rather than the
levels or strata of "lexical morphology" (Kiparsky 1982). Morphological
rules, of either sort, describe relations between two types of morphological
entities, which I will refer to as the "input" and the "output". The
temporal metaphor is dispensible, but it is nevertheless useful, since it
suggests that the relations are in fact functions (in the mathematical
sense), and I want to hold as much as possible to the position that outputs
are unique.
Derivational rules embody generalizations about the whole stock of
lexemes; they predict phonological, morphological, syntactic, or semantic
properties of output lexemes on the basis of properties of input lexemes.
Inflectional rules embody generalizations about the forms of lexemes;
they predict, on the basis of phonological properties assembled in a stem
for a lexeme, phonological properties associated with combinations of
grammatical categories. It follows that there must be a third set of rules,
call them "shape rules", that embody generalizations about shapes. Fi-
nally, there are other generalizations, beyond those in inflectional and
shape rules, about the properties of individual lexemes; these "lexical
redundancy rules" (as they are most commonly called) relate phonological
properties to morphological properties (predicting, say, the paradigm
class of a lexeme from the number of syllables in its primary stem),
semantic properties to morphological properties (predicting, say, the
paradigm class of a lexeme from the fact that it denotes a female human
being), morphological properties to other morphological properties (pre-
Inflectional morphology as a (sub) component of grammar 219
dieting, say, the applicability of one inflectional rule from the applicability
of another, and thus describing the clustering of inflectional rules that
goes under the name of "paradigm class"), syntactic properties to syn-
tactic properties (predicting, say, one subcategory of a lexeme from
another), and so on. These can be systematized as in Table 1.
Table 1. Types of rules and their corresponding input and output entities
Type of rule Type of input entity Type of output entity
Derivational Lexeme Lexeme

Inflectional (Stem of) lexeme Form
Shape Form Shape
Redundancy Lexeme property Lexeme property
Derivational and inflectional rules have a special status in this frame-

work. Like syntactic constructions, they pair semantic content (for deri-
vational rules, this is embodied in a function deriving the semantics of
the output lexeme from the semantics of the input; for inflectional rules,
this is the set of grammatical categories expressed, or realized, by the
mapping from stem to form) with a set of formal conditions, which in
the case of morphological rules are embodied in functions mapping one
phonological entity into another (for derivational rules, mapping one
primary stem into another; for inflectional rules, mapping a stipulated
stem into the phonological information associated with a form).
2. Inflectional rules
An inflectional rule has three parts: a context, which can be viewed as a

set of conditions on the input, including stipulations of the relevant
category and the affected stem; the realized feature, which can be viewed
as a set of conditions on the output; and the realization itself, which can
be viewed as a set of conditions on the association between input and
output, stipulating a set of mapping rules (which I called "allomorphy
rules" in Zwicky 1985), each mapping rule involving a phonological
operation and a slot in which this operation applies, plus possibly one
or more conditions on the operation itself. Note that despite the dynamic
metaphor in the word "realization", such rules can be — and I propose
that they should be — conceived of as a set of necessary and sufficient

conditions on certain properties of lexemes.
The general scheme must have a place for several stems for a lexeme;
for operations that predict alterations within stems as well as the affixing
of material to them; and for a number of affixal slots, giving the essentially
flat structure that is typical of inflection. The scheme is summarized
below, along with two examples from English.
A scheme for inflectional rules
Necessary conditions:
Context Input conditions, each stipulating: category; af-
fected stem
Sufficient conditions:
Realisata Output conditions
Realization Association conditions (mapping rules), each stip-
ulating: operation; slot; conditions
Example I of an inflectional rule: English genitives

Context 0; 0
Realisata genitive
Realization affix /z/; slot B; unless slot A = jzj
Example II of an inflectional rule: English noun plurals

Context noun; 0
Realisata plural
Realization affix /z/; slot A; 0
Example I concerns the genitive in my children's ideas, the chairman
of the department's proposals, the person I talked to's objections, and the
person I saw's hat. The rule realizes "genitive" on any category whatsoever
(hence, there are no conditions on the relevant category) and without
any conditions on the affected stem, and it says that "genitive" is realized
by affixing jzj in a particular slot (identified here arbitrarily as "B") so
long as another specified slot ("A") is not filled with the phonological
content jzj. Given independent stipulations that slots A and Β are suffixal,
and that A precedes B, this rule describes forms like child-ren-'s, and its
condition on realization correctly "suppresses the Z" (Zwicky 1987b) in
examples like my kids' ideas. Example II concerns the somewhat simpler
case of the plural in my kids, the two cats, and those remarkable churches,
the allomorphy in these examples being a matter of automatic phonolog-
ical rules rather than morphological rules.
3. Some details
3.1. Defaults and overrides

Inflectional rules specify conditions all of which must be met by forms.
But sometimes the application of a rule will be blocked, either because
some more specific rule is applicable (as when /n/-suffixation for the
English past participle, which is available only for a particular paradigm
class of verbs, overrides the default rule calling for past participle identical
to the past form), or because contradictory features are stipulated for
individual lexemes (as when the stipulation of the past thought for the
lexeme T H I N K overrides the default /d/-suffixation for past).
Overrides are predicted only when there are conflicts in the specifica-
tion of phonological properties for forms; here my assumptions differ
from those of Anderson (1986). When a more general rule is formally
compatible with a more specific one, then both apply, as in the German
form Kind-er-n, with both plural -er and dative plural -«; or the Swahili
form ha-wa-ku-soma 'they didn't study', with both negative ha- and
negative past ku- (vs. affirmative past Ii-) (Hinnebusch 1979: 256); or the
Hua form bau-ne 'we (plural) stay', with both bau (in which first person
is realized by backing the front vowel of bai) and the first person plural
(and second singular) declarative -ne (vs. -e) (Haiman 1980: 47 — 50).
3.2. Slot competition

In some other instances where two inflectional rules are in competition
with one another — in particular, because they call for the affixation of
different phonological material in the same slot, even though the gram-
matical categories they realize are compatible — it seems that universal
principles governing the overriding of defaults cannot be appealed to,
and the interaction between the two rules must simply be stipulated as a
language-particular fact. As Anderson (1986: 8) says of Georgian, "the
formal markers v- [marking 1 SUBJ] and g- [marking 2 OBJ] are mutually
exclusive by virtue of their "competition" for the same formal position";
as it happens, the v-prefix wins this particular contest.
3.3. Multiple mapping rules

The scheme permits multiple operations within a rule, as when umlaut
and suffixation combine to realize plural in German forms like Stühle.
Indeed, such multiple operations can compete with one another, with the
more general operation taking precedence. This is the type of analysis I

would give to (some of) Carstair's (1990) examples of "phonologically
conditioned suppletion", for instance, the realization of second person
singular indefinite present indicative in Hungarian as ~(a)sz in general,
but -ol after sibilants and affricates; the two suffixations — the first
without constraints and the second with a phonological constraint on it,
but both filling the same slot — belong to the same inflectional rule.
3.4. "Zero morphs"
What happens when a condition in the realization portion of a rule is

not satisfied — when, for instance, we check to see how the English
genitive rule above applies to the lexeme K I D with the feature "plural"
as well as "genitive" and discover that the plural rule requires a jzj in
slot A — is that the operation simply fails to apply, and the input is
unaltered (by that operation, at any rate). The rule applies, and it predicts
a "zero morph", but without any explicit stipulation to this effect.
There can be paradigm classes defined in part by the inapplicability
of particular mapping rules: e.g., the English noun class QUAIL, PHEAS-
ANT, ELK, MOOSE, ..., for which the affixation of jzj in slot A is
inapplicable. A redundancy rule presumably relates (at least as a default)
the semantics of these noun lexemes, involving reference to creatures
hunted for sport, to this morphological property of them; but there are
also individual nouns like SHEEP that idiosyncratically have the prop-
erty.
3.5. Gaps
What happens when one of the conditions in the context portion of a

rule is not satisfied — when, for instance, we check to see how the English
plural rule above applies to a preposition — is that the rule simply fails
to apply, predicting in this example that prepositions have no plural
forms. The nonapplication of a rule thus predicts systematic gaps in the
lists of forms. (There are also idiosyncratic gaps, of course: combinations
of grammatical categories for which no phonological shape is listed for
some lexeme.)
Gaps will be predicted when a lexeme lacks the stem that is called for
by an inflectional rule. The Latin verb lexeme COEP 'begin', for instance,
has a Stem 2 (which is used for perfect forms like coepi Ί have begun')
but in the Classical language it lacks a Stem 1 (which is used for present
forms: *coepio Ί begin') (Hale - Buck 1903: sec. 199.2).
And there can be paradigm classes defined in part by the inapplicability
of particular inflectional rules: e.g., the English adjective class FA-
THERLY (vs. WORLDLY), VISIBLE (vs. ABLE), CROTCHETY (vs.
DUSTY) ..., for which inflectional rules realizing comparative and su-
perlative are inapplicable: *fatherlier vs. worldlier, etc. A redundancy rule
presumably relates (at least as a default) the phonology of these adjective
lexemes, involving stems with more than two syllables, to this morpho-
logical property of them; but there are also individual adjectives like ILL
that idiosyncratically have the property.
3.6. Systematic identity

One side effect of the set of inflectional rules for a language is to group
together forms that are systematically rather than accidentally identical
in phonological shape: English first person singular present and second
person singular present rhyme, singular and plural moose, past participle
and past bought, for instance, vs. present run and past participle run,
where the identity is fortuitous. (See also Carstairs 1987: Section 4.2.2).
I take the default situation to be that all forms of a lexeme are identical
to the primary stem; this is the situation that actually obtains in languages
without inflectional morphology, like Mandarin Chinese. When rules do
not alter this state of affairs — either because none is applicable, as for
first person singular present and second person singular present rhyme,
or because an applicable rule's mapping rule is blocked, as for singular
and plural moose — we have systematic identity. But motivated identities
also result from the character of rules themselves: some rules apply to a
disjunction of grammatical-category features, as when a rule realizes
nominative or accusative singular for German weak adjectives by suffixing
-e; other rules explicitly refer the realizations for one set of features to
those for another, as when past participle is referred to past for English
verbs (in regular verbs like jumped as well as in many irregulars) or third
to first person for German verbs (for modals like kann, in imperfect
singulars like sah and machte, in plurals like present sehen and imperfect
machten, etc.). There is then a place for both directional relationships
(described by referral rules) and nondirectional ones (described by rules
realizing feature disjunctions).
I must observe that the term "form" is ambiguous in the common
usage of linguists, since it can refer either to an individual pairing of
grammatical categories with a phonological shape or to a "form set", a

set of systematically identical forms. Talk of the "form see", embracing
the base form and five of the six present forms (all except the third person
singular) of the verb lexeme SEE, is really a reference to a form set, not
a form.
3.7. Alternative shapes for a given feature set

The description of lexemes in Section 1 above allows, and correctly so,
for more than one phonological shape within a form: a regular-conju-
gation past dreamed and a special-conjugation past dreamt for the lexeme
DREAM, for instance. Among the morphological properties of such a
lexeme, then, is a disjunction of paradigm class information.
3.8. Paradigm classes

The primary units of paradigm information are inflectional rule features,
which can be thought of as pointing to individual inflectional rules via
arbitrary indices: + Rule 26, —Rule 77. These features can be grouped
into sets in two different ways.
First, there are features for rules that are in exclusionary relationships
with one another (because they realize the same grammatical categories
in formally incompatible ways). Within any such set there can be an
ultimate default (for the English past, suffixation of /d/; for past participle,
referral to past) and "subdefaults" at one or more further levels (for the
English past, suffixation of /t/; for past participle, suffixation of /n/). The
applicability of one rule within such a set for a particular lexeme ordinarily
excludes the applicability of all the rest for that lexeme. This blocking
effect is not invariable, of course, since alternative shapes are possible.
Second, there are rule features that define paradigm classes, each class
being a cluster of rule features that together ordinarily predict a full set
of forms for a lexeme. (Individual lexemes, or subregular sets of lexemes,
can of course diverge from this pattern by exhibiting gaps, idiosyncratic
forms, or exceptional rule features.) The redundancy provided by a system
of paradigm classes supplements the redundancy provided by inflectional
rules realizing disjunctions of features, by rules of referral, and by default
settings to yield the sort of usefully impoverished inventory of paradigms
that Carstairs (1987) has made so much of.
The existence of a "molecular" level of paradigm-class features, in
addition to the "atomic" level of inflectional rule features, is not a logical
necessity, and serious attempts have been made to do without it, as in

Wurzel's (1990) proposal to rely entirely on "paradigm-structure condi-
tions" (redundancy rules that relate inflectional rule features). For Wurzel,
some of the work of paradigm class features is done by canonical "Kenn-
formen", particular forms from which the other forms within a paradigm
can be predicted via the redundancy rules. I make no commitment to the
existence of "Kennformen" here; they obviously facilitate the learning of
inflectional morphology, but I see no reason to suppose that the theory
of grammar always makes them available.
3.9 Stem rules
In a very simple world, there would be a single assemblage of phonological

information about each lexeme, a single stem (in the terminology I have
been using here; for others, this is a "base", a "root", a "grade", a "basic"
or "underlying" representation — or, alas, "form") which provided the
phonological content for the inputs to both inflectional and derivational
rules. But the world of morphology is often more complex than this, as
is recognized in traditional references to constructs like the "perfect stem"
(vs. the "present stem"), in the positing of "template" or "pattern"
morphology in non-concatenative approaches to phonology, and in
Spencer's (1988) revival of "morpholexical rules" applying at "level 0".
I assume, then, that several stems might be available for particular
lexemes, and that individual inflectional or derivational rules can call for
specific stems. These stems I will identify arbitrary as "stem 1", "stem
2", and so on, since there is no guarantee that a particular stem can be
identified with a unique set of grammatical categories. But how are the
different stems related to one another phonologically? Apparently, in just
the same ways that an input stem can be related to its output form — in
every way from suppletion, at one extreme, to complete predictability by
rule, at the other. What is needed is a set of stem rules, expressing default
generalizations about stem-to-stem relationships and using the same set
of operations (prefixations, suffixations, vowel shifts, consonant shifts,
reduplications, metatheses, subtractions, and so on) as inflectional and
derivational rules. The question then arises whether the same stem serves
as input for all such rules, or whether, say, stem 2 is built on stem 1 for
some lexemes and stem 1 on stem 2 for others; Stump (1984) has argued
for the latter position.
3.10. Slot calculus

Slots cannot be identified (in general) with specific sets of fillers, though
in simple systems of inflectional morphology it might be convenient to
use a shorthand reference like "the plural slot" (instead of saying "slot
A, which is filled only by material realizing the grammatical category
plural"). Slots constitute another dimension of abstraction, independent
to some degree of the dimension of grammatical categories.
Slots also cannot necessarily be identified with particular locations
with respect to the stem. In English, for instance, when nothing happens
to fill slot A, material in slot Β will end up immediately after the stem
(as in the genitive singular child's). Slots are only potential locations. If
there is a slot that must be filled in some language, then this fact must
be stated explicitly, as part of a tactics, or calculus, of slots.
Such a slot calculus is also needed to provide for the possibility that
slots might occur in one order in the context of certain features (say, in
non-finite forms) but in a different order in other contexts; slot-ordering
conditions will do the descriptive work that might otherwise be attributed
to "morph(eme) metathesis" rules. Thus, in the spirit of generalized
phrase-structure grammar (Gazdar et al. 1985), I propose that the stip-
ulations about what can fill particular slots be separated in principle from
stipulations about the ordering of those slots.
3.11. Cross-lexeme referral

Finally, we need the ability to stipulate that the stems and forms of one
lexeme are identical (either fully or by default) to those of some other
lexeme — a sort of cross-lexeme referral. This is what it means to say
that the English lexemes progressive-BE (/ am singing), passive-BE (/ am
attracted by penguins), and obligative-BE (/ am to leave for Vienna soon)
are, for the purposes of morphology, default-identical to the lexeme
copulative-Β Ε (/ am happy). The different BE lexemes have quite different
syntax and semantics, but essentially the same morphology.
4. The lexicon
Here I would like to encourage a view of the lexicon of a language as
the set of all relevant information about its words, hence as the domain
of facts to be described by morphological rules for the language. The
domain of facts to be described by the syntax of a language is the set of

all relevant information about its phrases and larger expressions; this has
sometimes been (confusingly) referred to as the "language", but I suggest
the term "syntacticon", as an evident parallel to "lexicon".
Like the syntacticon, the lexicon in this sense is highly redundant, and
is either not finite or at least astronomically large. The rules of syntax
extract generalizations about the syntacticon, describe what is predictable
or redundant in the contents of the syntacticon; the rules of morphology
correspondingly extract generalizations about the lexicon, describe what
is predictable or redundant in the contents of the lexicon. Both sets of
rules must be supplemented by a list of stipulated eccentricities, which
we might refer to as the "idiosyntacticon" and "idiolexicon", respectively.
Linguists sometimes talk as if there were a substantive issue as to
whether the lexicon contains all relevant information about words or just
what is idiosyncratic. Perhaps some confusion has resulted from different
metaphors supporting references to "lists", but there are simply two
different concepts here. Given the demonstrations of Jackendoff (1975)
that we cannot expect there to be a unique idiolexicon for any particular
lexicon, however, I must admit that it is the (full) lexicon that must be
the primary object of theoretical interest. And I must stress that the
lexicon in this sense, like the syntacticon, is an abstract object; no claims
are being made about what the "mental lexicon" is like, or what a
computationally tractable model of a lexicon would be like, important
though these questions are. Nor are any claims being made about the
character of the sets of primitive signs that serve as the bases for con-
structing the lexicon and syntacticon — "the list of primes, which are the
words in syntax and the morphemes in morphology", as Di
Sciullo — Williams (1987: 21) would have us believe. Nor are any claims
being made about whether the theory of grammar should accord any
special status to those subsets of the lexicon and syntacticon whose
members have some idiosyncratic properties (the "listemes" of Di
Sciullo — Williams).
5. Inflectional morphology in relation to derivational

morphology and phonology
As in Zwicky (1988), I assume that derivation shares its operations with

inflection (and stem rules), though without the slot organization char-
acteristic of inflection. The phonological effect of a derivational rule is
to predict stems of one lexeme on the basis of information about other

lexemes.
The simplest way to achieve this effect is to stipulate that the primary
stem of the output is identical to the primary stem of the input; this is
"zero derivation" or "conversion", and it is amply illustrated in languages
of all morphological types. Slightly more complex is the conversion of a
non-primary stem. The next simplest way is to stipulate that the primary
stem of the output results from applying a set of operations to a specified
stem (normally the primary stem) of the input, and this too is amply
illustrated, especially in affixal derivation. There is at least one other
possibility, namely derivation that uses a stipulated form of the input,
say the feminine form of an adjective used in the derivation of an adverb
(as in French FAUSSEMENT, built by suffixation on the feminine form
fausse of the lexeme FAUX); derivation of this sort can produce the
apparently paradoxical effect of inflection "inside" derivation.
As for phonology, I assume a component division between "morphon-
ology" and phonology proper, akin to the distinction in Dressier (1985)
and ultimately based on the natural phonologists' differentiation between
the "rules" of non-automatic phonology and the "processes" of automatic
phonology (Donegan — Stampe 1979), the functions of morphonology
being primarily demarcative and culminative, while automatic phonology
serves the more strictly phonological functions of pronounceability and
perceptibility.
Automatic phonological rules describe shape adjustments with direct
phonetic motivations, and they apply within "prosodic" (that is, purely
phonological) domains, like the syllable, foot, phonological word, and
phonological phrase. Morphonological rules draw their operations from
the same inventory as inflectional and derivational rules (and stem rules
and shape rules) — these operations varying in phonetic naturalness,
though their historical origins in automatic phonology are usually visible
— and apply within morphosyntactic domains and across morpheme
boundaries.
6. The syntax of inflectional morphology

I adopt here, without further argument, the Principle of Morphology-
Free Syntax, also known as the Lexicalist Hypothesis (see Zwicky (in
press) and Scalise (1984: 101 — 102, 191-196)) for discussion and refer-
ences), according to which syntactic rules placing conditions on syntactic

representations are blind to the internal structures and derivational his-
tories of the words occurring in those representations. A careful formu-
lation of the principle must not, of course, outlaw all sorts of reference
to matters morphological in syntactic rules; rather, it should confine such
reference to (abstract) grammatical-category features, while forbidding
mention of particular morphological rules or operations within syntactic
rules. On this view, no syntactic rule can be sensitive to the application
of the regular realization rule for plural in English, to the application of
an inflectional rule suffixing /z/, or indeed to the paradigm class of a
lexeme.
The syntactic component of a grammar, then, incorporates a number
of conditions on the distribution of grammatical category features, via
what are traditionally called "government" rules (saying, for example,
that in German the default case of a direct object is accusative, and that
the case of the direct object of HELFEN is dative) and "agreement"
rules (saying, for instance, that adjectives must have person, number, and
case values consistent with those of the nouns they modify). The usual
situation is for grammatical-category features to be located on the head
word of the relevant constituent (V in a VP composed of V + NP or V
+ VP, for instance), but we must apparently also allow for edge place-
ment, in which features are located on the first (or last) word of the
relevant constituent, whatever that word might be (see the discussion of
Tongan in Zwicky 1987b). Except insofar as the values of grammatical
category features are constrained by such rules, they can be freely in-
stantiated.
As a general principle, material satisfying the conditions of syntax can
be paired with material satisfying the conditions of morphology whenever
these are compatible (and, as we shall see, in certain other instances as
well). This principle allows for the matching of material that is more
highly specified in morphological representations than in syntactic ones
(as when a lexeme with various "purely morphological" features, such as
paradigm-class features, can be paired with syntactic representations that
are innocent of such features) and for the converse (as when a lexeme
that lacks number specification, as I would claim the English expletive
THERE does, can serve as subject both when syntactic conditions call
for a singular subject and when they call for a plural, as in There is no
answer and There are no answers).
An expression will then be ill-formed if it fails to satisfy either a
syntactic condition (e.g., the condition that verbal complements of English
modal verbs must be in their base form: We will be / *being / *been /

*are quiet) or a morphological condition (e.g., the stipulation that obli-
gative-BE, like the English modal verbs, has only finite forms; compare
* We will be to be quiet with We are to be quiet and We will have to be
quiet).
6.1. Inflectional morphology and particle lexemes

It might seem that inflection is a topic of rather limited interest to the
theory of grammar, given the fact that so many languages have little or
no inflectional morphology. However, there are special relationships
between inflection and a type of lexeme that is, so far as I know,
exemplified in all the world's languages, so that inflectional morphology
cannot be easily dismissed as a collection of exotica. These lexemes are
variously (sometimes rather desperately) labeled as "grammatical words",
"particle words", "nonlexical items", "little words" or "particles"; I will
call them "particle lexemes".
Particle lexemes serve as marks of syntactic constructions in the same
way that inflection does. What one language does with inflection, another
does with particle lexemes. Within a single language, such lexemes can
occur in alternation with inflection (the discovery OF the city vs. the
city'S discovery, MORE handsome vs. handsome-R) or in combination
with it (Swedish DET store hus-ET 'the big house', with both a definite
article and a definite suffix on HUS, or English a friend OF Robin'S).
I have treated inflection as the realization of grammatical category
features on individual lexemes, and I now propose that particle lexemes
are also realizations of grammatical category features — pure combina-
tions, in fact, of syntactic and grammatical category features, with no
other semantics (see Zwicky (to appear) for further development of this
proposal). Syntactically, they are simply words — that is, "word-rank",
"lexical", or "zero-bar" syntactic categories — with grammatical category
features, like other syntactic words. Morphologically, they are lexemes
with all the properties of other lexemes except that they have defective
semantics.
Particle lexemes vary enormously in their integration with the rest of
the lexicon. At one end of the scale are those that have only a single
form and also serve as unique bearers of a set of grammatical categories.
In English, for instance, the infinitive marker TO is a particle lexeme —
suppose it is lexeme # 26 in the English lexicon — belonging to the
auxiliary subcategory of the verb category, but with only one form,
realizing the grammatical category infinitive, and it is the only verb

lexeme having this form; this is what saying that TO is the mark of the
infinitive amounts to. Similarly, the dative marker TO is a particle lexeme
— lexeme # 253, say — belonging to the particle category, but with only
one case form, realizing the grammatical category dative, and it is the
only particle lexeme having this form; this is what saying that TO is the
particle marking the dative case amounts to.
Other particle lexemes serve as representatives of (sub)categories and
have more or less full form sets. Copulative-BE, for instance, is a particle
lexeme — lexeme # 3, say — representing the copulative subcategory of
the verb category (a subcategory with such other members as BECOME,
STAY, GET, and SEEM), and it has a complete set of verb forms.
Similarly, the English personal pronoun lexemes like WE and IT are
particle-lexeme representatives of a subcategory of nouns, with no se-
mantics beyond that associated with this subcategory and with the gram-
matical categories of person and number.
A property of expressions that is distributed via grammatical category
features is a property — like constituency, linear ordering, and certain
shape properties — that is available to serve as a mark of particular
constructions. It is just one item in the tool kit that the grammar of a
particular language makes available for this purpose. As Sadock — Zwicky
(1985) observe, a yes-no question construction might be marked, in one
language or another, by a constituent ordering, by an intonational con-
tour, by a verbal inflection, by a particle lexeme, or by several of these
in concert.
Such a mark might have a characteristic semantics of its own, but this
is only a default and can be overridden by the semantics associated with
the construction. As a result we cannot expect always to be able to find
a "meaning" for such a mark in all of its occurrences. Grammatical
categories, in particular, might serve a number of syntactic functions, as
when the English present participle is used both in progressive VPs (They
were playing Mozart) and in postnominal VP modifiers (Anyone having
a hat on will be arrested), or when a past participle is used both in passive
VPs (They were praised by everyone) and in perfect VPs {They have been
to Vienna many times). The English base form has an extraordinarily
wide range of uses, illustrated in I made them be quiet, To be quiet is
impossible, You can be quiet, Be nice to your guests!, Go be nice to your
guests, and I watched you be nice to your guests.
Saying that particle lexemes as well as inflection are realizations of
grammatical categories then predicts that the same sort of diversity in
syntactic function is possible, and this prediction is correct. In English,

for instance, we have the superlative particle lexeme MOST used not
only in the true superlative construction (You are the most worldly person
I know), where it is in alternation with an inflectional variant (You are
the worldliest person I know), but also as the mark of an "absolute"
construction (You are most worldly), where the inflectional variant is
unavailable (*You are worldliest). The English infinitive particle lexeme
TO has an extraordinarily wide range of uses, illustrated in It's too heavy
to lift, I intend to talk, It's hard for us to hear the orchestra, The person
for you to see is Kim, Oh to be in Vienna!, and To understand this you
have to be able to read Hungarian.
6.2. Clitics
The simple picture sketched so far is known to be inadequate. At least
some of the items that have been labeled "clitics" require genuine com-
plications in the scheme of component interfacing.
At the outset, I must discard a collection of "leaning" elements as
being beside the current point. These are elements that are simply pho-
nologically dependent on adjacent material, forming prosodic units —
phonological words or phonological phrases, in particular — with them
(see Kaisse — Zwicky 1987 for a compact discussion and references). The
English complementizer particle lexeme THAT, for instance, can belong
to the phonological phrase of material following it, so that that Chris
can make a phonological phrase within the sentence I know that Chris
has gone. So long as the syntax and morphology of these elements present
no special features, they are not of interest in the present context, though
they do require stipulations via principles of prosodic domain formation,
principles describing the interface between morphosyntax and automatic
phonology.
It is also true that "syntactic dependence" does not in itself necessarily
cause difficulties for the description of the morphology-syntax interface.
Syntactic rules must be able to locate certain classes of lexemes .by
reference to the contents of some host constituent, in particular by
reference to the host's head or edge words — to stipulate, for instance,
that some class of adverbs is limited to occurrence after the first word
(that is, after the first unit of word rank) in a clause. In many cases this
is all that need be said.
What then of "phrasal affixes"? (The contrast is with "bound words",
the terminology being that of Nevis 1986 rather than of my own earlier
work on clitics.) Consider the English genitive. So far as its syntax goes,
it works in the same way as an affix; it realizes a grammatical category
feature, here with edge location. The fact that the realization of genitive
interacts with the phonological shapes of ordinary inflection leads us to
assign it to an outer layer of inflectional morphology. (This is not Nevis's
treatment, but it is the one advanced by Kanerva 1987 for Finnish
possessives and Zwicky 1987b for English genitives.) It appears that in
general phrasal affixes are to be analyzed syntactically as grammatical
category, features distributed like other such features (though the usual
situation is for phrasal affixes to be located at constituent edges, for
ordinary inflection to be located on heads) and morphologically as
constituting an outer layer of inflectional morphology.
We are left with bound words, like the English reduced auxiliary clitics
(auxiliary reduction: I'd be quiet, It's been noticed). Each bound word
instantiates a lexeme (WOULD, HAS) and so should be treated syntac-
tically as a formative, as a word, rather than as features.
It is sometimes suggested that bound words are just phonologically
dependent words, but several facts suggest otherwise: (a) The set of
bound-word clitics is often lexically idiosyncratic, as when the forms was
and were fail to participate in auxiliary reduction; (b) bound-word clitics
are often subject to surface filters (the "surface-structure constraints" of
Perlmutter 1971 and consequent literature) that limit their combinations
with one another and with their hosts, and constrain the ordering within
these combinations; and (c) bound-word clitics often show special mor-
phophonemics, as in the auxiliary-reduction shape alternations /wud/ vs.
/d/ and /haez/ vs. /z/. Instead, we need to say that bound word clitics in
combination with their hosts make a new sort of word-like unit — what
I will call a "morphosyntactic word" — for the purposes of morphology,
while they are simply independent words for the purposes of syntax. (The
idea that the words of syntax and the words of morphology need not be
coextensive is now a familiar one, thanks to such works as Sadock 1985
and Di Sciullo-Williams 1987.)
The treatment of bound words as part of morphosyntactic words
means that they too fill slots, and we can expect the principles of inflection
to carry over to bound words. The sort of morphosyntactic word that
deserves the label "clitic group" is then in effect an inflected word built
on an inflected word as stem, and the surface filters that constrain the
constituency of these clitic groups and the ordering of their parts are
nothing more than a slot calculus at a new level.
I must point out that which lexemes are bound-word clitics, or have
bound-word clitic alternants, in a language cannot necessarily be pre-
dicted on the basis of other properties of the lexemes. Bound words can
be spread across the syntactic categories of a language; some particle
lexemes are bound words, but then some (like dative TO and the personal
pronouns in English) are not; some closed (sub)category lexemes are
bound words, but then some (like the directional particle TO and the
auxiliary SHOULD in English) are not; and some minor category lexemes
are bound words, but then some (like the English quantifiers EACH,
ANY, EVERY, ALL) are not.
Bound-word clitics are, of course, not the only phenomena that appear
to require a divergence between the representations appropriate for syntax
and those appropriate for morphology, and so to require that we for-
mulate principles describing the interface between morphology and syn-
tax, principles (whether parochial or universal) constraining the associ-
ations between morphosyntactic words and syntactic words. There is in
fact a panoply of such phenomena, including portmanteaus (French du),
compounds (English apple eater), serial verbs (English go look), clause
unions (French faire partir), and incorporations (West Greenlandic ga-
mutegarpog 'sled-have'). Bound-word clitics are also not the only in-
stances of word-like units (whether in syntax or in morphology) that
properly contain units of similar type; so do compounds, serial verbs,
clause unions, and incorporations.
7. Conclusion
I have tried to give some of the flavor of the Interface Program as it

concerns inflectional morphology. The program relies heavily on the
assignment of rules to components or subcomponents, the interactions
between which are largely determined by universal principles (but partly
by language-specific conditions). Within the components of inflectional
and derivational morphology, as well as within syntax, rules are viewed
as pairings of semantic content with sets of formal features, and on the
formal side, as static conditions on the well-formedness of the lexicon
and the syntacticon. A logic of defaults and overrides plays a major role
in determining the interactions between these conditions, and the sepa-
ration of declarations about the combinability of material from decla-
rations about ordering that has been explored in the recent syntactic
literature is carried through to morphology, where I have suggested that
slot filling and slot ordering should be separated as well.
References
Linguistic Theory 4: 1—31.
1988a "Inflection", in: M. Hammond —M. Noonan (eds.) Theoretical morphology
(Orlando: Academic Press), 23—43.
1988b "Morphological theory", in: F. J. Newmeyer (ed.) Linguistics: The Cam-
bridge survey (Cambridge: Cambridge University Press), vol. 1, 146—191.
Carstairs, Andrew
1987 Allomorphy in inflexion. (London: Croom Helm).
1990 "Phonologically conditioned suppletion", this volume, 17 — 23.
Di Sciullo, Anna Maria —Edwin Williams
1987 On the definition of word (Linguistic Inquiry Monograph no. 14) (Cam-
bridge, MA: MIT Press).
Donegan, Patricia J. —David Stampe
1979 "The study of natural phonology", in: D. Dinnsen (ed.) Current approaches
to phonological theory (Bloomington: Indiana University Press), 126—173.
1985 Morphonology: the dynamics of derivation (Ann Arbor: Karoma).
Gazdar, Gerald — Ewan Klein — Geoffrey Pullum — Ivan Sag
1985 Generalized phrase structure grammar. (Oxford: Basil Blackwell).
Haiman, John
1980 Hua: a Papuan language of the Eastern Highlands of New Guinea. (Amster-
dam: Benjamins).
Hale, William G . - C a r l D. Buck
1903 A Latin grammar [reprinted 1966] (University of Alabama: University of
Alabama Press).
Hinnebusch, Thomas J.
1979 "Swahili", in: T. Shopen (ed.) Languages and their status (Cambridge, Mass.:
Winthrop), 209-293.
Jackendoff, Ray
639-671.
Kaisse, Ellen M. —Arnold M. Zwicky
1987 "Introduction: syntactic influences on phonological rules", Phonology Year-
book 4: 3 - 1 1 .
Kanerva, Jonni M.
1987 "Morphological integrity and syntax: the evidence from Finnish possessive
suffixes", Language 63: 498 — 521.
Kiparsky, Paul
1982 "From cyclic phonology to lexical phonology", in: H. van der Hulst — N.
Smith (eds.), The structure of phonological representations, part 1, (Dord-
recht: Foris), 131-175.
Matthews, Peter
1972 Inflectional morphology: a theoretical study based on aspects of Latin verb
conjugation. (Cambridge: Cambridge University Press).
Nevis, Joel A.
1986 Finnish particle clitics and general clitic theory [PhD dissertation, Ohio State
University 1985] (OSU Working Papers in Linguistics 33).
Perlmutter, David M.
1971 Deep and surface structure constraints in syntax. (New York: Holt, Rinehart
& Winston).
Sadock, Jerrold M.
1985 "Autolexical syntax: a theory of noun incorporation and similar phenom-
ena", Natural Language and Linguistic Theory 3: 379—439.
Sadock, Jerrold M. —Arnold M. Zwicky
1985 "Speech act distinctions in syntax", in: T. Shopen (ed.), Language typology
and syntactic description (Cambridge: Cambridge University Press), vol. 1,
155-196.
Sapir, Edward
1921 Language: an introduction to the study of speech. (New York: Harcourt,
Brace).
Scalise, Sergio
Spencer, Andrew
1988 "Arguments for morpholexical rules", Journal of Linguistics 24: 1 —29.
Stump, Gregory T.
1984 "Two approaches to predictive indeterminacy", Linguistics 22: 811—829.
Wurzel, Wolfgang U.
1990 "The mechanism of inflection: lexicon representations, rules, and irregular-
ities", this volume, 203-216.
Zwicky, Arnold M.
1985 "How to describe inflection", Proceedings of the Berkeley Linguistic Society
11: 372-386.
1986 "The general case: basic form versus default form", Proceedings of the
Berkeley Linguistic Society 12: 305 — 314.
1987a "Phonological and morphological rule interactions in highly modular gram-
mars", Proceedings of the Third Eastern States Conference on Linguistics
(Columbus: Ohio State University), 523 - 532.
1987b "Suppressing the Zs", Journal of Linguistics 23: 133-148.
1988 "Morphological rules, operations, and operation types", Proceedings of the
Fourth Eastern States Conference on Linguistics (Columbus: Ohio State
University), 318-334.
to appear "Jottings on adpositions, case inflections, government, and agreement", in:
D. Brentari et al. (eds.), The joy of grammar: A Festschrift for James D.
McCawley.
in press "The morphology-syntax interface", Oxford International Encyclopedia of
Linguistics.
Topic 4: Computer morphology
Morphology in LDOCE and in the ASCOT system
Willem Meijs
1. The ASCOT project
The main goal of the "ASCOT" research project 1 was the development,
on the basis of the computerized Longman Dictionary of Contemporary
English (LDOCE for short, editor-in-chief Paul Procter, published 1978),
of an automatic tagging system, i.e., a software package which can
"process" an uncoded text word by word and attach one or more gram-
matical code-tags to every word that it can "recognize".
ASCOT consists of three interacting components: a scanner, a lexicon
(Aslex), and a morphological component (Reroot). The scanner inspects
incoming text-sentences word by word. For each word-form it encounters
it makes a search through Aslex to see if there is an entry identical in
form to the text-word. Since ASCOT must make many passes through
the lexicon, Aslex is stored in the form of an L-tree (Skolnik 1980), which
allows very fast access. If an entry (or entries) identical to the text-word
is/are found, the relevant information stored under the entry or entries
is retrieved and copied onto the text-item. We call this a "straight hit".
If no such entry/entries can be found control is passed to Reroot. By
peeling off possible affixes and comparing the remainder with the entries
in Aslex, Reroot tries to establish whether the text-word is perhaps an
inflected and/or derived form based on an Aslex-entry (for instance, a
plural or a past participle, or a form prefixed with non-, pseudo-, etc.). If
this turns out to be the case, information associated with the base-form,
insofar as it is still relevant to the inflected or derived form, is copied
onto the text-item, along with an indication of the inflection or derivation
involved. We call this an "indirect hit". If neither the straight search via
the scanner, nor the indirect one via Reroot is successful, a "no success"
code is attached to the text-word and the scanning passes to the next
text-word.
Often a straight hit does not exhaust all of the information that could
be retrieved. Thus it can happen that some item may be an inflected/
240 Willem Meijs
derived form or a base form. Thus the noun fitter would be found as a
straight hit (since it has an entry of its own), but if the system were then
to pass on straightaway to the next word, the information (retrievable
via Reroot) that this word might also be the comparative form of the
adjective fit would go unnoticed. Similarly, the straight hit moped as noun
(meaning "bike with small engine") would block its identification as past
tense or participle of the verb mope, the form summons could be either a
noun or third person present tense of the verb summon, etc. (cf. Akkerman
et al. 1985:51, for some more examples). Since one cannot predict be-
forehand for which straight hits further morphological analysis via Reroot
might reveal additional information, the scanner should in fact always
make two passes, one to scoop up the straight hits, the other to reap the
indirect ones.2
2. The L-tree data-structure
The basic data-structure in the ASCOT-system is a "letter-tree" in the

so-called L-tree format (Skolnik 1980). Figure 1 gives a minute but
representative portion of the Aslex L-tree. (Naturally the example is not
meant to be exhaustive.)
The organization of the L-tree is such that at any stage in the matching-
process you know "where you're at". Each string of characters in Figure
1 represents an "element" of the structure. Such an element thus contains
an entry or part of an entry, and in addition one of the following: a
"continuation link" to the first element in a linked list of continuations
(represented by hyphens in Figure 1), a "successor link" to the next
element in the linked list of which the element is a member (vertical lines),
and/or a pointer to "information" on an entry or prefix (superscripts).
At each point in the tree-structure it is thus clear whether the string
developed up till then constitutes an entry in its own right or not, whether
it forms a first part (possibly a prefix) of some entry or entries, and
whether there are alternative branches to some (other) entry or entries.
Trying to find an input-word from a text in Aslex thus means trying to
find the largest possible letter-by-letter match against the Aslex L-tree.
If no complete match can be found, the program must check whether
the input-word could possibly be a derived and/or inflected form of an
entry-word that does occur in the L-tree structure. So this is where
morphology enters the picture.
abacus1"
re + -a-d v -able A
I
iA
gret v/N -ful A
s-cue v/N -er N
earch v/N -er N
se-a N -kale N
I I
e V !V/N. e r N
un + -der + p -mine v
ite v wear v
just A
re-quitedA
I
s-ervedA
I
jV/N
zulu N
Figure I.
3. Morphology in ASCOT
Morphology played an important role in the ASCOT project in a number

of ways. On the input side there was the question of abbreviated complex
and derived forms as we found them in the computerized dictionary. On
the output side there was the requirement of a morphological component
which could recognize inflected and derived forms that were not to be
found as such in the dictionary.
242 Willem Meijs
3.1. Morphological input

In LDOCE many compound and derived forms are entered in full.
However, often they are entered in an abbreviated form under the head-
word on which they are based. 3 Thus the noun philanderer is indicated
as -derer in the entry for the verb philander, tidiness as -diness in the entry
for tidy, librarianship as ~ ship under librarian, bearably as -bly under
bearable, balefully as ~fully under baleful and so on. In the ASCOT
project we wanted to restore such entries automatically to their full form.
Now while human users on the whole find it easy enough to interpret
such abbreviated forms, it turned out to be quite difficult to devise
efficient algorithms to make the computer do this automatically.
In principle the forms with a tilde should be a matter of simple
arithmetic, rather like doing sums: just add the ending to the base-form
of the entry, as illustrated in (la):
(la) crowded librarian opaque
~ness ~ ship ~ly
crowdedness librarianship opaquely

In practice we sometimes came across inconsistencies, however. Cf. (lb):
(lb) baleful smutty existentialism
~fully ~~ tiness ~ist
*balefulfully * smutty tiness *existentialismist

The forms in (lb) were basically due to errors in the input — the endings
should have had hyphens rather than tildes.
For the forms with a hyphen ("-") we developed a different algorithm,
which tried to establish the largest sequence of letters that the "suffix"
(looking at it from left to right) and the headword (at its rightmost side)
had in common, subsequently deleted all letters in the right part of the
headword, starting with the identical ones, and then added the "suffix".
Thus in the case of -ation in the entry for overcompensate the sequence
at is found as the largest common righthand substring in overcompensate,
so all letters, starting with at are deleted, leaving overcompens to be
correctly combined with ation to form overcompensation. Similarly phi-
lander would first be truncated to philan (on account of the common
righthand substring der), and derer added to it to yield philanderer. In
addition we had to reckon with certain systematic spelling variations,
such as the change of y into i, the disappearance of mute e etc.; cf. (2):
(2) tidy iy > i) bearable (e > 0 ) philander

-diness -bly -derer
tidiness bearably philanderer

After a few modifications the algorithms for restoring the abbreviated
forms with hyphens and tildes worked quite satisfactorily. Thus the final
algorithm produced incorrect output in only 29 out of 3,819 cases. All
remaining incorrect forms (less than 1%) were corrected manually.
3.2. Morphological output

Clearly a tagging system that can only recognize base-forms of entries in
the dictionary is of limited value (even if it includes the expanded abbre-
viated forms discussed in the preceding section), since obviously any text
to be tagged will also contain many inflected and derived forms. On the
output side the development of a morphological component was therefore
one of the most important subgoals of the ASCOT project.
The basic procedure is simple and straightforward; successively peeling
off letters, working both from right to left and from left to right in a
recursive fashion, it tries to establish whether the letter(s) peeled off
constitute an inflectional or derivational affix, and whether the remainder
corresponds to an entry in the dictionary. The results are constantly
checked against general rules for spelling changes and against specific
information contained in the entries encountered. Thus in a case like
taped general rules about amalgamation of final mute e versus doubling
of final consonants, as well as specific information about such doubling
in the entry for tap lead to the outcome that this is a past tense or past
participle form of tape rather than of tap. In a case like putting two
analyses are correctly provided, one relating it directly to the entry putt,
the other relating it to put, via consonant doubling. The form putted, on
the other hand, is related to the entry putt only, since information specific
to the entry for put tells the analyzer that this is a strong verb with
irregular past tense and past participle forms identical to the stem, hence
putted cannot be past tense or past participle of put.
From the programming point of view it would not have been too
difficult to also include a compound analyzer. A simple algorithm to
check whether a given unlisted input form could be split up into, say, a
listed noun (or adjective) followed by another listed noun, would take
care of the most common nominal types of compounds, for instance.
244 Willem Meijs
However, assuming a system which does not stop at "straight hits", the
number of perfectly ordinary words that would get an additional com-
pound analysis in a second pass turned out to be astounding. To give a
few examples, the adjective barbed (as in barbed wire) would be analyzed
as a compound of the nouns bar and bed, the adjective paintable as a
table having something to do with pain, the verb dampen as some sort of
pen somehow connected with a dam, while ticking would presumably be
interpreted as "the king of tics"! Similarly disastrous results would ensue
for words like beefish, flushed, rampant, reddish, and innumerable other
ones; cf. the bracketings in (3):
[|pain]^[table]^]N vs [\paint]vable]A
vs [[barb]êd\A
Pe*>]N[/i's/2]N]N vs [[beef\Nish]A
[[/7w]N[s/ii?i/]N]N vs [[flush]ved]v.e(i
vs [[red(d)]Aish]A
[[ram]N[/?fl«i]N]N vs [rampant]A
vs [[damp]Aen]v
[[fic]N[fci#ig]N]N vs [[tic(k)]ving]v_ing
The morphological analyzer, Reroot, as contained in the present ASCOT

package, is thus fairly restricted. Inflection is covered fully, derivation is
restricted to one affix layer, and compounding is not included. At the
moment, Reroot is rather slow, thereby undoing some of the speed of
the L-tree access. We are therefore working on an improved morpholog-
ical analyzer, Morphgram, which will speed up morphological processing
considerably and at the same time bring it more in line with the theoret-
ical-linguistic and psycholinguistic views regarding the lexicon and its
interaction with morphology developed in Meijs (1975, 1981, 1985). A
more detailed discussion of these considerations can be found in Meijs
(in press).
4. Conclusion
What distinguishes ASCOT from other approaches is that it does not

restrict itself to a theoretical sample-lexicon with a few hundred items,
but that it constitutes a realistic lexicon making some 50,000 entries, each
with a wealth of grammatically relevant information, available for natural
language processing tasks. And via its morphological component the

number of items that it manages to process can in fact be multiplied by
a~ factor four or five.
Notes
1. "ASCOT" (which stands for "Automatic Scanning System for Corpus Oriented Tasks")
is the name of a project supervised by the author and funded by the Dutch Organization
for the Advancement of Academic Research (ZWO) under proj. no. 300-169-004. This
project was carried out between 1-3-1984 and 28-2-1987, with a three-month extension
financed by the Amsterdam University Arts Faculty. Research assistants were: Pieter
Masereeuw (until 1-7-1985), Hetty Voogt-van Zutphen, and Eric Akkerman. For details
about ASCOT see Akkerman — Masereeuw — Meijs (1985) and Akkerman — Meijs —
Voogt-van Zutphen (1987, 1988).
2. Actually, we have made the choice between one or two passes a "user option", with two
passes as the default situation. The "one-pass" option is provided for syntactic analysis
systems that have a morphological analyzer of their own.
3. As it turns out, there are three different formats in which morphologically complex
words can occur in LDOCE: (1) in full, as headword, (2) in full, within (usually at the
end o f ) the entry for the base-word, and (3) in truncated form, with hyphen or tilde,
within the entry for the base-word. What motivates the choice of the particular format
adopted is not always very clear.
References
Akkerman, Eric —Masereeuw, Pieter—Meijs, Willem J.

1985 Designing a computerized lexicon for linguistic purposes: ASCOT Report
No 1 (Amsterdam: Rodopi).
Akkerman, Eric —Willem J. Meijs —Hetty J. Voogt-van Zutphen
1987 "Grammatical tagging in ASCOT", Corpus linguistics and beyond, edited
by W. J. Meijs (Amsterdam: Rodopi), 181-193.
Akkerman, Eric —Hetty J. Voogt-van Zutphen —Willem J. Meijs
1988 A computerized lexicon for word-level tagging: ASCOT Report No 2 (Am-
sterdam: Rodopi).
LDOCE
1978 Longman dictionary of contemporary English (editor in chief: Paul Procter)
(Harlow —London: Longman).
Meijs, Willem J.
1975 Compound adjectives in English and the ideal speaker-listener. (Amsterdam:
North-Holland).
1981 "Synthetische composita: Voer voor morfologen", Spektator 10: 250 — 291.
1985 "Lexical organisation from three different angles", ALLC Journal 6:1 —10.
in press "Morphology and word-formation in a machine-readable dictionary: prob-
lems and possibilities", Folia Linguistica 24/1 —2.
Skolnik, J.
1980 L- trees [Technical Report, Amsterdam Arts Faculty Computer Department],
Topic 5: The psycholinguistic study of
morphology
Morphology and the mental lexicon: psycholinguistic
evidence
Bruce L. Der wing
1. Introduction
As the problem of the mental lexicon is a massive one, by all accounts,

this paper will limit itself to the following two controversial distinctions
related to it: (1) storage vs. retrieval and (2) rules vs. analogical networks.
Both of these will be shown to bear importantly on the question of the
overall character and organization of the mental lexicon, on the place of
morphology in relation to it, and on the fundamental issue of the proper
characterization of the phenomenon of morphological productivity. In
all cases the conclusions drawn will be based on what I perceive to be
the best available psycholinguistic evidence, with some speculation also
allowed regarding what I see as the most promising directions for future
research in this area.
2. Storage vs. retrieval
In an extensive critique published fifteen years ago (Derwing 1973), I

noted that, from an information-processing perspective, the key principles
of evaluation invoked in classical generative phonology related exclusively
to the storage side of the ledger, implying that the best linguistic system
was the one that maximized generality (or specification by rule) and
minimized the amount of information specified in the lexicon. I then went
on to say that (1973: 154, n. 2):
What is ignored in this approach is the problem of retrieval: how

difficult is it to learn, recall and use [the resulting] very skeletal and
abstract lexical representations and ... rules in order to produce or
comprehend speech?
250 Bruce L. Derwing
In her recent book, Words in the mind, Jean Aitchison (1987: 9) makes
much the same point in terms of a very telling analogy. Suppose, she
suggests, that we viewed the mental lexicon as something analogous to a
library, in which information was stored in the forms of books. From
the standpoint of economy of storage alone, the most efficient approach
would obviously be to stack books of the same size together, filling each
room from wall to wall and floor to ceiling with tightly packed piles.
From the standpoint of ease of retrieval, however, this would constitute
perhaps the least efficient approach: to retrieve a book from such a
library, we would need information about its size (rather than its content)
in order to know what room to go to, and then we would have to extract
the books, one-by-one, pile-by-pile, working from the door to the back
wall, until we chanced to hit upon the particular volume of interest. Thus
we find that the real libraries that we are familiar with are organized on
quite different principles, placing a premium on ease of retrieval, with
few concessions to economy of storage.
Likewise, in real mental lexicons, too, we have no reason to think that
a heavy premium need be placed on economy of storage: the human
mind seems capable of storing almost unlimited amounts of information,
certainly far beyond the average person's normal everyday or even lifetime
needs, and in the case of words, can easily manage the 250,000 or so
entries that Diller (1978) estimates for the typical educated English-
speaking adult. (And this is in the monolingual case; the addition of one
or more other languages does not seem to impose any significant new
burden on the normal memory.)
On the other hand, we do have very good reasons for thinking that a
premium must be placed on the process of retrieval, in view of the fact
that the normal processes of speech production and (especially) compre-
hension must operate within time constraints that are extremely severe.
In one study, for example, it has been shown that native speakers can
recognize a word in their language in approximately one-fifth of a second
from its onset, or can reject a sound sequence that is a non-word in as
little as half a second (Marslen-Wilson — Tyler 1980) — and some of that
time involves the motor reaction after the decision is made. I conclude
that there can be no serious question that the human word-store is
primarily organized to enhance rapid and accurate retrieval and so we
should not be in the least bit surprised if a considerable amount of
redundancy in storage might be tolerated in order to expedite this process.
As Aitchison (1987) also reports in her excellent survey, the psycholin-
guistic evidence to date points consistently in this same general direction.
Whether taken from speech errors, aphasic speech, word-games, or from
Morphology and the mental lexicon: psycholinguistic evidence 251
controlled psycholinguistic experiments (as nicely illustrated by Stan-

ners—Neiser—Painton 1979), the evidence confirms what Butterworth
(1983) calls the full-listing hypothesis, that is, it presents a picture of a
lexicon that stores whole words, in sharp contrast to conventional wisdom
in linguistics.
The traditional linguistic view, of course, is that the lexicon should be
a repository of idiosyncratic, unpredictable information; that predictable
information ought to be characterized by rule; and that the basic lexical
and syntactic unit must therefore be the morpheme, not the word. I have
already emphasized, however, that the lexical storage economies so
achieved are done so at the very heavy cost of extra processing prior to
access. Moreover, as many observers have noted, including several in this
volume, these economies "cannot be extended to the semantic system,
since the meaning of derived words frequently cannot be recovered from
the meaning of their constituent parts" (Henderson 1985: 223; see also
Butterworth 1983: 264). So there is more than one good reason for a
lexicon of whole words.
3. Rules vs. analogies

In fact, there is only one good reason to question the full-listing hypothesis
for English, and this is the undisputed fact of morphological creativity
or productivity. We know from the naturalistic investigation of child
language (e. g., overgeneralization errors, such as foots and goed), as well
as from the results of controlled experimentation with both children and
adults (as in Berko's early work on English morphology [1958]7, that at
least some morphological operations are highly productive. For most
linguists, this has implied that "rules" of some kind must necessarily be
learned. And perhaps they are, for by now we recognize that predictability
by rule and specification in the lexicon are by no means incompatible or
mutually exclusive aims. The error, fundamental to classical generative
phonology, was to take the function of rules to be precisely that of
simplifying the lexicon (the economy-of-storage syndrome again). The
implicit argument was, why bother to learn rules, if everything has to be
listed in the lexicon, anyway?
There are, however, some perfectly sensible answers to this question,
as well. One, suggested by Vennemann some years ago (1974), was that
the function of rule-learning might be to help organize the lexicon, to
give it structure, which is something that is obviously going to have to
be done in one way or another, anyway, if efficient retrieval is to become

a realistic possibility. From this perspective, then, a rule does not simplify
the. lexicon by removing general information from it (e. g., by accounting
for differences in morpheme variants), but rather expresses generalizations
about what remains included in it. And learning such a rule involves
increasing the totality of one's knowledge, rather than decreasing it.
Moreover, it would be highly convenient to have at the ready a "fall-
back procedure" (Butterworth) or "lexical tool-kit" (Aitchison), which
would enable a speaker or hearer to reconstruct words in emergencies,
as when an old word is forgotten or the need arises to deal with some
new or borrowed word not yet present in the lexicon. So one can have
one's large lexical cake and productivity, too, as the data suggest. Thus
even with the full-listing hypothesis, the linguists' notion of "rule" is still
with us, at least for the time being — and it was this perspective, in fact,
that motivated my own early work on the English inflections, which
sought, through the systematic elimination of all the various alternatives,
to arrive at the one true form of "the" rule (or rules) involved.
One aspect of this particular line of research was an attempt to
ascertain the appropriate conditioning environments for the three regular
variants of the English plural suffix. One logical possibility, consistent
with Berko's results, was that the choice of the appropriate allomorph
was determined on the basis of the final rhyme element of the stem, i.e.,
that, on the basis of exposure to such real singular : plural word pairs as
bug : bugs, rug : rugs, plug : plugs, etc., a rule was learned that if the
singular form ended in /\g/, the suffix j-zf should be added to form the
plural. In this framework, an entirely separate rule would presumably
have to be learned for singulars ending in /ig/, based on such pairs as
pig : pigs, fig: figs, etc., another for /og/-stems, based on such pairs as
dog : dogs, etc. See analysis 1 A below, where some of these possibilities
are indicated.
Analysis 1A (Rhyme)
An alternative possibility, however, was that a single, more general,

rule might be learned for all such words, according to which the affix
was selected on the basis of the final consonant of the root alone, ignoring
the vowel, i. e., taking in the entire set of /g/-stems in one fell swoop, as
it were, as in analysis 1 B. (Though linguists seem almost constitutionally

prone to assume that it is the broadest generalization that is always
learned, it is, of course, an empirical question whether real language
learners do what linguists like to think they do. Hence the experiments.)
In any event, to check these two alternatives out, I tested some local
children on nonce-forms that didn't have any rhymes and, for the fuller
picture, I systematically varied the nonce-words used in such a way that
only the final consonant was left unchanged (as in the set wug / droig j
oog) and learned that it didn't make any (significant) difference. For each
stem-type, so long as the final consonant (or consonant cluster) was
preserved, the results were generally the same (see Derwing — Baker 1980
for a summary of these results).
At the time, the conclusion my colleagues and I drew from these data
was that the rhyme analogy was wrong and that the final-segment strategy
was right (or was at least still a candidate for the "one ultimate truth").
A new theoretical perspective has emerged in recent years, however, which
puts the preceding discussion in an entirely new light.
This approach goes informally by the designation of "connectionism,"
or, more technically, "parallel distributed processing", the most familiar
version of which is closely associated with the names of Rumelhart —
McClelland (1986). As the name implies, what all connectionist models
have in common — including a recent linguistic alternative to Rumel-
hart — McClelland proposed by Skousen (1989) — is the attempt to
account for morphological productivity through the establishment and
subsequent utilization of a network of analogical connections among full
lexical items.
One interesting innovation of the analogical approach is that one form
of a rule need not be chosen over another. In fact, in the familiar linguistic
sense of the term, there are no rules per se involved in such accounts at
all. As seen in Fig. 1, both the rhyme and final-segment analogies are
supported by sufficient connections to allow for a solution to the problem
in a Berko-type task.
Thus, from this perspective, Berko's subjects could have used either
the rhyming strategy (by tracing the dotted connection lines) or the final-
'bugs' 'rugs' 'pigs' 'dogs'
consonant strategy (by tracing the dashed lines in Fig. 1) in order to

pluralize the nonce form wug, and this fact is in no way compromised
by the Alberta subjects' performance on droig, for which the rhyme
strategy would fail.
Attention should now be directed to the morpheme boundaries ( + )
which appear in the node-labels of Fig. 1, which have been included to
emphasize two points: first, that, just as there is no essential incompati-
bility between a massive, redundant lexicon and linguistic productivity,
there is likewise no reason to think that full-word storage need imply
that lexical representations must remain unanalyzed. The experiments
show, in fact, to cite Cutler (1983: 73), that "morphologically complex
words are ... no more difficult to access from the lexicon than morpho-
logically simple words ... [and there is] abundant evidence that the mental
representations of words contain information about morphological struc-
ture, and that speakers draw on this information in creating new words".
Consider, for example, the results of some early lexical decision tasks
(e.g., Murrell — Morton 1974 and Stanners —Neiser —Hernon —Hall
1979), which showed that "morphological relatives act as effective primes,
and that regular relatives are as effective as the base [word] itself if the
base is the target" (Butterworth 1983: 288). Speech errors, too, point to
"close connexions among regularly related forms" (Butterworth 1983:
283). These data point to a set of linkages among lexical items that can
be schematically represented as in Fig. 2, where I have crudely indicated
the increasing strengths of the connections by simply increasing the
number of lines that link the words.
Thus, as the priming studies indicate, a base word like sing is connected
most closely to its inflectional variants (such as sings or singing), less
closely to its derivatives (such as singer), and is connected only very

weakly to any irregular forms that it might have (such as sang). Clearly,
the analogical-network framework would seem to be a quite natural way
to express a set of relationships of this kind, which are apparently going
to have to be represented in some way or another, in any event.
If we may return now to the case of the English plural inflection, there
is another widely overlooked characteristic of a fully adequate connec-
tionist account that should also be pointed out. The critical data are
partially shown in Tables 1 and 2 below, which represent successive
"stages" in a Berko-style study of the acquisition of the plural inflection
by English-speaking children. 1
It is evident from these data that performance on the nonsense /z/-
stem is quite comparable with — and in some cases even superior to —
that on the other sibilant-final stems. 2 What is noteworthy about this is
that, unlike the previous examples discussed, there seemed to be no
plausible /z/-stem plurals that might serve as analogical models in this
case. 3 In other words, it looked as though the children had learned to
pluralize this particular type of nonce-stem not by comparing it with
other /z/-stem plurals, which did not exist for them, but rather by
comparison to other (i. e., non-/z/) sibilant stems. The implied rule in a
linguistic account was thus along the lines of analysis 1 C below:
Analysis 1C (Feature)
— iz / [+sibilant]
— sibilant
-s/
(PI) < — voiced
— sibilant
-z/ + voiced
Table 1. Object clusters for group IV subjects (η = 25)
Responses
Cluster Phoneme
Irr. Null /IZ/ /s/ Μ Dupl.
Μ 0 4.0 0 0 96.0 0
Μ 0 0 0 0 100.0 0
Μ 0 0 0 0 100.0 0
/ö/ 0 0 0 0 100.0 0
IV 0 0 0 0 100.0 0
Μ 0 0 0 0 100.0 0
Μ 0 0 0 0 100.0 0
Η 0 0 0 0 100.0 0
Μ 0 0 0 0 100.0 0
AV 0 4.0 0 0 96.0 0
/g/ 0 4.0 0 0 96.0 0
Μ 0 8.0 0 0 92.0 0
Ν 0 12.0 0 0 84.0 4.0
IV 4.0 0 8.0 88.0 0 0

/Ρ/ 4.0 0 0 92.0 0 4.0
Ν 0 4.0 0 92.0 0 4.0
Ν 0 4.0 0 96.0 0 0
/θ/ 4.0 8.0 16.0 72.0 0 0
Μ 0 68.0 28.0 0 0 0
Μ 0 92.0 8.0 0 0 0
1)1 0 72.0 20.0 8.0 0 0
ßl 4.0 48.0 28.0 20.0 0 0
Icl 0 44.0 44.0 12.0 0 0
HI 8.0 52.0 36.0 4.0 0 0
M e a n percent correct: Μ 97.2

Ν 88.0
/IZ/ 27.3
Total 77.8
M e a n age 4.92
Table 2. Object clusters for the " > 2 1 correct" group (n = 18)
Responses
Cluster Phoneme
Irr. Null Μ Ν Μ Dupl.
Μ 0 0 0 0 100.0 0
Λ)/ 0 0 0 0 100.0 0
Ν 0 0 0 0 100.0 0
/»/ 0 0 11.1 0 83.3 5.6
Ν 0 0 0 0 100.0 0
Μ 0 0 5.6 0 94.4 0
/η/ 0 0 0 0 100.0 0
/m/ 0 0 0 0 100.0 0
Μ 0 0 0 0 100.0 0
/d/ 0 0 0 0 100.0 0
/g/ 0 0 0 0 100.0 0
Μ 0 0 0 0 100.0 0
Ν 0 0 0 0 100.0 0
/f/ 0 0 0 100.0 0 0
/Ρ/ 0 0 0 100.0 0 0
Ν 0 0 0 100.0 0 0
Ν 5.6 0 0 94.4 0 0
ßl 0 0 5.6 83.3 0 0
Μ 0 22.2 77.8 0 0 0
Μ 0 33.3 66.7 0 0 0
1)1 0 11.1 83.3 0 5.6 0
β/ 0 0 94.4 5.6 0 0
/δ/ 0 5.6 88.9 5.6 0 0
m 0 0 100.0 0 0 0
Mean percent correct: Μ 98.3

Ν 95.6
Μ 85.2
Total 94.7
Mean age 5.78
Actually, we now have data to suggest that the analysis 1 D is the best
one, which posits two lexical variants and one low-level phonological
rule:
Analysis I D (Hybrid)
— iz / [ +sibilant]
(PI) =
-ζ / Ε
+ obstruent
R l . [ +obstruent] —> [ — voiced] /
— voiced
This conclusion is based on a recent miniature artificial language study

carried out by one of my students (Dennis 1988) on grade-school children,
which confirmed an earlier study of mine with adults in showing that,
while obstruent-final devoicing was largely automatic in appropriate
novel environments (e. g., 92% for /z/), subjects did not know what to
do when faced with novel sibilant + sibilant (e.g., /gAc + z/) or homor-
ganic stop + stop clusters (e.g., /bek + g/), where lax-vowel insertion
occurred in only 3% —16% of the trials).4
In any event, it seems clear that feature-nodes, as well as morphological
analysis, will have to be included in any satisfactory connectionist or
analogical account, and, with this understanding, the long-term prospects
for analogical models are much better than one might gather from a first
reading of the Pinker—Prince (1988) evaluation of the specific Rumel-
hart — McClelland realization. Moreover, there is no real prospect that
the implementation of these design improvements will yield a model that,
according to Pinker — Prince (1988: 182) "may be nothing more than an
implementation of a symbolic rule-based account". The reason for this
is that, as noted by Skousen (1989), there are a number of absolutely
fundamental characteristics that distinguish the generative rule-based
approach to the description of language from the analogical one, and
these differences provide a potentially vast empirical ground on which
the relative merits of the two approaches might be systematically and
carefully weighed. I will briefly discuss only a few of these here.
A key contrast, of course, is that fundamentally different mechanisms
are posited to account for productivity in the two cases; thus, (1), while
the generative approach requires a problematic (innate) learning strategy
for discovering the "simplest" set of rules from the data, the analogical
alternative is predicated upon the existence of a large memory capacity
to store the data (combined, of course, with a capacity to analogize on
the basis of overt similarities and differences between stimuli, already

well known to exist5); (2), while the former must need to know how the
rules interact (consider the endless debate in linguistics on questions of
rule-ordering and the cycle), the latter needs only to be able to access
data quickly, another conditio sine qua non of the real-time language
processor; and (3), while the former requires finding the correct rule that
applies to a given context, the latter must instead find some appropriate
example after which to model behavior.
A good illustration of this third contrast is provided by the sound-
spelling correspondences of English. There are, for example, three com-
mon pronunciations of the English digraph ea, namely, as the j\j in bead,
the /ε/ in head, or as the /e/ in break. In the spirit of a rule-based account,
an attempt would presumably be made to isolate one of these as the
"regular" case, relegating the rest to the status of "exceptions" (cf.
Pinker-Prince 1988: 137). In a study by K a y - M a r c e l (1981), however,
it was shown that the reading of the novel nonce-spelling yead varied as
a function of the other items in the set. Thus reading gone, bead, shoe,
pour led to the reading /yid/, while gone, head, shoe, pour led instead to
the reading /yed/. In other words, the "rule" employed depended on the
examples consulted, in conformity with the assumptions of the analogical
approach. 6
By the same token, in her investigation of the acquisition of English
past-tense forms, Marchman (1988: 10—11) has found (a) that "children
are sensitive to several types of regularities that govern the formation of
[these forms]" and (b) that "there does not appear to be a common,
developmental^ related sequence in how children organize the categories
of regular and irregular verbs", leading to the conclusion that acquisition
might best be viewed "as a protracted resolution of several competing
and interdependent sub-systems". 7 Similarly, I have collected data which
show multiple, competing sources of productive morphological regularity,
even in adult speakers; in the most notable case, the past-tense treatment
of nonce-stems ending in the alveolar stops /t/ or /d/, such as /smit/ and
/koyd/, the results for our adult subjects show as much strength for the
null pattern of hit, let, cut, put, bid, rid, etc., as for the so-called regular/-id/
pattern of verbs like hated or dreaded, and with plenty of inconsistency
and individual variation thrown in, to boot (Derwing — Baker 1974).8
This points to another distinguishing feature (4), namely, that, while
the generative rule-based approach implies categories whose boundaries
are well-defined, and transitions in predicted behavior that are sharp and
precise, the analogical approach implies categories and transitions in
predicted behavior that are gradual and imprecise. This follows as a

corollary of the inherent design feature of connectionist accounts that
paths to regularity are non-unique, a property that seems to me to be
perfectly compatible with the notion of a "scale of ruleness" (Mötsch
1990) and with the traditional notion of "analogical leveling" in language
change (see also Derwing et al. 1988 for a phonological illustration related
to the lexicon).
Finally, in (5) we find a global macro-difference that perhaps best
characterizes the essential contrast between the two approaches: while in
a rule-based system usage is a function of the description, in an analogical
account usage is the description (Skousen 1989, pp. 3 — 4, who lists many
other important differences, as well; see also the discussion by Winograd
1975: 185 — 191 on the distinction between the "declarative" and "pro-
cedural" knowledge). Interestingly, on the basis of this last feature,
connectionist or analogical accounts also show some promise of providing
an explanation for one of the most puzzling notions in linguistics, namely,
the concept of tacit knowledge, i. e., "knowledge that" of which a speaker
does not have any overt knowledge (Chomsky 1965). Rumelhart (1979:
2 — 3) has put it this way:
Perhaps the classical case of using knowledge how (procedural knowl-
edge) to produce knowledge that (factual knowledge) occurs in the
domain of grammatical judgements. The knowledge that we have about
language seems to be largely embedded in the procedures involved in
the production and comprehension of linguistic utterances. This is
evidenced by the relative ease with which we perform these tasks when
compared with our ability to explicate the knowledge involved in them.
Semantic knowledge would appear to be the same. Whereas we can
quickly interpret sentences, it is only with the most painstaking effort
that we can produce definitions of terms with any generality.
Though still in its infancy, the debate between the "rule" and "analogy"
perspectives is certain to generate a good deal of both interest and heat
in the years to come, judging by the size and scope of the opening volley
from Pinker — Prince (1988). At the very least, linguists now know that
there is a new kid on the block who is showing a surprising amount of
spunk and verve, who has a few bucks to throw around and, though not
necessarily from the very best of families, it can at least be said that he
comes from one with a lot of "connections". The competition, as usual,
will be good for both sides, forcing them not only to analyze their own
assumptions more carefully, but also to shore up their barricades with
Morphology and the menial lexicon: psycholinguistic evidence 261
more substantial, empirical stuff. And although connectionist accounts

in particular still have a long way to go to establish their place as a solid
alternative to linguistic rules, perhaps the most exciting thing about them
is that they are not only compatible with the notion of a highly redundant
lexicon, but actually require one in order to work. From this point of
view, at the very least, the approach can be said to have gotten off on
the right foot, in the sense that it can be said to have received a good
deal of implicit psycholinguistic support at the outset.
One final qualification still does need to be mentioned, however, which
relates to the fact that the vast bulk of research on which the full-listing
hypothesis is based has been research on English, or at least on a small
set of languages typologically very close to English. This is a potentially
very important limitation, since, in the general scheme of things, English
comes quite close to the morphologically "isolating" end of the morpho-
logical spectrum. What can we say, then, about the bounds of the lexicon
in a language of the agglutinative type, or any other (such as Eskimo) in
which morphological richness and productivity reach levels that we are
accustomed to in English only in syntax? To take even the relatively
familiar case of the Finnish language, for example, it has been estimated
by Karlsson (1986) that, under the strong form of the full-listing hypoth-
esis (which would claim that all real and potential word-forms of a
language are lexically stored as wholes), "Finns would have mental
lexicons that are (tens of) thousands of times larger than English-speaking
people's lexicons". In view of what he regards as the implausibility of
such a conclusion, Karlsson goes on to propose a modified version of
the full-listing hypothesis, suggesting that only the "prototypical forms"
are lexically stored, viz., only those forms "which speakers have heard,
actively used, and which make sense both pragmatically and semantically"
(Karlsson 1986: 28). This makes perfectly good sense. (Why create and
store words, after all,· that have never been used or heard?) Even for
English, this idea would allow for the possibility that some inflected forms
are listed separately in the lexicon, while others are not, a situation that
might help explain some of the inconsistent experimental results on this
particular point (see, for example, the discussion of inflectional affixes in
Aitchison 1987: 109 — 116 and the new data provided by Stember-
ger-MacWhinney 1988).
4. Conclusions
In sum, after a long period of neglect, rapid progress has been made in
recent years in the psycholinguistic investigation of morphology, both in
terms of the character of the mental lexicon and the ways in which
morphological productivity might be accounted for. As would be expected
from such a surge of research activity, new and even potentially revolu-
tionary theoretical ideas have also begun to emerge. At least as often as
not, these new discoveries have flown in the face of received knowledge
in linguistics (such as support for the full-listing hypothesis, as contrasted
to a notable lack of support for anything like the generative idea of a
morpheme-invariant underlying form); we have also seen some of the
pivotal rule proposals considerably weakened, such as English lax-vowel
insertion (or even completely undermined, as in the case of the phono-
logical analysis of English vowel-shift; see Wang —Derwing 1986); finally,
the challenge has even been raised that rule-based systems may be fun-
damentally misguided. Considering the overall scorecard to date, perhaps
the time is ripe to rethink some of our most fundamental and cherished
theoretical assumptions, now that evidence is accumulating from psycho-
linguistic research that they may be dead wrong. To be sure, progress is
not always painless, but it is very exciting nonetheless.
Notes
1. The issue as to how these "stages" were defined, and the justification for them, goes
well beyond the scope of this paper, but is dealt with in some detail in Baker — Derwing
(1982) and Derwing-Baker (1986).
2. Except the /z/-stems, whose special status can be explained (see Derwing—Baker (1979:
212).
3. Linguists often use the word rouges to illustrate this pattern in introductory courses,
but this is a highly marked plural of a mass noun which children are unlikely to know.
The most likely analog in many dialects is the plural form of garage (itself rather rare
a usage), but this is a /j/-stem in the dialect tested.
4. Note that analysis 1 D illustrates the notion of "phonologically conditioned suppletion",
as developed by Carstairs (1990).
5. See Derwing (1973) and especially (1976) for extensive illustrations of the analogy-
drawing capabilities of the normal child.
6. While few, if any, generativists would actually want to incorporate sound-spelling
relations within the same purview of phonological/morphological description, Skousen
shows that the analogical approach is equally appropriate to either domain.
7. Cf. Derwing (1979) and Derwing — Baker (1979) on the "competition" among alternative
formal patterns, and Bates — MacWhinney (1987) and MacWhinney (1987) for a more
general "competition model".
8. Grammaticality/acceptability judgements show a similar indeterminacy, as Ross (1979),
in particular, has emphasized.
References
Aitchison, Jean
1987 Words in the mind. (Oxford: Basil Blackwell).
Baker, William J. —Bruce L. Derwing
1982 "Response coincidence analysis as evidence for language acquisition strat-
egies", Applied Psycholinguistics 3: 193 — 221.
Bates, Elizabeth — Brian MacWhinney
1987 "Competition, variation, and language learning", in: B. MacWhinney (ed.)
Mechanisms of language acquisition (Hillsdale, N. J./London: Lawrence Erl-
baum), 157-193.
Berko, Jean
1958 "The child's learning of English morphology", Word 14: 150-177.
Butterworth, Brian
1983 "Lexical representation", in: B. Butterworth (ed.) Language production (New
York: Academic Press), vol. 2: 257 — 294.
Carstairs, Andrew
1990 "Phonologically conditioned suppletion", this volume, 17 — 23.
Chomsky, Noam
1965 Aspects of the theory of syntax. (Cambridge, Mass.: MIT Press).
Cutler, Anne
1983 "Lexical complexity and sentence processing", in: G. B. Flores d'Arcais —
R. J. Jarvella (eds.) The process of language understanding (New York:
Wiley), 4 3 - 7 9 .
Dennis, Diane
1988 Rule governed behaviour in English inflectional morphology [M. Sc. Thesis,
University of Alberta].
Derwing, Bruce L.
1973 Transformational grammar as a theory of language acquisition: a study in the
empirical, conceptual and methodological foundations of contemporary lin-
guistics. (London: Cambridge University Press).
1976 "What kind of rules can children leam?", in: W. von Raffler-Engel —
Y. Lebrun (eds.) Baby talk and infant speech (Amsterdam: Swets & Zeitlin-
ger), 6 8 - 7 8 .
1979 "Psycholinguistic evidence and linguistic theory", in: G. D. Prideaux (ed.)
Perspectives in experimental linguistics (Amsterdam: Benjamins), 113 — 138.
1987 "A cross-linguistic experimental investigation of syllable structure. Part I:
background and methodology", in: S. Delancey —R. M. Tomlin (eds.) Pro-
ceedings of the Third Annual Meeting of the Pacific Linguistics Conference
(Department of Linguistics, University of Oregon, Eugene, OR), 93 — 102.
Derwing, Bruce L. — William J. Baker
1974 Rule learning and the English inflections (Final report to the Canada Council,
File No. S72-0332).
1979 "Recent research on the acquisition of English morphology", in: P.J.
Fletcher—Μ. Garman (eds.) Language acquisition: Studies in first language
development (Cambridge: Cambridge University Press), 209 — 233.
1980 "Rule learning and the English inflections (with special emphasis on the
plural)", in: G. D. Prideaux — B. L. Derwing —W.J. Baker (eds.) Experi-
mental linguistics: integration of theories and applications (Ghent: Story-
Scientia), 247-272.
1986 "On assessing morphological development", in: P. J. Fletcher—Μ. Garman
(eds.) Language acquisition: Studies in first language development (Cam-
bridge: Cambridge University Press), 2nd ed., 326 — 338.
Derwing, Bruce L. — Maureen L. Dow —Terrance M. Nearey

1988 "Experimenting with syllable structure", in: J. Powers —K. de Jong (eds.),
Proceedings of the Fifth Eastern States Conference on Linguistics (Depart-
ment of Linguistics, Ohio State University, Columbus, OH), 83 — 94.
Diller, Karl C.
1978 The language teaching controversy. (Rowley, MA: Newbury House).
Henderson, Leslie
1985 "Issues in the modelling of pronunciation assembly in normal reading", in:
Κ. E. Patterson —J. C. Marshall —M. Coltheart (eds.) Surface dyslexia
(London: Erlbaum), 459 - 508.
Karlsson, Fred
1986 "Frequency considerations in morphology", Zeitschrift für Phonetik,
Sprachwissenschaft und Kommunikationsforschung 39: 19 — 28.
Kay, Janice —Anthony Marcel
1981 "One process, not two, in reading aloud: lexical analogies do the work of
non-lexical rules", Quarterly Journal of Experimental Psychology 33 A:
397-413.
MacWhinney, Brian
1987 "The competition model", in: B. MacWhinney (ed.) Mechanisms of language
acquisition (Hillsdale, N. J./London: Lawrence Erlbaum), 249 — 308.
Marchman, Virginia A.
1988 "Rules and regularities in the acquisition of the English past tense", CRL
Newsletter (Center for Research in Language, University of California, San
Diego, La Jolla, CA) 2: 4 - 1 3 .
Marslen-Wilson, William D. — Lorraine K.Tyler
1980 "The temporal structure of spoken language understanding", Cognition 8:
1-71.
Mötsch, Wolfgang
1990 "Problems of word structure theories", this volume, 79 — 85.
Murrell, Graham A.— John Morton
1974 "Word recognition and morphemic structure", Journal of Experimental
Psychology 102: 9 6 3 - 9 6 8 .
Pinker, Steven —Alan Prince
1988 "On language and connectionism: analysis of a parallel distributed proc-
essing model of language acquisition", Cognition 28: 73 — 193.
Ross, John R.
1979 "Where's English?", in: C. J. F i l l m o r e - D . Kempler-W. S.-Y. Wang (eds.)
Individual differences in language ability and language behavior (New York:
Academic Press), 127 — 163.
Rumelhart, D. E.
1979 Analogical processes and procedural representations (CHIP 81) (La Jolla,
CA: Center for Human Information Processing, University of California at
San Diego).
Rumelhart, David E. —James L. McClelland — the PDP Research Group
1986 "On learning the past tenses of English verbs", Parallel distributed process-
ing: explorations in the microstructures of cognition vol. 2: Psychological and
biological models (Cambridge, MA: KIT Press), 216 — 271.
Skousen, Royal
1989 Analogical modeling of language. (Dordrecht: Kluwer).
Stanners, Robert F.—James J. Neiser —Scott Painton
1979 "Memory representation for prefixed verbs", Journal of Verbal Learning
and Verbal Behavior 18: 733-743.
Stanners, R. F. —James J. Neiser — William P. Hernon —Roger Hall

1979 "Memory representation for morphologically related words", Journal of
Verbal Learning and Verbal Behavior 18: 399—412.
Stemberger, Joseph R — Brian Mac Whinney
1988 "Are inflected forms stored in the lexicon?", in: M. Hammond — Μ. Noonan
(eds.), Theoretical morphology: Approaches in modern linguistics (San Diego:
Academic Press), 101 — 116.
Vennemann, Theo
1974 "Words and syllables in natural generative grammar", in: A. Bruck et al.
(eds.) Papers from the parasession on natural phonology (Chicago: CLS),
346-374.
1988 "The rule dependence of syllable structure", in: C. Duncan-Rose —
T. Vennemann (eds.) On language: Rhetorica, phonologica, syntactica: a
Festschrift for Robert P. Stockwell, from his friends and colleagues (London:
Rontledge), 257-283.
Wang, H. Samuel —Bruce L. Derwing
1986 "More on vowel shift: the back vowel question", Phonology Yearbook 3:
99-116.
Winograd, Terry
1975 "Frame representations and the declarative/procedural controversy", in:
D. G. Bobrow —A. Collins (eds.) Representations and understanding: studies
in cognitive science (New York: Academic Press), 185 — 210.
Rule-creating creativity: analogy as a synchronic
morphological process*
Jaap van Marie
1. Introduction
In this paper, I will deal with the following two topics: first, the impor-
tance of the analysis of existing words for the study of morphological
creativity and, in relation to that, second, the role of analogy in the
process of the creation of new complex words. By stressing these points,
this paper joins in with the tradition established by (among others)
Humboldt, Paul, and Saussure, according to which existing words, both
simplex and complex, are subjected to a constant, never-ending in-
spection on the part of the speakers of the language. In this conception,
the lexicon must, in the words of Humboldt, not be equated with "eine
fertig daliegende Masse" since it is — even if we leave the coining of
new words outside consideration — "ein fortgehendes Erzeugniss und
Wiedererzeugniss des wortbildenden Vermögens" (Humboldt 1836:
109-110).
A recent rephrasing of this view can be found in Mötsch (1987: 24):
"The creation of new words (...) presupposes rules. But rules need not
have an existence of their own. We may conceive of rules as the result of
a process of analysis operating on similarity of items of the vocabulary".
Motsch's views come close to the stand taken in van Marie — Koefoed
(1980) and van Marie (1985) where it was argued that the speakers
of a language have the capacity to construct rules on the basis of the
existing words, a capacity we called "rule-creating creativity". This
ability, then, directly bears upon analogy as a synchronic morphological
force.
In the following, I hope to make clear that this never-ending inspection
of the words in the lexicon may have different results and may result in
* I am indebted to Geert Booij and Caroline Smits for their critical comments on an earlier
draft of this paper.
268 Jaap van Marie
different coining-devices. In my opinion, at least the following three

phenomena should be distinguished: (1) the incidental actuation of exist-
ing derivational patterns, (2) the coming-into-existence of new deriva-
tional patterns, and (3) non-derivational coining, particularly affix sub-
stitution. (Note that for reasons of space limitations the third phenom-
enon will not be discussed; but cf. van Marie 1985: chap. 8.) From this
it follows that analogy, even within the realm of derivational morphology,
is no homogeneous concept. Yet, however different these phenomena
may be, they have one thing in common: their starting point. All three
processes have their origin in the analysis of existing, complex words in
the lexicon.
The following attempt to come to grips with analogy as a synchronic
force is based on a large-scale experimental investigation of morphological
creativity and the different forms it may adopt. This investigation involves
the "feminization" of neutral personal names in Dutch by children at the
age of 11 to 17, which has resulted in a corpus of over 5,000 terms for
females. In modern Dutch, there is a whole series of such categories of
female personal names based on nouns, cf. the following list of morpho-
logical categories (not exhaustive):
(a) -in: ap-in, boer-in, cf. aap 'ape, monkey' and boer 'farmer'
-es: prinses, cf. prins 'prince'
-se: kasteleinse, cf. kastelein 'innkeeper'
-e: leerlinge, cf. leerling 'pupil'
-ster: wandelaarster, cf. wandelaar 'walker, promenader'
(b) -(t)rice: conductrice, cf. conducteur 'conductor'

-(t)euse: controleuse, cf. controleur 'controller'
(The suffixes in (a) are added to a base, whereas the suffixes in (b) have
the same position as the suffixes of their neutral counterparts.)
What is interesting about these categories of female personal names is
that most of them are decidedly non-productive. Only the categories in
-e and -ster are productive, though the derivational base of both categories
is of rather a restricted type (van Marie 1985, 1986). As the semantic
properties of these categories are very transparent, they constitute an
excellent starting point for an experimental investigation of morphological
creativity as the speakers are not hindered by semantic opacity in their
attempts to find the most suitable female counterpart of the given neutral
personal names.
Rule-creating creativity 269
2. The incidental actuation of existing derivational patterns
The never-ending inspection of existing words on the part of the speakers

of a language may result in the setting-up of completely incidental rules
in order to actuate a normally extinct word-type (van Marie —Koefoed
1980; van Marie 1985). These rules underlie the, often stylistically marked,
ad hoc formations which form part of a pattern whose characteristic
precisely is that it "normally" never gives rise to newly coined words. In
relation to these incidental rules several remarks are due.
First, these rules are of the "normal" additive type, i.e., they add an
affix to a base (of a complexity lower by one degree), which means that
they do not involve affix substitution. Put differently, as far as their
formal properties are concerned, these rules are completely identical with
the "permanent rules" which account for the actuation of the productive
word-types. From this it follows that, as far as their formal and semantic
characteristics are concerned, these ad hoc formations may completely
conform to their existing counterparts (from which the incidental rules
by means of which they are formed are deduced). Consider, e.g., vrekk-
in, which in the tests turned out to be by far the most popular female
counterpart of vrek 'miser'. As to both its formal and semantic charac-
teristics, this formation cannot be distinguished from its actual congeners
such as held-in {held 'hero') or vriend-in (vriend 'friend').
Note, however, that the ad hoc derivatives may also differ from their
actual counterparts, in consequence of the fact that the incidental rules
are often of a rather straightforward, "unsophisticated" type. Specifically,
all kinds of details characteristic of the actual words may be glossed over
in the process of rule creation. Consider the following cases which
illustrate the point I wish to make.
Ad hoc formations very frequently differ from their existing counter-
parts as to their formal make-up. As far as existing words are concerned,
-in (as a pure Germanic suffix) hardly ever figures after non-native stems.
In the experimental data, however, formations in which -in is added to a
non-native base do occur. Consider giraff-in as the female counterpart of
giraffe 'id.', in which -in is added to a non-native simplex base, a phe-
nomenon which has no parallel in the actual words of Dutch. Compare
also monteur-in (cf. monteur 'mechanic'), in which -in figures after the
non-native suffix -eur, a combination of affixes which does not occur in
the actual words of modern Dutch (and which, by the way, sounds rather
odd indeed).
270 Jaap van Marie
As mentioned above, what these ad hoc formations show is that the

incidental rules appear to be of a rather straightforward, oversimplified
type. All kinds of additional "details", particularly those relating to the
conditions on the base, may easily be glossed over in the process of rule
creation. Note that in a framework which involves rule creation, this can
be explained in a natural manner, since it is even hard to imagine how it
would be possible to detect all these details. As a first guess it does not
seem too far-fetched to start from the idea that it is one (or very few)
word-pair(s) to which the analysis of the actual words preceding the
creation of an incidental rule is restricted. That is, on the whole it will
be only one or very few word-pair(s) which are taken into account. From
this it follows directly that it is simply impossible to discover all the
details of the extinct word-type, which means that the deduced incidental
rules are of the straightforward, "degenerate" type more or less by
definition.
3. New derivational patterns
Let me next discuss the second manifestation of analogy as a synchronic

force. The experimental study of morphological creativity makes perfectly
clear that the inspection of the actual words in the lexicon may lead to
new derivational patterns which can be characterized as short-cuts in the
derivational system and which lead to (informally) affix clustering. In
this paper I will confine myself to those short-cuts which involve the
reinterpretation of a derivative of the second degree as a derivative of
the first degree (cf. van Marie 1984). Clearly, this type of reinterpretation
is nothing but the direct — synchronic — reflex of the constant inspection
on the part of the speakers of a language to which the words in the
lexicon are subjected. In order to come to grips with this type of reinter-
pretation, consider a formation like dans-er-es. In grammars of modern
Dutch, it is common practice to regard this word as the derivative in -es
of danser 'dancer', which in its turn is the derivative in -er of dans-en 'to
dance'. That is, danseres is usually considered a derivative of the second
degree. However, as was already pointed out by Paul (1920: 245) in many
cases it is questionable whether analyses of this type are justified, cf.:
Sehr häufig ist der Fall, dass eine Ableitung aus einer Ableitung in
direkte Beziehung zum Grundworte gesetzt wird, wodurch dann auch
wirkliche direkte Ableitungen veranlasst werden mit Verschmelzung

von zwei Suffixen zu einem.
The net result of this is clear: a formation like danseres can directly be
related to dans-en, resulting in the coming into existence of the new,
"secondary" suffix -eres.
This example suggests that we should distinguish synchronic mecha-
nisms of reinterpretation: speakers of a language have the ability to place
the actual words in different configurations, i.e., they may establish "new"
derivational relationships — relationships which are independent of fixed
derivational patterns. Note, moreover, that these mechanisms operate on
existing complex words.
It is my impression that this type of reinterpretation does not take
place at random but that it is particularly prominent in specific types of
complex words, i.e., those which, in one way or another, involve a non-
transpositional process. If this impression is correct, this finding is im-
portant in that it makes clear that the synchronic mechanisms underlying
this type of reinterpretation are, at least in part, governed by grammatical
factors.
On the basis of my above remarks, the synchronic mechanisms bringing
about the synchronic short-cuts in the derivational system can be defined
as follows:
(a) [[[BASE]X A]x B]y => [[BASE], AB]y
(b) [[[BASE]X C]y D]y => [[BASE]X CD]y
(c) [[[BASE]X E]x F]x [[BASE]X EF]X

Examples: groenigheid, the abstract noun in -heid derived from groenig
'greenish', which in its turn is based on groen 'green', exemplifies (a);
danseres, which illustrates type (b), is discussed above; lokettiste, the
female counterpart in -e of lokettist 'ticket-clerk', which in its turn is
derived from loket 'ticket window, ticket office', is an exemple of (c).
From the above it follows that reinterpretation of words which are
formed by means of combination of a transpositional and a non-trans-
positional process (i.e., (a) and (b)) have the same effect: in both cases
there arises a new transpositional suffix. All new suffixes, transpositional
or not, consist of the "condensed" combination of the former suffixes
(which may remain in the language as independent elements but this need
not be the case).
272 Jaap van Marie
As said before, the experimental study of morphological creativity

makes perfectly clear that these short-cuts in the derivational system are
far from rare, as many newly coined terms for female persons contain
such a secondary, condensed affix. See below for some examples:
* -aire (probably < neutral -air + female -e),
cf. montaire ( = female monteur 'mechanic')
* -ares ( < neutral -aar + female -es),

cf. nazares ( = female nazaat 'descendant')
* -(e) res ( < neutral -er + female -es),

cf. posteres ( = female post (bode) 'postman')
* -(e)rin ( < neutral -er + female -in),

cf. girafferin (= female giraffe 'id.')
* -eure ( < neutral -eur + female -e),

cf. collegeure {— female collega 'colleague')
* -iere ( < neutral -ier + female -e),

cf. makeliere ( = female makelaar 'broker')
*
-iste ( < neutral -ist 4- female -e),
cf. emiriste ( = female emir 'id.')
* -(l)inge (probably < neutral -ling -f female -e),

cf. emiringe female emir 'id.')
As can be observed in other types of creatively coined words, in many
cases phonological adaptation of either stem or affix is involved. This
means that there is some experimentation on the part of the speakers of
the language in order to arrive at words which "sound good".
4. Conclusions
In the preceding I have discussed two mechanisms which I consider to
bear upon analogy as a synchronic force. In my view this discussion
justifies two conclusions. First, what these mechanisms have in common
is that they take existing complex words as their starting point. I am, of
course, aware of the fact that the notion "existing complex word" is a
much debated one. Several linguists have doubted the relevance of this
notion to competence theories of word-formation. The consequence of
this stand is clear: such theories may provide insight in several respects,
but they are unfit to deal with analogy and with the type of word-
formation upon which analogy bears. To put this differently, within the
framework of theories which deny the notion "existing complex word"
all phenomena discussed here must be viewed as part of performance.
Secondly, even on the basis of the above we cannot but ascertain that
analogy is a rather heterogeneous phenomenon.
In my opinion it is the joint operation of these two factors which
underlies the fact that so many morphological theories have not succeeded
in coming to grips with analogy as a synchronic process.
References
Humboldt, Wilhelm von

1836 Über die Verschiedenheit des menschlichen Sprachbaues und ihren Einfluss
auf die geistige Entwicklung des Menschengeschlechts. (Berlin: Schneider).
Marie, Jaap van
1984 "Morfologische veranderingen in breder perspectief' [Morphological
changes in a broader perspective], in: Vorm en funktie en tekst en taal
(Leiden: Brill), 1 3 1 - 1 5 3 .
1985 On the paradigmatic dimension of morphological creativity. (Dordrecht:
Foris).
1986 "The domain hypothesis: the study of rival morphological processes",
Linguistics 24: 601 - 6 2 7 .
Marie, Jaap van —Geert A. T. Koefoed
1980 "Over Humboldtiaanse taalveranderingen, morfologie en de creativiteit van
taal" [On Humboldtian linguistic changes, morphology, and linguistic cre-
ativity], Spektator 10: 1 1 1 - 1 4 7 .
Mötsch, Wolfgang
1987 "On inactivity, productivity and analogy in derivational processes" [Paper
presented at the round table on word structure theories, 14th International
Congress of Linguists, Berlin, GDR],
Paul, Hermann
1920 Prinzipien der Sprachgeschichte5. (Halle: Niemeyer).
Topic 6: Typology and non-Indo-European
morphologies
Sapir's approach to typology and current issues in
morphology*
Stephen R. Anderson
The Sixth International Phonology Meeting began in July, 1988 with a

commemoration of the 50th anniversary of the death of Trubetzkoy. By
the time this paper appears, we will have passed another, similar anni-
versary: that of the death of Edward Sapir in 1939. I hope it is not
inappropriate to devote a paper read at that meeting to the work of
Sapir, who was one of the few North American linguists who had a
substantial acquaintance with — and an influence on — the work of
Trubetzkoy.
In any case, among the major figures of our discipline, Sapir surely
merits a place on the podium in any discussion of the typology of
morphological systems. In his little book Language (Sapir 1921), he
presented one of the most detailed typological schemas ever proposed
for the word structure of natural languages, basing his discussion pri-
marily on non-Indo-European material. He also believed deeply in the
importance of such an analysis. For example, much of his proposed
classification of the languages of North and South America rested not
on evidence of the traditional comparative sort, but rather on his per-
ception of typological similarities among languages. While he certainly
did not believe that such structural similarities were a sure guide to
genetic relationship, he did feel that once the core features of a language's
structural type had been identified, these were the properties that were
most likely to remain (and to have remained) stable over time. As such,
they could be a guide even to rather remote relationships, in the absence
of other evidence. "Typology" for Sapir seems to have been pretty much
the same thing as "morphological typology", and he would thus have
been an ideal person to discuss the topic of "Typology and non-Indo-
* The research reported in this paper was supported in part by grant number BNS84 — 18277
from the US National Science Foundation, and the paper was written while the author
was supported in part by a fellowship from the John Simon Guggenheim Foundation.
The support of both of these organizations is gratefully acknowledged.
278 Stephen R. Anderson
European morphologies". Since he is not available, though, I will at least

try in the present paper to take my inspiration from his text.
It is not immediately obvious just how to address this topic in a way
that makes contact with the current work of those dealing in morpho-
logical theory. While everyone talks about languages as belonging to
some or another particular type, and of such and such a feature as
"typologically significant", I know of no serious attempt since Sapir's in
1921 to articulate a comprehensive view of just what the range of linguistic
types might be in the area of word structure. Even today, when people
talk about the type to which a language belongs morphologically, they
are as likely as not to intend by this one of the nineteenth century labels
"agglutinating", "inflectional", "polysynthetic", and the like: labels that
Sapir had hoped to improve on. Perhaps my doubts about the vigor with
which typological studies are in fact currently being pursued can be taken
to be confirmed by the fact that apparently no one at all (apart from the
plenary speakers) proposed a paper on this topic for the Krems Mor-
phology Meeting, and there were no section or discussion papers on the
program of that meeting that were devoted to the topic.
Of course, everyone can agree that a valid classificatory scheme for
natural languages — one that succeeded in isolating the essential and
largely unchanging features of individual languages, as Sapir wished —
would indeed be a worthwhile accomplishment. But I submit that such
a high-minded goal has not in fact been achieved in the area of word
structure, by Sapir or by anyone else; and that it is in fact very unlikely
that it ever will be. Words have the form they do because of the syntactic
structures into which they are organized; because of the phonological
exigencies of the language; and especially (following the remarks of
Hagege 1990) because of the history of the language with respect both
to grammar and to sound. While it is certainly possible (and indeed
interesting) to study the properties of morphological systems, I think it
is unlikely that these systems per se, standing as they do in a position
where they are subject to so many other influences, will turn out to have
genuinely autonomous properties that will reveal the true "genius" of a
language.
In the earlier history of typologizing, though, at the time when a
linguistic type was of necessity a morphological type (because neither
syntax nor phonology in the modern sense had really been discovered
yet), there was a tendency to see the goals of typological study in relation
to a supposed teleology in language change. Some early (and not partic-
ularly lamented) typologists, who took the linguistic structure of Greek,
Sapir's approach to typology 279
Latin, and/or Sanskrit to represent the acme of human intellectual or-

ganization, saw the role of typology as providing an account of the
evolutionary stages by which the spirit of mankind had risen out of the
primeval slime of isolating-ness to pass through the stages of agglutination
and polysynthesis to arrive at the ultimate goal of inflection. Sapir
compared this approach to that of the "zoologist that sees in the organic
world a huge conspiracy to evolve the race horse or the Jersey cow"
(Sapir 1921:124), and there is no particular reason to soften that judgment
now. Especially in light of the fact that attested changes go in all directions
among the classical morphological types, there is no reason at all to
imagine that such a typology will serve as a guide to the orientation of
linguistic change, even if it does turn out to serve some other purpose.
But if it is indeed so unlikely that typology will show us the path of
a stadial evolution of languages, why do we bother to do it at all, and
what does a typology really amount to? While one sometimes gets the
impression that a typology is simply a taxonomy — a classificatory
scheme intended to provide a convenient set of labels for roughly equal
subclasses of languages — that obviously cannot be correct; or at least
if it is, it describes an enquiry of singularly little intellectual interest.
To see this, suppose someone were to propose the following classifi-
cation: (a) those languages in which the root meaning 'tongue' begins
with a [ -I- Coronal] segment are to be called "iconic"; (b) those languages
in which the root for 'tongue' begins with a [ — Coronal] segment are to
be called "anti-iconic"; and (c) those languages in which there is more
than one root for 'tongue', belonging to different classes, or in which
there is no basic root for this notion at all, are to be called "transitional".
Now in fact I have not tabulated the languages of the world that would
belong to each of these classes, but suppose for the moment that such a
count revealed roughly equal numbers of languages in each set. Still, no
one would (or should!) take this proposal seriously. But why not?
Presumably, what makes such a proposal blatantly silly is the fact that
nothing at all follows from the phonological composition of the root for
'tongue' in a language, except for something about the way sentences
sound which involve talk about tongues. The early typologists at least
had the idea that something would follow from determining whether a
language was "isolating", "agglutinative", etc.: namely, some predictions
about what was likely to happen to it in the next round of linguistic
change. If we now have serious reservations about the utility of that
terminology, it is presumably because we are pretty sure that those
consequences do not, indeed, follow: that is, that the traditional terms
do not really characterize a teleology inherent in language development.

Our doubts do not spring from problems about how to apply the labels
(although this may in fact be a far from trivial matter as many writers,
including Hagege 1990, have made clear).
So we can conclude that the parameters of a typology ought to be
ones from which something follows: that is, they ought to identify groups
of properties that co-vary with one another, so that knowing how one
thing works entails knowing about others as well, as a direct consequence
of whatever it is that motivates the typological labels. This is certainly
the sense that typology (at least good typology ...) has had in syntax.
For instance, the alacrity with which the syntactician strives to know
immediately, for any language under discussion, whether it is SVO, SOV,
or VSO (or perhaps one of the other possibilities brought forcefully to
our attention by Derbyshire —Pullum 1981), comes from the fact that at
least since the work of Greenberg (1963), we have felt confident that
knowing the relative order of the main constituents of a simple transitive
main clause will allow us to predict a whole host of other things, ranging
from the relative order of modifiers and their heads all the way down
(for some, at least) to whether the language is more likely to have a rule
of vowel harmony or one of palatalization.
But the best typological work in syntax certainly does not stop at
identifying an inventory of correlated properties. When we find a number
of things that appear to go together in a "typologically significant" way,
inquiring minds want to know: Why? And the form that the answer to
this question most generally takes is the development of a framework for
grammatical description that will provide a relatively limited number of
dimensions along which variation is possible in the system of individual
languages, with each dimension allowing only a small number of possi-
bilities. Such a framework should then make a number of individual
consequences follow from the setting of each of these "parameters" in a
given way. Sticking to the example of word order, the X-theory of phrase
structure is a descriptive schema for the internal organization of syntactic
units as all being in a sort of harmony with one another, subject only to
a few options, such as "heads are phrase-final" (vs. "heads are phrase-
initial"). If this is indeed true, the X-theory of phrase structure can be
regarded as attempting a genuinely explanatory account of how and why
particular sub-facts correlate in the way they do within individual lan-
guages.
When we have achieved such a result, we are entitled to say we have
(at least part of) a substantial typology, where the range of language
types is then the same as the range of available parameter settings. Insofar
as there are not many possible settings, there will not be very many
possible types; and insofar as the languages avail themselves in roughly
equal numbers of the available options, they will distribute in roughly
equal fashion among the types. We can then use the parameter values
themselves as convenient labels for the resulting typological categories.
It makes excellent sense, however, to ask just how likely it is that a
situation comparable to that found in syntax will in fact turn out to
obtain in morphology. How plausible is it, that is, that the system of
word-formation rules in a language will turn out to be globally para-
metrized in the way syntactic structure seems to be, as implied in any
serious search for a morphologically based typology of languages? Some
efforts have been made along these lines by, for instance, those who think
of morphology as the syntax of words and who focus on a program of
extending essentially syntactic insights to domains within the word. Re-
gardless of how one feels about the ultimate productivity of this enterprise
on its own terms, it still leaves most of the more mechanical side of word
formation — getting the segments right and getting them in the right
places — largely untouched.
When one turns to the details of how words are composed of pho-
nological material, my own impression is that the number of substantial
correlations within a language that could give rise to a productive para-
metrization of the sort sketched above is actually very limited. Much of
the reason for this (as stressed by Hagege 1990) is the fact that word
structure is not really an autonomous domain in the same way syntax
and phonology are. Much more of what one finds in morphological
systems is the result of the interplay of other areas of grammar, and
especially of diachronic change; and it is in these other domains that the
basic parameters of a language's structure are presumably to be sought.
However interesting it may be to study, much of morphology may in
some sense be an epiphenomenon.
To shift the emphasis of the discussion somewhat, though, and despite
the rather pessimistic attitude suggested above toward the potential sig-
nificance of any typology of morphological systems, the search for such
a typology has an important role to play; and indeed, when put into
proper perspective, it constitutes another way of posing the most fun-
damental questions there are about word structure. This is because, beside
the search for overall correlations among phenomena that might yield a
minimal specification of the range of languages in the world, there is
another methodology available for pursuing typological studies. This is
to attempt to ensure that whatever the account that is given of some area
of linguistic structure (morphology, for instance), it is one that will be
adequate to accommodate all of the systems that might be encountered.
One can do "typology", that is, in the sense of exploring the full range
of diversity in the languages of the world, and without necessarily feeling
that the effort has been a failure if that diversity is not ultimately reducible
to some small number of binary- or ternary-valued parameters.
This is actually a perfectly respectable goal to take along in exploring
the typology of morphological systems (among other parts of language),
even though it is not what most people think of as the goal of typology.
But now notice that on this reading, the development of a genuinely
adequate typology is a project which is essentially co-extensive with the
development of a generally applicable morphological theory. That is, a
theoretical descriptive framework which is really adequate to all of the
world's languages, can be seen as constituting in itself the most central
kind of a "typology", even if its descriptive dimensions do not reduce to
some small number of parametric possibilities. Of course, where corre-
lations are to be found, they ought properly to be incorporated into the
theory; and if enough are found, maybe it will be possible once again to
interpret the theory as a set of labels for language types. But this is only
one of many possible outcomes, and in the meantime there seems no
particularly good reason to distinguish fundamentally between linguistic
typology and linguistic theory, construed as a general enterprise.
Let us, then, turn to the most fully developed notion of word-structure
typology in the traditional literature, taking this to be (at least the outline
of) a theory of the range of constructs that are necessary in principle for
the description of the morphological systems of all possible languages.
As I suggested at the outset of this paper, the system outlined in Sapir's
Language goes as far as anything in this direction, and it is that framework
that I would like to examine briefly here.
Following a tradition represented also by his teacher Franz Boas,
Sapir starts from the notion that the word structure system of a language
is composed of two kinds of object: a set of grammatical processes and
a set of concepts expressed by the application of those processes. There
must also be, of course, a basic word stock that serves as the foundation
for morphological elaboration. Now the important thing about these two
sets is the fact that they are in principle quite independent of one another:
The question of form in language presents itself under two aspects.

We may either consider the formal methods employed by a language,
its 'grammatical processes,' or we may ascertain the distribution of

concepts with reference to formal expression. What are the formal
patterns of the language? And what types of concepts make up the
content of these formal patterns? The two points of view are quite
distinct. (Sapir 1921: 57)
The independence of the two sorts of object means that the relation
between processes and concepts is (at least in the general case) many-to-
many, with the same process associated with multiple concepts and/or
the same concept expressed by more than one process. For example, in
English the single process "suffix /z/" (more formally, /Χ/—>/Χ + ζ/) may
be invoked to express plurality, third singular present in verbs, possessive,
etc.; while the same concept, e.g., 'plural' may be expressed in a number
of different ways ("suffix /z/", "umlaut stem /au/ to /ai/", /Xum/—»/Xa/,
etc.). This is a notion of morphological structure which has also been
adopted in the more recent literature in one form or another by, among
others, Robert Beard, Brian Joseph, Arnold Zwicky, and myself (to
mention only some of the authors represented in the present volume). As
these and other writers have stressed, it differs in quite fundamental ways
from views based on the traditional conception of the "morpheme", an
exaggeratedly minimal Saussurean sign composed of the unity of a con-
stituent of form with a constituent of meaning. As opposed to the one-
to-one relation between elements of form and of content that ought to
obtain if words were really composed exhaustively of such "morphemes",
the picture presented by Sapir (and its successors in the current literature)
predicts the rather more complex picture that seems actually to charac-
terize much of natural language (and of which, of course, the simple one-
to-one association is merely a special, limiting case). Reasons to believe
in such a position have been developed elsewhere (see, e.g., Anderson
1988); the present remarks are intended only to point out that this was
also Sapir's view.
With respect to the typology of morphological systems, Sapir proposed
to replace the traditional one-dimensional classifications (such as that
into isolating, agglutinating, inflecting, etc., languages) with a number of
logically independent classificatory parameters. Given his view that mor-
phology is made up of a set of processes and a set of concepts, it is quite
natural that the basic division within his scheme is between one dimension
relating to the type(s) of processes a language employs and another
referring to the sorts of concepts it expresses morphologically. A third
dimension, referring to the degree of internal complexity of words, cor-
responds to the traditional scale that runs from "isolating" to "polysyn-

thetic". Sapir actually has little to say about this last dimension (which
for him runs from "analytic" languages through "synthetic" to "polysyn-
thetic" ones), and since it is less obvious how it might correspond to a
distinct aspect of the grammar, I will ignore it here. Our present focus is
on Sapir's typologies of processes and of concepts, and on the presup-
positions about morphological structure which appear to underlie them.
The types of process that Sapir recognizes can be described in com-
paratively brief terms, since the inventory he gives is a rather familiar
one. He notes that concepts can be expressed by any of the following
techniques:
(la) Word order
b) Composition (compounding of stems)
c) Affixation (including prefixation, suffixation, and infixation)
d) Internal modification (vocalic or consonantal ablaut, consonant
mutation, etc.)
e) Reduplication
f) Variations in accent (pitch, stress, etc.)
This catalog of possibilities is not a particularly revolutionary one,
and in fact the main point to note about it is the fact that affixation is
treated as just one kind of process, on a par with other changes of a non-
affixal character. Morpheme-based views of morphology, in contrast,
tend to consider affixes as the only really legitimate kind of morpheme,
with everything else described as some sort of morphologically condi-
tioned phonological concomitant of a (possibly null) affix.
A contemporary account of the formal processes that play a role in
morphological structure would look quite like Sapir's, though the distinct
status of some of his types within morphological theory can perhaps be
questioned: we might well omit "word order" for instance, since that is
not really an aspect of morphology. Reduplication can probably be
subsumed at least partly under the heading of affixation, following the
now familiar line of McCarthy (1981), Marantz (1982), and others. 1 The
category of variation in accent probably involves some instances of
affixation and some of internal modification; in Sapir's days accentual
phenomena (especially pitch) were treated as somewhat more exotic and
less connected with the segmental phonology than we would accept today.
On the other hand, we might well add to Sapir's list the possibility of
circumfixes (simultaneous prefixation and suffixation corresponding to a
single unit of morphological form). This particular kind of 'discontinuous
affixation' is particularly well developed in some Indonesian and Philip-

pine languages. For example, in Indonesian ke-...-an seems to constitute
a unit, as indicated by the statement that the prefix ke- "is seldom used,
except in conjunction with the suffix -an" (MacDonald 1976: 63). Thus,
kebisaan 'capability' is derived directly from bisa 'be able' rather than by
prefixation of ke- to a hypothetical *bisaan, or by suffixation of -an to
an equally hypothetical *kebisa. This circumfix can in fact be applied to
more complex bases, as in the case of ketidakmampuan 'impotence',
derived from tidak mampu 'not be able' (cf. mampu 'be able'). Its com-
ponent parts are only marginally and unproductively attested as inde-
pendent affixes, and not with these roots. The same circumfix also appears
in formations like kehujanan 'be caught in the rain' (cf. hujan 'rain') and
kelihatan 'be visible' (cf. lihat 'see') where the verbal status of the deriv-
atives is completely incompatible with any claim that the prefix and the
suffix represent independent processes that happen to be cumulated in
these forms. 2
Among the non-affixational possibilities, it is worth noting further
that there are at least a few languages in which metathesis or reordering
of segments serves as the marker of some grammatical information. An
example is the aspectual category in several of the Salish languages
illustrated by the Clallam contrast between ckwu-t 'shoot' and cukw-t
'shooting', based formally on the interchange of the root vowel and the
consonant preceding it. Some controversy has surrounded the issue of
whether metathesis is involved in such examples as a primitive, but it is
argued in Anderson (1988) that at least some of the Salish languages
(notably Saanich: see Montier 1986) must be so analyzed.
A particularly interesting possibility from a formal point of view is
that of marking a grammatical category not by the presence of some
overt formative but rather by the (anti-iconic: see Dressier 1985, 1987)
deletion of phonological material from the base. As in the case of
morphologically motivated metathesis, the number of clear examples of
this phenomenon are quite limited, but it must apparently be recognized
in some instances. Apart from examples cited in Anderson (1988), a case
of this sort has recently been described in some detail by Martin (1988),
on the basis of work by Broadwell (1987). Languages of the Southwestern
Muskogean group show a process of "syllable (or rhyme) dissociation"
illustrated below:
(2a) [Alabama]
i. balaa-ka 'lie down (sg.)'; bal-ka 'lie down (pi.)'
ii. batat-li 'hit once'; bat-li 'hit repeatedly'
iii. kolof-li 'cut once'; kol-li 'cut repeatedly'
b) [Choctaw]
i. bonot-li 'roll up (sg. obj.)'; bon-li 'roll up (pl. obj.)'
ii. bakaaf-li 'split (sg. obj.)'; bak-li 'split (pl. obj.)'
c) [Koasati]
i. atakaa-li 'hang (sg.)'; atak-li 'hang (pl.)'
ii. lataf-ka-n 'kick (sg.)'; lat-ka-n 'kick (pl.)'
Martin shows that the operation involved in these forms is one which
derives the plural (or iterated, etc.) form from the singular by dissociating
the final syllable from the prosodic pattern and then re-associating its
onset consonant melody into the preceding rhyme (if possible, consistent
with the syllabic structures permitted in the language). Such a process
would have been called a "subtractive morph" in Hockett's (1947) guide
to the structuralist morphological zoo; what is important for our purposes
is that it has no apparent (coherent) reformulation as the addition of an
affix.
With the modifications suggested above, then, a list3 like Sapir's can
serve as a description of the grammatical processes available to individual
languages. Each particular word-formational process invokes one of them
(or perhaps more than one at a time, as in the case of German suffixes
accompanied by stem umlaut). As the basis of a typology, such a list can
be used to characterize the subset of what is formally possible that is
actually instantiated in a particular system. Where a language has no
complex words at all, of course, it requires no word-formation processes,
and this can be regarded as the limiting case which is identified by the
term "isolating". Where all of the rules involve affixation, pure and
simple, with no other changes, this is the defining characteristic of an
"agglutinating" language. If at least some of the word-formation rules of
a language involve non-affixational processes (internal change, metathe-
sis, subtraction, accent shift, etc.), this is the basis for calling the mor-
phology of that language "symbolic" (to use Sapir's term).
Finally, the phenomenon that was classically supposed to make Indo-
European languages of the ancient type so special was the presence of
complex affixation: affixes that are accompanied by some morphologi-
cally motivated phonological change. Languages with such internally
complex processes are called "inflecting" in an earlier terminology, one
that seems sufficiently misleading to require replacement, as for instance
by Sapir's term "fusional". The defining characteristic of this type is
primarily the fact that segmentation into neat, discrete morphemes is
rendered more difficult by the extent to which other effects accompany
or obscure affixation. The overvaluation of this supposed linguistic "type"

seems to be an instance of making a virtue of necessity, and there is no
particular reason to recognize it as distinct. Of course, the degree of
internal complexity of those word formation rules that cumulate two or
more processes as a unitary indication of some category is evidently a
possible dimension of the formal structure of a morphological system.
This gives us a set of categories (isolating, agglutinating, symbolic,
and fusional) that can be laid out (somewhat uncomfortably) on a sort
of scale. But does this serve as a "typology" for language? It would, if
languages were homogeneous, such that for instance the presence of a
single "symbolic" process entailed the exclusivity of this formal type of
word formation rule. But as everyone knows, that is not the way the
facts actually do turn out. Of course a single rule in a single language
does have a particular type within this scheme, but languages as wholes
tend to involve greater or lesser mixtures of different sorts of processes.
In the rare case where all of a language's morphology is limited to a
single formal type, as for instance where Turkish is typically claimed to
be limited to suffixation, we can call such a language, e.g., "agglutinating"
as a whole. Most languages will not be eligible for such comprehensive
labels, however. In general, it is individual rules and not entire languages
that can meaningfully be called agglutinating, symbolic, or fusional in
any precise, categorial sense.
There are clearly generalizations to be made over the sets of word-
formation processes that constitute the morphological systems of (entire)
languages, but these are only very rarely of the sort captured by labels
like "agglutinating", etc. Much more common is the highly local sort of
generalization typified by the fact that English verb inflection is based
entirely on the rules "suffix /z/" and "suffix /d/". In a given language
there are typically some rather specific, language-particular statements
to be made about the range of its grammatical processes, perhaps in the
form of some sort of redundancy rule over morphological rules or "meta-
rule" as has occasionally been suggested.4 The rule classes identified in
such generalizations, however, tend both to be much more specific than
those delimited in (1) and also to involve overlap among the classical
formal types. Indeed, it is questionable whether a label like "agglutinat-
ing" actually captures a significant reality in the cases where it can be
consistently applied: surely it is at least as important that all of the
morphological material of Turkish is borne by suffixes as that the rules
are all ones of affixation.
If these observations are correct, there is no reason to expect any

(suitably modified) version of the rule typology in (1) to serve as the
basis of a typology for languages, as implied by the traditional labels.
This is confirmed by the fact that when Sapir, who attempts to use such
labels as one dimension of his typology, actually gets around to classifying
languages on the basis of their morphological "technique", the values he
assigns tend to be expressions like "fusional-agglutinative, with a symbolic
tinge". Such a formulation may reflect well the overall statistical balance
of different process types in a language, but it is not at all the sort of
categorial description we expect of a typology, and its actual significance
is far from clear. We conclude that it is important to study the range of
grammatical processes found in the languages of the world, so as to
understand how morphological rules operate; but that this effort is
unlikely to pay dividends of the sort imagined by classical typologies.
Passing from the study of process types to that of concepts, we reach
what can be argued to be the most interesting and innovative part of the
discussion. Sapir notes first of all that there is a range of concepts which
can be formally expressed in language, and suggests that these fall into
a small number of categories, as in:
(3 a) Basic [radical] concepts
b) Derivational concepts
c) Concrete-relational concepts
d) Pure-relational concepts
For a fuller discussion of just how these are delineated by Sapir, the
reader must of course consult the original text, but the essential content
of the categories in (3) is as follow. The "basic" concepts of (3 a) are
those involved in the fundamental semantics of basic lexical roots, which
characterize the meanings of the language's elementary word stock (e.g.,
the properties representing the basic meaning of 'cat'). The "derivational"
concepts of (3 b), in contrast, are those introduced by a set of semantic
functions which operate on lexical items to yield new lexical stems from
existing ones (e.g., the semantic correlate of the diminutive process which
maps German Katze 'cat' into Kätzchen 'kitten'). The "pure-relational"
concepts of (3d) are a set of intrinsically a-semantic features having
purely grammatical significance, but nonetheless formally represented in
the shapes of individual words as a reflection of syntactic structure (e.g.,
the role of the nominative case in Latin to mark the subject); while the
"mixed" or "concrete relational" concepts of (3 c) are those that function
syntactically, like the pure-relational concepts, but which can be projected
in some positions from the semantics of some part of the structure (e.g.,
grammatical number in English, which is grammatically relevant as the
basis of the agreement relation between subjects and verbs, but which is
a function of the semantics of the subject NP).
Sapir then observes that the members of these categories are not all
equally necessary to the structure of language in general. In order to say
anything at all, of course, the speaker of a language must have at hand
some meaningful words, and so every language must express "basic" or
"radical" concepts. Furthermore, these words must be capable of being
put together syntactically, and the representation of syntactic structure is
the business of the "relational" concepts: indicating subject-hood and
object-hood, status as modifier vs. modified, etc. Much of the structure
indicated by pure relational concepts may be reflected by word order
alone, and so is of no particular interest to a theory of word structure.
Recall, however, that Sapir himself includes word order as a "grammatical
process": in consequence, he can claim that every language must of
necessity have some formal expression of his pure-relational concepts,
since otherwise it would have no syntactic structure. Abstracting away
from word order, then, we can see that "pure-relational" concepts will
be represented morphologically exactly where information of a purely
syntactic nature is carried by the forms of words.
As opposed to "basic" and "pure-relational" notions, however, which
must necessarily be formally represented in some fashion in every lan-
guage, there is no necessity for a language to have any "derivational"
concepts at all. It is always possible to represent a complex meaning as
a (syntactically formed) combination of two or more basic meanings, like
'little tree', instead of as a derivational function modifying a single basic
meaning (as in 'tree-let'). Similarly, it is not necessary for a language to
make use of any semantically significant property, like number or (nat-
ural) gender, as one of the devices that indicate grammatical structure
such as the modifier-modified relation or that between a predicate and
its subject. Mixed-relational concepts as a category are thus dispensable
to particular languages as well.
Sapir thus suggests that we can classify languages according to what
elaborations of the basic, irreducible inventory of concept types ("basic"
and "pure-relational") they employ. That is, a language may optionally
make use of derivational concepts; mixed-relational concepts; both of
these; or neither. It might seem that the resulting four classes of language
would provide a rather clear-cut categorial distinction, and that any given
language would belong to exactly one of the four set types; but in fact
when Sapir actually applies this classification in concrete cases, it too

turns out to be somewhat more of a scale than might be expected.
Languages turn out to develop, say, derivational concepts to a greater or
a lesser degree. Again, it seems more useful to think of a particular rule
as "realizing" or "implementing" a concept of a particular type, rather
than of a whole language as doing this.
What is particularly interesting about this schema, and what is relevant
to current morphological discussion, is the framework within which the
distinction between relational and non-relational concepts is introduced.
Limitations of time and space preclude a detailed rehearsal of Sapir's
analysis here, but its essence is an argument that the grammatical pattern
of a sentence is to be described in a way that is completely independent
of the degree of morphological complexity (indeed, of the morphological
or semantic content) of the words that actually compose it. This is
suggested fairly well by the following quotation from Language. After
comparing the two sentences The farmer kills the duckling and The man
takes the chick, he goes on to observe (Sapir 1921: 85)
We feel instinctively, without the slightest attempt at conscious anal-
ysis, that the two sentences fit precisely the same pattern, that they
are really the same fundamental sentence, differing only in their ma-
terial trappings. In other words, they express identical relational con-
cepts in an identical manner.
We can interpret this, only a little bit anachronistically, as a sort of
incipient statement of the lexicalist hypothesis: the claim that the syntax
neither manipulates nor has access to the internal structure of words.
Rather, it derives a grammatical pattern in which constituency, word
order, and the distribution of "relational" concepts are specified — in
contemporary terms, an S-structure with no concrete lexical items con-
tained in it. These actual lexical items come from the word stock of the
language, which is made up of its basic words together with the results
of applying to these any derivational functions the language may employ
(and insofar as the rules of the language permit, further derivations from
derived stems).
Since derivational concepts are integral parts of the lexical items that
constitute the "material trappings" of sentences, they clearly must be
introduced within whatever we want to call the dictionary or word stock
of a language. 5 Relational concepts, on the other hand, are part of the
grammatical pattern: in fact, it is precisely the property of being part of
the grammatical pattern that makes a concept "relational" in the first
place. This is true regardless of whether a given relational concept is

"pure" (i.e., exclusively of syntactic import) or "mixed" (i.e., related to
semantic as well as syntactic properties). We must imagine that the mixed
relational concepts are indicated in grammatical structure, and serve to
constrain the implementation of that structure in concrete words by
requiring that only items whose semantics is consistent with the features
present may interpret the relevant positions in the structure.
The Janus-like property of mixed relational concepts, which have both
a syntactic and a semantic side, leads to the rather interesting consequence
that it is not possible to determine the role a given concept plays in the
grammar just by knowing its signification. 6 The same concept (in terms
of signification alone), that is, may either be relational or not depending
on whether it has a role to play in the grammar as well as a meaning.
Consider the property 'plural', for instance, whose signification we
can agree is clear and more or less identifiable across languages. Sapir
compares the role of the "plural" in English with its role in Nootka 7 in
a form like that below:
9
(4) inikw-ihl- ?friinih- is- Ή
fire-in house-pl-dim-def
'the small fires in the house'
Sapir argues that in Nootka the plural is a derivational category, as
opposed to English, where it is a (mixed) relational one. There are two
parts to this argument: "First, the plural element precedes the diminutive
in Nootka" (Sapir 1921:105). Now this is only an argument if we assume
that concepts of the relational sort form layers of structure that are
strictly external to any layers formed by derivational concepts. On the
(unchallenged) assumption that the diminutive in Nootka is in fact
derivational, this entails a similar derivational status for any concept
represented by a structural layer internal to that of the diminutive. Since
Sapir's relational concepts are just the ones that correspond to inflectional
morphology (as opposed to derivation) in traditional and current writing,
the argument thus rests on the traditional observation that inflection
comes "outside o f " derivation — an observation made more precise and
demonstrated to be a theorem in the morphological system outlined in
Anderson (1988).
This argument would only be of interest if we could find some inde-
pendent confirmation of the claim that only the English plural, and not
that of Nootka, actually meets the criteria for being a relational category.
This is in fact provided as the second step in Sapir's argument (1921:

105):
What more than anything else cuts off all possibility of comparison
between [the plural in English and in Nootka] is this, that in Nootka
neither the plural nor the diminutive affix corresponds or refers to
anything else in the sentence. In English 'the house-firelets burn' (not
'burns'), in Nootka neither verb, nor adjective, nor anything else in
the proposition is the least concerned with the plurality or the dimin-
utiveness of the fire.
That is, neither "plural" nor "diminutive" is distributed by a rule of
grammatical patterning in Nootka. In English, "plural" is a mixed rela-
tional notion, while in Nootka it is a derivational one, not because the
category involved means different things in the two languages but rather
because English has a syntactic rule of agreement that makes reference
to it while Nootka has no such rule. Because of this difference, it is
perfectly coherent for the Nootka plural marker to come inside of other
clearly derivational material, while this would not be possible in English.
The defining property of relational concepts is given rather discursively
by Sapir, but this argument (in the context of the rather extensive
discussion in chapter 5 of Language) makes it clear that what is meant
is the following:
(5) Relational concepts are those that are relevant to the grammatical
structure of the sentence, independent of its lexical interpretation.
Let us now sum up Sapir's framework for grammatical description,
which has several interesting properties. Formally, the morphology of a
language consists of an inventory of processes that can affect the shapes
of words so as to express concepts. In particular, morphology does not
consist of a collection of form/meaning pairs (minimal signs) that can be
concatenated to form words, as in later models taking the morpheme as
a primitive notion. Furthermore, the "syntax" (or the "grammar") is a
system that describes abstract, neutral grammatical structures without
regard to the specific words which will interpret the structural positions
within them. The dictionary of a language contains its basic word stock,
together with the results of any derivational processes it may utilize. Such
derivational processes (better, grammatical processes expressing deriva-
tional concepts) apply "in the dictionary" in this sense. In contrast,
inflectional grammatical processes (i.e., those distributing and expressing
relational concepts) must apply "in the syntax", in the sense that they
must apply to a lexically interpreted grammatical structure. Despite this

difference in their interaction with the rest of the grammar, a single set
of generalizations may characterize the set of "grammatical processes"
(or morphological rules) of a language, whether derivational or inflec-
tional.
This framework should seem fairly familiar to readers acquainted with
the theory which has sometimes been referred to as "extended word and
paradigm" morphology, but which has more recently come to be called
"a-morphous" morphology (for several reasons, most notably because it
tries very hard to get along without morphemes). Naturally, this will
seem like a felicitous result to an adherent of this view, since there are
many scholars whom it would be a lot worse to have on one's side in a
difference of intellectual opinion than Sapir.
However, the main point here is the slightly more modest one that
such a view of the matter seemed for Sapir to flow rather naturally out
of an attempt to explore the limits of the diversity to be found in the
word-structure systems of natural languages. He himself said he was
talking about the "types of linguistic structure" in the culminating chapter
of his discussion of morphological theory. It is fairly clear that when
typological investigation is carried out in a genuinely universalizing way,
it is not really distinct from the development of a general theory of
(morphological) structure. In other words, responsible typology is no
different from "doing theory".
As a matter more of the sociology of the field than of its substance,
people who say they are doing "typology" often are more concerned with
catalogs of surface phenomena than with explanatory frameworks, and
they have a bias toward descriptions that put everything into a small
number of discrete categories; while people who say they are doing
"theory" often do most of it on the basis of finding out more and more
about a very few languages. Everyone knows that there are more than
enough both of abuses and of exceptions on each side, but the goal of
this discussion has been to show that the dichotomy is a fundamentally
illusory one, since there is no substantial difference between typology and
theory when correctly viewed. Of course, if it turns out that the correct
descriptive framework admits of only a very few dimensions of variation
for languages, with few possible values on each, some will say that we
have discovered a typological framework while others will say that we
have found the right set of parameters for universal grammar. There is
no reason to think that what would make the one set happy shouldn't
make the others happy too, though. And seeing the typological and the
theoretical enterprise as basically the same should provide a worthwhile
insight for all concerned.
Notes
1. It is not actually obvious that the affixation analysis solves all the problems involved in
reduplication. In particular, the actual mechanics of association of melodic material with
the empty skeletal positions analyzed as an affix of reduplication are not completely
straightforward. It remains to be seen whether all instances of apparent reduplication
can be successfully reduced to the simple addition of an affix which happens to be
melodically underspecified.
2. This is because both ke- and -an, in the limited circumstances in which they are
independently attested as affixes, produce nouns from various sorts of bases.
3. Naturally, we would like to go beyond listing to provide a more explanatory account
of the class of grammatical processes. Along with the apparent seductiveness of the
classical morpheme, this seems to be some of what lies behind attempts to reduce all of
morphology to affixation. It does not seem possible to accomplish that reduction,
unfortunately; but the alternative of saying simply that the class of grammatical processes
is delimited by the possibilities of a rich tranformational formalism is unsatisfactory as
well. One possibility suggested by Martin (1988) is the claim that the operations available
to morphology are exactly those available to the phonology, thus reducing one unsolved
problem to another. A complicating factor is the role of historical change in shaping
synchronic morphologies: this may have the consequence that actually attested systems
contain an unrepresentative (or at least seriously skewed) selection from among the
theoretically possible processes. For some discussion of this, see Anderson (1980) and
Janda (1984).
4. The notion of "meta-rules" over the morphology was suggested in Anderson (1986).
Janda (1982 and elsewhere) has explored in some detail the notion that individual
languages make extensive use of a small number of processes.
5. The word "lexicon" has been used in a number of rather distinct senses in the recent
literature: as the domain of the "lexical" or "cyclic" rules of the phonology; as the set
of surface-grammatical word forms of a language (not necessarily the same as the set
of its surface-phonological words); as the set of its uninflected (but possibly complex)
stems, perhaps including compounds; etc. It is roughly this last sense of "lexicon" that
we would like to invoke here, but given the confusion that exists in the literature about
what it means to say something is "in the lexicon" we would simply avoid using the
word as much as possible. Aronoff (1988) makes some particularly sharp comments
concerning confusions in the use of "lexical" in recent linguistic discourse.
6. Actually, this is already clear from the fact that the same sense can be conveyed either
by a basic concept associated with a root or by a derivational function. Since both basic
and derivational concepts are localized within the dictionary, however, the indeterminacy
involved here is less radical than the point referred to in the text.
7. Although this ethnonym has a securely established status in the linguistic literature, it
is actually not an accurate name for the people Sapir referred to (more accurately known
as the Tsishaath or Tseshaht — [c'lsa-âth]) or for their language. Since this paper is not
fundamentally concerned with the ethnography of the northwest coast of North America,
however, we will perpetuate Sapir's usage here in the spirit of quotation.
References
1980 "On the development of morphology from syntax", in: J. Fisiak (ed.) His-
torical morphology (The Hague: Mouton), 51 —69.
Linguistic Theory 4: 1 — 32.
1988 "Morphological theory", in: F.J. Newmeyer (ed.) Linguistics: The Cam-
bridge Survey (Cambridge: Cambridge University Press) I: 146 — 191.
Aronoff, Mark
1988 "Two senses of lexical" [Unpublished paper, SUNY, Stony Brook].
Broadwell, George Aaron
1987 "Subtractive morphology in Southwest Muskogean" [Paper presented at
the 40th Annual Kentucky Foreign Language Conference].
Derbyshire, Desmond — Geoffrey K. Pullum
1981 "Object-initial languages", UAL 47: 192-214.
Dressier, Wolfgang Ulrich
1985 "On the predictiveness of natural morphology", Journal of Linguistics 21:
321 - 3 3 7 .
1987 "Subtraction in a polycentristic theory of natural morphology", in: E.
Gussmann (ed.) Rules and the lexicon: studies in word-formation (Lublin:
Wydawnictwo Katolickiego Uniwersytetu Lubelskiego), 67 — 77.
1963 "Some universale of grammar with particular reference to the order of
meaningful elements", in: J. H. Greenberg (ed.) Universals of language
(Cambridge, MA: MIT Press), 7 3 - 1 1 2 .
Hagege, Claude
1990 "Do the classical morphological types have clear-cut limits?" this volume
297-308.
Hockett, Charles F.
1947 "Problems of morphemic analysis", Language 23: 321 —343.
Janda, Richard
1982 "On limiting the form of morphological rules: German umlaut, diacritics,
and the 'cluster constraint'", NELS 12: 140-152.
1984 "Why morphological metathesis rules are rare: on the possibility of historical
explanation in linguistics", PBLS 10: 8 7 - 1 0 3 .
MacDonald, R. Ross
1976 Indonesian reference grammar. (Washington: Georgetown University Press).
Marantz, Alec
1982 "Re reduplication", Linguistic Inquiry 13: 435-482.
Martin, Jack
1988 "Subtractive morphology as dissociation", Proceedings of the 7 th West Coast
Conference on Formal Linguistics.
McCarthy, John J.
1981 "A prosodic theory of non-concatenative morphology", Linguistic Inquiry 12:
373-418.
Montier, Timothy
1986 An outline of the morphology and phonology of Saanich, North Straits Salish
(Occasional Papers in Linguistics 4, University of Montana).
Sapir, Edward
1921 Language (New York: Harcourt, Brace & World).
Do the classical morphological types have clear-cut
limits?
Claude Hagege
0. More or less implicitly, many human languages have been assigned

to one of the three, or four, main language types proposed by classical
nineteenth century typologists, i.e., fusional, agglutinative, isolating, and
incorporating. Only in the twentieth century was attention focused on an
objective fact, which by now has become common knowledge among
linguists dealing with typology: these classical types are mere abstract
constructs, and no language corresponds in its entirety to a given type;
rather, most languages evince a certain affinity for one type, but not to
the exclusion of others, and some languages seem to present features
equally distributed between two or more types.
It is now necessary to go one step further. In other words, two
directions of research deserve to be explored. First, to what extent does
a language belong to one type or another, i.e., what are the details of
poly typical complexity? Second, why does such a situation occur, i.e.,
what are the reasons for polytypical complexity? In order to answer these
two questions, I will examine a sample of languages which appear to
shed some light on one of the most-debated issues in morphological
typology: the exact boundary between the so-called agglutinative and
fusional types. Section 1 will present some cases in which two related
principles, generally held to be characteristic of the "pure" agglutinative
type, are violated: one is the "one form —one meaning" principle, the
other is the "no alteration at morpheme boundary" principle. Section 2
will propose an interpretation of these violations: it will show by what
process an agglutinative language can acquire fusional features. Finally,
we will examine what conclusions can be drawn from these facts as
regards the theoretical status of morphology and man's place in nature
as a language builder. In other words, my aim here is to test the validity
of a theoretical framework (the above-mentioned two principles) and,
having shown its inadequacy, to substitute for it another one, whose
explanatory power is wider.
298 Claude Hagege
1. Details of poly typical complexity

Let us examine, to begin with, three languages traditionally assumed to
be agglutinative: Hungarian, Turkish, and Kannarese.
1.1. Hungarian
Consider a form häz-a 'his house', and compare it with häz-a-i 'his houses'
and with häz-a-i-k 'their houses'. Judging only by these three forms, we
could immediately state that there is a perfect application of the "one
form —one meaning" principle typical of agglutinative languages, since
we can posit k for plural of possessor, i for plural of the possessed object,
and a for third person singular (of possessor). However, the same a vowel
also appears in häz-a-m 'my house', häz-a-d 'your (sg.) house', häz-a-i-m
'my houses' and häz-a-i-d 'your (sg.) houses', where the possessor is
singular, but not third person, as well as in häz-a-i-n-k 'our houses', häz-
a-to-k 'your (pi.) house' and häz-a-i-to-k 'your (pi.) houses', where the
possessor is neither singular, nor third person, these forms, along with
häz-un-k 'our house' and häz-u-k 'their house', only providing confir-
mation as to the status of k and i.
Consequently, as far as a is concerned, we are left with a dilemma.
Either we analyze it, in all the forms where it appears except in häz-a,
häz-a-i and häz-a-i-k, as a simple link vowel, and in these three latter
forms as a marker of the third person (singular) of the possessor, with
such an analysis implying, in a very uneconomical way, that the speaker,
within one paradigm, uses two different kinds of a, which have unrelated
functions. Or we posit am, ad as markers of the first and second person
singular of possessor respectively, un, ato, u as first, second, and third
person respectively when there is more than one possessor, these markers
being combined with k, the common marker for plural possessor and
a...n, resp. a...to as variants for first and second person plural possessor.
But since we have already assigned a meaning to i and k and since there
is no distinctive function attributable to a in the first and second person
singular when there is a single possessed object, nor in the whole paradigm
when there is more than one possessed object, we could also say that,
for instance, the markers of the first and second person of the possessor
are m and d respectively, in which case the one for third person singular
is zero. 1 But this conclusion is itself untenable; although it might seem
justified to the extent that we take into account the paradigm and its
pressures, it makes the simple description of a central feature of Hun-
Do the classical morphological types have clear-cut limits? 299
garian morphology impossible: any noun, within the limits imposed by

meaning compatibilities, can be followed by a third person singular
possessor suffix, i.e., a vowel (preceded by j- in some definite cases),
which is front or back in harmony with the root vowel(s): asztal-a 'his
(her, its) table', ösztön-e 'his instinct', kert-j-e 'his garden', kalap-j-a 'his
hat', etc. Who would say that -a or -e here should be interpreted as zero?
It appears from the above that blending of affixes and discontinuity
of some morphemes, even in a language known as agglutinative, give it
fusional features which are far from what we would expect of the type
to which most of its morphology assigns it. Not leaving Hungarian, one
could also mention other phenomena which violate the "no alteration at
morpheme boundaries" principle, held to be characteristic of the "pure"
agglutinative type. For instance, if we add a postposition like the inessive
suffix -banj-ben after häz-a 'his house' or kert-j-e 'his garden', we get häz-
ä-ban and kert-j-e-ben respectively. In other words, the short vowels a
([a]) and e ([ε]) have been replaced by the long vowels ά ([a:]) and e ([e:]).
1.2. Turkish
I will briefly describe here some phenomena which all tend to contradict
the widely shared assumption of the "perfect" agglutinative character of
Turkish morphology. Consider the following list of alternations (from R.
Underhill (1986: 14)):
Nominative Definite accusative
sebep sebeb-i cause
kelebek <— kelebeg-i 'butterfly'
hak <r- hakk-i 'right'
§ehir §ehr-i 'city'
zaman ([zamän]) zaman-i ([zama:nui]) 'time'
These examples illustrate various phenomena which occur at mor-
pheme boundaries. These phenomena should not occur if it were true
that Turkish is purely agglutinative. We see that a syllable-final voiced
stop is devoiced when it is followed by zero (sebep j sebeb-i, from Arabic
sabab); corresponding to a velar consonant at the end of a word, we have
a fricative or zero in an intervocalic environment (kelebek/kelebeg-i)·,
two syllable-final geminate consonants alternate with a word-final simple
consonant (hakk-ijhak, from Arabic haq) — if the consonant cluster is
made of two sonorants, then, in word-final position, an epenthetic vowel
will be inserted between them (§ehr-ij§ehir, from Persian sahr); a short
vowel in a word-final closed syllable alternates with a long vowel in the
syllable preceding -ι (zamanjzaman-i, from Persian zamän).
300 Claude Hagege
1.3. Kannarese
The agglutinative character of Kannarese appears clearly in its nominal
system, but there are some irregular facts even there. 2 Thus, given katte
'donkey', -galu plural suffix (preceding the case marker), and -0, -annu,
-a, -ige, and -inda, markers of nominative, accusative, genitive, dative,
and instrumental respectively, we get the following paradigm:
Singular Plural
Nominative katte-0 katte-galu-0
Accusative katte-y-annu katte-gal-annu
Genitive katte-y-a katte-gal-a
Dative katte-ge katte-gal-ige
Instrumental katte-y-inda katte-gal-inda
Several facts deserve to be noticed here. First, a -y- is inserted in the
singular before -annu, -a, and -inda, but the -i- of -ige, which appears in
the plural, is absorbed in the singular at the boundary between katte and
-ige. Second, we observe that the -u of -galu is deleted before the initials
of the case markers in the plural (except for the nominative, which is 0).
One might say that these small "accidents" do not detract much from
the generally agglutinative nature of such a system, as also evidenced by
the transparency of nominal groups with multiple modifications, like
huduga-r-ibbar-ig-öskara-vägi-yü
(boy-pl.-two-dat.-goal-purposive-inclusive)
'for the two boys too'
But if we examine the verbal system, we find a much more complex
situation. The present of the verb ele 'pull', for instance, is as follows:
Person Number Gender
ene 1 sg
eve 1 pl
iye 2 sg
iri 2 Pi
ele-y-utt- äne 3 sg masc.
äle 3 sg fem.
ade 3 Sg neut.
äre 3 pi masc./fem.
ave 3 pl neut.
From this paradigm, we can infer that -utt- is the present tense marker
(linked to the verb ele by -y- since the latter ends in a vowel), and that
e...e is the 1. person discontinuous marker,

i...e 2. person sg. discontinuous marker,
ä...e 3. person non neut. discontinuous marker,
a...e 3. person neut. discontinuous marker
So far so good. However, if we carry the examination a little further,

we see that the marker of the singular is -n- in the first person but -y- in
the second, -n- again in the third person masculine (implying that the
first person, although without gender distinction, retains the form of the
masculine), -/- in the third person feminine, and -d- in the third person
neuter. In other words, even if it is true that -n-j-l- is a gender opposition,
the "one form —one meaning" principle3 is violated here as far as -n- and
-y- are concerned, since these are two allomorphs of the singular. Should
we say, then, that -iye comes from *ine by palatalization of -n- between
the two palatal vowels i and e l But even if this were true (and there is
no synchronic demonstration of its truth), how are we to explain the
allomorphs for the plural? Should it be represented by -ν-, as is suggested
by -eve and -ave, or by -r-, as is indicated by -iri (where the final -i
remains to be explained) and by -ärel There is no synchronic solution
to this problem, even if we posit that *nv or *dv v, and that *nr or *lr
—> r. The only conclusion is that the system is partly agglutinative and
partly inflectional, or, at least, much less agglutinative than in the nominal
paradigm.
2. Reasons for polytypical complexity
First of all, two series of examples can be given here, which shed light
on the reasons for the situation presented above.
2.1. Estonian
Grade alternation is a well-known feature of Estonian morphology, just

as it is of the morphologies of other Finno-Ugric languages, such as
Finnish or Lapp. Estonian has three different types of grade alternations 4 :
302 Claude Hagege
2.1.1. Consonant mutation

The strong grade has a stop or an s at the onset of the second syllable,
and in the weak grade, these consonants are either lost (with a fairly
frequent vowel alteration), or assimilated to a preceding sonorant, or
replaced by a semi-vowel, for example:
Strong Weak Gloss
rege (part.sg.) ree (gen.sg.) 'sledge'
viga (nom.sg.) vea (gen.sg.) 'error'
madu (nom.sg.) mao (gen.sg.) 'snake'
tigu (nom.sg.) teo (gen.sg.) 'snail'
kulda (part.sg.) kulla (gen.sg.) 'gold'
soda ([saeda]) (nom.sg.) söja (gen.sg.) 'war'
2.1.2. Alternation of geminate stops

This takes place after a long first syllable: the strong grade has a geminate
stop and the weak grade a single stop, for example:
nurka ([nurkka]), part.sg,/nurga ([nurka]), gen. sg. 'corner'
taipu ([taippu]), part.sg./ta/6« ([taipu]), gen.sg. 'understanding'
2.1.3. Late prosodic quantity alternation

In the strong grade, traditionally called the third degree of quantity or
overlength, the long first syllable is pronounced with extra intensity and
greater length (noted here by/), whereas in the weak grade, also called
the second degree of quantity, there are no such features:
Quantity 3 Quantity 2 Gloss
meltsa (part.sg.) metsa (gen.sg.) 'forest'
lilnna (part.sg.) linna (gen.sg.) 'town'
laullu (part.sg.) laulu (gen.sg.) 'song'
soolla (part.sg.) soola (gen.sg.) 'salt'
The conditioning factor of these alternations, in Estonian as in other
Finno-Ugric languages, and especially Finnish, is the final -n marker of
the genitive case. Before disappearing, this marker had brought about a
weakening of the consonant which, due to the presence of this -n, turned
out to be at the beginning of a closed syllable. The weakened consonant
was finally deleted: e.g., in Finnish, jalka 'leg' nominative; the genitive is
*jalka-n —> *jalyan —> *jalan and in Estonian, jalg ( n o m . ) / j a l a

—> jala,
(gen.). Thus the addition of a final consonant, producing a closed syllable,
made the word quantitatively heavier, but this was later compensated for
by weakening of the first consonant of this syllable. When, by the
subsequent deletion of some of the weakened consonants, it happened
that two vowels became contiguous, high vowels were lowered by assim-
ilation to a preceding or following a (e.g., madujmao and vigajvea re-
spectively), and two adjacent high vowels were equally lowered by analogy
with the general lowering of high vowels (e.g., tigu/teo). The result of
these historical processes, which have long ceased to be productive in
Estonian and which today have a purely morphological character, at least
as far as consonant mutation is concerned, 5 is the fusional aspect of the
genitive of the words to which qualitative consonant mutation applies.
This is made still clearer by comparison with the nominative, but also
with the comitative and dative cases, which follow an agglutinative
strategy, ending with the suffixes -ga and -ile, respectively.
2.2. Palauan
This Austronesian language, spoken on the westernmost island of the
Carolines, would be assigned to the agglutinative type considering that
many forms, in the noun and in the verb systems, are derived through
suffixation of clearly identifiable morphemes. However, some of the
suffixes are stressed, or enclitic, which results in strong alterations of the
root vowels of stems combined with such suffixes. Let us take as an
example the noun phrase expressing inherent possession. 6 The possessive
suffixes constitute the following paradigm (V = stressed vowel):
sg. 1. -Vk pi. 1. inclusive -Vö
1. exclusive -(m)äm
2. -Vm 2. -(m)iu
3. -VI 3. -(r)ir
Here is a sample list of some of the main phenomena that can be

observed:
Noun in isolation Possessed noun
— Long vowel: short vowel:
e.g., bahι: 'country' balu-el 'his country'
— Short vowel following a sono- <-*• no vowel:
rant:
e.g., rasm 'needle' rsm-em 'your needle'
304 Claude Hagege
— Short vowel adjacent to a semi- <-> no original vowel, but vocaliza-

vowel: tion of the semi-vowel7:
e.g., 9jull 'cushion' ΉΠ-ek 'my cushion'
— Short vowel in environments <-> central vowel:

other than the last two:
e.g., ker 'question' ksr-iö 'our (incl.) question'
As is evidenced by these examples, the stem vowels, becoming un-
stressed by shifting of the stress to the possessive suffixes, undergo various
kinds of alteration: shortening, deletion, centralization. A comparable
situation occurs in the verb system. For instance, corresponding to
lisdb 'to burn (transitive)', m^-luôs 'to write' or md-gimö 'to trim', which
are imperfective verbs with the general Austronesian verb marker moN-,
indicating here that the process is going on, we have passive obligative
participles, formed with the enclitic suffix -/. This suffix attracts the stress
to the last syllable of the stem to which it is attached. Therefore, we get,
respectively, sdsob-l 'having to be burnt', h9ük-l 'having to be written'
and kmuö-l 'having to be trimmed'. Various phenomena have occurred
here. The verb marker msN-, in Palauan as in other Austronesian lan-
guages, brings about the nasalization of the initial consonant of the verb
root, hence mo-qimd, as opposed to kmuö-l, where the original k- is
retained. But later in the history of the language, Austronesian η yielded
Palauan /. This / assimilates initial dentals such as s or I itself, hence mo-
lesab and ma-lii99s as opposed to sosob-l and h9uk-l respectively. As for
the k of the latter, it can be explained by a general rule of Palauan
morphology, according to which no si cluster is admitted, and s dissim-
ilates, instead, into k* With respect to the vowels, we see that unstressed
vowels become centralized (sasob-l and h9uk-l) or even deleted (kmuö-
9
l). The full vocalic grade appears only when the vowel is stressed, as is
attested, for the first syllable of the verb root, by ms-lisab, ma-lü^ds, and
ma-rjimö, and for the second one, by sssob-l, h^ük-l, and kmuö-l. Both
vowels can also appear in their weakest grade (i.e., centralized) if none
is stressed. This is the case with the other passive obligative participle of
Palauan, semantically equivalent to this one. It is formed by adding a
stressed -äl suffix. The variants of the forms given will therefore be, as
expected: sasab-äl, b93s-äll, and ksmö-äll (epenthesis of 9 in the last form
is explained by another rule of Palauan: no three-consonant cluster is
accepted, unless one of the three consonants is r, I, or τ;10).
Obviously, Palauan has acquired fusional features which make it very
different from the general agglutinative type characteristic of other Aus-
tronesian languages, in which the root is much easier to identify, remain-

ing the same through the morphology, with a low grade of allomorphy.
Palauan turns out not to be very distinct, from a typological viewpoint,
from a strongly fusional language, such as, e.g., Israeli Hebrew.11 Allo-
morphy in this language happens to characterize, also, the possessive
paradigm and, even better, it is also due to a shifting of stress from the
root to the possessive suffix! For example, we have:
tsorex 'need' < * tsork-ό 'his need'
röxav 'width' < * roxb-äm 'their width'
lexem 'bread' < • laxm-ί 'my bread'
sefer 'book' < > sifr-ä(h) 'her book'
When the possessed and/or the possessor are plural, allomorphic
variation goes even further, e.g.,
tsrax-äv 'his needs'
tsraxe:-nu 'our needs'
tsorxe-xem 'your (pi.) needs'
tsorxe ha-9is 'the man's needs'
which, along with tsorex and tsork- above, give altogether six allomorphs.
3. Conclusion
Two logically related conclusions can be drawn from the present study.
Firstly, polytypical complexity is the result of phonetic evolution. Con-
sequently, this study shows that morphology is not an autonomous
domain: word structure cannot be analyzed short of having recourse to
historical phonetics, or if it is, the kind of analysis that such a choice
implies remains at the quite insufficient stage of mere description, and
does not teach us what we can expect to learn about the characteristics
of human languages. Secondly, and as a consequence, if we do not try
to delve into the reasons for polytypical complexity, then we cannot meet
the main requirement which prompts every linguist to do linguistics: to
know more about man's nature. Linguistics is a social science, and as
such, it has something to teach us about human qualities.12 Speakers-
listeners, who are far from being "ideal", as was assumed in Chomsky
(1965), build their language, through generations, less unconsciously than
306 Claude Hagege
is generally suggested. They do not care for abstract principles such as

the purity of a given morphological type, be it agglutinative or flectional.
What they tend to realize is ease of articulation at the lowest cost, and,
when possible, regular correspondence between form and meaning. But
languages are not frozen systems. They constantly change. As a conse-
quence, what at a given time was a means to simplify communication
becomes a burden as time passes. Languages are made of an interplay
between various levels. What is convenient for the speaker-listener from
the phonetic viewpoint may become a complication from the morpho-
logical one, as is evidenced by the Turkish facts presented above and by
the fate of Estonian consonant mutation, which today is frozen, unpro-
ductive, and devoid of clear motivation. The Hungarian and Kannarese
blending of agglutinative and fusional features may also be explained by
the freezing of productive processes which, at an earlier stage, were
motivated. The speaker-listener leaves a living trace of his activity. The
spectacular evolution of Palauan morphology towards a fusional type,
which makes it so strikingly different from classical Austronesian lan-
guages and gives it a Semitic physiognomy, bears evidence of a cultural
(gestural) habit which cannot pass unheeded by any careful observer as
soon as he arrives on the archipelago: Palauans speak very fast, they
strongly stress stressed vowels, and this, unavoidably, has a devastating
effect on the unstressed ones. The study of the limits between morpho-
logical types gives clues to the investigation of man and society.
Notes
1. A zero-morpheme is proposed for the third person singular suffix in the "several
possessions" paradigm {-(j)aimj-(j)eim, -(j)aid/-(j)eid, -(j)ai/-(j)ei, etc.), in Kiefer
(1985: 89).
2. Cf. Aronoff-Sridhar (1984: 7, 9).
3. This principle is violated even in "perfectly agglutinative" languages, like those in the
Turkic family: in Azeri, the -am and the -ssn of kalir-am and kdir-sm, respectively Ί
come' and ' y ° u come' are blendings of person and number, since the plural has forms
which are not analyzable synchronically: -ik and -siniz. Likewise, in Osmanli, for the
same verb, we have, in the present, gel-i-yor + -um, -sun, -uz and -sunuz respectively.
Furthermore, in Salar and in Saryg-Yugur, Turkic languages spoken in the Chinese
province of Kansu, there is no mark for the plural possessive, kak-am being, for
example, in Salar, 'my brother' as well as 'our brother'. In the declension of the personal
pronouns of these languages, we find inflectional features: in Salar, Τ is men, 'of me'
is mi, 'to me' is maga or ma. All these phenomena are not particularly "agglutinative",
to say nothing of the «regular existence» of characteristic features considered by
Austerlitz (1970) as defining an agglutinative type ("developed participial system, the
finite verb as cloture marker at the end of the sentence", etc.).
4. The following data are taken from Hint (1981), and from A l l i k - H e l p - P a k k (1987).
5. The other two processes, early alternation of geminate stops and historically late
quantity alternation, are still productive in the modern language as is evidenced by the
treatment of recent loanwords (cf. Hint (1987)).
6. All the Palauan material comes from my own fieldwork; the results are published as
Hagege (1986).
7. The detail of this process, rule 6a of the ten rules characteristic of Palauan morpho-
phonemics, is presented in Hagege (1986: 30). It is one of the most idiosyncratic and
striking features of this language: the unpronounceable CwC group (where w =
semi-vowel), which results from the deletion of the root vowel whose stress has been
shifted to the suffix, yields a pronounceable word by vocalization of w: taut 'aim' +
el 'his' —> *twt-el —• tut-el.
8. For some exceptions to this rule, see Hagege (1986: 23).
9. By full vocalic grade, I do not mean "underlying form", since I reject this notion,
especially when it is confused with an historical restitution. An abuse of that kind leads
Wilson (1972: 47) and Josephs (1975: 497) to posit, in order to "explain" the i in the
possessive inflection of ker 'question' (cited above), a "derivation" such as keri —> keri
—» ker, by rules of "stress assignment" and "unstressed vowel deletion". As a conse-
quence, Dressier (1974: 139) mentions Wilson and Josephs' "data" as illustrating his
generalization on the vowel types that are the most threatened with deletion by
apheresis, syncope, and apocope. As a matter of fact, this is by no means a productive
rule in modern Palauan, but only a very old and reconstituted historical process.
10. Cf. Hagege (1986: 22).
11. However, there might be some discussion on whether the Hebrew facts are really mere
violations of the "one form —one meaning" principle.
12. The linguistic theory which underlies this view is expounded in Hagege (1988).
References
Allik, Jüri — Help, Toomas —Pakk, Heiti

1987 "Psycholinguistic evidence for linguemes: the Estonian rective", Symposium
on language universal (Tallinn: Academy of Sciences of the Estonian SSR),
8-13.
Aronoff, Mark — Sridhar, S. N.
1984 "Agglutination and composition in Kannada verb morphology", in: Papers
from the parasession on lexical semantics (Chicago: Chicago Linguistic
Society), 3 - 2 0 .
Austerlitz, Robert
1970 "Agglutination in northern Eurasia in perspective", in: R. Jakobson —S.
Kawamoto (eds.) Studies in general and oriental linguistics, presented to
Shirö Hattori on the occasion of his sixtieth birthday (Tokyo: TEC), 1 —5.
Chomsky, Noam
1965 Aspects of the theory of syntax. (Cambridge, MA: MIT Press).
1974 "Phonologische Prozesstypologie", Linguistica generalia I: Studies in lin-
guistic typology (Praha: Acta Universitatis Carolinae, Philologica 5), 1 3 5 -
144.
Hagege, Claude
1986 La langue palau. Une curiosite typologique. (München: Fink).
308 Claude Hagege
1988 Lefon inaugurate de la chaire de theorie linguistique. (Paris: College de

France) ( = "Linguistic theory. A contribution to an anthropological pro-
ject", Diogenes 145, 1989, 1 7 - 3 5 ) .
Hint, Mati
1981 "Neodnorodnost' sistem ceredovanija stupenej estonskogo jazyka" [The
nonuniformity of the gradation systems of Estonian], Sovetskoe
finno-ugrovedenie (Tallinn) 17: 247 — 265.
1987 "Loan words and the Estonian grade alternation", in: R. Channon —L.
Shockey (eds.) In honor of Ilse Lehiste (Dordrecht: Foris), 415—432.
Josephs, Lewis S. —Emesiochel, M. — Tmodrang, M. —Wilson, H.
1975 Palauan reference grammar. (Honolulu: The University Press of Hawaii).
Kiefer, Ferenc
1985 "The possessive in Hungarian: a problem for natural morphology", Acta
Linguistica Academiae Scientiarum Hungaricae 35: 85 — 116.
Underhill, Robert
1986 "Turkish", in: D. I. Slobin —K. Zimmer (eds.) Studies in Turkish linguistics
(Amsterdam/Philadelphia: Benjamins) (Typological Studies in Language,
8), 7 - 2 1 .
Wilson, Helen
1972 The phonology and syntax of Palauan verb affixes (Honolulu: University of
Hawaii Working Papers in Linguistics vol. 4, no. 5).
Index of languages
Alabama 285 Hebrew 305, 307

Albanian 100 Hua 221
Arabic 299 Hungarian 14, 17, 21, 6 9 - 7 6 , 150, 203, 222,
Arawak 150 298 f., 306
Azeri 306
Icelandic 205
Basque 151 Indonesian 285
Bengali 3 - 1 5 Irish 109
Breton 62 Italian 9, 18, 20 f., 62, 8 7 - 9 7 , 9 9 - 1 0 8 , 162,
166
Chinese 14, 150, 223

Japanese 14, 142
Choctaw 286
Clallam 285
Czech 35 f. Kafa 150
Kalkatungu 18
Kannarese 298, 300 f., 306
Danish 108
Koasati 286
Dutch 9, 100, 2 6 8 - 2 7 2
Korean 150
English 8 - 1 2 , 14, 19, 2 7 - 3 1 , 37, 6 1 - 6 7 ,

Lapp 301
8 0 - 8 3 , 87, 9 9 - 1 0 8 , 1 1 1 - 1 3 1 , 134f., Latin 36, 72, 97, 1 1 3 - 1 1 6 , 1 2 6 - 1 2 9 , 145,
138, 142, 144, 147 f., 150 f., 154, 172, 156, 161, 184 f., 204 f., 209, 222 f., 279,
204 f., 210, 212 f., 2 1 8 - 2 3 5 , 2 3 9 - 2 4 5 , 288
2 5 1 - 2 6 2 , 283, 287, 289, 291 f.
English, Middle 30
Nootka 291 f.
Estonian 3 0 1 - 3 0 3 , 306
Palauan 3 0 3 - 3 0 8
Finnish 100, 150, 233, 261, 3 0 1 - 3 0 3 Pashto 166 f.
French 14, 18 f., 4 6 - 5 2 , 5 5 - 5 7 , 63 f., 66, Persian 299
100f„ 106, 108, 1 3 3 - 1 3 8 , 161 - 1 6 5 , Polish 36, 164
168, 1 9 7 - 2 0 2 , 228, 234, 268 Provencal, Old 201
Fulfulde 18, 21
Romanian 1 8 3 - 1 9 4
Georgian 221 Russian 32, 148, 162, 164, 168, 203, 205,
German 5, 9 - 1 1 , 32, 6 9 - 7 6 , 8 0 - 8 3 , 100, 210, 213
107, 148, 156, 1 6 0 - 1 6 4 , 1 6 6 - 1 6 8 ,
2 0 4 - 2 1 5 , 221, 223, 229, 286, 288 Saanich 285
Greek, Ancient 71, 1 1 3 - 1 1 6 , 1 2 6 - 1 2 9 , Salar 306
145, 147, 155, 175 f., 278 Sanskrit 18, 21, 279
Greek, Modern 1 7 1 - 1 8 0 Saryg-Yugur 306
Greenlandic 150, 234 Serbo-Croatian 36, 1 6 4 - 1 6 6 , 168
Gujarati 150 Sen 17, 21
310 Index of languages
Slovak 36 Ukrainian 164

Swahili 221 Ute 142
Swedish 149, 230
Vietnamese 62
Tagalog 62
Turkana 17 Welsh 217
Turkish 18, 22, 287, 298 f., 306 Wolof 150
Subject index
ablativus modi 96 clipping 125

absolute neutralization 37 f. clitics 3 - 1 5 , 171-180, 183-194, 232 ff.
abstractness 37 f. coherence 25
accent variation 284 ff. combining forms 111 — 131
ad hoc formations 269 f. competence theories (of word formation)
adjacency condition 92 f. 273
affix clustering 270 f. complex adjectives 88 ff.
— homonymy 48 compound(ing) 88, 9 9 - 1 0 8 , 133-138,
— substitution 268 160 f.
afterthoughts 192 connectionism 253, 258 ff.
agglutinating 286 f., 297 connotative effect 74
agreement 145 ff., 153, 292 control 63 f.
a-morphous morphology 293 conventional component 54
analogical models 255, 258 conversion 81
— word formation 85 cross-classification of affixes and meanings
analogy 85, 251 ff., 2 6 7 - 2 7 3 50
analytic (type) 284 cross-lexeme referral 226
anaphor 188 cyclic rules 35 ff.
anti-iconic 285
argument structure 61—67, 83 deadjectival nominalization 162 fT.
articulated declension 185 ff. defaults 221
aspect 151 f. defective paradigms 198
associativity 43 — 57 derivational component 54
atom condition 94 derivative of the first/second degree 270
attested form 50 determination of gender 166 f.
— meaning 44, 46, 52 diachronic change 26, 281, 294
attribute markers 183 diagram 25
automorphism 26 diminutives 72 fT.
discontinuous affixes 81
baby talk 73 f. discourse function 71 ff.
backgrounded episodes 192 discourse function strategies 184, 191 — 194
base form inflection 204 drift 26
— component 54 dvandva 5 f., 131
base rule 54 dynasty of governors 9 — 12
binding 64 f.
borrowing 9 9 - 1 0 8 echo words 7 f.
bound word 232 f. emphasis 3 — 15, 71
bounding domain 9 empty morphs 20 f.
bounding node 9 f. — morpheme entailment 159—168
endoclitics 172 f., 178 f.
categorial markers 212 f. endophoric function 184, 194
chiasmus 28 evaluative suffixes 89
circumfix 284 f. excessive 69 ff.
312 Subject index
extended word-and-paradigm model 20, 293 Kennform 207, 211 ff, 225
extrinsic rule ordering 37 f.
language of love 73 ff.
fallback procedure 252 learnability 37 — 39
feature-node 258 left branch condition 11,15
feminization 268 left-dislocated objects 191, 194
final-segment strategy 253 f. lexeme-morpheme based morphology 168
focalization 188, 190 f., 194 lexical component 54
foregrounded events 193 — decision tasks 254
full-listing hypothesis 251, 261 — insertion 201
function composition 65 f. — phonology 20
function of rules 251 — pragmatics of morphology 69
fusional 286, 297 — semantics of morphology 69
lexical-morpheme hypothesis 264 f.
gaps 222 f. lexicalist hypothesis 228, 290
government 9 — 12 lexicon-external/internal inflection 197 ff.
grammatical concepts 282 f., 288 ff., 294 locality 20
- basic 288 f., 294 locative nominalization 164 ff.
— concrete-relational 288 ff. logical form 3, 8 — 11
- derivational 288 ff., 294 loose compounds 105
— mixed-relational 288 ff.
— pure-relational 288 ff. mapping rules 219, 221 f.
- radical 288 f., 294 markedness 24, 79 ff, 204 f.
grammatical processes (sensu Sapir) 217, markedness isomorphism 31
282 ff., 293 markers of discourse prominence 192
meta-rule 287, 294
hailstone models 20 ff. metaphorization 119f.
head 61 f., 82ff., 102-107, 134f. metathesis 285
homonymy 46 ff., 52, 163 f. middle (voice) 145
minor rule 51, 56
iconicity 25 modification 81, 284
idiosyncrasy 46, 55, 201, 213 modular grammar 217
incidental rules 269 ff. morpholexical phonology 35 — 39
incorporated markers 213 f. morphological creativity 267 ff.
indexicality 185 morphology-free syntax 228
inflection-class membership 206 morphopragmatics 69 — 76
inheritance of argument structure 83 f. morphosemantics 69
integration of foreign compounds 109 f. morphosyntactic representation 209
interactive orientation 71, 74 — word 233
interface program 172, 217 multiple inflection 207
intermediate stems 199 f.
intermorphs 20 ff. negation 174 ff.
intimacy 72 neoclassical compounds 112
irregular(ity) 205, 210-214 nonmorphological word-creating rules
"is a" condition 101 136ff.
island 9 - 1 2
isolating 284, 286 object clitics 191 ff.
isomorphism 25 one form — one meaning principle 297
Subject index 313
opacity 37 f. Saussurean sign 283

orientation of government 9 f. scope assignment 9, 11
overrides 221 secondary suffix 271
secretion 116ff., 125
paradigm class 219, 224 f. selector 55
— structure conditions 206 ff., 225 semantic drift 93 ff.
paragraph topic 193 — interpretation 44 f., 47 f., 52, 79, 83 f.
parasitic formation 198 — 200 — presupposition 70 ff.
particle lexemes 230 ff. semiosis 28
passive 141 — 156 semi-suffix 113 f.
past-tense inflections 259 ff. separation hypothesis 159 IT.
percolation 13, 79, 102 f., 160 shortening of compounds 100ff., 107 f.,
perfect 147 ff., 151, 154 119 f.
perfective 151, 154 slot 219
phrasal affixes 230 f. — calculus 226
plural inflections 252 ff. — competition 221
polysynthetic 284 speech situation 73 ff.
polytypical complexity 298 ff. stem inflection 204
possessive markers 188 f., 194 — rules 223 ff.
postpositions 10 storage vs. retrieval 249—251
poverty of the stimulus 38 f. stratification of the lexical component 43,
pragmatic reversal 72 52-57
strict cyclicity condition 37, 39
— strategies 183-194
structural coherence 29
predictable form/meaning 44 — 52
structure preservation 37
prefixation 81 f., 93 ff., 122ff.
stylistically marked formations 269
principles-and-parameters approach 38 f.
subtractive techniques 81
proclitic case morpheme 188 superlative 70 ff., 89
proclitic definite article 184 suppletion 1 7 - 2 2 , 209fT.
productivity 87, 91, 100-108, 251 f., 258 f. SVO language 105 f.
prominent discourse entities 194 syllable dissociation 285 f.
symbolic 286
quantificational domain 9 ff. synonymy 160 ff.
quantifier raising 10—12 syntactic typology 280
synthetic (type) 284
redundancy relations 35 — compounds 80, 82, 84
redundancy rules 79, 83 systematic identity 223 f.
reduplication 7 f., 14, 81, 284, 294
referral rules 223 telos of change 30 f.
regular-marked/unmarked 205 f. topic framework 187
reinterpretation 270 topicalization 188, 190, 194
resultative 148, 153 truncation 50 ff.
rhyme analogy 252 typological parameters 280 f.
— strategy 254
rule coherence 25 — 34 valence 150 f., 154
rule-creating creativity 267 — 273 verbal noun 5
rule features 206 ff. voice 152fT.
rules vs. analogies 251 ff. vowel shift, English 262
314 Subject index
Wackernagel's Law 172 f. X-theory 80, 280

weak pronouns 173 — 179
word order 99, 101, 105-109, 280, 284, zero derivation 228
289 f. — lexemes 160
word-structure rules 80 — morphs 222
word syntax 79
List of contributors
Paul Kent Andersen

Fakultät für Linguistik und Literaturwissenschaft, Universität Bielefeld,
D — 4800 Bielefeld, Federal Republic of Germany
Stephen R. Anderson
Cognitive Science Center, The Johns Hopkins University, Baltimore,
Maryland 21218, USA
Joseph Bayer
Max-Planck-Institute for Psycholinguistics, Wundtlaan 1, Postbus 310,
NL —6500 AH Nijmegen, The Netherlands
Robert Beard
Department of Modern Languages, Literatures and Linguistics,
Bucknell University, Lewisburg, Pennsylvania 17837, USA
Andrew Car stairs
Department of English, University of Canterbury, Christchurch 1,
New Zealand
Danielle Corbin
Universite de Lille III, CNRS, URA DO 428 "SILEX", France
Bruce L. Derwing
Department of Linguistics, University of Alberta, Edmonton,
Alberta T6G OZ1, Canada
Departement de linguistique, Universite du Quebec a Montreal,
Case postale 8888, Succursale "A", Montreal, P.Q. H3C 3P8, Canada
Wolfgang U. Dressier
Institut für Sprachwissenschaft der Universität Wien, Berggase 11,
A - 1 0 9 2 Wien, Austria
Claude Hagege
College de France, 11 Place Marcelin-Berthelot, 75231 Paris Cedex 05,
France
Brian D. Joseph
Linguistics Department, The Ohio State University, 204 Cunz Hall,
1841 Millikin Road, Columbus, Ohio 43210-1229, USA
316 List of contributors
Ferenc Kiefer
A Magyar Tudomänyos Akademia, Nyelvtudomanyi Intezete,
Budapest I., Szenthäromsag utca 2, H —1250 Budapest, Hungary
Aditi Lahiri
Max-Planck-Institute for Psycholinguistics, Wundtlaan 1, Postbus 310,
NL —6500 AH Nijmegen, The Netherlands
Maria Manoliu-Manea
Department of French and Italian, Sproul Hall, University of California,
Davis, California 95616, USA
Jaap van Marie

Koninklijke Nederlandse Akademie van Wetenschappen,
PJ. Meertens-Instituut, Diabetologie, Volkskunde en Naamkunde,
Keizersgracht 569-571, Postbus 19888, N L - 1 0 0 0 GW Amsterdam,
The Netherlands
Willem Meijs
Engels Seminarium, Universiteit van Amsterdam, Spuistraat 210,
NL —1012 VT Amsterdam, The Netherlands
Yves-Charles Μ or in
Departement de linguistique, Universite du Quebec ä Montreal,
Case postale 6128, Succursale "A", Montreal, P.Q. H3C 3J7, Canada
Wolfgang Mötsch
Akademie der Wissenschaften der DDR, Zentralinstitut für Sprachwis-
senschaft, Prenzlauer Promenade 149 — 152, 1100 Berlin,
German Democratic Republic
Sergio, Scalise
Dipartimento di Italianistica e Filologia Romanza, Universitä di Venezia,
Dorsoduro 3246, Ca' Foscari, 1 — 30123 Venezia, Italy
Michael Shapiro
Department of Slavic Languages, Brown University, Box Ε, Providence,
RI 02912, USA
Andrew Spencer
The Polytechnic of Central London, Faculty of Languages,
9 - 1 8 Euston Centre, London NW1 3ET, Great Britain
List of contributors 317
Irene Vogel
Department of Linguistics, College of Arts and Sciences,
University of Delaware, 46 E. Delaware Avenue, Newark,
Delaware 19716, USA
Beatrice Warren
Stockholms Universitet, Engelska Institutionen, S —106 91 Stockholm,
Sweden
Wolfgang U. Wurzel
Akademie der Wissenschaften der DDR, Zentralinstitut für Sprachwis-
senschaft, Prenzlauer Promenade 149 — 152, 1100 Berlin,
German Democratic Republic
Wiecher Zwanenburg
Vakgroep Romaanse Taalen en Kulturen, Rijksuniversiteit te Utrecht,
Kromme Nieuwegracht 29, NL —3512 Utrecht, The Netherlands
Arnold Μ. Zwicky
Department of Linguistics, The Ohio State University, 204 Cunz Hall,
1841 Millikin Road, Columbus, Ohio 43210-1229, USA
m Eung-Do Cook * Keren Rice
m Athapaskan Linguistics
m Current Perspectives on a Language Family
1989.15.5 χ 23 cm. VIII, 645 pages. With 1 map. Cloth.
m ISBN 311011166 7
[Trends in Linguistics. State of the Art Reports 16)
m
This volume represents an attempt to show the pre-
m sent state of the art in the study of this group of
Native American languages spoken in a large part of
m Alaska, as well as on the California coast and in the
American Southwest (including such languages as
m Navajo, Apache, and Hupa). These languages are
characterized by a particularly complex verb mor-
m phology.
m The lengthy introduction by the editors gives a

general overview of areas that have been important
m in the field of Athapaskan studies in the past 25
years, and provides the reader with the context in
m which the following contributions can be seen. The
papers themselves deal with diachronic linguistics,
phonology and morphology, syntax, discourse and
m ethnolinguistics, and have been prepared by lead-
ing scholars in the field.
m
m
mouton de gruyter
m Berlin · New York
ra George Horn
im] Lexical-Functional Grammar
1983.14,8 x 22,8 cm. IX, 394 pages. Cloth
[m
ISBN 90279 3169 0
(Trends in Linguistics. Studies and Monographs 21)
\m The analysis outlined in this monograph is formulated in
ι the context of the major developments in linguistic theo-
mi ry stemming from the proposal of the so-called Lexicalist
Hypothesis by Chomsky. The most significant product
of linguistic research during this period has been the
development and expansion of the lexical component
and consequent reorganization and reformulation of the
m rules of the transformational-generative model, in which
this component has been assigned many of the tasks for-
ΐγγΐ merly associated with the syntactic component.
More recently, various counterproposals to Chomsky's
analysis have been suggested. Perhaps the most impor-
tant of these was developed by Bresnan, the key feature
of which is the virtually complete reduction of the syn-
m tactic component.
This work is an attempt to extend and reformulate cer-
IYIJ tain of Bresnan's and Chomsky's ideas, combining the
basic organization of Chomsky's model, in which lexical
[m and non-lexical operations are clearly distinguished,
with a non-syntactic account of bound anaphora, con-
trol, and NP movement phenomena. The proposed
ΓΤΊ model provides a framework in which universal general-
izations can be captured, and language variation can be
|γγι accounted for without the complex machinery of
Chomsky's current analysis, at the same time maintain-
ing distinctions that are obscured in Bresnan's purely
mi lexical analysis.
\m\
[mi mouton de gmyter

Berlin · New York · Amsterdam

Morpho 123456

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Morpho 123456

Uploaded by

Copyright:

Available Formats

Contemporary Morphology

® Printed on acid-free paper which falls within the guidelines of the

Library of Congress Cataloging-in-Publication Data

Contemporary morphology / edited by Wolfgang U. Dressier

Deutsche Bibliothek Cataloging-in-Publication Data

Contemporary morphology / ed. by Wolfgang U. Dressier ... —

© Copyright 1990 by Walter de Gruyter & Co., D-1000 Berlin 30.

inflection from the history of French, P. K. Andersen against a universal

2. Morphology, Phonology and Aphasia (New York: Springer).

Topic 2: Word formation 41

Topic 3: Inflectional morphology and clitics 139

Topic 4: Computer morphology 237

Topic 5: The psycholinguistic study of morphology 247

Topic 6: Typology and non-Indo-European morphologies 275

Do the classical morphological types have clear-cut limits? . . . . 297

Index of languages 309

Subject index 311

List of contributors 315

In this paper, we advocate the view that to provide an adequate account

2. The core facts

(la) babul- ο kha- be

(4a) mar- ch- i- ο

(10a) [bap ma]-o

The interpretation of dvandvas such as (10a) suggests that /o/ attaches

3. Emphatic clitics in the lexicon

3.1. Bengali has a rule which degeminates syllable-initial geminates. The

(11a) kha- cchi —• khac.chi (closure of preceding syllable)

Resyllabification is, however, blocked outside the domain of a phonolog-

c) Inadmissible verb stems:

(15a) mereche tereche '[past part + 3rd pers] beat'

g) no:t '(male) dancer'

4. Emphatic clitics in syntax and logical form

have argued, a phrase narrowly focused by only, even, etc., must be

(17a) we saw only John

(19a) John would [even [talk to MARY]] (adjoined to VP)

(adjoined to N P inside PP)

(20a) weil Hans [sogar [mit MARIAN sprechen würde

b) *weil Hans [mit [sogar [MARIAN sprechen würde

(adjoined to NP inside PP)

Adopting the rule of quantifier raising (QR) as suggested in May (1977),

(21a) Focus-sensitive quantifiers {only, even; nur, sogar; -ο etc.) must

(22a) ami fipra- r- ο fange kotha bolbo

There are other postpositions, however, which do not allow similar

4.2. Like in German, Bengali adjectives take NP-complements to the

(23) weil das Gericht []\dem Patienten] sogar] bekömmlich] ist

(24a) ram- er- ο prio

Exactly the same constraint seems to be at work in logical form, as

the bare noun. The dominating minimal maximal category is the NP

The results achieved in Section 3 left us with an apparent paradox: when

1. mar is the V-stem.

In some morphological alternations, the alternants are dissimilar pho-

(2) Turkana: deverbal abstract nouns: suffixes -isi, -u and -f^C,

(3) Seri: Passive prefix on verbs: p- before vowel-initial roots, a:9-

(4) Kalkatungu: Dative suffix on nouns: -ku after consonant-final

(5) Fulfulde (Southern Zaria district): General future active tense:

(6) Turkish: 3rd singular possessive marker on nouns: suffix (subject

(11) Italian masculine definite articles: lo or /' (singular) and gli

as Zwicky (1985: 435) puts it; for phonologically conditioned suppletion,

and morphosyntactically empty intermorph whose sole function is to

One of the factors commonly retarding progress in linguistic theory is a

in Prague and Copenhagen) of the foundations of linguistic theory. What

[-Θ] mouth [-Ö] mouth

that the mapping of sound/meaning relations is one of simple replication;

cally to produce diagrams of the markedness values of the terms of