e

&
*.
,|:

96

Roronoucos

Diane. 1985. Case Th,eorg a,nd the Projecti'on Principle. Doctoralrr

r-lissertation. I\{IT.
McCtoskey, James. 1983. A VP in a VSO Language? In Order Concord,
and. Consti,tuency, ed.. Gazdar Gazdar, Ewan I{lein, and Geoff Pullum.
Dordreci-rt: Foris.
McCloskey, James. 1984. Raising, Subcategorization and Selection in Iv
ern Irislr. Natural Lanquaqe and Linquist'ic Theory 1:441-485.
Ranrchand, Gillian C. 1997. Aspect and Predtcatton:The Semanti,cs oJ A
i
ment Structure. Oxford: Oxford University Press'
1\{s.,
Univerand
Case.
Agreemcnt
N{ovement,
1990.
Dominique.
Sporticire,
sitv of California, Los Angeles.
Tenny, Carol. 1994. Aspectual Roles and the Syntctr-Semant'ics InterJace,
Dordlecht: Klurver Acadernic Publishers
Tbavis, Lisa. i991. Derived Objects, Inner Aspect and the Structure of VP.
Paper presented at the North Eastern Linguistics Society (NELS) 22':
Vendler, Zerto. 1967. Linguistics in Plt'ilosophy. Ithaca, New York: Cornell

Ntlassam,

Building Verb Meanings

ALKA R,RPPRpOnT HOVAV AND BETH LnvTN

University Press.

Verkuyl, Henk J. 1972. On tlte Compositional Nature of the Aspects.

Netherlands: Dordrecl'rt.
Verkuyl, Henk J. 1993. ,4 Theory of AspectualitE. Cambridge: Cambr
University Pr,:ss.
Zaenen, Annie. 1993. Unaccusativity in Dutch: Integrating Syntax and
ical Semantics. In Semantics and the Lericon, ed. James Pustej
129-161. Dordrecht: Kluwer Academic Publishers

syntax-lexical semantics interface in the eightpoint the hypothesis that many aspects of the
tsyntactic structure of a sentence-in particular, the syntactic realizetion of arguments-are projected from the lexical properties cf the
verbs and ot hcr predicators in it (see Wasow 1985). This hy1;oihcsis is
made explicit, for example, in the GB framework's Projection Princi(Chomsky 1981, 1986). On this approach, the Iexical property of
a verb that is taken to determine its syntactic behavior is its rieaniog (..g., Levin 1993; Levin and Rappaport Hovav 1995; Pinker 1989).
Much effort was therefore devoted to developing lexical semantic repre,.Much research on the
ies took as its starting

sentations wirich couid serve as a basis for the perspicuous formulatioo

rules-often called linking rules-which determine the realization of

ments in svntax.
Close scrutiny of possible argument expression options reveals thiit
verbs exhibit a range of argument expressions, a phenomelrou
has more recentiy been the focus of attention in work on arguexpression. Sornc illustrations of this variation are given beloiv,
the verbs sweep, wh,istle, and run.

ill

a.
b.
c.
d.
e.

Terry
Terry
Terry
Terry

swept.
su.cpt thc floor.

swept tlie crumbs into the corner.

swcpt thc leaves off the sidewalk.
'Ierry swepi the floor clean.

i;.We have presented this research in a rrumber of different forurns and would like to
:. thank the audiences for thcir questions and contrilrutions. We aro also grateful to
I{ari Olsen, Grace Song, ancl two reviewers lbr their cornments on arr carlier versicn
+,_ii:,lof

j.l}l:l

..

this paper. This rcscarch w:rs supported by NSI:'grant DBS-9221993 to Levin

The Projection oJ

Argunents: Lerical and Composit.ional Factors

Miriam Butl arrd Willrnlrn Gouder (e,Js.)

r998, CSLI PrrLlic:rtions
tr) Qopyriglrt @)

l
l

s8/

M.qr,x,{ R.,q.ppapoRr'Hovev AND BDTU LEVIN

f.
(2)

i
;l!
I

i
t

f,

(3)

BuIr,orNc VpRs MplNrNcs

Terry swept the leaves into a pile.

Kim whistled.
Kim whistled at the dog.
Kim whistled a tune.
Kim whistled a warning.
Kim whistled me a warning.
Kim whistled her appreciation.
Kim whistled to the dog to come.
The bullet whistled through the air.
The air whistled with bullets.
Pat ran.
Pat ran to the beach.
Pat ran herself ragged.
Pat ran her shoes to shreds.
Pat ran clear of the falling rocks.
The coach ran the athletes around the track.

This variation involves not only the number and syntactic type of
plements that a verb can take, but also the allowable combinati
these complements. For instance, although sweep may simply
direct object as in (1b), the NP the crumbs is possible as a direct obj
of sweep only if followed by a PP complement; thus, the acceptable,
contrasts with the unacceptable *Terry swept the crumbs.
the variation in syntactic context correlates with variation in
Consider the verb sweep agait. Sentence (La) shows the propert
an activity, while (1c)-(1f) show propeities of accomplishments.
specifically, (1c) and (1d) describe the bringing about of a c
location, (1e) describes the bringing about of a change of state,r
(1f) describes the creation of an artifact. A second example rei
this point: the verb run cafi be found in contexts describing an
motion to a goAl aq ijl;b)
as in (3a), dlrected
lirected mo_UgL.!e*A
iu$!), a caused change of
------i#
as in (3c) or (3d), or a caused change oflocation as in (3f).
Such variation poses a serious problem for theories of
expression. On an approach which takes argument expression
determined from a lexical semantic representation via linking
verbs with multiple options for the expression of arguments
have multiple lexical semantic representations. If such variatioil
is
rule rather than the exception-and recent studies show that t\e,
nomenon is indeed widespread-then the lexicon must contain a
number of verbs with multiple lexical entries. This result is
able, however, if only that it seems counterintuitive to have to
that, for example, in sentences (1)-(3), we are dealing with six d

99

sueep, nine different verbs whistle, and six different verbs rzn.

,l

Then

ver

bel

determine the range of meana verb can have, and the existence of these multiple lexicr,l entries
uld be accounted for in a principled way.
Although a first glance at examples such as (1)-(3) might give the
ion that a verb's meaningl can vary in almost unlimited ways, a
look at the attested variations in meaning reveals that this is not
There are several striking properties of such variation which suggest

t--rd-,thuunay-Ue."-..qpsfi.i;i-qd--ag j_B!-Eli.g-'o.,flhti]til-GT
c-aUq. :t-e_S-r+"lgl.R9.lys=egty"," For example, the Beliavior oil sweep
cal of verbs of surface contact through motion; thus, the verb of
contact w'ipe is found in the same range of contexts as sweep,
,..

- ,.,r

.i-.+,i,Y'4.!Gf:

illustrated below.
a.
L
U.
j

C.

''

d.
e,

f.

Terry
Terry
Terry
Terry
Terry
Terry

wiped.
wiped the table.
wiped the crumbs into the sink.
wiped the crumbs off the table.
wiped the slate clean.
wiped the crumbs into a pile.

, the verb run

shows a pattern of behavior characteristic of

of manner of motion, while the range of contexts availabie
l/e is typical of verbs of sound emission (for a discussion of the
er, see Levin 1991). Thus, verbs naming similar concepts appear in
ppme range of syntactic contexts, with verbs in each class showing a
pattern of behavior. The fact that the range of meanings availa verb seems to be determined by its semantic class membership
four presentation of this phenomenon we

attribute the variation in meaning in

such as (1)-(3) to the verb, characterizing it as verbal polysemy; others
found in.ffi
!89*--v.yiation i" -gililg lg" |-h9 9o-nstruc!_i9n the verb--is-thii
tiid pErs"p-ttines. i ri t [e ippendi*l wheie we-'poirif oiit
T"ir, t"""
lem of accounting for the variation in meaning. We also argue that the
incorporate similar assumptions about the nature of verb meaning,
a consequence the insights that we present in this paper into the nature of
ariation in meaning carry over in large part to the constructional approach,
r;phough we will describe the problem in terms of verbal polysemy in the body
paper.

i :i ,",t

100

Mer,xn Reppe.potr Hovev

.q,No

Bo'ru Lovrw

BurlrrNc VpRs

that there are priqqlples which govern variation ir, _ygfb

as illustrated in deffiif6y-Sl-otiln (1987, 1996),
(1-975, 1985, 1991), Wienold (1995), and others, the range of mr
available to members of a particular verb class in one language
not be available to the members of the corresponding class in
Ianguage, with systematic differences attested.

in& Furthermore,

1- Manner

i
versus Result Verbs
It is striking that the verbs whose impressive elasticity in
illustrated in (t)-(3) all lexically specify or "lexicalize" the m
in which the action denoted by the verb is carried out. These

contrast with verbs such as break and open which lexicalize the

jf ,!]ffi

e.1,."a-".ff

ff[i_[-iJs_i1,..p".*1_";ettrre_qlade..Wetfr

2We a.e not concerning ourselves with variations in meaning that involve
or metaphorical uses of a verb (e.g., eat as in The acid ate the metal).

MnA.NTNGS

101

to verbs like sweep ar'd run as "manner verbs,, and to verbs such
After clarifying the nature of the
nded semantic distinction, we demonstrate a correlation between it
the range of syntactic and semantic variation available to a verb:
ilt verbs show a much narrower range of variation in meaning and
tactic context than manner verbs.
As noted above, the verb,jpeep_gxhibits behavior tha" is represenive of that of other verbs of6urface contact, including
ry.tb g-1d wdpe.
three verbs are distinguished from each other in tfie maniilT?
contact they specify but none of these verbs, in its most basic
entails
a resulting change in the contacted surface. Thus, although
,
ror is typically swept in order to remove dirt and debris, a floor thai
t need not end up being clean. Although a hearer will infer that
floor is a clean floor because the conventional goal of sweepis to clean a floor, there is nothing contradictory in saying Zl.acg
break, and open as "result verbs.,,3

ISn Ver
, and perhaps most importantlv, no
the same range of flexibility in meaning and argument expression,
again this property is correlated with a verb's semantic class. For
ample, as we illustrate extensively in section 1 of this paper,
change of state such as break do not permit anything like the r
argument expressions that verbs ofsurface contact such as streep
A prerequisite to the development of an explanatory theory of
sible variation in verb meaning is the documentation of allowable
tions and, more important, the constraints on such variation.2 To t
end, we present a case study of the two verbs break and sweep,
representative of a well-established semantic class: verbs of ex
'caused change of state and verbs of surface contacTTIiiiffi i
rr rgsprecti-vely. We choose these verbs because they serve to illustr
dichotomy in the English verb lexicon that turns out to be ce
understanding the nature of the elasticity of verb meaning. As we
verbs show a striking contrast in the rar
, I i, section 1, these two
'i-, ij --.-. -^-- :-- . - , ,-:r^l-la, rr--,,
nr -r
rr---- -:-- ,-r-,rr
meanings
available
to
them.
First, we- carefully
review
the ar
f''
u-expression options available to each; then, we identify a fu
difference in their meaning. In section 2 we use this difference d
clue to the nature of the process which gives rise to variation in v
meaning. In section 3 we sketch the rudiments of a theorv whichl
predict the exact range of meanings and syntactic contexts which
dividuai verbs may be associated with. In section 4 we show how
theory accounts for the difference between the two verbs under sr

"_i

!.r'E

achieved location (a kin-d of result)

ir ggtAilS-d=-by_s..1=rch verbs unless

goal,"phlaCelF--+,"{ded.
fur exa[tple,
example, the
lne sen[ence
sentence P
6v@l_.I-4rlroEtr_rD*duu!u. For
rat
at rans
runs
states that Pat is moving in a particular way. Typically, we

r.pru,
95121ip1!-.

that Pat is also undergoing some displacement, but in the abof a goal phrase this displacement need not be oriented towards
particulg,L_ggal, apd in fact there need not be any displacement at
WS!*fq!".y^_p!gc-e\ The verb whistle shows the behavior typical of
rbs of sound emission; such verbs lexicalize the manner in which a
is emitted (and perhaps properties of the sound emitted) and
from each other with respect to manner of sound emission. Thus,
Ie is emitted by forcing air through pursed lips, while a grunt
ves the vocal tract.
[4anner verbs like sweep, run, and. whi,stle can be contrasted with
verbs,4 which lexicalize a particular result, but more often than
are vague as to how the result is achieved. There are tye-lyp-gl
type lexic4lizes a qqsu-lti.ng stare and tne-oih*er a
i*Ly"fpUgle.
-a,,
i"At i;;:- T-;;,;
."Cf,' *
ii*lai o,'
s; ;i
"h
""
"i
TIG*ifrme implies, lexicalizes a particular achieved

;i

;;;;:

state, and

,v.erb denotes

the bringing about of this state. But though the

manner/result verb distinction has also received independent support from

in child language acquisition, including Behrend 1gg0, Gentner 1g7g, and
et al. 1991.
are some verbs which lexically specify both manner and resultl czt is
such a verb.

102

M.a.r,xe Rapp.q,poRt' Hovev aNo

BBru

Luvrr.r

BUTLDTNG

verb itself denotes the bringing about of this state, it leaves the nat
of the causing activity involved unspecified; that is, such verbs do n(
Iexicalize a manner. For example, clothes may be dried by putting t
into a dryer or by putting them out in the sun; the verb d.ry may be
no matter how the dry state is brought about. Likewise, a vase may be
broken in many ways; again the verb break,s own meaning contributes
nothing to the specification of how the vase came to be broken. Thg
verb clean also specifies a resulting state, while leaving open how it ls.
achieved. In fact, how a clean state is obtained depend.s on the surfi
involved: floors are cleaned by sweeping or washing, while counters
tables are wiped. VS:bp__y.hiSh lexis_A]ize..ap.aqhtgygd tocation are t
t"qglg.Up_Lgf lesuit verbs. These are verbs of directed"motion su

SJ9ru,'-sr,;;Ailr.dft ,*,w,[l,sJ.,-.te"l;iilr-;q"+e'U*is-Ci-tpdig;""f
Ug**9ly*qhp.__a-dueciio.s)-,..b* *_+g!_-q-gn-?-ngel of.,rpo*,trs.r1", r,oi eiim

their syntactic behavior than the manner verbs by presenting a vari

of contrasting properties of the two verb types.+First, two-argume
manner verbs more readily allow the omission of their direct ob.j
than
-twg-algilment result vetbs. Contrast the manner verb sweep a
the result verb break in (5). Although sueep may occur without
object even in the absence of any context, break cannot, and it is e
difficult to think of a context that would improve an example such
(5b).

(5)

Leslie swept.

a.

xKelly broke.

b.

6)

a.

b.
a.

7)

b.

Vpna MpeNrNcs

108

Cinderella scrubbed her fi.ngers to the bone.

xThe clumsy child
broke his knuckles to the bone.
The child rubbed the tiredness out of his eyes.
xThe clumsy child
broke the beauty out of thu ,ur".

I,. Although sweep and. other verbs of surface contact can be used

as

or change of location, change of state, and creation

as illustrated
u verb of change of state like break cannot show sr,,h
a range
1 {f),
,f uses, as illustrated in (8)-(f0).
In each pair, the (a) sentence with
'eok cannot be assigned an interpretation
to parallel the (b) sentence

19191

ith a manner verb; the intendedlnterpretation ofeach break

sentence
ven in parentheses.

*Kelly broke the dishes

off the table.
(meaning: Kelly removed the dishes from the table by
breaking the table)
b.
Kelly swept the leaves off the sidewalk.
xKelly broke the
a.
dishes off the table.
(meaning: Kelly broke the dishes and as a result they
went off the table.)
b.
Kelly shoved the dishes off the table.
xKelly broke the
a.
dishes into a pile.
(meaning: Kelly broke the dishes and made a pile out
of them)
b.
Kelly swept the leaves into a pile.
ilar distinction is found among verbs of motion. Directed motion
glg_qlqq lgl4
-g,:.Ilei.. behaytp_ri they do not exhib!t-:b.Le_;cange
iqqs- av.eila}-le.to verbs of manner of motion. Thus, the verb
oo
found in t[e;;ffi;;ng ;ffi;f;#it ;,%a* as the attempt *'#
to
it in contexts parallel to those in (3) shoffi*
tf
.h uet}.]'
a.
-..
The students went.
i
b.
The students went to the beach.
gh +i#1,.{.s.-.*
*The jetsetters went
;ar
c.
themselves ragged.
*The runner went his
d.
shoes to shreds.
*The pedestrian went
i: e.
clear of the oncoming car.
*The coach went the
t.
athletes around the tr"ack.
L, the impressive flexibility of manner verbs
with respect to art expression contrasts with the rerative rigiditv of result verbs.
a.

^t

-r i

Second,

rygllglJgrb!_can readily appear with

a wid.e range

of

,;4

subcategorized," gbjegts. whereas such objects are not available

sult verbs. In (6a) her fi.ngers is a nonsubcategorized object si

is not the surface that is being scrubbed. Although this senten

understood to describe the scrubbing of a surface, the surface
not mentioned. Thus, the sentence means that Cinderella scrubi
something, perhaps the floor, until her fingers were raw; however,'(
cannot have a parallel interpertation: the child broke many thir
and as a result of handling the broken things his knuckles were
An additional pair of examples contrasting manner and result vert

given in

(7).

he Monotonicity of Verb Ivleaning Construction

this section we provide an intuitive explanation for the distinctive
tavior of manner and resrlt verhs. Then. in the followinri scction

104

Mer,re Rlppaponr Hovav aNo Bptu LovrN

BUTLDTNG

derives the desired results.

As a first step towards understanding why the difference in

of meaning component lexicalized in the verb should be corre
with such a striking difference in verb behavior, we point out
manner verbs and result verbs have different lexical aspectual c

i,

manner and result verbs. For example, as illustrated in (1), the vt

sweep has a variety of uses that quaiify as accomplishments, as well
uses

that qualify as activities; nonetheless, we assign this verb a

classification as an activity verb because all its uses entail an acti

involving a particular manner of surface contact, but only some of
uses entail a result. We assume that the basic classification of a
is determined by the nature of the concept which the verb lexi
see section 3.1
Research on lexical aspect has brought out intenelationships
tween the various aspectual classes. These interrelationships have
captured in the predicate decompositions assigned to the members
the various aspectuai classes by certain theories of lexical se
representation. For instance, accomplishments are generally
a complex event structure. (For arguments in favor of such an
structure see Dowty 1979; Parsons 1990; Pustejovsky 1991, 1995; V

Valin

1993.)

(12)

[ [ x ACr ] CAUSE I BtrcoME

ly <SrArE > ])l

'i
L'i'

Accomplishments are complex events composed of two subevents:

causing event-typicaliy an activity-and the change of state it bri
about. Thus, in the representation (12) the first subevent of an
complishment has the representation also associated with activiti
given in (13), while the second subevent has the representation
associated with achievements, given in (14).
'

(13) [x ACr
(14) [ eocor,to lx

105

meanings which thg verb is associated with involve the adof a resulting state: a change of location, a change of state, or

we outline a theory of verb meaning and argument expression w

Vpns Mp,qNtucs

<STATE> ])
The possible variations in verb meaning illustrated in (t)-(3)
involve the "expansion" of an activity to yield various kinds of
plishments. For example , if sweep is basically an activity verb, all

Ilr r-

coming into existence of some artifact. In each instance, the adi

n of the resulting state is signalled by an element in the syntax. iL
!
example, in all the examples in (1), the verb sweep itself signals \
activity part of the event structure, and in (tc)-(tf) the resulting i
is signalled by some other element in the sentence. Th', pattern iI
J
that verb meaning is built up incrementally.
If we take this idea further and assume that verb meaning is built
fashion, in
pln
we have a natural explanation for
#ior of the result veibs and the manner verbs.s If
uid that many of the result verbs are inherently classified as
Iishments, then the rigidity in their behavior can be attributed
the nature of the lexical representation of accomplishments: accomIqglts have fully lexica,lly-specifi ed r -epresenJations. They cannot
on activity readings without eliminating a lexically-specified com-.
of verb meaning (the resulting state): F\rrthermore, we proj3" tt ut no additional expansion of their representation is possible
he accomplishment representation is the most complex representa\l
available. Consequently, if the accomplishment verb is one that
a change of state, no change of location can be added to the
!r
ng of the verb, and vice versa.

,k ,,

li

Components of the Theory

this section we sketch the components of a theory intended to acrnt for the possible meanings available to different types of verbs.
is theory incorporates the insight of the previous section that verbs
ve basic classificationq which can then be- expanded in a monotonic

friffi.'ln seCiion 3.1 we introduce the representations of verb meaning

iiiume, Then in section 3.2 we present well-formedness conditions
e are some types of meaning change that are not monotonic, but these apto represent phenomena distinct from that under study in this paper' An

iears

is the phenomenon sometimes referred to as "semantic bleaching," which

to involve the loss of some element of word meaning. Illustrations of se-

bleaching are the use of the verb break it Th,e news broke or the use of the
JaIl in The baby tell asleep. lt is significant that semantic bleaching always
ves the loss or weakening of the idiosyncratic aspect of verb meaning (what we
the "constant" in section 3.1), and to our knowledge, never involves removal
.grammatically-relevant aspects of verb meaning (what we call the "template"
i:rsection 3.1). F\rrthermore, "semantic bleaching" is quite idiosyncratic, being
iated with individual verbs rather than with grammatically-relevant semantic
of verbs in the sense of Levin 1993. Both these properties set it apart from
type of variation in meaning represented bv examples such as (1)-(3).

ii
:11

il

/.1 -- \I lUO /\MALKA RAppApost Hovlv aNo BBru LBvrN

\ggrr"qfrirg the association
of a syntactic expression with the meani

\r

refr6sentation.

3.1 The Basic Elements of Verb Meaning

Much recent work in lexical semantics either implicitly or explici
, iie'cognizes a distinction between two aspects of vlrb meaning,- wh
rwe term the "structural" and the ,,idiosyncratic.,, This distincti
1so central to our theory of the repres..rtutio, of verb meaning.
structural part of a verb's meaning is that part which is relevant to
termining the semantic classes of verbs that are grammatically rele
while the idiosyncratic part of a verb's meaning distinguishes that
from other members of the same class. The recognition of this subdi
sion in verb meaning is a maior achievement of recent lexical sem
research, and it is assumed, although not necessarily made explicit,
the work of many researchers.6 For example, Grimshaw (fOOe) c
tinguishes between "semantic structure,,, roughly the structural
ponent of meaning, and "semantic content,,, roughly the idiosyncr
component of meaning. ,{ale an_d Keyse. (1ggl) 4ssociate fixed me
ings with certain structuhi6fifigurations which occur in the lex
representation of many verbsl these configurations correspond to

structural componelt of meaning. In their representations, the

tgYg:r,uri**pec!-sf pq3nips. a-rct rese
\ggd" which is
serted into these structures and then undergoes Hei-ElMovement i
an empty V position, thus integrating the idiosyncratic component i
the structural component of verb meaning.
It is usually assumed that the structural component of verb meani
fl .
is that aspect which is grammatically relevant-for example, relevr
{/
to .lg"T"nt
argument realrzation-and
realization-and defines
defi.nes the grammatically_relevant
grammatically-relevant ir
f,
i\ ,o
mantic classes of verbs- that is, those semantic crasses of verbs who
share syntactically- and morphologically-salient properti
1 r-members
fact, in most current theories, the aspects of meaning wticn i
\,\In
l\"Brammatically relevant usually are those which define the various c
toloSical types of events, which correspond roughly to the recognizr
| |'
vendler-Dowty aspectual classes of verbs. It is for this reason, that t
grammatically-relevant lexical semantic representation is often cai
an "event structure." In contrast, the idiosyncratic aspect of v
rri.eaning serves to differentiate a verb from other verbs sharing
,

6For more extensive discussion of this

distinction and its conseque[ces

Grimshaw 1993. For a presentation of the alternative point of view-that a v,

meaning cannot be carved up according to whether or not it is grammaticar
relevant-see Taylor 1996, and Jackendoff 1g96 for a rejoinder.

Burloruc VpRe MBeNrNcs

102

structural aspects of meaning; the idiosyncratic aspect is not

vant to the verb's grammatical behavior.
In many current theories, verbs are given an articulated lexical senantic representation taking the form of a predicate decomposition
Jackendoff 1990; Levin a,d Rappaport Hovav lggb; pinker rodo; Rapaport Hovav and Levin in pressl van valin 1gg3). A predicate decomPi.tiol ig made up of.two major types of componen#, p'rim'iti"" prAaates and what have been called "constants." bpecific'co-binations of
ffiitiv6 predidafe-s represent the structuiar dspect of verb m,.r,ning,
the constants represent the idiosyncratic element of rneanin[.
various combinations of primitive pred.icates constitute the bastock of lexical semantic templates of a language (the ,,thematic
:s" of Pinker 1989). A verb's.meaning consists of an association of
golgtant with a particular lexical seinantic tdmplate. For example, ii,
ve change of'state verbs share the lbxical semantic iemplate in rl'
!5a), with the various members of the class being derived by a particinstantiation of the constant representing the state, as illustrated
the representation of causative drE in (15b).
.e

) a

[[xACr

] CAUSE IBECOME ly <STATE>l]l

[ [ x ACr ] oAUSE I encotro I y <a,Br> ] l'1"
As discussed in the introduction, verbs with multiple meanings are
rciated with multiple lexical semantic representations. These mulmeanings usually arise from the association of a single constant
more than one lexical semantic template. For example, some of
uses of the verb stoeep illustrated in (1) can be derived by the asiation of the constant <swU0p> with the three temprates in (to);
present fully-instantiated representations that include the actual
tant in section 4.

b.

6)

a.

b.
c.

xACTI
I x ACr I CAUSE I BECOME ly <SrArE> )l)
Ix ACr ] causn Inocoun ly <pLACE>']'l'l

we assume that universal Grammar provides an inventory of lex'.semantic templates consisting

of various combinations of primipredicates, which correspond to a large degree to the generally
nowledged event types. For this reason, henceforth we ,efe. to a
ca.[ semantic template as an ,,event structure template.,, The in_
tory of event structure tempiates includes those listed below.T (The
lhe proposed activity event structure template has a singre argument, arthough
recognize that there are two-argument activity verbs such
we discu"ss

^"i.""p;

108

i
i

N,Ialxe R,rplaponh Ilovev .qno Bultt LuvtN

Burr,orNc VBRa MoaNtNcs

x ACTar1aa17rrr,,

x <STATE>

BECOME lx <STATE>))
I x ACTaIaaNNEB> ] C,tUSO

Iencor'tn Iy <57,,{78> ] I l
x CAUSE I encotvto Iy <S7:l7E> ] I l

'

(activity)
(state)

lrr

include those given below:e

7)

(achievement)

(accomplishment)
(accomplishment)

Althougir the set of event structurc templates is fixed, we

109

imal elements of meaning encoded in the constants must be given

ntactic expression. In this way, these rules also ensure the monotonicity of verb meaning. The fundamental canonical realization rules

italicized rnaterial in angle brackets represents the constant, which we

discuss imrnediatcly below.)

18)

(e.g., brush, hammer, saw, shouel, ...)

placeable object -+
Ix CAUSE IBtrCoME Iy WITH <7I11I'rG> ] I l

),

:.

assume

(e.g., butter, oil, paper, tile, war,

thal tlrr- set o[constants is open-cncled. Each constant has an ontd

Iogical categorization (Jackendoff 1990; Pinker 1989), drawn from; all

fi.xed set of types (e.g., state, thing, place, manner, etc.). Each con:
'stint
is also associated rvith a name (i.e., a phonological string). Asl
the discussion above suggests, we assume that the ontological t
of a constant determines its basic association with a particular
struciure tempiate; these associations are specified via what we call
'lcanonical realization rules," whose narne 1s inspired by the canonical
structural realizations of Grimshaw 1981. In addition, each constanti
also determines the basic number of participants in the event it is a&l
sociated with.8 For example, although both running and sweeping F.el
7- activities, an event of running minimally involves the runner' while an
,/ event of swceping minimally includes a swccper and a surface
l
\ of I he nature of swccping itsclf.
\
Tlrerc are t$.o ways irr whiclr constanLs aro intcgratcd into event
structure ternplatcs: eithcr they may be modifiers of predicatesiloi
they may serve as arguments of predicatcs, filling a particular argu'
ment slot in a ternplate. A modifier constant is associated with the
predicate in an event stmcturc template that it rnodifies. For example,
a constant categorized as a manner usually modifies an activity, and
it is thcrefore associated with the predicate ACT in the activity event'
structure template. Likervise, an instrurnent constant modifies an raqi,i;
tivity (since it specifies the generic activity associated with using tt
instrument), so it too u'ill be associated with the predicate ACT

1)

3.2 ancl 4.1.

sGoldbe.g (1995) and varr llout (1996) also propose that the idiosyrrcratic com
nent of a vcrb's mearring, what we call the "constant," determines the basic number
of partir'i1,;rnts itt an lvenl

...)

place -+ [x CAUSE I BtrCoME ly <PLACE> l)]

(e.g., bag, bor, cage, crate, garage, pocket, ...)

internally caused state -+

lx <STATE>

(e.g., bloom, blossom, decay, fl,ower, rot, rust, sprout, . . .)

externally caused state -+
[ [ x ACr ] CAUSE I nocor,to ly <STATE>
(e.g., break, dry, harden, melt, open, . . .)

ll)

rule pairs a constant of a particular ontological type, specified

the Ieft of the arrow, with the event structure template to the right
the arrow. In these rules, the place of the constant in the associated

structure template is indicated in capital italics and placed between angle brackets. Argument constants appear in tiie appropriate
L position in the templates. Modifler constants appear as subts to the appropriate predicate in the event structure templates
We call the pairing of a constant and an event structure template
]irn lqyqnt structure." We consider the pairings cf constants and event
lghstead of the event structure template in (19), the canonical realization rule
ior placeable objects might give rise to the ternplate: I x CAUSE I BECOME
<THING> AT V ] ] ]. There is some debate as to which template is the more
appropriate one; see Hale and Keyser 1993 for discussion. We also leave it to further
to determine whether the templates in (19) and (20) should be of the form

activity template.
TIie basic idea behind the canonical realization rules is that
the treatment of the nonactor argulrlent of tlvo-argtrrnent activity verbs in sectiong

manncr + [ x ACTa *.q.*ruaa, ]

(".g., jog, run, creak, whistle, . )
instrurnent -+ [ x ACTa TNSTR1MENT> l

fcvent cause event" rather than the proposed "individual cause event."
Our representations for activities use the primitive predicate ACT modified by
rnanner constant; other representations for activities use a predicate DO taking
coustant as an argument or, if not, take a form that is suggestive of this analF.r{, as in Hale and Keyser 1993. For example, our proposed analysis of laugh
xACIalaLtcs;, ] could be contrasted with the following alternative analysis,
xDO <l,q,UGI1> l. The representation in terms of DO appears to receive support from languages like Basque where the counterparts of English one-argument
rctivity verbs are expressed by ttre verb do plus a noun (Levin 1989); however, this
rcpresentation does not seem appropriate for two-argument activity verbs. Clearly,

ftrther investigation irrto the proper represerrtation of activity verbs is necessary.

110

N4ar,xe R.qppaionr Hova.v nNo

Bllrtt LDvIN

Burr,nrNc VoRs MTINTNGS

lr
cou1 l.

structure templates effected by the canonical realization rules to

stitute a basic verb meaning; we return to derived meanings below'r1i
We assume th.at the "name'i associated u'ith a particular verb
ing is contributed by the constant (see Rappaport Ilovav and Levin iul;
press for discussion). A verb's lexical entry, thcn, consists of the name;
contributed by the constant together lvith the meaning, represented'
i, lcllosyn;L;
idiosyn-l
*t ut IS
specifies what
as an evcnt structure. Since tlie constant specthes
say
thafli
will
meaning-we
its
"corc"
particular
verb-i.e.,
a
to
cratic
verb
sueep
the
exarnple,
an
To
take
constant.
the
verb
lexicalizcs
the
takes its nanre from a constant whicli specifi.cs a certain u]&nllr ofr ,
surface contact involving motion; due to the nature of this type of suqi
face contact, this constant is basically associatcd with an activity evep,t;;
stiucture templatc by the canonical realization rule (17), and thus, thei,
verb strreep is an activitY verb'
We irave just suggcsted that a simplc verb rncaning involves tt
association qf a conqlqq! *i!.h _at gv-e-ry 9t1y9tqr9 tgmBllte by a canou:i;
ical realization rule to p.oduCe *f,uf *" ni"" cilled an event structuid;l
horvever, there is more to "buildlng" tlie basic event structure of a verb;
than filling the constant position in the event structure template with .
the constant. wherr a constant is associated r'vith an event structurp,
ternplate by a canonical realization rulc, thc participants associatedil
with the constant rnust bc matched up, if possible, with appropri-,;
ate vzrriables in the event structurc templatc. \A/e assume that, api
Goklberg (1995:50) puts it, "semantically compatible" participants
paired with each other.10 For examplc, consider the manner of I
tioo ,rerb rzn, wiiich takes its namc from a constant that specifit
certain irrorrrr., of motion; tliis constant is associated witir an
iL

that involvcs a singlc participalit, a runner. As a manner

this constant is associatcd by canonical realization ntle (17) with
activity template; the single participant associated with the c
is scmantically compatible with the single variablc associated with
activity event structurc templatc, so the two can be matched up'
There is, howcver, one complication in tliis proccss' In certain il
stances the constant has morc associatcd participants than there
variables in the corresponding event structure template. In such
stances, some participants are not paired with variables in the evel
strr:cturc ternplatc. Tltis situation arises with two-argument actil

10ure d.o not spell out the details of the process that intr:grates the participarits'
associatecl rvith the constant rvit,h the variables of tlte event structure
here, but see Goldberg 1995, particularly chapter 2, for some relevant discuss
which though coucherl in terms of the constructional approach discussed in
appendix. can be translated into the perspectivc that rve are taking here

111

ity verbs, such as the vcrb sueep. we have suggested that the verb
sweep is derived from a constant associated with two participants, although the activity event structure template this constant is associated
with has only one variable. The actor participant is matched up with
the variable of the a,ctivity template, but the nonactor argument is
not matchcd up with a variable in the activity template and must be
integrated into the resulting event structure in some other way. Its

presence, we contend, is simply licensed by the constant.

As a result, we can distinguish between two types of participants in
an event structurc: those that are licensed by virtue of both the event

'

st-rqcture ternplate and the constant and those that are licensed by
virtue of the constant alone. In fact, Grimshaw (1gg3) argues convincingly for making a dichotomy among arguments along roughly
these lines based on differences in the behavior of the two types of
argumentsl she refers to them as ,,structure,, and ,,content,, argu_
mel!s-, respectively; we will refCr to them as ,,structure,, and .,coosturrt,,
iparticipa,ts:'t yn relurn to the differe,.ces in behavior in section 4.1,
atter proposing in scction 3.2 that there are distinct conclitions govern_
ling the syntactic realization of the two types of participants. In the
.eve[t structures in the remainder of the paper, constant participants
be undcrlined to distinguish them from srructure participanis.
"wjll
we
have just desciibed the construction of a basic verb meani,g.
":i
The widespread variation in verb meaning indicates that rnany,
__*_-J','^ if not
most, verbs are also associated with derived verb meanings. we attribute much of the variation in verb meaning, specifically that variation which is monotonic in nature, to Temprate Augmentation, which
allows more complex event structure templates to be built on simpler
bnesl however, Template Augmentation can only create meanings that
are consistent with the basic inventory of lexical event structure templates.

[?,)

Template Augmentation: Event structure templates may

be freely augmentcd up to other possible templates in the basic

inventory of event structure templates.

order to show how Template Augmentation accounts for tire range of

nantic and syntactic variation exhibited by different verbs, we first

need to make explicit the well-formcdness conditions governing the syrtactic realization of arguments and the predicates they are associated
with.
IWe leave it for further study to
determine whether the notion of structure argu_
meut might overlap with the notion of aspectual or event roles in the work of Ritier
and Rosen (1996, this volunre) and Tenny (1995).

,r

,fl
Iir
,I
ii
l:

112

/ Malxe R,qppaponr

ButlorNG VoRe MoA,NlNcs

Hovav eNo Boru LuvtN

113

(25) ArgumentRealizationCondition:
a.
There must be an argument XP in the syntax

3.2 well-formedness conditions on syntactic Realization

we posit two well-formedness conditions on the syntactic realization of
event structures. Their statement reflects assumptions common in the'
in the
uuu semantic representation must be synelements lrf
LlldL clsulErrlr
Iiterature
ltLeldLLrru that

b.
I

tacticallycncodedandtliatthesyntacticlcpresentationmustbefully
interpreied semantically. This idea has generally been lnstantiated
with respcct to the arguments of predicatcs, as in the Theta-criterion
of chomsky 1981, but in the subevent Iclentificzrtion condition in (2a)
we extencl this iclea to the major components of the event structure,
the subevents that constitute it.
,ll
(24)
$urlevent Identification Condition:, Each subevent in the
*"rrt .t.,r.iure must be identificd, by a lexical head (e'g', . v.l
an A, or a P) in the syntax.
In most instances this iclentification comes about as a side cffect of
association of the constant with an event structure template via o
of the canonical realization rulcs: the verb takes its name from t
Constant, and we assume the verb also idcntifies any subevents in the
event structure that results from tlie association of the constant wit\
an event structure template by the relcvant canonical realization ru
If Ternplate Augmentation contributes anotl'rer subevent to an even
structure previously created via a canonical realization rule, then thi
additional subevent must be identifred by another Iexical head in
syntax for the subevent Identification condition to be mct. As state
the Subevcnt Identification Conclition allorvs for a single verb in t
syntax to identify rnore than one subevent when a canorrical realizatioqi
rule associates tlte constarrt with a complex event structure template;
as we discuss in cletail in scction 4.1, cxternally caused change of statq
verbs, rvhich are built using canonical reallzation rule (22) are a casq
,trii
in point,.
Argument
two-part
the
Tlie second u''ell-formedcss condition,
alization Condition, involves tlle syntactic rcalization of tlie
participants in the event structure.

for each
structure participant in the event structure.
Each argument XP in the syntax must be associated
with an identified subevent in the event structure.

The first part is a reformulation of well-formedness conditions recently

proposed in the work of Grimshaw and Vikner (1993) and van Hout (in
press). Grimshaw and Vikner introduce their well-formedness condition to account for an obligatory adjunct requirement in some passives.
Van Hout appeals to such a condition to accounb for the interaction
between aspectual prefixes or particles and transitivity in Dutch and
Russian. The second part of the Argument Realization Condition is
reminiscent of a range of cond.itions in the literature that ensure that

all constituents in the syntax are properly licensed, including GB's

Theta-Criterion (Chomsky 1981, 19Ll\ and LI'G's Completeness and
Coherence Conditions (Kaplan and Bresnan 1982).
The first Argument Realization Condition makes reference only to
structure participants. Therefore, !-f an activity (sub)event includes
loth a structure participant and a constant participant, the Algument Realization Condition requires only that the former be realized
in [he syntax. Nevertheless, we suggest that there is a recoverabilr-ty'condition on constant participants, which in certain circumstances
requires their realization. This recoverability condition is not a wellformedness condition on the syntactic realization of the event structure,
since such constant participants are not essential to event structure (see
also Grimshaw 1993), rather it simply requires that the content of a

constant participant be recoverable. Their content is trivially recoverable when they are syntactically expressed, but such participants may
sometimes also be rccoverable when unexpressed from the context via
pragmatic inferences; see section 4.1 for further discussion. Thus, the
.conditions on the syntactic realization of constant and structure par-

,ticipants differ.

i'

Besides the well-formedness conditions, we assume the existence of

i"theory of linking that determines the specific syntactic expression

of the participants in tlie event structure. We assume that this theory takes the form of a set of linking rules such as those in the work
of Carter (1988), Jackendoff (1990), or Levin and Rappaport Hovav
(1995). Such rules determine the precise syntactic expression of participants based on their function in the lexical semantic representation.
Nothing, however, hingcs on our use of linking rules, and we acknowledge that the bcst forurulation of a linking theory is still a matter

tL4

MlLt r<t R,q.ppapdnr Hov.qv .q.No

Betn Levru

BurlorNc VpRs MnaNrNcs

of debate (see Levin and Rappaport Hovav to appear for discussion

Linking rules have been open to criticism since they refer to partic
positions in lexical semantic representations. Some researchers ha
tried to avoid making such explicit statements, for instance, throug
the use of thematic hierarchies (Belletti and Rizzi 1988; Grims
1990; among others) or even by taking linking to be a reflection
depth of embedding in a lexical semantic representation (Joppen
Wunderlich 1995; Wunderlich 1997; they adopt this idea from the wr
of

Bierwisch).

In the next section we show how the assumptions laid out in

section, aiong with the representations assumed for the various
of verbs, account for the differences in behavior of manner and r
verbs. To do this, we show how the multiple uses of sweep adse
how such multiple uses cannot arise for the verb breo,t.
E.

4 The Representation and Derivation of

4.L

Simple Verb Meanings

Sweep is a manner verb which takes

its name from a manner co

fl

(26) [xACTasyypnp>y)

It
E]

There are two argument expressions that can be associated

in (27).

sweep's event structure; these are given

within Role and Reference Grammar (Van Valin 1990, 1993). In the activityt

of these verbs, only the actor argument is assigned a macrorole, while the ot
nonactor argument is not; this assignment contrasts with two-argument
ment verbs, each of whose arguments is assigned a macrorole. Although the
of macrorole is very different from the notion of structure argument, like the
ofstructure argument, this notion has repercussions for syntactic realization.

12The two arguments of two-arfument activity verbs also receive distinct treat

f.

both there is a predicate in the syntax, the verb sweep, which by

ue of the canonical realization rule associating mauner constants
activity templates, identifies the sole subevent in the event struc, ensuring that the Subevent Identification Condition is satisfied.
only the actor is a structure participant, the first Argument Ren Condition, (25a), is met as long as the actor is expressed,
it is in both (27a) and (27b). 44{,lie!s[y,-Lhere;is_t}-r_.r9g*qirgo!
This requirement
-t!3t--!bq-"-o,*.stapt-pgr.!rsiPg$..bs*t-esglexuhk.
in the transitive sentence, (27a); the acceptability of
intransitive use of sweep means that the constant participant is
in (27b). We now elaborate on this condition,
i While the appropriate formulation . r the recoverability condition is

.r

with
participants: a sweeper and a surface. (We assume that the stuff t
might be on the surface is not among the minimum set of particip
since it is possible to sweep a surface without anything being on
This constant, which we represent as <SWEEP>, is associated w
an activity event structure template by canonical realization rule (1
The single variable of the activity template will be matched up w
the sweeper participant associated with the constant since they
semantically compatible. However, the constant is also associated u
a second participant, the surface, and so this participant is I
only by the constant, making it a constant participant, not a
participant.l2 Thus, sweephas the basic event structure in (26).

Phil swept the floor.

Phil swept.

Because of the nature of sweeping, this constant is associated

$

a.

b.

115

ber ofdebate, we adopt the condition proposed by Brisson (1gga):

bonstant participant can be u
uderstood as "
:a1." We ilEsTiffie this condition by showing how iTTEffit-#or
nces among verbs of surface contact with respect to unexpressed l
:cts. The intransitive use of the verb sweep is felicitous since there
prototypical surface associated with a sweeping event: a floor. The.l,
g association between sweeping and floors means that no partic- I
context needs to be specified with intransitive sweep to ensure \.
appropriate interpretation of the unexpressed participant. And in i
when used in isolation (27b) cannot describe a situation in which

Verbs

Multiple Meanings
$

t)

in the context of dish-washing. The surface cortact verb rub

cannot be used intransitively without context, and an even more
ated context appears to be necessary with it than with w,ipe, if
intransitive use is to be felicitous.
Our analysis can account for the striking contrast in the behavior
perbsof r" f .gC_qS!q$.s!9!assweepand.exgg-._l_ally-"-"*A!!g4_-vglh
gha-sce--ql,:iatesu.g! -?s bie;T; the 611di;-u"iike the former, r"q"ire
object no matter *hailllii! tontext. The account hinges on the
in the event structures associated with these two types of
: break, unlike sweep, is basically associated with a complex event
re with two subevents. First, though, we review a subdivision

116

among verbs of change of state which is relevant to determining

basic event structure.
In Levin and Rappaport Hovav 1994, 1995 we argue for two ty
of changes of state: those like blossom, grow, rot, and trlilt, which
conceptualized as internally caused, and those like break, ope\ or
which are conceptualized as externally caused.13 As their names
gest, the source of an internally caused change of state is internal,
the entity that changes state, while externally caused changes of
have a source outside the entity that changes state. The notion ofex

nal causation is reflected in the complex nature of the event structr

template that the canonical realization rules associate with cons
naming externally caused changes of state. Specifically, the external
in (26J.
(28).verb break
nas the
tne event structure ln
state verD
1reat has
eqused cnange
change ol
of staLe
reEused
/
,l

Burr,orNc VpRe MoA,NINGS / 117

MALKA Rapp.a,pbrur Hov.qv nNo BBIH LsvtNI

(za) [ [x.ACT.y;Nff;,

and the number of variables associated with the event structemplate; however, a comparable discrepancy does not arise for
ally caused verbs of change of state. TbS"Iglg_ylq!"-9o-Sp-!-a"gjs
asqo_cg1!-ellyith tw.o.par.:-reip^s_qtF"i-,tlr"q. _q-aus,.er aud the ,9p!tly--.tbat

state-what is typically refe11-qd !-o- 3s the patienl. These conare piired with an event structure template with two variables'
in each subevent, and each of these two variables is semantically
ble with, and thus matched up with, one of the two constantiated participants. Thus, there are no participants in the event
bure whose presence is purely sanctioued by the constant; that is,
participan[s are stryct_urp -par!i-crp3nt1.
'Ith-lals-baCkfir?r"rind, we now look at the application of the Ar-

causo IBECoMtr ly <BRIKEN>)ll

This event structure has two subevents: a causing activity
J

change of state. This analysis is argued for by many, including

ll
ii

(1979), Levin and Rappaport Hovav (1995), Parsons (1990), Pur

jovsky (1991, 1995). In the case of break, the constant lexically
ifies the change of state involved, but provides no information
the causing subevent beyond its existence. It is for this reason
hre-ak is a result verb and not a manner verb. While the nature
the causing subevent is not specified by the constant, the event str'

ture associated with breo,k must nonetheless include a represen

of this causing subevent since the change of state break denotes
colcep,lg{-ized-as}eyllg 3q.gl9tgl_ggs"9_-g_tAgg_-loj.gS"-ci+r-rjrr"c-spn
the constant that gives its name to the verb breakl,,
'taneously. Since
y with a complex event structure template by the relevd
canonicai realization rule, the verb break identifies both subevents
this complex event structure, ensuring that the Subevent Identificati
Condition is met.la
An important part of our analysis of verbs of surface contact
the difference between the number of participants associated with
t3In Levin and Rappaport Hovav 1995 we support the distinction between the
kinds of changes of state and the difference in representation which we attr
to them with datp concerning the morphological encoding of the relationship
tween the members of causative-noncausative pairs of both types of verbs c
linguistically. We discuss the event structure template associated with i
laThis description of the lexical semantic representation of externally caused
of change of state brings out an asymmetry in possible verb meanings: there
result verbs which imply the existence of a causing activity which is left u
ified, but there are no manner verbs implying an unspecified result. We leave
exnlanation of this asvmr-netrv to ftttttrr: research.

I
I
I

,*
rt

Realization Condition to the event structure of an externally

change of state verb. This con-.i--n requires the obligatory exof both the subject and the direct object of such verbs, a prop
of these verbs noted in section 1, since each realizes a structure
ipant, one associated with each subevent of the complex event
of these verbs. Our analysis of this property adopts the ceninsight of the analyses proposed by Brissou (1994) and Grimshaw

'L*-*-*'

)i

,l

\t
"l
tl

ii

it,

Iike these two earlier aualyses, also ties the obliga),

iness of the verb's arguments to the licensing of its complex event
ucture. Q1 9-u1 account, lhe cause! arggment realizgs thg. !-IT._!"*1J
:ticipant of []ie first.ggbqvC{Ij and the pqtjeq.I-arg-qm;nt ie*q-1-!7ej"iffi
:uct ure p articip ant of the g.e-9914-. q." b,eye$!:.-.".
'Thus, botli ai6'obligatorily expressed. In fact, Brisson (1994) notes
no matter what the context, the direct object of break cannot be
since

)a.
'b,

ti
i!
$
!

ri
I
ril

, as shown in (29).
Tlacy broke the dishes.

*tacy

broke.

to the
in
their
difference
riuess of the direct object arises from the
t structures.
-The
bdii-ffiown property of externally caused verbs of change of
d a property that sets them apart from internally caused
of change of st ate-is th9i1. nqrficip+tias. ill ..!!.e_."9qu-.s-g, iyq-.pl_
ion, illustrated in (30); that is, they allorv transiiive causative
intransitive noncausative uses.
, the sharp contrast between break and sweepwith respect

30)

a.

b.

Tlacy broke the dishes.

The dishes broke.

Levin and Rappaport Hovav 1995 we argue that even in the nonwith the same complex event

carrsative rrse srrch vcrhs are a,ssociated

ii
ii

118

/ Me.lxe

structure that we associate with the causative use. That is, D

example, would have the event structure in (28) whether it is t
sitive or intransitive. We also present several types of evidence.l
this analysis, which we do not repeat here. The question is how i
weli-formedness conditions on syntactic realization are met with
causative break. We assume that the Subevent Identification Condi
is met in the intransitive use in the same way it is met iu the

tive use, since the constant is still the same, and the same
realization rule associates this constant with an event structure
plate. But the intransitive use of brealc appears to be problematic
the perspective of the Argument Realization Condition. Only a sin
argument is expressed, and yet the verb is associated with an
structure with two subevents, so that Argument Realization Condi
(25a) appears not to be met.
In ggst l_q.ngg,_1g:*h:.g-l*:."ji"_ge-l&r+
t5rg .Ls-.ild iraled _by, :g1nq,typ eTfHorpirbto gicat markin g, and in ma:
'5I1,hese
languages, the lniransiii.,e usg is moiph-oiogicaffy qrore.-cgj
pt.q" ghu{,

BurloINc Vpne MeaxrNcs

Rene6eoRr Hovev AND BETH LEVrN

ift. qia;sde

_ule (see

uaspLrmitii'hOE,"UaJ;u;-i

Often the intransitive use involves a reflexive morpheme, as in

Romance and Slav-ic- l"inguages. Wq assrrme tfiaj-ilft_iii6*;;;pf

jfi :
yhichse1,vres.Iq.S9.[:$^-l_4!.A_!g!i;nentReatizatioii"'bonditic
on t his account, ii ;#;i al i; ridT iifiinffi riAliE'ffiffiteme.,e u

o f int ransi iy-els-dtte.rent

d,1$_q1e nt trom
fro m that
t h at of
o f intri
i nt rans i t ive
!
lfgg.ft,*_t--h.Sll, "1p
9_j.I9..fr_"!--a.9ll,
eitd-in-f;ai; iJ rai u' *".
t
gbisgt*r+i."
ats ti"si
xe, \ *gy:-g"lgplis$
"oli]il;friEseaTbfi;=i"t;
aoiltipiiiliv m;ffi16ft;ffii$-il;iLdffi;'o#iilguffiC;iili;
qicallv
marked cioss-linsuisticallv in accrrsp
guages';'- nevefta6t-6$,'t'liGE-an
Ian guages';nevef'ra6t-6sd:'thfi -is
;iea ilq"i.i"g
fr-6rt
6{h..
h"ru,
st udy.

;;

119

, and the NP the floor is an argument of this subevent, so that

Argument Realization Condition is also met.
r)

srArg> )l)

)
)

event structure in (32) is that of an accomplishment, in contrast

the basic event structure, (31), which is that of an activity. T].tS.-g-ig
LcS."_t^bp-t-ggn!_elr'a-eS-,ryith

the*ttryg*uFsq -o-!--s_-tu-g9?-.dp-.reeer-vc -d-if-er--e3.I

bual classifiq.-di-g,p1**.3or exampie, (33), like accomplishments in

rtral, is incompatible with durative time adverbials, while (27a),
activities in general, is compatible with them.

Phil swept the floor for an hour.

*Phil
swept the floor clean for an hour.
b.
two uses of sweep show different interpretations with almost that
consistent with their distinct aspectual classifications. As illusin (35), the accomplishment use of sweep is ambiguous with
as expected of an accomplishment (Dowty 1979), while the acvity use is not. Sentence (35a) can mean either that Phil was going
begin sweeping but did not, or that Phil began srveeping, but he
ped before he swept the whole floor clean; (35b) can only mean
Phil was going to begin the activity, but it cannot mean that he

)a.

"tr;;;;;ffi

4.2 The Effects of Template Augmentation

l
I

tl
i!
,i

ii
t!
:i
L:

The multiple meanings of a verb like sweep can be derived via Temp
Augmentation on the basic event structure template associateid
this verb, given in (26) and repeated in (31). As an activity tem
this basic template is a subpart of several of the other possible
structure templates listed in section 3.1; thus, this template can bd
augmented to give these other templates as long as the resulting coml
plex event structure meets the well-formedness conditions on syntactil
l-i:lll
o-^^:c--rr
-^^l:-^r:^realization.
Specifically, their satisfaction will require a predicate in
syntax that can identify the second subevent introduced via
Augmentation, as well as an argument in the syntax that is
with this subevent. For example, one potential event structure t
could be derived by applying Template Augmentation to (31) is (32)p
This event structure is associated with the resultative construction as
in (33), where the head of the resultative phrase identifies the second
subevent, ensuring the satisfaction of the subevent Identification Coni

nt is identified by the pl;1pSl|;y5*.plSU-tr The causing subevent is

witn tfre sub;ect argument and the change of state subevent

36)

*Phil swept clean.

As Brisson (1994) points out, the simple transitive use of verbs like sueep can
ippear with telic time adverbials as in Phil swept the fi,oor in ten minuteg However,
we agree with her in assuming that this, compatibility does not necessarily imply

ffi

120

BurlurNc Vpns MplNtNcs

/ Melxa ReppepoRr: Hbv,cv eNo BprH LovrN

There are several other possible meanings of sweep that can arise via

5i"a *ittr sentences like those in (38).

(37) [ [x ACTasyz6 np>.Y ] CAUSE Inncortn lz <PLACE> l)

(38) a. Phil sweptSe crumbs onto the floor'l
b. Phil swept[he crumbs off the tableJ
In order to understand how the syntax of this pair of sentences is assg
ciated with the particular event structure we attribute to them, recall
that we assume that the constant associated wit]n sweep involves only
two participants: the sweep*g1agfuhe syrface'Jhese. ty3 participa
are present in the Easic event structure for sweep, and this event str
ture is the causing subevent in the more complex event struct*u uf{ii
associated with this ve.b. Ig.(3-B),rrulile (sql ilg q+".t gbj.* it**
o-!.9. _qf

.the-

ar.

gu mrJ', Lq*aI-thp - 9 qusi

+S FS-1Ye

ond ar gu
.
be expressed since it

n!.: T he

sec

oi lfr. .uuti.rg subevent, the suifiie,

"".a "ot
constant participant. The second subevent, which is not identified
the verb iiself, ii identified by the prepositiqt onf,-tg"!k-b)-19$9li
', and of,in $g(hpUsqse rth6inioie, tlieie is an argument XP,
I tn" rrlil" *.;;irt;d with this subevent, the crumbs, as required I
\t ;;" X;-"*"* ,i""itr"ion condition. Interestingiv, in (38b) the su
\\fu." u*r*ent is expressed within the XP headed by the prepositil
ia."tifies the second subevent, so that the constant' participant,

\\tf,ut

| !ecoverable.

r.ii

Contrasting with the acceptability of (38) is the ungrammatic

of (39). This sentence is ungrammatical for the same reason that (
above is, namely, that there is no argument XP in the syntax
the second structure participant, so that the Argument
Condition is violated.

121

it

there is
is a participant in the second su
Alternatively,
second
tifv this
simply with an activity event structure,
we associaLe the verb in
as the stuff removed, would not be
if
it
is
interpreted
the crumbs,
in this event structure, Ieading to
single
subevent
ated with the
Condition (25b). In fact, the
Realization
Argument
violation of the
to associate with (40) is the
able
are
that
speakers
interpretation
jnsensical one where the crumbs are the surface that is being swept. lt
reading is precisely the expected one if the sentence is associated
the activity event structure.lo
\Me see, then, that activity verbs, typified by the verb stlreep, have
-stants which are modifiers in an activity event structure, which can

il

il

f -"er!l,tlrl!*,t,ss-ager,*s"F-pEer3:i-si5rirJ9m".re3"Ditrer=r nnnm nl qh m ort

- o, i. o*,.i se .1..,e" rtii, s-'ffi*lfiE*5l6fr&t which
i

nt. Each of these meanings

class of verbs. The second
grammaticaily-relevant
a
with
ffiociat"a
nt can no! -only {"".g!gg:kg"_oJ tlt"_9l*b*+Se-sf lg.-Uql,
eafiffi"Aenote ailffi64 tffit;;*# into existence via the
tr;et'ff;-f**;'irio o p'ile, in which a pitej-i6'ffiAs
result
of the sweeping. English, but perhaps not
as
a
existence
the
same
allows
languages,
"name" (derived from the constant) to
(Rappaport Hovav and Levin in
templates
all
such
with
,bssociated
from
the activity of the causing
is
derived
name
Because
the
).
t, the direct object is omissible when it is recoverable, and the
tic direct object of the verb in the accomplishment uses of the
account also explains a constrast noted in Levin and Rappaport Hovav
1). There we pointed out that when verbs l\ke sweep are used as verbs of
rval, they cannot express their arguments in exactly the same way as verbs

-H:*":".t1,!:
:lg:gptabte """:-"S.:rgp, Bigrc:lq@*Y9l!,Jsrg-19!a state conbtant ln rts oasiE evenT-dtruEtflfr-so that crannbs h?s no subevent
ugmentat
ing Phil swept the fl'oor clean of crumbs, where the adjective cleon. identifies
added change of state subevent'
dilected motion use of verbs of manner o,[.n-nSl|Sn'39i\t Pat ran to fu-hsa.ch'

an-Ecfinfffo-aAfiripiGEffi"t-"t-irt;
acllvrty-f,o-accorrrPrrslllrlellu Dllll!i
v6lies
ves an

;;;;i ;61;fi i;ffi T"i1;646coE;fffi ilC

i

ciated with the same type of accomplishment event structure as (

or (38b). If so, its ungrammaticality follows because the crumbsis
r participant in the activitv srrhevent in the event stnlctrrre, (37),

as clear or clean. Clear and clean enter into the construction illustrated in I
*Ph!*sw.e2\
ed ths.tshlp-gfulisbS,s. but verbs like sweep do not:
1trg.flooL ,
. The reason for the contrast is that in the acceptable sentence dishes is I
ant associated with the state constant <CLEAR>, which appears in the I
cfrrre of
nf .a
a. verb
verh like clear-.slnce
cl.eo,r since it-_E:e!.-g2siernallyJAugegghglg="-9f-s!e!S
it is an externallv caused change of state I
strueture

h!y9!1,,1ftr SEtgiJ*,
qh thev also consist

account we present in this sEtion olthe l

contact may not extend to verbs of manner i
|We leave the exact representation and derivation of the accomplishment i

arla;Esulf-f-so, ihe
ffi uses of verbs of surface

122

Burlpu.tc Vnns MrnNrxcs

M,lr-xe R,teRaeopr Hov.q,v auo BBru LBvtN

123

rh-e-res-ultaliveph1q,,s9-${hgrspeci-$gs- j}9-g[a1ee*at"lre'3d'y
t x o ud'aitio"uf t"."tt i"g st ge- n*:.uS:-" 39i.*

'iln"

examples in section 1.
We now show how the rigidity of externally caused change of
verbs with respect to argument expression arises from the'inte

ofTemp}ateAugmentationandthecomplexeventstructuretempla
a1
associated with externally caused state constants' The natient
particip
sole
is
the
break
as
such
verb
ment of a change of state
the second subevent in its event structure, and, as such, it is a str
participant and must be realized by an XP in syntaxl the linking
d.etermine that the patient is expressed as a direct object' fhiff-rcI

*Kelly broke the dishes off the table'

Kelly wiped the crumbs off the table'
b.
Furthermore, because the template associated with a verb like br
cannot be augmented further, no other achieved state or locatio
o
be added to a sentence with break, evert with the normal direct
(42)
alq
in
(a)
sentences
the
as
such
no
sentences
Thus, there are
with interpretations parallel to the (b) sentences' That is, the
(a) sentence cannot be used to express 'Kelly broke the dishes anr
a ,errlt they went off the tabie', and the second cannot l" :tfif
value
express 'Kel1y broke the dishes and as a result they were

(41)

";;b.i
Kelly broke the dishes to Pieces'
caused
nother possible way to vary the meaning of an externally
causing
the
vary
le of siate verb such as break would be to
,t, 5rrst as the meaning of a verb of surface contact can be
of
by varying the nature of tn" resulting state in the change
lr,rU"rr"i. in ord", to obtain such meanings there would have
;a1e*icul way of adding information about the causing subevent'
ao.t oot ,""* to be one. In fact, we know of no verb that
tn"."
tf"ftrpr" meanings that are differentiated
I
lilh':a sharp
11Y"::.?::]:
,break yith
instrument'
ffi-o.o;",6-irrui "u" mean both
'.exicalized in the verb, the
i1;; with the hand,. untess it is
lit*r, ,f tr""ifying this information in English is periphrastically'
-instance,
through the use of. a by clause'
lil;;, th" p.opJrti". that distinguish the verb break from the verb
of their event struceD canbe accounted for through the interaction
and the
Augmentation'
!'i.p*r".rtutions, the operation of Template

a,

*KeIi1- broke the dlshes ofi the table,

-i::L:

-.f -\ --.J -r:t>- -:s teli

--

:,;==::

lrOther Consequences
yements show the same rigidity in behavior as accomplishments'

with a
iccomplishments, achievements are lexically associated
I resulting state. The state associated with the achievement
iw is a particulil (deictically detarnirued) location, whtle t'be

)iffi.e.-w ry

tr-fidwtr Ey6

riq@ffi

6 g.

I Mrr.r<t Repplpdn'r Hovev aNo Bors LPvIN

(45) [BECOME lx <STATE> ]l
It might appear that Template Augmentation
I24

Burr,orr.rc VpRn MpaNrNc

could be appli

the basic event structure template of achievements, adding a

ing subevent and giving rise to an accomplishment event
template. However, this possibility does not seem to be
and it is probably ruled out because there is no way to identify,l
added subevent. An achievement verb does not itself identify the
ing subevent since it is not conceptualized as externally caused;
thermore, there appears to be no way to properly identify a
subevent via the addition of another predicate. Because of this,
which are inherently classified as achievements do not have
<

variants:

ii (46)
ti

a.

[ BECoME Ix ACTayarrra, ] l
Finally, we illustrate how the theory presented here accounts in a
al way for an interesting difference between verbs which denote
d-slas"c9s"p.l. gfiple and
9l$glS-Ar mentioned in section 4.1, verbs hke breff,
gh they lexicalize only a resulting state, involve states that are
ualized as being externally caused; thus, they are basically as_
. with a complex event structure template which
involves both a
subevent and a change of state subevent. "ihese verbs contrast

7)

The train arrived.

xThe conductor arrived the train.
The letter came.
*The mailman came the letter.

When achievement verbs do have causative counterparts, -the

name, as in the pair come and bring. We p
verb has a different''*-*%..*+
the causative verb denot?ls -dif-EFt-Frnally
is
because
that this
verb bring, for example, Iexically specifies d
location;
the
of
change
subevent (i.e., that the agent accomtr
the
causing
thing about
location), as well as the resulting loi
of
in
its
change
the theme
imaginable application of Tem
another
Furthermore,
subevent.
for basic achievement verbsJ
option
not
an
is
also
Augmentation
state
specified in the event str
achieved
is
an
there
already
cause
states or locations
achieved
verb,
no
other
achievement
of an
just
like
break.
verbs
accomplishment
as
with
added,
Achievements cannot assume activity readings for the same
that accomplishments cannot: this would involve the eliminati
resulting state which is an inherent part of the meaning associ
achievement verbs. Activities, on the other hand, cannot be
with achievement readings since there is no way to composi
derive an achievement from an activity, there being no event str
templates of the form in (47).1e
r;l
lgThis is not to say that there are no sentences describing the inch
activity; there are after all sentences such as Pat began to walki however,
knowledge there are no simple verbs which lexicalize such an event type, a pi

made bv Dowty (1979:124). Presumably, there is no such event type in

inventory of event types. It is also true that achievements can sometimes
certain properties of activities, such as when the vetb notice is used durativt
assume,-however, that such shifts are not structurally represented in changtie
event strrrctr:re template. We intend to elaborate on this point in future

"gJ.@

i*sl""q"llv-g*sssd*
}jS"t"?llts.*sssd*bapse*sls-t..atp*.yer!:Ly-.ichn_qiies{ffi ilJE;
about naturally in an entitv. The*. *i,il,J;;;;;*
_9"h"-g*!..1e!,*lplly_",+."g+*e+tj!X.Thesesiites*a*c""ffiffi
a"
ving their source internal to the entity that changes siate, and
tly the constants naming them cannot be associated with
plex event structure template that involves a causing subevent.
, Levin and Rappaport Hovav (lggb) note that these verbs
,be_in_state,
a systematic ambiguity:
reading and
{loyl"S both a
l"rsad:qs. rhififfireffie Eet"rieen t[;-tn;o
the fw6*i6Effiffiilis
r
in (48).
a.

The amaryllis blossomed for ten days.

b.

The tree blossomed in a day.

Itehc nature of (aSa) is highlighted by the durative time adveriAu interpretation involving iteration of telic events is excluded
since an amaryllis has only one flower. In contrast, (4gb) has a
interpretation, as indicated by its compatibility with it ,r, ti*"

bial [lrF

gg!i$g$1lP*g".9=!-9-i!!-elt^tjy--abg9-gtjlg*n.externally " caused

";6;-il-rt;t;

;;;i6;-il-'t#a"re'ea:

: verbi like . break never have- a

p qentence like The uase broke can nevei^.nie"aii-tliat-iE6 vase w*
rtate of being broken.
gp,$ifference in the behavior of the two types of change of state
can be accounted for if we assume that internally causeJ change of
Iike blossom are basically associated with a state template,
-g{ sta!_e verbs

lx

<IN-BLOSSOM> l
we assume the basic analysis of such .rerbs is as internally
state verbs: they take their name from an internallv caused
r.,constant, which is associated with a state template by canonical
ion rule, (21). The internally caused change of state interpre_
then, is the result of Template Augmentation, which derives an
rent template from the state template through the addition of
icate BECOME, as

in

(50).

f,fl

ilfl

ir
r,

I
I

k
E
$
h

li
$

E-

&

&

B
$
E

I
I

126

/ Melx.r, R.qppap6nr Hovlv nxo BsrH

(50)

[ BECoME

lx

<IN-BLOSSOM>

Burr-oruG VpRs MoINTNGS

LpvtN

my

Since no additional subevent is introduced, no additional

XP need be added to identify the added predicate and no additio

predicate is needed to meet the Subevent Identification Condition.
fact, it is a widespread phenomenon tha-t-.slgtj
"iiffirmrat
isilflftfiK"p]ienomenon is particularly
ffi;;;;i"A'frilf;effi of perception and cognition such as see,
ard understond (Dowty 1979), but it extends into other verb
This ambiguity is not available to externally caused change of
verbs since they, by hypothesis, are basically associated with an
plishment representation, and indeed it is not attested with exter
caused change of state verbs.

phenomenon of verbal polysemy treated

The research reported in this paper is clearly just a start. For exampl
not all the diathesis alternations listed in Levin (1993) necessarily lel
themselves to the analysis given above. The case study presentgd
a first attempt at making explicit how verb meanings are structurt
how these meanings are expressed in syntax, and how verbs assume
tended meanings. In this regard it is an attempt at a generative
of verb meaning. Although a first glance at the range of meaninl
verb can assume gives the impression that verb meanings can
aimost unlimited ways, a closer look at the attested variations
that this is not so. As in the case of syntax, much is to be learned'fi
the constraints on these phenomena. But these constraints can only
uncovered once the descriptive work of detailing the kinds of attest
alternations is done. We hope that our work, built on the pioneerj
work of others, will provide a framework for doing just this'
more, as in syntax, our understanding of the phenlm"oo, ,rtd", Ji'
can be greatly enriched by cross-linguistic comparison. It is well-kr
that verbs in English show a greater elasticity in meaning than
verbs
inotherlanguages.4.o;--"-*"u"qple,,-!"h-""3?**gl3j.
do
so
in
cannot
rases,
di
ish can appear with
1

in this paper has been

ussed in the literature in the context of the problems which this

non poses for the model of the syntax-semantics interface deibed in section 3.1, which we have elsewhere called the "projectionist
(Levin and Rappaport Hovav to appear). As already men, this model assumes that verbs have strrctured lexical semantic
tations from which syntactic structu'es are projected (ChomfifOaf, 1986; Levin and Rappaport Hovav 1995; Pinker, 1989).20 In
i,!

Conclusion

oTfiers). Presumably, the comparison of patterns attested in d

languages will help lead to the development of more principled
predictive accounts of lexical phenomena.

127

ndix: The Constructional Approach to Verbal Pol-

l)

Ianguages (SIoEi-n 1987, 1996; Talmy 1975, 1985;

model, verbs with multiple meanings must have multiple lexical

tic representations, one for each meaning. These meaniugs, in
determine the various syntactic structures that the verb can be
in. Given the pervasiveness of verbal polysemy, the projectionist
results in a significant proliferation of lexical representations
with individual verbs, as mentioned in the introduction.
Many researchers have looked for an alternative approach that obthe need for multiple lexical representations; these researchers
, Borer 1994, in press; Erteschik-Shir and Rapoport 1995; Ghomeand Massam 1995; Goldberg 1995; Hoekstra 1992; Hoekstra and
1990; Ritter and Rosen 1996, this volumel among others) sugq that the phenomenon of verbal polysemy points instead to a model
syntax-semantics interface which denies that verbs ha're struclexical semantic representations from which syntactic structures
projected. Elsewhere we have termed this alternative approach
('constructional" approach.2l Proponents of the constructional ap-

Projection Principle conflates two separate claims. One involves the idea
verbs have structured lexical entries which register the number and the types of
arguments. The second is that these properties are configurationally encoded
levels of syntactic representation. LFG's Lexical Mapping Theory (Bresnan
iKanerva 1989) is aiso projectionist in the sense that the number and type of
is determined by the lexical entry of a verbl however, it denies the second
of the Projection Principle, namely, that these properties are configurationaiiy
at all levels of syntactic representation.
al researchers have adopted constructional approaches independently in reryears, most notably Goldberg (1995), Hoekstra (1992), and Borer (1994, in
). There are significant differences in the various instantiations of this aph, reflecting differences in the proponents' theoretical frameworks and goals.
gxample, Borer and Hoekstra assume more articulated syntactic representations
Goldberg. Goldberg clearly delineates the aspects of meaning which reside in
verb and those that are associated with each construction; she also investigates
tional polysemy, paying special attention to extended meanings which are
across constructions. Since our purpose is to contrast the constructional ap.
with the projectionist approach, we focus on the fundamental assumptions
fihe constnrctional approach, which the various instantiations share. However,

flil

iil

ifi
[ifl

fiil

;il

ifl
rffi

{[
r{

f;fi
)il
r{I

;ll
'iI
i$
fiil

,f,W

;fr

128

r,
B

I
k

t
$

I
I

fr

t
F
F

I
b

tI
I
L
D

$
B

$
F

6.

&

l1

i;

/ Malrce RappapoRr Hovav

ButloINc VPRn MrA,NtNcs

aNo BsrH LnvrN

proach claim that only some aspects of meaning reside in verbs

selves. Certain syntactic constructions or syntactic positions
associated with elements of meaning. One common proposal (H
1992; trrteschik-Shir and Rapoport 1995) is that different cons
correspond to different event types in the Vendler-Dowty sense.
Iated proposal is presented by Borer (199a): certain syntactic
assign the NPs occupying them a salient role in a particular event
On the constructional approach, verbs name basic concepts, whi
freely inserted into syntactic structures, whose meanings are then
bined with the core meanings of the verbs. The meaning of a ver\
given use is determined coripositionally from the meaning of the,:
and the meaning of thd constructiorl. The major constraint on
grating a verb into a construction is the "compatibility"
meaning of the verb and the meaning associated with the
structures into which the verb is inserted.
The constructional approach assumes that verbs do not haye
tiple meanings; rather, a given verb names a concept which i
those aspects of meaning which are common to the verb in all its
Many verbs name concepcs which are compatible with the mea^nir
of more than one syntactic construction, and these verbs can be
in more than one syntactic context. The meaning of the verb,
appears io vary according to the meaning of the various constru
it is found in. trYom the perspective ofthe constructional approac\,
terms "verb meaning" and "verbal polysemy" are actually misnqn
in the sense we have been using them in this paper.
To iilustrate this point, we draw on the work of Hoekstra (1992) j
his approach, the structure below in (51a) is associated with an acti'
interpretation, while that in (51b) is necessarily associated with
accomplishment interpretation, where PRED can represent eithgr
achieved state or an achieved location. The concept named by
verb sweep is compatible with both an activity meaning and a vari
of accomplishment meanings. Therefore, as shown in (1), a verb
sweep is compatible with either activity or accomplishment coutt
and may be inserted into either syntactic structure. Thus, there isi
need to give sweep distinct lexical entries. The distinct meanings
li
from the distinct syntactic representations the verb can be i
i

&

$
H.

ri
$

we want to point out that there is no reason to assume that all alternations in
gument expressions have a single source. In fact, Jackendoff (1990), among

attributes some alternations in argument expression to the existence of mult

Iexical entries for certain verbs (mediated by some kind of lexical rule) and oth
to the interaction between a verb's lexical entry and certain constructions, wh

themselves have associated meanings.

129

In fact, on Hoekstra's approach, the lexical entry for the verb

does not ind.icate the number of associated arguments; this is
ned by the structure the verb is inserted into'
a.

INPV]

[s" NP PRED ] I (Hoekstra 1992)

i'is not our intention in this appendix to choose between the con:tional and projectionist approaches to the relation between the
,al representation of verbs and the syntac'ic expression of their
. Our main purpose is to point o .t that there are many
shared by both approaches, that the challenges facing both
the same, and that the initial suggestions we have made in this
li coocurrrirrg how to meet these challenges can be integrated into
aches of both sorts. Therefore, although proponents of the conional approach have presented their approach as a significant
lrture from existing theories, we believe that the choice between
two approaches is not clear at this point, though we believe that a
between them will ultimately be possible.
As already mentioned in section 3.1, most current lexical semantic
rries recognize a dichotomy iu verb meaning and distinguish what
Larr" ul."udy referred to as the structural and the idiosyncratic
bs of verb meaning. We have shown ttrat inilg-q-ry;9:t]915!3p;
encode{ in some
[-1,-he st-ruet-ur.+-1..a_qp_ect gf pean-ing--.ip--gl"-qJly
b.

INe

of skeletal event struCiure-i6liresentation, whil-e. the idiosy-nctAiic

ent of meanirrg i, ,"pr"tented by the c6nstant' The constructional

...og.rirL the same diitinci'ionl alt[ough it is represented

ntly. In the constructional approach, the"idi"osJ.-!.9.1q.t&-9-q"lgP..9=

of meaning i!,se!! -coptilutes the le-xical representatigl of th,g y.e1,b,

iL ,t struciural aspects of meaning do not reside in the lexical en"
of individual -ve.r..!s, but rather are aJs-ifiateQ wit[ cer-tpln b.,.5;c
qlfUgtqtps, ihose which are associatea wiifr skeletal eveni in-

dince both projectionist and constructional approaches

this basic distinction among the elements of verb rneaning,
[e iraior difference between the two concerns whether the association
,.t*."o the constant and the template is registered in the Iexicon or

Both approaches to verb meanings are clearly capable of handling

:s withmultiple meanings, but, as we have shown throughout this
per, the major challenge facing any effort to handle verbal polysemy
ihe delineation of the exact range of meanings available for any
ividual verb. Most current approaches to verb meanings have yet
take up this challenge. The constructional approach, as mentioned,

130

/ M.tlx,t Rappapoh:r Hovlv

Reronorvcos

eNo Bpru LpvrN

appeals to the compatibility between the verb and the construction, b

this compatibility has not been explicated by most of its proponents.
Once the details of compatibility have been worked out, the t
of verb meaning that emerges might well resemble the one we have
gun to articulate in this paper. In fact, although the theory
in this paper has been cast in projectionist terms, the components
the theory can easily be incorporated into a nonprojectionist
work. Template Augmentation, the canonical realization rules,
the well-formedness conditions on argument realization,
the Subevent Identification Condition, can be seen as initial stepsl
ward explicating what it means for constants to be compatible
il
the syntactic structures they appear in. That is, the major
nents of our theory can be adopted by a constructional approach ifi
is assumed that the event structure templates are not registered in lei
ical entries and the association between the constant and the temp
i

is not lexical, but rather is checked interpretively using appropri

reformulations of the rules and conditions we have articulated.23
22Goldberg's research (1995) is clearly in the spirit of the program we ad
in this paper. She presents a well-articulated theory of the representation of
meaning and the ways in which verb meaning and constructional meanings'
integrated, and we are in fact indebted to her work for an understanding of
of the issues involved in variation in verb meaning. It is not clear that Gold
theory as it stands accounts for the contrasts which we elaborate on in this
although there is nothing in principle which prevents these insights from bei
incorporated into it. We leave a fuller comparison of Goldberg's approach and i
for future work23Some proponents of the constructional approach have claimed that the
straints on the syntactic structures which verbs can appear in can be reducedr
well-known syntactic constraints (see, for example, Hale and Keyser 1993; HoeJ
stra 1992), although this claim is not necessary to the constructional approach siil
there are versions of this approach, such as Goldberg's (1995), which Jo not requi
the constraints to be syntactic. If it can be shown to be true in general
constraints are syntactic, then there would be strong support for the par
syntactic version of the constructional approach. As should be clear, we hav6'b)
fered a nonsyntactic explanation for the constraints on verb polysemy which w
have uncovered, but we have not yet dealt with the constraints discussed by
and Keyser. See Kiparsky in press for a critique of the syntactic explanation of
constraints which Hale and Keyser describe and see Rappaport Hovav and
1996 for a discussion of certain problems with the syntactic explanation Hoek
offers for constraints of the sort which we discuss in this paper.
i

131

rences
Jurij D. 1973. Regular Polysemy. Linguistics

142:5-32,

A. 1990. The Development of Verb Concepts: Children's

Use of Verbs to Label Familiar and Novel Events. Child Deuelopment
, Douglas

61:681-696.

i, Adriana, and Luigi Rizzi. 1988. Psych-Verbs and O-Theory. Natural

"Language and, Linguistic Theory 6:291-352.
, Hagit. In press. Passive without Theta Gricis. ln Morphological Interfaces, ed. Steve Lapointe. Stanford, California: CSLI Publications.
, Hagit. 1994. The Projection of Arguments.In Functional Projections,
Un'iuersity of Massachusetts Occasional Papers -17, ed. Elena Benedicto
and Jeff Runner. Amherst, Massachusetts: GLSA, University of Massachusetts.
, Joan, and Jonni

M. Kanerva. 1989. Locative Inversion in Chichefra:

in Grammar. Linguistic Inquiry 20(1):1-

Case Study of Factorization

50.

isson, Christine. 1994. The Licensing of Unexpressed Objects in English

'Verbs. In Papers from the 30th Regional Meeting of the Chicago Linguis':'tic Societg (CLS), Volume 1: The Main Session, 90-102.

, Richard J. 1988. On Linking:

rff
'rffi

,fl

Papers by Richard Carter. (Lexi-

con Project Working Papers 25). Cambridge, Massachusetts: Center for

Cognitive Science, MIT

Noam. 1981. Lectures on Gouernnlent and Bi,nding. Dordrecht:
Foris.

ky, Noam. t986. Knowledge oJ Language: Its Nature, Origin and Use.
New York: Praeger.
ty, David. 1979. Word Meani,ng a,nd Montague Grarnm,ar. Dordrecht:

y, David. 1991. Thematic Proto-roles and Argument Selection.

'm
.Lan-

ru

guage 67(3):547 6L9.

ik-Shir, Nomi, and Tova R. Rapoport. 1995. A Theory of Verbal

'Projection. Ms., Ben-Gurion University of the Negev, Beershera, Israel.
ner, Dedre. 1978. On Relational Meaning: The Acquisition of Verb
Meaning. Child, D eaelopment 49:988-998,
i, Jila, and Diane Massam. 1995. Lexical/Syntactic Relations without Projectio n. Lingui,stic Analy sis 24:.tT 5-2L7 .
berg, Adele E. 1995. Constructions: A Construction Gramrnar Approach to Argument Structure. Chicago: University of Chicago Press.
mshaw, Jane. 1981. Form, F\rnction and the Language Acquisition Device. In Th,e Logi,cal Problem of Language Acquis'ition, ed. C.L. Baker
and John .I. McCarthy. 165-182. Cambridge, Massachusetts: The MIT

ffi

fl
ffi

Press.

mshaw, Jane. 1990. Argument Stracture. Cambridge, Massachusetts:

MIT Press.

The
il

i
tf,

ffi
132

/ RnranoNcns

RoreneNcos

Grimshaw, Jane. 1993. Semantic Structure and Semantic Content.

Rutgers University, New Brunswick, New Jersey.
Grimshaw, Jane, and Sten Vikner. 1993. Obligatory Adjuncts and the
ture of Events. In Knowledge and Language II: Ler,ical and,
Stntcture, ed. Eric Reuland and Werner Abraham. 143-155.
Kluwer Academic Publishers.
Gropen, Jess, Steven Pinker, Michelle Hollander, and Richard Gold
1991. Affectedness and Direct Obiects: The Role of Lexical
in the Acquisition of Verb Argument Structure. Cognition 4L:153Hale, Kenneth, and Samuel J. Keyser. 1993. On Argument Structure and
Lexical Expression of Syntactic Relations. In The View from Building i
Essays in Honor of Sgluain Bromberger, ed. Kenneth Hale and Samuel
Keyser. 53-109. Cambridge, Massachusetts: The MIT Press.
Haspelmath, Martin. 1993. More on the Typology of Inchoative/
Verb Alternations. In Causatiues and Transi,tiuity, ed. Bernard
and Maria Polinsky. 87-120. Amsterdam: John Benjamins.
Hoekstra, Teun. 1992. Aspect and Theta-Theory. In Themati,c Stracture;

MIT

Press.

Jackendoff, Ray. 1996. Conceptual Semantics and Cognitive

C ognitiu e L'insuisti,cs 7 :93-129.
Joppen, Sandra, and Dieter Wunderlich. 1995. Argument Linking in Basq

Lingua 97:123-169.

Kaplan, Ronald M., and Joan Bresnan. 1982. Lexical-Functional

mar: A Formal System for Grammatical Representation. In ?he
Representation ol Grammatical Relati,ons, ed. Joan Bresnan. 173Cambridge, Massachusetts: The MIT Press.
Kiparsky, Paul. 1997. Remarks on Denominal Verbs. ln Compler
ed. AIex Alsina, Joan Bresnan, and Peter SeIIs. 473-499.
California: CSLI Publications.
Levin, Beth. 1989. The Basque Verbal Inventory and Configurationality.
Configurationah,ty: the Typology oJ Asymmetries, ed. Laszlo Maracz
Pieter Muysken. 39 62. Dordrecht: Foris.
Levin, Beth. 1991. Building a Lexicon: The Contribution of
Tlreory. Internation al J o urnal of L exicog raphy 4:205-226.
Levin, Beth. 1993. English Verb Classes and, Alternation: A
Inuestigation Chicago: University of Chicago Press.
Levin, Beth, and Malka Rappaport Hovav. To appear. Flom Lexical
mantics to Argument Realization. In Handbook of Morphosyntar
Argument Structure, ed. Hagit Borer. Dordrecht: Kluwer AcaCemic

lishers.

133
fr[

in, Beth, and Malka Rappaport llovav. 1991. Wiping the Slate Clean:
A Lexical Semantic Exploration. Cognition 4Lzl23-15L.
in, Beth, and Malka Rappaport Horav. 1994. A Preliminary Analysis of
Causative Verbs in English. Lingua92:35-77.
, Beth, and Malka Rappaport Horrav. 1995. Unaccusatiuity: At the
Syntar-Ledcal Semantics Interface. Cambridge, Massachusetts: The

MIT

ru
ii

iI
1il

ut
$t

Press.

, Vladimir Petrovich. 1969. Nekotorye Verojatnostnye Universalii

v Glagol'nom Slovoobrazovanii. In JazylcouAe Uniuersalii i Linguistiche'
skaja Tipologija, ed. I.F. Vardul'. Moscow: Nauka.
, Terence. 1990. Euents in the Semantics of English. Cambridge,

iri

ti

Massachusetts: The MIT Press.

, Steven. 1989. Learnabilitg and Cognition: The Acquisition ol Arguiment Structure. Cambridge, Massachusetts: The MIT Press.
iovsky, James. 1991. The Syntax of Event Structure. Cognition4L4T-

James.

1995. The Generatiue Lexicon.

Cambridge,

1\1[as-

sachusetts: The MIT Press.

t Hovav, Malka, and Beth Levin. In press. Morphology and Lexical
Semantics. It Handbook ol Morphology, ed. Andrew Spencer and Arnoid
Zwicky. Oxford: Blackwell.
t Hovav, Malka, and Beth Levin. 1996. Two Types of Derived Acplishments. In Proceedings oJ the First LFG Conference, erl' Miriam
Butt and Ttacy H. King, 375-388. Grenoble, Flance: RANK Xerox
, Elizabeth, and Sara Thomas Rosen. 1996. Strong and V'/eak Predicates: Reducing the Lexical Burden. Linguistic Analgsis 26:29-62.
in, Dan I. 1987. Thinking for Speaking. In Proceeilings of the Thirteenth
Annual Meeti.ng pt the Berkeley Lingu'istics Society (BLS), 435'445,
Two Ways to tavel: Verbs of Motion in English
.ffGrammatical Constntct'ions: Their Form and Meaning,
ed. Masayoshi Shibatani and Sandra A. Thompson. 195-219. Oxford:

Oxford University Press.

, Leonard. 1975, Semantics and Syntax of Motion. ln Syntar and
Semantics 4, ed. J.P. Kimball. 181-238. New York: Academic Press
, Leonard. 1985. Lexicalization Patterns; Semantic Structure in Lex.'ical Forms. In Language Tgpology and, Semantic Description 3: Grami matical Categories and, the Lericon, ed. Timothy Shopen. 57-149. Cam, bridge: Cambridge University Press.
, Leonard. 1991. Path to Realization-via Aspect and Result. In' Proceed,ings of the Seuenteenth Annu,a! Meeti,ng oJ the Berkeley Linguistics
Society (BLS), 480-5i9.
ior, John R. 1996. On Running and Jogging. Cogni.tiue Linguistics
7:2L-34.

I
,i

'.1[

li

{I

tl

,ii
ii

i
I