You are on page 1of 42

d:/96-1/aarts.

3d 31/3/98 17:42 disk/mp

Transactions of the Philological Society Volume 96:1 (1998) 117158

BINOMINAL NOUN PHRASES IN ENGLISH1


By Bas Aarts
University College London

(Received 20 June 1997; revised 16 January 1998)

Abstract
This paper discusses the structural and semantic properties of
Binominal Noun Phrases (BNPs) in English. BNPs involve two
nominals, N1 and N2, which are in a Subject-Predicate relationship with each other, such that N1 is the Predicate and N2 the
Subject. Examples are a hell of a problem, a wonder of a city, that
idiot of a prime minister etc. On the basis of various types of
syntactic evidence, it is argued that they are headed by the
second of the two nominals, not by the first one, as has often
been claimed. It is further argued that BNPs do not involve
movement, as has recently been suggested in the literature. A
consequence of this study is that it supports viewing syntax as a
flexible system, in which there may be a tension between a rigid
arrangement of elements into categories and constituents, and
the occurrence of unexpected configurations, or of shifts in
patterns taking place diachronically or synchronically.
1
This article is an extensively revised and expanded version of a study published
under the title `The syntax of Binominal Noun Phrases in English' in Dutch Working
Papers in English Language and Linguistics (no. 30, 1994, 128). Different versions of
this paper have been presented in the Department of English, University of
Massachusetts at Boston, the Linguistics Department, University of Manchester,
the Department of Language and Linguistic Science, University of York, and in the
Research Centre for English and Applied Linguistics, University of Cambridge. I
would like to thank the following people for comments: David Adger, John Anderson,
Kersti Borjars, Gillian Brown, Keith Brown, Bill Croft, David Denison, Nik
Gisborne, Steve Harlow, Christopher Lyons, Donaldo Mercedo, Chuck Meyer,
Sharon Millar, John Payne, Bernadette Plunkett and Nigel Vincent. I would also
like to thank Martin Everaert and Ad Foolen for pointing out some useful references,
and Flor Aarts, Valerie Adams, Marcel den Dikken, Sidney Greenbaum, Dick
Hudson, Richard Larson, Elizabeth Closs Traugott, Randolph Quirk, And Rosta
and anonymous referees for reading, and commenting on, earlier drafts of this paper.
Special thanks are due to James McCawley for commenting perspicaciously and
constructively on different successive versions of this paper. Needless to say that none
of the above necessarily agrees with my views.

# The Philological Society 1998. Published by Blackwell Publishers,


108 Cowley Road, Oxford OX4 1JF and 350 Main Street, Malden, MA 02148, USA.

d:/96-1/aarts.3d 31/3/98 17:42 disk/mp

118

transactions of the philological society 96, 1998

1. Introduction
In the structuralist linguistic tradition, and in the modern frameworks that stem from it, syntax is viewed as the component of the
grammar that deals with the segmentation of lexical linguistic matter
into hierarchical configurations of elements that are grouped
together into constituents, more specifically phrases. Constituency
lies at the heart of most current approaches to syntax, and with good
reason, as it can be shown, through such processes as movement,
substitution and ellipsis (see e.g. Aarts 1997), to play a crucial role in
language. However, while recognizing the need for syntactic theory
to be firmly based on phrase structure and constituency, there are
phenomena that pose serious problems for a grammar that places
too rigid and dogmatic an emphasis on segmentation into constituents. This paper discusses NPs of the type in (1) below which will be
seen to be intractable as regards a straightforward phrase structure
treatment.
(1) a.
b.
c.
d.
e.
f.
g.
h.

a hell of a problem
that plonker of a plumber
her nitwit of a husband
those fools of doctors
a wonder of a city
that idiot of a prime minister
some rotten little fig of a human being (W.5.1)2
a colourless mouse of a woman (adapted from S.1.3)

In these examples we have two nominals (italicized in (1) above),


which I will refer to as N1 and N2, respectively. I will call NPs of this
type Binominal Noun Phrases (BNPs). BNPs are of interest from
both a semantic and syntactic point of view. A semantic characteristic is that the first noun denotes a property or quality that is
predicated of the referent of the second noun. Interestingly, this
subject-predicate relationship is internal to a nominal maximal
projection, and it is reversed: the subject expression follows, rather
2
Sentences or phrases marked in this way are taken from the Survey of English
Usage at University College London.

d:/96-1/aarts.3d 31/3/98 17:42 disk/mp

aarts binominal nps

119

than precedes, the predicate. Syntactically, the interest of these


constructions lies in the fact that it is not immediately obvious
which of the two nominals is the head of the construction. While at
first blush N1 appears to be the head in a simple N+PP construction,
it can in fact be shown that N2 functions as the head in BNPs, both
syntactically and semantically. Also of interest is the question of the
role played by the element of: while it looks like a preposition, it can
be shown that it does not introduce a typical PP complement to N1.
English BNP constructions have not received very wide attention,
though there have been a few recent discussions (Kayne 1994, den
Dikken 1995).3 In this paper I will present a new analysis of English
BNPs that will be seen to be recalcitrant from the point of view of
strict constituency, but that can be motivated using distributional
evidence. Specifically, I will argue that a BNP like (1a) is structured
as in (2) below:
(2) [NP a [hell of a] problem]
Here N2 is the head, modified by hell of a. Before turning to the
particulars of this analysis in sections 38, I will have a few words
to say about the history of the BNP construction in the next
section, where I will also present a descriptive account of their
properties.

3
Their counterparts in Dutch have been extensively discussed. Among the first
Dutch linguists to discuss the construction were de Groot (1949) and Royen (1947
54). The latter provides an impressively long list of examples of Dutch BNPs
(Vol.III.2: 117131). However, it was Paardekooper (1956) that became the classic
BNP reference in Dutch linguistics. Its publication inspired a spate of further
discussion on the topic notably van den Toorn (1966) van Caspel (1970), Klein
(1977), Paardekooper (1984) and, more recently, Pekelder (n.d.), Everaert (1992), den
Dikken (1995) (who also discusses English BNPs), and Haeseryn et al. (1997). There is
also an excellent article in the Woordenboek der Nederlandsche Taal (`Dictionary of
the Dutch Language', the Dutch equivalent of the OED, vol. XVIII, columns 410
413). Van Caspel's article contains useful additional references to articles and
grammar sections dealing with the construction both in Dutch and in other languages.
For French, important studies of BNPs can be found in Milner (1973, 1978) and
Ruwet (1982).

d:/96-1/aarts.3d 31/3/98 17:42 disk/mp

120

transactions of the philological society 96, 1998

2. The history of BNPs and their properties


It has been claimed that the English BNP construction was
borrowed from Latin via French (Einenkel 1901, mentioned in
Austin 1980; Curme 1914), the model being examples like monstrum
mulieris (monster woman.GEN; `a monster of a woman'), flagitium
hominis (disgrace man.GEN `a disgrace/crook of a man') and scelus
viri (crime man.GEN `a crime/criminal of a man') (Curme 1914: 56).
They are not attested in Old English (Curme 1931: 85).4 In the above
examples it is clear that we are dealing with a subjective genitive, and
most linguists would agree that N2 is the subject of N1 (`the woman
is a monster'; `the man is a crook/criminal'), notable exceptions
being van Caspel (1970: 281), who regards N1 in monstrum mulieris
as the subject of N2 (`a monster in the guise of a woman', `a monster
who is a woman'), and Littre, who maintains that in French `un
fripon d'enfant, c'est un fripon qui est enfant; mon bourreau de
matre, c'est mon bourreau qui est mon matre, et ainsi de suite' (`a
rascal of a child, it is a rascal who is a child; my tormentor of a
teacher, it is my tormentor who is my teacher, and so on') (195658:
1307, quoted in Austin 1980: 359). The earliest citations for English
BNPs in the Oxford English Dictionary date from the fifteenth
century, though I find that the Middle English Dictionary (Kuhn
1980) lists thirteenth-century instances.
Turning to the present-day English construction, we can represent
the skeletal structure of BNPs as in (3) below:
(3) Det1 N1 of Det2 N2
Preceding N1 there is a determiner position which Austin (1980:
361f.) claims can be filled by any determinative element. However,
pace Austin, there appear to be some restrictions on Det1. For
example, the definite article the sounds distinctly odd in certain
combinations such as e.g. ?the lout of a businessman, ?the nincompoop of a civil servant, though not in others, cf. the rascal of a
landlord (Curme 1931: 85). John Payne (p.c.) has suggested that the
before N1 becomes more acceptable if an NP with the same referent
4
The Woordenboek der Nederlandsche Taal (Dictionary of the Dutch Language)
conjectures that the Dutch construction was also probably borrowed from French
(van Caspel 1970: 281). Royen (194754: III.2:8) agrees.

d:/96-1/aarts.3d 31/3/98 17:42 disk/mp

aarts binominal nps

121

as the BNP is introduced into the discourse first: e.g. I went to see
my bank manager, but the sly fox of a man had just left. Also, as has
been noted by various linguists (e.g. van Caspel 1970, Everaert
1992, den Dikken 1995), if N2 is a proper name, then N1 cannot be
preceded by an indefinite article (cf. e.g. *a creep of a James, *an
egotist of an Alex). I will return to this restriction below. The N1position can be filled by proper nouns. Two interesting examples I
came across recently are a Miss Havisham of a piano (The Independent on Sunday, 5.1.97) to describe an old piano, and a Kate Moss of
a wine (BBC Food and Wine, 4.2.97), used, presumably, to describe
a wine with not much body to it. As for N2, it is always preceded by
the indefinite article a in Det2 position, except in plural NPs, where
it is zero (see (1d) above). N2 can never be a pronoun (cf. *those
bullies of them).5
Binominal NPs are often used as insults, though not always, as
(1e) shows. What does seem to be the case is that N1 is invariably
evaluative. BNPs are infrequent, as the investigation of two corpora
reveals: the corpora contain only a handful of instances, predominantly from spoken material.6 The most frequent type involves the
sequence hell of a as in (1a), and these are what Austin (1980: 361f.)
calls `figurative' in that a simile or metaphor is involved. She
contrasts them with a `literal' type where N1 is the category to
which N2 is assigned (e.g. that miser of a manager). It seems that
among the figurative types we need to recognize a gradient from
idiomatic coinages (e.g. a hell of a . . ., a heck of a . . .) to free
coinages (a skyscraper of a man, a rat of a schoolkid). The idiomatic
coinages are relatively fixed. For example, neither N1 nor N2 can be
5
The restrictions on the determiners preceding N1 and N2 also hold for Dutch
(Paardekooper 1956) and German. In Portuguese, however, the following is possible: o
estupido do rapaz (the stupid of-the boy, `that idiot of a boy') (Mateus et al. 1989: 195).
And in Spanish we can have el bribon del abogado (the scoundrel of-the lawyer, `that
scoundrel of a lawyer'), el charlatan del medico, (the charlatan of-the doctor, `that
charlatan of a doctor') (Ramsey 1962: 548549). In all these cases a definite
determiner precedes N2.
6
The corpora were the Survey of English Usage (SEU) and the British component
of the International Corpus of English (ICE-GB). I used WordCruncher software in
the case of the SEU corpus to search for the string `of a', and in the case of ICE-GB I
used ICECUP (International Corpus of English Corpus Utility Programme). Both
corpora contain written and spoken material. I thank Nick Porter for helping me
collect my data.

d:/96-1/aarts.3d 31/3/98 17:42 disk/mp

122

transactions of the philological society 96, 1998

pluralized (cf. *those hells of problems). Furthermore, in idiomatic


BNPs, unlike in less established instances of the construction,
adjectival premodifiers of N1 are limited in occurrence. Thus,
although we can have e.g. an absolute hell of a problem, with an
intensifying adjective, we can hardly say *a dreadful hell of a problem
or *a tiresome heck of a journey which contain descriptive adjectives.
Premodification of N1 in the literal types is unproblematic (cf. that
useless prude of a counsellor, that dirty digger of a gutter journalist),
as is premodification of N2 in both the figurative and literal types (cf.
a hell of a nice guy, a heck of an awful job, that dickhead of an
inarticulate tourguide). It thus looks as though the left-hand portion
of BNPs is more fixed than the right-hand portion, at least in the
figurative BNPs.
3. The head in binominal NPs
An important issue, both from a semantic and a syntactic point of
view, is which of the two nominal elements inside the BNP is the
head of the overall construction. I have italicized overall here
because, however we analyze BNPs, both nominals are heads at
some level, but only one of them can be the head of the construction
as a whole.
If we treat N1 as the head, an NP like (1a) would be analyzed as
in (4):
(4)

NP

Spec

N'
N

hell

PP
P

NP

of

a problem

d:/96-1/aarts.3d 31/3/98 17:42 disk/mp

aarts binominal nps

123

Among the studies in which N1 is treated as head are van Caspel


(1970), Abney (1987), McCawley (1988), Napoli (1989), Pekelder
(n.d.) and Everaert (1992).
If we treat N2 as the head we end up with a structure like that in
(23):
(5)

NP
Spec

N'

MP

N'
N

hell of a

problem

Here the string hell of a has provisionally been assigned the label
`MP' (Modifier Phrase).
In resolving the issue of headedness in BNPs we might ask which
of the two nominals in these constructions satisfies the selectional
restrictions imposed on the construction as a whole. Napoli has
observed that both in Italian and English BNPs it is N2 that satisfies
selectional restrictions (1989: 224). In this connexion, consider the
sentences below:
(6) She's a summer's day of a girl.
(7) #She's a summer's day.
(8) My uncle is a rock-hard stone of a man: he is completely
inflexible.
(9) #My uncle is a rock-hard stone.
In (6) and (8) the selectional restrictions imposed on the italicized
NPs are met by N2, not by N1, as is shown by the pragmatic oddity
of (7) and (9).

d:/96-1/aarts.3d 31/3/98 17:42 disk/mp

124

transactions of the philological society 96, 1998

However, selectional restrictions are known to be an unreliable


test for determining headedness, at least syntactic headedness, as the
sentences in (10) and (11) show (McCawley 1968: 133):
(10) My buxom neighbor is the father of two.
(11) My neighbor is the father of two.
The oddity of (10) is caused by the modifier inside the subject NP,
not by its head, witness (11) (see also Jackendoff 1972: 18, Abney
1987: 268269 and Everaert 1992: 50). Consider also:
(12) I multiplied what I had computed/*sliced by 2.71.
(McCawley, p.c.)
In (12) there is not even a head noun present in the nominal string
following multiply to satisfy this verb's selectional restrictions.
In view of the unreliability of selectional restrictions in determining headedness we have to look for some other way of establishing
what the head is in BNPs. Hudson (1987), modifying proposals
made in Zwicky (1985), has suggested six criteria for establishing
headedness, which I will apply in turn to the NP construction under
investigation.7
The first, semantic, criterion for headedness asserts that the
element which determines what the overall phrase is a `kind of' is
the head; (see also Zwicky 1993). With regard to this criterion,
consider the following data:
(13) I am getting fed up with this flop of a TV programme.
(14) He's a shining light of a seminar leader.
From (13) and (14) it becomes clear that in both cases it is N2 that
determines the reference of the overall NP. In (13) this flop of a TV
programme refers to a type of TV programme, not to a kind of
flop. Similarly, in (14) a shining light of a seminar leader refers to a
type of seminar leader, not to a shining light. Also, in (6) a
summer's day of a girl refers to a type of girl, not to a summer's
7
I have left out a possible seventh criterion which pinpoints the head as the ruler of
a constituent, i.e. the word on which other elements depend. As this notion is
controversial (see Zwicky 1985: 15), and Hudson (1987) no more than mentions it,
I have ignored it here.

d:/96-1/aarts.3d 31/3/98 17:42 disk/mp

aarts binominal nps

125

day, and in (7) a rock-hard stone of a man refers to a type of man,


not to a stone.8
As we have seen, Napoli has argued that N1 is the head in
binominal constructions. In order to deal with facts like those in
(6)(14) she argues, following an idea floated in Abney (1986), that
we need two head notions, namely syntactic head and semantic
head.9 For Napoli, then, in BNPs N1 is the syntactic head, while
N2 is the semantic head. Implicitly, McCawley has made a similar
claim: he posits the structures in (15a) and (15b) below for the BNP
a bitch of a problem, and comments that these structures
can be viewed as a compromise between the semantic and
syntactic demands of the [Adjectival Noun] bitch, in that [a
bitch of a problem] can be viewed as having both a structure
in which bitch is treated as an A and the of and the second a
are ignored [(15a)], and one in which bitch is treated as a N,
of as a P, and both a's as articles [(15b)]. It should be noted
that while [(15b)] has the appearance of a structure in which
bitch is the head, it is really problem that is the head (1988:
741).10

See also Napoli (1989: 193), and, for a diverging view, van Caspel (1970).
See also Everaert (1992: 5253). It has been pointed out to me that N1 is a
predicate which `is subcategorized to take the second N'. However, N2 is a subject
expression, and these never appear in subcategorization frames.
10
McCawley's trees are an idiosyncratic adaptation of Chomsky's (1970) proposal.
I have used the prime notation instead of the bar notation.
9

d:/96-1/aarts.3d 31/3/98 17:42 disk/mp

126

transactions of the philological society 96, 1998

(15) a.

NP

Det

N'

N'
N

(of a) problem

bitch

b.

NP

Det

N'

P'
P

bitch

of

NP

Det

N'

problem

In (15b) N1 is the syntactic head, while N2 is the semantic head.


Adjectival Nouns (ANs), a term borrowed from Ross (1973), are
intermediate between adjectives and nouns, more specifically, they
are `. . . adjectives with respect to semantics but nouns with respect
to morphology and surface syntax' (McCawley 1987: 459).11
11
The term Adjectival Noun is an unfortunate one to my mind. As McCawley
admits, N1 is morphologically and syntactically a noun. Surely, this is all that is
relevant. The fact that it shares with prototypical adjectives the property of being able

d:/96-1/aarts.3d 31/3/98 17:42 disk/mp

aarts binominal nps

127

Allowing for doubly headed phrases is conceptually unattractive


for two reasons. Firstly, from the point of view of the so-called
`logical problem of language acquisition' (cf. Hornstein and Lightfoot 1981) it is reasonable to suppose that a unified head analysis
makes the task of explaining language acquisition easier. A child
would then need to acquire only one concept of headedness. Any
such assumption concerning learnability would naturally need to be
substantiated by empirical research, but it is plausible, as a null
hypothesis, to suppose that learning two notions of headedness is
more difficult for a child than learning just one.
More importantly, on general grounds of elegance and economy,
an analysis in which semantic and syntactic heads coincide is
preferable to an analysis in which the two are distinguished. Ceteris
paribus, the single head analysis is superior. We will see below that it
can in fact be shown that N2 takes on both head roles.12
A second criterion for identifying the head of a constituent is to
find its morphosyntactic locus, i.e. the place where any inflections are
located. Which constituent is the morphosyntactic locus in binominal NPs? Is it N1 or N2 that takes inflections? The most important
inflection in the English NP is the plural marker. If we pluralize a
particular instance of a BNP, e.g. (16) below, the result is an NP in
which both N1 and N2 take the plural -s ending, witness (17).
(16) that fool of a doctor
(17) those fools of doctors (=(1d) )
to have a modifying function does not make it an adjective, or even adjectival, if we
simply assume that nouns can premodify other nouns, just like adjectives can. The
problem is that McCawley is confusing the notions of form and function: only if we
wrongly take the prenominal modifying function of adjectives to be the pivotal part of
the definition of that word class does it make sense to talk of `adjectival nouns'. The
issue at hand is identical to the problems posed by the locution `an X used as a Y' (in
this case: `a noun used as an adjective'), succinctly discussed in Huddleston (1984: 93):
Is it being claimed that an X used as a Y actually is a Y or that it is merely
functionally like a Y? If it actually is a Y, then in what sense is it also an X? If it is
not a Y but just functionally like one, how can this be reconciled with a definition of
Y in terms of function if Y is defined as a word having a certain function, how can
a word occur with that function and yet fail to be a Y?
12
On the issue of one or more heads in Noun Phrases see the opposing views of
Andrew Radford (1993) (`Head-hunting: on the trail of the nominal Janus') and John
Payne (1993) (`The headedness of noun phrases: slaying the nominal Hydra').

d:/96-1/aarts.3d 31/3/98 17:42 disk/mp

128

transactions of the philological society 96, 1998

We have no way of pinpointing the inflectional locus here and


conclude that this criterion does not apply in determining headedness in BNPs.13
The third criterion for headedness pinpoints the head as the
subcategorizand, i.e. the constituent which is subcategorized with
regard to its complements. We might argue that N1 takes on that
role, the following of-phrase being a complement (see (4) above).
Notice, however, that the of-phrase in BNPs in no way resembles
what are normally analyzed as PP complements. Compare (18) and
(19):
(18) a voracious reader of books
(19) a despicable racist of a policeman
In (18) we have a prototypical noun complement in the form of a
Prepositional Phrase. The grounds on which the PP is regarded as a
complement are that we can point to an analogical V+NP construction in which the verb corresponds to the noun in (18), and the NP
corresponds to the PP. Thus, reader of books corresponds to read
books. Observe that in (19) clearly such a correspondence cannot be
established. In fact, it is difficult to establish any sort of semantic
connection between N1 and the of-phrase. We are certainly not
dealing with partitive or possessive meanings, which are typically
expressed by of-phrases. The general observation seems to hold that
in BNPs there is no semantic relationship between N1 and the
following of-phrase. Of course, we can establish a relationship
between N1 and the noun inside the of-phrase, in that N2 is the
subject of N1, as we have seen. The contrast between (18) and (19),
then, shows that N1 cannot be regarded as the subcategorizand in
BNPs. This is evidence that it cannot be the head of the overall NP.
Although N1 is not the head of the NP as a whole, data such as
(20) and (21) show that it is evidently a head:
(20) that destroyer of education of a minister
(21) this manipulator of people of a mayor
13
(17) could be taken to suggest that both nominals in BNPs are heads (McCawley,
p.c.), but for reasons given above, such an analysis is unattractive.

d:/96-1/aarts.3d 31/3/98 17:42 disk/mp

aarts binominal nps

129

In these BNPs N1 takes a PP complement (of education and of people


respectively; cf. he destroys education; he manipulates people). N2 can
also take PP complements:
(22) a virtuoso of a director of finance
(23) that genius of an instructor of music
(20)(21) and (22)(23) show that both nominals in BNPs are
heads.14
The examples in (20) and (21) are instructive, and will be seen to
constitute crucial evidence against the N+PP analysis of BNPs. The
reason is that if the PPs of a minister and of a mayor are
complements in BNPs such as that destroyer of a minister and this
manipulator of a mayor, as proponents of the analysis in (4) above
have argued, then they will be forced to say that destroyer and
manipulator take two PP complements in (20) and (21), because of
education and of people in these phrases are nominal complements in
anybody's book (cf. he destroyed education/he manipulates people).
Clearly, this cannot be correct, because the nouns destroyer and
manipulator, like their verbal counterparts destroy and manipulate,
are two-place predicates, not three-place predicates. In an analysis
like (5), which will be elaborated in section 7 below, this problem
does not arise, simply because the strings of a minister in (20) and of
a mayor in (21) are not complements of N1.
Hudson's fourth criterion identifies the head as a governor, that is,
the unit in a constituent which can determine the morphosyntactic
form of a sister unit. However, with regard to BNPs neither N1 nor
14
Napoli (1989: 214215, 220221) notes that NPs like (20) and (21) are totally
unacceptable in American English (AmE), but good in British English (BrE). She puts
two question marks in front of NPs like (22) and (23), but observes that they are
acceptable to British speakers. In BrE (20)(21) are slightly less good than (22)(23),
so that both in American and British varieties of English (20)(21) are relatively worse
than (22)(23). The difference in acceptability between these two sets of sentences is
plausibly due to stylistic factors. In any case, the differences between BrE and AmE do
not really concern us here. What is important is that NPs like (20) and (21) are possible
in at least one variety of English.
Notice that in both AmE and BrE a prenominal complement to N1 is dubious, but
unproblematic for N2:
(i) ?that education destroyer of a minister
(ii) that genius of a music instructor

d:/96-1/aarts.3d 31/3/98 17:42 disk/mp

130

transactions of the philological society 96, 1998

N2 can be said to determine the morphosyntactic shape of any of its


neighbouring constituents. This criterion is therefore irrelevant, as is
the fifth criterion Hudson puts forward, which states that the head is
the distributional equivalent of its mother. In the constructions we are
investigating the distributional equivalent is NP, but because BNPs
contain two nominal expressions, either could be the head. This state
of affairs is what prompted us to investigate the question of headedness in BNPs in the first place.
The final criterion defines the head as the obligatory constituent in
a structure. This seems to be closely related to the identification of
the head as that constituent of which the construction as a whole is a
kind. Thus we saw that the NP a rock-hard stone of a man refers to a
kind of man, not to a kind of stone. It is only a small step to drawing
the conclusion that man must therefore be the obligatory element in
this NP. However, if we look at BNPs in context, a more complex
picture emerges. Consider the NPs in (24):
(24) a. A hell of a problem came up in our meeting yesterday.
b. That plonker of a plumber fixed the shower to the wrong
wall.
c. I told those fools of doctors a lie.
d. I refuse to talk with that idiot of a prime minister.
e. I detest her nitwit of a husband.
f. I consider Istanbul a wonder of a city.
In (24a) clearly the nominal problem is obligatory as it would be odd
to say ?*A hell came up in our meeting yesterday. The other NPs of
(24) are more problematic. In (24b) we can leave out either noun
because we can say That plumber fixed the shower to the wrong wall,
but equally well That plonker fixed the shower to the wrong wall.
Similar considerations apply to (24c) and (24d). In (24e) husband
would seem to be the nominal that cannot be left out (cf. ??I detest
her nitwit), but in fact, if we change the determiner to that, then
leaving out N2 is unproblematic: I detest that nitwit. (24f) is a special
case. Here it would be odd to leave out wonder as we would end up
with the sentence ??I consider Istanbul a city, which is not ungrammatical, but pragmatically odd, simply because Istanbul is unquestionably a city. Notice that, if we leave out the Adjective Phrase
wonderful in the sentence I consider Istanbul a wonderful city, the

d:/96-1/aarts.3d 31/3/98 17:42 disk/mp

aarts binominal nps

131

same oddness occurs.15 All (24f) shows, then, is that N2 cannot occur
as a `bare' nominal. It appears, then, that it is precisely in the
idiomatic cases that N2 is obligatory, whereas in the literal cases
either N1 or N2 can be left out.
The conclusion we can draw from applying the Zwicky/Hudson
tests is that while some are not applicable or inconclusive, others
offer support for the contention that N2 is the head in BNPs.
4. Specifiers and modifiers in binominal NPs
We turn now to some further properties of binominal NPs, specifically the leftmost specifier position inside them. As we will see, this
position must be construed to enter into a relationship with N2, not
N1. Consider the following examples (Napoli's 418):
(25) a. our miserable sod of a cleaner
b. your jerk of a brother
In these particular cases, the specifiers our and your before N1
unquestionably belong with N2: in (25a) and (25b) the `possessed'
entities are not sod and jerk, but rather cleaner and brother,
respectively.16 These facts become particularly clear when we paraphrase (25ab) as in (26ab) (McCawley, p.c.):
(26) a. that miserable sod of a cleaner of ours
b. that jerk of a brother of yours
Notice also the ungrammaticality of the following:
(27) *our miserable sod of a cleaner of ours
(28) *your jerk of a brother of yours
These are bad because of the double possessive marking on N2.
15

This was pointed out to me by And Rosta.


See Austin (1980: 361) and also Napoli (1989: 212), who have both made this
observation. Something similar seems to be the case in the following examples from
Denison (forthcoming):
16

(i) these sort of ideas


(ii) those sort of jokes
Here the plural pre-N1 determiner clearly specifies N2, not N1.

d:/96-1/aarts.3d 31/3/98 17:42 disk/mp

132

transactions of the philological society 96, 1998

Consider next (29) which contains an example of a demonstrative


determiner that can be said to specify N2:
(29) that clever little wretch of a Rebecca (Austin 1980: 361)
This NP is especially interesting: despite the fact that the proper
name Rebecca is preceded by an indefinite article, it refers to a
definite person. Analysing (25) and (29) as in (4) above does not bring
out the fact that Det1 specifies N2, not N1. Our analysis of BNPs
should reflect this. Conversely, we should ensure that Det2 should not
be taken to specify N2. Given that Det1 specifies N2, there would then
occur a clash of definiteness between the higher and lower determiners in NPs like (29), because Det1 is definite, while Det2 is indefinite.
The fact that the higher determiner belongs with N2 can also be seen
from the impossibility of a combination like *a clever little wretch of a
Rebecca (den Dikken 1995: 16) where indefinite Det1 a clashes with
definite Rebecca. See also Haeseryn et al. (1997: 854).17
Turning next to modifiers that precede N1 in BNPs, consider the
following NPs:
(30) a. this oceanic barge of a woman (adapted from S.1.10)
b. some shrinking violet of a civil servant (adapted from S.5.1)
c. a curate's egg of a book (from a recent Language book
notice)
The most likely interpretation of the NPs in (30) is one in which the
expressions oceanic in (30a), shrinking in (30b) and curate's in (30c)
modify N1, i.e. barge, violet, and egg, respectively. It is pragmatically
unlikely that oceanic modifies N2 in (30a), because of the oddness of
17
Den Dikken cites interesting data from Dutch which confirm that Det2 does not
specify N2. Consider (i) and (ii) below, taken from den Dikken (1995: 10), which are
possible for some speakers of Dutch, myself included:

(i) ?(?) dat schandaal van een directeurssalarissen


that outrage of a managers' salaries
(ii) ?(?) die ramp van een getalscongruentiefeiten
that disaster of a number agreement facts
In these NPs plural N2 is preceded by the singular form of the indefinite article (een).
Royen (194754: III.2:9) cites dat aap van een meisje (`that monkey of a girl') which is
noteworthy because the neuter demonstrative pronoun dat can only go with the neuter
N2 meisje. N1 (aap) is a masculine noun, which requires the demonstrative pronoun
die.

d:/96-1/aarts.3d 31/3/98 17:42 disk/mp

aarts binominal nps

133

the expression oceanic woman, at least outside a context. In (30b) it is


difficult to imagine what sort of person `a shrinking civil servant'
could refer to, and in any case, shrinking violet is a fixed expression,
just like curate's egg in (30c), and we can therefore be sure that
shrinking modifies violet and curate's modifies egg. In (30) oceanic,
shrinking and curate's can be regarded as adjuncts of N1. The
examples in (30) suggest that pre-N1 modifiers always modify N1,
which is what one would expect.
However, consider next the NP in (31):
(31) another bitchy iceberg of a woman (S.1.3)
Here, clearly, bitchy is pragmatically an inappropriate modifier of
iceberg. What (31) shows, then, is that we should allow for the
possibility that modifiers that immediately precede N1 modify N2.
We must therefore assign a structure to BNPs that can accommodate
this fact.
While in (30) and (31) it is clear which nominal constituents are
being modified by expressions that immediately precede N1, there
are BNPs in which this is not at all obvious. Consider the NPs in (32)
below:
(32) a. a crescent-shaped jewel of an island (from Austin 1980)
b. that clumsy oaf of a newscaster
c. that senseless maniac of a driver
In (32a) the most likely interpretation is for crescent-shaped to
modify island, as Austin (1980: 366) points out; cf. (33a). It is,
however, not entirely ruled out for this AP to modify jewel. In (32b)
and (32c) it is much more clearly the case that pre-N1 modifiers can
modify either N1 or N2, as (33bc) show:
(33) a. a jewel of a crescent-shaped island
b. that oaf of a clumsy newscaster
c. that maniac of a senseless driver18
18
The NPs in (33) are not exact paraphrases of those in (32). Thus in (32b) there is
no presupposition that the newscaster is clumsy, whereas in (33b) there is (James
McCawley, p.c.). NPs like those in (32) are versatile from a communicative point of
view, in that language users can exploit the ambiguous status of the pre-N1 modifier.

d:/96-1/aarts.3d 31/3/98 17:42 disk/mp

134

transactions of the philological society 96, 1998

Two important facts come out of this discussion; one is that the
highest specifier in BNPs, Det1, determines N2, the other is that preN1 modifiers sometimes modify N1, sometimes N2.

5. The of-NP sequence in binominal NPs


As we have seen, in previous studies the of-NP sequence in BNPs has
been analyzed as a constituent, more specifically, as a complement to
N1. Data such as (20) and (21), repeated here, are problematic for
this view:
(34) that destroyer of education of a minister
(35) this manipulator of people of a mayor
In these examples of education and of people are clearly complements
of N1 in both cases (see section 3), but this being so, of a minister and
of a mayor cannot also complement this nominal, because destroyer
and manipulator are two-place predicates, which take only one
internal argument. As we have seen, these data militate against the
N+PP analysis of BNPs, and are conclusive in themselves in
rejecting it. However, there is further evidence which suggests that
the of-NP sequences in BNPs are not even constituents, and hence
cannot be complements of N1.
First, consider the following data, taken from Abney (1987:
297):
(36) a. a monster of a machine
b. a fool of a lawyer
c. a little slip of a girl
d. *[of a machine]i, it was [a monster ti]
e. *[of a lawyer]i, he was [a fool ti]
f. *[of a girl]i, she was [a little slip ti]
g. *[a monster ti] was delivered [of a machine]i
h. *[that fool ti] showed up [of a lawyer]i
i. *[a little slip ti] came in [of a girl]i

d:/96-1/aarts.3d 31/3/98 17:42 disk/mp

aarts binominal nps

135

The data in (36di) show that movement of of-NP sequences from


BNPs, to the right and to the left, is not possible.19
By contrast, movement is not problematic for of-NP sequences
which are uncontroversially complements to a head noun. Compare
the phrases in (37ac) with those in (37df):
(37) a. a copy of the exam regulations
b. a new publication on Russia
c. a reading of contemporary poetry
d. [a copy ti] was received [of the exam regulations]i
e. [a new publication ti] was issued [on Russia]i
f. [a reading ti] was organized [of contemporary poetry]i
In (37df) the of-NP strings have moved to the right resulting in
acceptable sentences. Leftward movement of such strings is less good,
but if a proper context is constructed the results are acceptable:
(38) They didn't send us a copy of the exam REGULATIONS, but
[of the exam PAPER]i they did send us a copy ti.
The second piece of evidence suggesting that of-strings in BNPs
are not constituents, concerns coordination facts: notice that we
19
See also Paardekooper (1956: 9394) who must have been one of the first
linguists to make the observation that rightward movement of the `PP' is not possible.
He offers the following data from Dutch:

(i) *Heb je dat niet altijd een schat


gevonden van een kind?
Have you that not always a treasure found
of a child?
`Haven't you always found her/him a treasure of a child?'
(ii) *Toen zag ik daar toch een boom staan van een matroos.
Then saw I there PTCL a
tree stand of a sailor.
`Then I saw a tree of a sailor standing over there.'
See also Klein (1977: 3339).
In this connexion, consider also the following (from Napoli 1989: 220):
(iii) a. a crook of a chairman of unspeakable stupidity
b. *a crook of unspeakable stupidity of a chairman
Napoli uses the data in (iii) to prove that the PP of a chairman is a complement of
the head noun crook. The ungrammaticality of (iiib) is then explained by observing
that the head is divorced from its complement by the phrase of unspeakable stupidity.
However, in Napoli's analysis the structure of the NP a crook of a chairman of
unspeakable stupidity would be as in (iv) below (irrelevant details omitted):
(iv) [NP a [N' crook [PP of [NP a [N' chairman of unspeakable stupidity] ]] ]]
Clearly, of a chairman is not a constituent here, and cannot be moved for that reason.

d:/96-1/aarts.3d 31/3/98 17:42 disk/mp

136

transactions of the philological society 96, 1998

cannot coordinate two of-NP strings in BNPs, as the following


example shows:
(39) *She called him a bastard [of a husband] and [of a father].
Such coordination is not problematic for straightforward PP complements, as expected:
(40) They sent us a copy [of the exam paper] and [of the exam
regulations].
Thirdly, consider (41) and (42):20
(41) [The exam paper]i, they sent us a copy of ti.
(42) *[An exam]i, we had to take a bitch of ti.
Evidently, regular complements of of can be topicalized, stranding
the preposition, but Det2N2 sequences cannot be moved in this way.
These facts constitute strong support for an analysis along the
lines of (5) where hell of a is treated as a complex modifier parallel to
hellish. Notice that in AP+N sequences it is also not possible for an
adjectival modifier to precede an extraction site:21
(43) a. We saw a hellish movie.
b. *[movie]i, we saw a hellish ti.
We conclude that of-NP sequences in BNPs have a different status
from of-NP sequences in NPs involving regular nominal complements.
Abney analyzes the BNPs in (36) as involving a PP complement,
which he claims is not assigned a thematic role by N1. His explanation of the starred data in (36) is a stipulative theory-internal one. It
involves the claim that in each case the trace left behind by the
moved PPs is not properly governed. The most immediate theta
governor is N1, but because it does not assign a Y-role, and because
we cannot have antecedent government either, the trace violates the
Empty Category Principle. Abney's explanation is unsatisfactory,
because the question arises of what the motivation is for analyzing
20
21

Pointed out to me by James McCawley. See also den Dikken (1995: 8).
Pointed out in den Dikken (1995: 8, fn 7).

d:/96-1/aarts.3d 31/3/98 17:42 disk/mp

aarts binominal nps

137

the PP as a complement in the first place, if N1 does not assign a


thematic role to it.
A far more plausible way to interpret the data in (36)(42) is to say
that the PPs cannot move because they are not constituents.22
6. Movement accounts of BNPs
Recently it has been proposed in the generative literature to deal
with the peculiarities of BNPs in terms of movement. Den Dikken
(1995) has observed that the reference of N1 is always indefinite. In
this connection, notice that the NPs in (1) can be paraphrased as in
(44):
(44) a.
b.
c.
d.
e.
f.
g.
h.

the problem is a hell


that plumber is a plonker
her husband is a nitwit
those doctors are fools
the city is a wonder
that prime minister is an idiot
that human being is a rotten little fig
that woman is a colourless mouse

Den Dikken accounts for this observation by hypothesizing that in


BNPs the indefinite article preceding N2 underlyingly belongs to N1,
but gets separated from it through movement. In an ingenious
analysis he proposes that the BNP a hell of a problem is derived as
follows:

22
The same conclusion is drawn by Akmajian and Lehrer (1976) and Selkirk (1977)
in discussing partitive NPs such as a number of stories: the of-phrase in such
constructions is not a constituent because it cannot be moved:

(i) *A number soon appeared of stories about Watergate.


(Akmajian and Lehrer 1976: 398)
In Klein's 1977 analysis of BNPs the of-NP string is also not a constituent. Instead,
the preposition is rather unsatisfactorily analyzed as standing on its own.

d:/96-1/aarts.3d 31/3/98 17:42 disk/mp

138

transactions of the philological society 96, 1998

(45)

DP

Det

F'

F'

NumPi
F
Agr

Agrk
a

[tj hell]i

of

AgrP
NP

Numj
a

problem

Agr'

Agr

NumP

tk

ti

(den Dikken 1995:11)


The nominal problem is the subject of a small clause (SC) whose
maximal projection is AgrP, and whose predicate is the NumP a hell.
The head of the SC, Agr, adjoins to the functional head F. The
indefinite article a, regarded as the head of NumP, then adjoins to
the moved head of the SC, while the remainder of NumP, the
nominal hell, moves up to the specifier position of FP. For a definite
BNP like that crazy crackpot of a caretaker, which contains a pre-N1
modifier that really modifies N2 (cf. the discussion of (31) above),
five movements are required:
movement
movement
movement
movement

of
of
of
of

Agr to F
the indefinite article a preceding N2 to F
NumP to Spec-of-FP
AgrP to Spec-of-DP at LF via Spec-of-AP

The stopover in Spec-of-AP effects agreement between Det1 and N2.


Crucially, den Dikken regards of in this process of `Predicate
Inversion' (PI) as a `nominal copula' (1995: 8) spelled out in F, and

d:/96-1/aarts.3d 31/3/98 17:42 disk/mp

aarts binominal nps

139

he assimilates his analytical account of the BNP construction to


other PI structures (including structures like e.g. French un drole de
type, which is a DP-internal adjectival PI-construction).
Another account based on movement is Kayne (1994: 106), for
whom of is a `prepositional determiner' (D/P) which is `comparable
to a prepositional complementizer' (ibid.). It projects into a prepositional determiner phrase D/PP (1994: 102). Kayne analyses the
phrases in (46a) and (47a) as in (46b) and (47b) (1994: 106):
(46) a. cet imbecile de Jean
b. cet [D/PP [NP imbecilej] [de [IP Jean I0 [e]j . . .
(47) a. that idiot of a doctor
b. that [D/PP [NP idiotj] [of [IP a doctor I0 [e]j . . .
Like den Dikken, Kayne posits an underlying Small Clause whose
predicate is preposed into Spec-of-D/PP.23
Presumably, the rationale for both movement analyses is the desire
to have the subject expression precede the predicate at a deeper level
of representation.24 There are problems, however, with movement
accounts of BNPs.
In den Dikken's analysis we might wonder what exactly the nature
is of the functional head F and its maximal projection FP. He
suggests that F is an aspectual head, but does not work out the
details. It is also not obvious what licenses all the posited displacements, though one of them, namely movement of the head of the SC
(Agr) to F is motivated, but only on theory-internal grounds:
according to the tenets of the version of the minimalist framework
that den Dikken adopts, the position that NumP trancends (Spec-ofAgrP), and the position it lands in must be made `equidistant' from
the position inside the SC that NumP moves from. To bring this
situation about the head Agr must be moved to F; see Chomsky
(1993: 17, 1995: 184185) for details.
For Kayne there are difficulties of a different nature. First, the
de+IP sequences in (46) and (47) are constituents. We have seen that
23
Cf. also Milner's (1973/1978) account, as expounded in Ruwet (1982: 240), where
N1 is base-generated in Spec-of-NP.
24
And for nothing to `mediate' between the subject and predicate underlyingly
(Kayne, lecture on `Prepositions as complementizers', School of Oriental and African
Studies, London, July 1995).

d:/96-1/aarts.3d 31/3/98 17:42 disk/mp

140

transactions of the philological society 96, 1998

this cannot be correct for various reasons. A second objection


against Kayne's analysis concerns NPs like (48):
(48) a hot summer
Notice that here hot is predicated of summer. As there is a
predication relation between summer and hot (`the summer is
hot'), there is nothing to stop us from positing a derivation of the
NP in (48) along the following lines:
(49) a [ [hotj] [SC summer [e]j]
And indeed, such a derivation is exactly what Kayne proposes for an
NP like the yellow book (1994: 101). Problems arise for NPs like the
following:
(50) a distant relative
Here, under one reading, we understand the AP distant to be used
metaphorically, i.e. the relative is not distant geographically, but
distant in the abstract network of family relations. In Bolinger's
(1967) terms, distant is then a reference-modifying, not a referentmodifying, adjective phrase. Such APs cannot occur in predicative
position salva veritate:
(51) *the relative is distant
For this reason a derivation like (49) is ruled out, given that relative
cannot function as a subject expression in a clause which has distant
as its predicate. In other modifier-head cases an AP in attributive
position can have a meaning that differs radically from the meaning
of that same AP in predicative position. Thus, in the NP a real shame
the meaning of real differs from the meaning of this adjective in
predicative position (e.g. my shame was real).
The upshot of all this is that a transformational account like
Kayne's for predicational modifier-head sequences in NPs of the
type a hot summer is excluded because no single rule can account for
all the permitted instances, and because changes in meaning occur in
certain derivations. The burden is on Kayne to explain why predicate
preposing is blocked in the case of AP-head sequences like (50),
while being permitted in (48), in French phrases like un drole de type,
discussed by den Dikken (1995: 23f.), and in BNPs in general. In the

d:/96-1/aarts.3d 31/3/98 17:42 disk/mp

aarts binominal nps

141

absence of such an explanation, a movement account like Kayne's


for (46a) and (47a) is seriously undermined.25
7. A synthesis
We are now in a position to synthesize the evidence of the last few
sections and to take a fresh look at BNPs. As we have seen, we have
been led to conclude that N2 is the head in binominal NPs, and that
the highest specifier position modifies N2, rather than N1. Furthermore, pre-N1 modifiers can modify N1 or N2. We have also seen, on
the basis of a variety of empirical data, for example (20) and (21), that
of-phrases in BNPs cannot be taken to be complements of N1, and, in
addition, that they are not moveable, which strongly suggests that
they are not constituents. These facts lead to an analysis along the
lines of (52) for BNPs, taking your brat of a brother as an example.
(52)

NP
Spec

N'
MP

N'
N

your brat of a

brother

Apart from the fact that N is the head of the NP, this analysis also
bears out the fact that the determiner your determines N2 (brother),
not N1 (brat), and that the of -N2 sequence is not a constituent. I will
briefly return below to the string brat of a, provisionally labeled `MP'
25
This situation is reminiscent of early generative proposals to handle nominalizations transformationally, which had to be abandoned in favour of a lexical account,
for similar reasons, namely the fact that a nominalization transformation could not
apply across-the-board.
If it is the case, as has been suggested to me by an anonymous referee, that yellow in
a yellow book is a different kind of adjective from distant in a distant relative, then it is
incumbent on Kayne to elucidate this.

d:/96-1/aarts.3d 31/3/98 17:42 disk/mp

142

transactions of the philological society 96, 1998

(Modifier Phrase). First I will deal with some matters concerning the
interpretation of BNPs, and present further evidence for (52).
The structure in (52) is reminiscent of the analytical proposals
made in Akmajian and Lehrer (1976: 399), and in Selkirk (1977:
313), where analyses like (53b) and (53c), respectively, are proposed
for NPs of the type in (53a) (a number of problems):
(53) a. a number of problems
b.

NP
QP

N'
N

QP
NP
a number

problems

of

c.

NP
N''
N'

NP
Det

N''
N'

number (of)

problems

d:/96-1/aarts.3d 31/3/98 17:42 disk/mp

aarts binominal nps

143

As in BNPs, in (53a) too we cannot move the string of problems (cf.


e.g. *a number emerged of problems), and this is one of the main
reasons for Akmajian and Lehrer, and for Selkirk, not to analyze it
as a PP constituent; (see also footnote 22). Selkirk presents further
evidence for her analysis in (53c). She discusses the sentences in (54)
and (55) (1977: 307):
(54) She bought him [NP dozens [PP of [NP those daffodils] ] ], only
two of which were faded.
(55) She bought him [NP [dozens of] daffodils], only two of which
were faded.
The only difference between these sentences is the presence in (54) of
the determiner those. Selkirk analyzes the NP dozens of those
daffodils as involving a head noun dozens with a PP complement
of those daffodils. The NP dozens of daffodils, by contrast, is
analyzed with daffodils as head (cf. (53c) ). She argues as follows:
in (54) there are two NPs, namely the overall NP dozens of those
daffodils and the NP those daffodils, the latter a prepositional
complement. In (55) there is only one NP, namely dozens of daffodils,
headed by daffodils. Now, Selkirk observes that the relative clause in
(54) can be related to both NPs, namely dozens of those daffodils and
those daffodils, yielding two interpretations. Under one reading the
two faded daffodils were among the dozens of daffodils she bought,
under another reading they were among the daffodils referred to by
the phrase those daffodils. In (55), by contrast, the relative clause can
only be related to the NP dozens of daffodils as a whole, not just to
daffodils, simply because the latter is not an NP, but only a noun,
assuming an analysis like (53c).
Consider now (56):
(56) We saw many idiots of football hooligans, two of whom were
English.
The relative clause can here be related only to the NP many idiots of
football hooligans as a whole, not to football hooligans alone. If this is
correct, then a structure like (57a) is to be preferred over a structure
like (57b). The reason for this is that (57b) incorrectly allows a
reading in which the relative clause can refer to football hooligans.

d:/96-1/aarts.3d 31/3/98 17:42 disk/mp

144

transactions of the philological society 96, 1998

(57) a.

NP

Spec

N'
MP

N'

NP

N'
N

many

idiots of

football

hooligans

b.

NP
Spec

N'
N

many

idiots

PP

NP

of

football hooligans

The evidence we have looked at so far in favour of an analysis of


BNPs along the lines of (52) is not exhaustive: there are three
additional types of evidence.
First, consider again the BNP in (32b), repeated here for convenience:

d:/96-1/aarts.3d 31/3/98 17:42 disk/mp

aarts binominal nps

145

(58) that clumsy oaf of a newscaster


We saw earlier that (58) is ambiguous because the modifier that
precedes N1 can be interpreted as modifying either N1 (`the newscaster is a clumsy oaf'), or N2 (`the clumsy newscaster is an oaf').
This difference in interpretation can be represented by two different
tree structures, namely (59) and (60):
(59)

NP
Spec

N'
MP

N'

N
that
(60)

clumsy oaf of a

newscaster

NP
Spec

N'
N'

AP
MP

N'

N'
N

that

clumsy

oaf of a

newscaster

d:/96-1/aarts.3d 31/3/98 17:42 disk/mp

146

transactions of the philological society 96, 1998

In (59) clumsy modifies oaf, whereas in (60) it modifies newscaster.


The N+PP analysis cannot account for the ambiguity of (58),
because the only structure available is (61), where the pre-N1
modifier modifies N1. The interpretation represented by (60)
cannot thus be accounted for in a phrase marker.
(61)

NP

Spec

N'
N'

AP

that

clumsy

PP

oaf

of a newscaster

Secondly, there is a close parallel interpretation between the


sequence N1 of a in BNPs and AP modifiers, as is shown by (62)
(64) below:
(62) That clown of a milkman vs. That clownish milkman
(63) An oaf of a bus conductor vs. That oafish bus conductor
(64) This fool of a pilot vs. This foolish pilot26 27
In this connexion, recall that in section 3 above I discussed the
sentences in (65) and (66):
(65) I consider Istanbul a wonder of a city.
26
See also McCawley's comments re (15a), quoted above, and the parallel
ungrammaticality of (42) and (43b).
Notice that a phrase like a foolish doctor does not entail that the doctor is foolish,
but it has been suggested to me by an anonymous referee that a fool of a doctor does.
The native speakers I asked about this all rejected this view.
27
Consider also the string too big (of) a house in the sentence I live in too big (of) a
house (Abney 1987: 325), to be interpreted as `I live in a house which is too big'.

d:/96-1/aarts.3d 31/3/98 17:42 disk/mp

aarts binominal nps

147

(66) I consider Istanbul a wonderful city.


We saw that leaving out wonder of a in (65) or wonderful in (66) is
pragmatically odd: ??I consider Istanbul a city. The oddness is the
result of the fact that our world knowledge tells us that Istanbul is
without doubt a city. Therefore stating that you consider it a city is
strange. The fact that omitting wonder of a and wonderful in (65) and
(66), respectively, results in the same pragmatic deviance can be
explained if we analyze both as pre-N2 modifiers.
Consider next the following data, from McCawley (1987: 461):
(67) a. Ulysses is a tough book for freshmen to read.
b. Ulysses is a bitch of a book for freshmen to read.
For McCawley (67b) is an example of a noun (bitch) mimicking
adjective behaviour, but notice that we can also employ these data to
show that both tough and bitch of a have the same (modifier)
function. Example (67b) is a problem for linguists who claim that
bitch of a book is a regular N+PP construction, where the PP is a
complement of N, because the burden will be on them to explain why
such strings do not otherwise occur in tough constructions:
(68) *Ethnic cleansing is [a violation of human rights] for the UN to
combat.
Treating N1 of a in BNPs on a par with AP modifiers also
accounts for the fact that for some speakers N1 can be preceded
by intensifiers, which are typically modifiers of adjectives, cf. that
very wonder of a city.28 In this connexion, compare (62)(64) with
Portuguese a estupida da Flora (the stupid of-the Flora `that stupid
Flora'), o pobre do Manuel (the poor of-the Manuel `poor Manuel'),
28
This was pointed out to me by John Payne. Compare also the most devil of a
predicament, cited in Jespersen (190949, VII: 342). Randolph Quirk tells me that in
AmE one can say (i), but not (ii):

(i) She was making the very hell of a problem.


(ii) *She was making a very hell of a problem.
Consider next (iii), also pointed out by Quirk, which appears to be more acceptable
than (i) in BrE:
(iii) Dad, I'm in the very devil of a hole!
It could be the case that the very that we have in these constructions is the very that we
have in phrases like his very soul, as James McCawley has pointed out to me.

d:/96-1/aarts.3d 31/3/98 17:42 disk/mp

148

transactions of the philological society 96, 1998

Spanish la tonta de Juana (the silly of Juana `that silly Juana'), el


pobre de Benito (the poor of Benito `poor Benito') (Ramsey 1962:
57), or French son perfide d'epoux (her treacherous of husband `her
treacherous husband'), une drole d'idee (a funny of idea `a funny
idea') (Littre 195658: 1307; see also Ruwet 1982: 251252). Of
interest in these cases is the fact that adjectives occupy the N1
position.
In general, a one-to-one matching of a particular meaning with a
particular structure is desirable, and the analysis of BNPs proposed
here treats both N1 of a sequences and APs as modifiers occurring in
structurally the same position.
A final piece of data that might be put forward to support the
analysis of BNPs presented here concerns the most common type of
BNP, that involving the sequence hell of a. Notice that we can
phonologically contract the words hell, of and a, such that /hel@v@/
results. This contraction is so common that it has an orthographical
reflex in the spelling helluva. However, this fact should perhaps be
regarded with some suspicion, in view of the currency of such
phrases as sunnuva bitch (or sunavabitch). Clearly, in this case we
would not want to argue for an analysis where son of a is taken to
modify bitch. Furthermore, the indirect relationship between syntactic structure and phonological/intonational structure makes predictions about the former difficult.
A residual issue is the nature of the string brat of a in (52)
repeated below as (69) which I labelled `MP'.

d:/96-1/aarts.3d 31/3/98 17:42 disk/mp

aarts binominal nps


(69)

149

NP
Spec

N'
N'

MP

N
that

brother

brat of a

As we have seen, brat of a must be a unit of some sort, but clearly it


defies a straightforward phrase-structure treatment. This is a
significant point to which I will return below. In the mean time I
propose the following refinement:
(70)

NP
Spec

N'
NP

N'

NP
Det

N'

your

indef

brat

of a

brother

d:/96-1/aarts.3d 31/3/98 17:42 disk/mp

150

transactions of the philological society 96, 1998

In this tree diagram brat is analyzed as a Noun Phrase involving a


zero indefinite determiner. The string of a is Chomsky-adjoined to it.
That brat is indeed a regular nominal head becomes clear by
considering data such as (20), (21) and (30) above, which show
that N1 is unexceptional in its complement-taking and modificational properties. Positing a zero indefinite determiner before N1 is
motivated by the fact that this noun is always interpreted as
nonreferential and indefinite. This can be shown by paraphrasing
your brat of a brother as your brother is a brat (Austin 1980: 359).
Similarly, the NPs in (1) can be paraphrased as in (44), repeated here
as (71) (cf. also den Dikken 1995: 910):
(71) a.
b.
c.
d.
e.
f.
g.
h.

the problem is a hell


that plumber is a plonker
her husband is a nitwit
those doctors are fools
the city is a wonder
that prime minister is an idiot
that human being is a rotten little fig
that woman is a colourless mouse

I will regard of a as a syncategorematic formation in adjunct


position. There is some evidence for treating this string as a unit,
and this concerns the fact that in many BNPs the of a sequence
seems to have become dysfunctional, and can often be left out
altogether. For example, for a simpleton of a judge we can also have a
simpleton judge, and instead of her nitwit of a husband we can say her
nitwit husband (see also Quirk et al. 1985: 1285 who discuss the
contraction of some fool of a policeman to some fool policeman).29
However, there do seem to be restrictions on of a-deletion. For some
BNPs this process is unlikely to take place. In these cases of a seems
to function as a pragmatic marker, which signals that phrases that
contain it should receive an evaluative reading. For example, the
phrase a barge woman is not equivalent to a barge of a woman.
29
McCawley (p.c.) disputes that a contraction like some fool of a policeman 4 some
fool policeman is significant, since the plural these fool policemen is also possible. And
Den Dikken (1995: 12) suggests that some fool policeman could be a compound or
appositive construction. However, this does not seem to constitute evidence against
contraction.

d:/96-1/aarts.3d 31/3/98 17:42 disk/mp

aarts binominal nps

151

Notice that the latter is evaluative and metaphorical, and removal of


the of a sequence seems to result in semantic indeterminacy: a barge
woman could mean anything from `a woman who lives on a barge' to
`a woman who rents out barges' and so on. It seems that non-literal
BNPs resist deletion of the of a-sequence, precisely because this
would result in the loss of metaphorical meaning. By contrast, for
literal BNPs, such as fool of a policeman, compression is unproblematic because they do not lose their evaluative feel when of a is
deleted. There is some evidence that the counterparts of of in BNPs
in languages other than English have also lost their original
function. Thus, von in German BNPs seems to have gone this way.
Van Caspel (1970: 286), citing other sources, notes that apart from
ein Schurke von einem Bedienten (`a villain of a servant') with von
governing dative case, German can also have ein alter Schelm von
Lohnbedienter (`an old villain of a waged servant', from Heine)
without an article before N2 and with nominative case on N2. It
would appear that in this last example von has lost its case governing
properties.
The analysis in (69) is open to the reasonable objection that brat of
a is an atypical modifier. However, there is some independent
evidence for structures like (69) which concerns NPs like a number
of problems, discussed above, and also NPs like these sort of ideas
and those sort of jokes (mentioned in note 16, from Denison,
forthcoming). In the latter cases it makes sense to analyse the
string sort of as a modifying phrase, given the fact that the plural
determiners these and those cannot be associated with the singular
noun sort. I will attempt to explain in the next section how the MP in
(69) came to be there, and what its function is.

8. BNPs and the N+PP analysis: some speculations


So far I have argued for a treatment along the lines of (69) for the NP
your brat of a brother. We were inexorably led to this analysis by a
number of considerations based on headedness, constituency and
configurational possibilities, as well as on the interpretation of
determiners and modifiers. However, it seems that the N1+PP
analysis, as shown in (4), somehow plays a role in the way we

d:/96-1/aarts.3d 31/3/98 17:42 disk/mp

152

transactions of the philological society 96, 1998

process BNPs. The question is `how'? In this section I will look at


two possible answers. What follows is speculative.
One possibility is that (69) is the result of a process of grammaticalization, here regarded as the development of grammatical
function in this case a modifier from lexical elements, or from
constructions (see, for example, Heine et al. 1991: 2, Hopper and
Traugott 1993: xv). BNPs are certainly characterized by what many
linguists agree is one of the hallmarks of grammaticalization, namely
idiomatization. Another feature of grammaticalization is phonological reduction (Hopper and Traugott 1993: 6465). We have already
seen that the most commonly occurring BNP is the type that
involves the sequence hell of a, which can be pronounced as
/hel@v@/. The trigger for this grammaticalization process could be
the fact that N1 has at some point in time lost its ability to assign a
theta role. This resulted in realignment of of a with N1. Notice that
crucially this account presupposes that N1+PP was available at some
point in the history of English. The grammaticalization hypothesis is
attractive, but evidence for it is hard to come by. I did find the
following in Kuhn (1980: 76):
(72) Out of his sepulture Ther sprong. . . Of floures such a wonder
syhte (1393, Gower, Confessio Amantis)
Within the italicised portion of this Middle English sentence the ofsequence has been topicalised, which may suggest that at the time it
was a regular PP. However, this example is far from straightforward.
It is not clear whether a wonder syhte of floures is in fact a BNP, or
simply some sort of genitival construction. The latter option is more
plausible, given that N1 is singular in this string, while N2 is plural.
Perhaps a more plausible way of tackling the problem at hand is to
discuss BNPs in terms of processing. Kajita (1977) has argued for a
dynamic model of syntax where certain syntactic groupings are
rearranged. One example given by him is the following:
(73) Those people are far from innocent.
In this sentence the adjective head innocent appears to be modified
by far from. In order to account for this Kajita proposes a rule of
syntactic reinterpretation such that (74) is reanalyzed as in (75), by

d:/96-1/aarts.3d 31/3/98 17:42 disk/mp

aarts binominal nps

153

analogy to (76) (these data, and those that follow are taken from
Kajita's paper):
(74) [AP [Adj far] [PP from innocent] ]
(75) [AP [Adv far from] [Adj innocent] ]
(76) [AP [Adv hardly] [Adj innocent] ]
The string far from is reinterpreted as an adverb. Kajita's analysis is
supported by such data as (77), where far from appears in a distinct
syntactic environment, modifying a verbal projection:
(77) It far from exhausts the relevant considerations.
The motivating factor for reinterpretation is the existence of what
Kajita calls a `head-nonhead conflict' (i.e. a conflict between
innocent and far). This is understood in the following way. Suppose
that the grammar generates two structures, one in which X is
unambiguously the head, and one in which Y is unambiguously
the head, for example (78) and (79) below:
(78) [AP far [PP from the city] ]
(79) [AP [Adv hardly] [Adj innocent] ] (=(76) )
Assume, furthermore that there are situations where headedness is
not immediately determinable, as in the phrase under investigation,
far from innocent. We then have a case of head-nonhead conflict
which is resolved through a process of syntactic reinterpretation.
In Kajita's terms we might speculate that in BNPs there is also a
head-nonhead conflict such that an NP like (80) is reinterpreted as in
(81), by analogy to (82):
(80) [NP a [fool [PP of a solicitor] ]
(81) [NP a [fool of a] solicitor]
(82) [NP a [foolish] solicitor]
We could formalise Kajita's idea of a head-nonhead conflict by
looking at BNPs in the context of a theory of parsing. Assuming that
processing takes place linearly in a left-to-right fashion, as in the

d:/96-1/aarts.3d 31/3/98 17:43 disk/mp

154

transactions of the philological society 96, 1998

theory of Hawkins (1995), the sequence in (3), repeated here as (83),


would initially be parsed as in (84):
(83) Det1 N1 of Det2 N2
(84) [NP Det [N' N1 [PP of a N2] ] ]
Such a parsing procedure would conform to Hawkins' principles
of Mother Node Construction (MNC) and Immediate Constituent
Attachment (ICA), which assert that in the left-to-right parsing of
a sentence a mother node M must be constructed over a category
C, if a word of category C uniquely determines M (MNC), and
that if an immediate constituent does not itself form a mother
node M, but can be attached to it in accordance with the phrase
structure rules of the grammar, then it must be attached as quickly
as possible (ICA) (1995: 62). In (83) Det1 would instruct the parser
to construct an NP over it, headed by N1, after which of would
cause a PP to be formed deriving the N+PP complement structure
of (84). My hypothesis is that at this point, that is, when the N2head is reached, backtracking ensues and the structure is reanalyzed as in (85):
(85) [NP Det [N' [MP N-of a] [N' N] ] ]
Backtracking occurs because the structure in (84) is semantically
uninterpretable: as we have seen, N1 does not assign a thematic role
to the PP, so unless reanalysis takes place the PP remains thematically `in the air'. Essentially, (83) is a garden path structure. This
account should in principle be testable: (85) is structurally more
complex than (84), and hence should take longer to process,
especially if backtracking is indeed necessary for correct interpretation. An experiment could be devised that measures the time it takes
to parse and interpret structures like (85), as contrasted with (84).
Clearly, such an experiment cannot be conducted within the bounds
of this paper and will be left as a topic for further research.30 This
purely synchronic parsing account, unlike the grammaticalization
hypothesis, does not assume that the N1+PP analysis in (4) is
available at any time, merely that the grammar initially erroneously
30
Givon (1993: 267) also talks of Reanalysis, but suggests that all the lexical
material before N1, including the determiner, is reanalyzed as a complex modifier.

d:/96-1/aarts.3d 31/3/98 17:43 disk/mp

aarts binominal nps

155

assigns this structure to BNPs. Of course, this processing account


raises the question why the grammar makes available garden path
BNP structures. I will not attempt to answer this question here,
because it has a much wider scope, in that it also applies to some of
the other constructions we looked at (e.g. a number of books, these
sort of stories etc.).
Whatever the exact provenance of English BNPs, on which future
(psycholinguistic) research will have to shed more light, the main
aim of this paper with regard to the synchronic analysis of BNPs has
been to show that N2 is the head (see (69) ) and N1 performs a
modifying function. This has an important consequence for the way
in which syntax should be viewed. The results of this study endorse
Heine et al.'s desiderata for the study of sentence structure (1991:
233):
What is required . . . is a framework for linguistic descriptions
that is not confined to static, discrete units such as word classes
or constituent types but rather includes dynamic entities such as
chains of grammaticalization as well as other types of continua,
among the fundamental taxa of linguistic analysis.
Syntax, then, should be seen as a flexible system, in which there may
be a tension between desiring to arrange elements rigidly into
categories and constituents and recognising the possibility of
unexpected configurations, or of shifts in patterns taking place
diachronically or synchronically.
Department of English Language and Literature
University College London
Gower Street
London WC1E 6BT
e-mail: b.aarts@ucl.ac.uk
References
Aarts, Bas, 1997. English Syntax and Argumentation, London: Macmillan.
Abney, Steven, 1986. `Functional elements and licensing', Paper presented at the
1986 Girona GLOW conference.

d:/96-1/aarts.3d 31/3/98 17:43 disk/mp

156

transactions of the philological society 96, 1998

Abney, Steven, 1987. The English Noun Phrase in its Sentential Aspect, PhD thesis
MIT, distributed by MIT Working Papers in Linguistics.
Akmajian, Adrian and Lehrer, Adrienne, 1976. `NP-like quantifiers and the
problem of determining the head of an NP', Linguistic Analysis 2.4, 395413.
Austin, Frances O., 1980. `A crescent-shaped jewel of an island: appositive nouns in
phrases separated by of', English Studies 61, 357366.
Bolinger, Dwight, 1967. `Adjectives in English: attribution and predication', Lingua
18, 134.
Caspel, P. P. J. van, 1970. `Een schat van een (niet meer zo jong) kind', De Nieuwe
Taalgids 63, 280287.
Chomsky, Noam, 1970. `Remarks on nominalization', in R. Jacobs and P. Rosenbaum
(eds), Readings in English Transformational Grammar, Waltham MA: Ginn and
Co., 184221.
Chomsky, Noam, 1993. `A minimalist program for linguistic theory', in K. Hale and S.
J. Keyser (eds), The View from Building 20, Cambridge MA: MIT Press, 152.
Chomsky, Noam, 1995. The Minimalist Program, Cambridge MA: MIT Press.
Curme, G. O., 1914. `The development of the analytic genitive in Germanic', Part II.
Modern Philology, 11.3, 3357.
Curme, G. O., 1931. A Grammar of the English Language. Vol. III Syntax, Boston etc.:
D. C. Heath and Co.
Denison, David, (forthcoming). `Syntax', in S. Romaine (ed.) The Cambridge History
of the English Language, IV, 1776present day, Cambridge: Cambridge University
Press.
Dikken, Marcel den, 1995. `Copulas', ms. Vrije Universiteit Amsterdam.
Einenkel, E., 1901. `On the history of the x-genitive', in An English Miscellany
Presented to Dr. Furnival in Honour of his Seventy-fifth Birthday, Oxford: Oxford
University Press, 6875.
Everaert, Martin, 1992. `Nogmaals: een schat van een kind', in Hans Bennis and J. de
Vries (eds), De Binnenbouw van het Nederlands: Een Bundel Artikelen voor Piet
Paardekooper, Dordrecht: JCG Publications, 4554.
Givon, Talmy, 1993. English Grammar, 2 volumes, Amsterdam: John Benjamins.
Groot, A. W. de, 1949. Structurele Syntaxis, Den Haag: Servire.
Haeseryn, W., Romijn, K., Geerts, G., de Rooij, J. and van der Toorn, M. C., 1997.
Algemene Nederlandse Spraakkunst, 2nd edition, Groningen: Martinus Nijhoff/
Deurne: Wolters Plantyn.
Hawkins, J. A., 1995. A Performance Theory of Order and Constituency, Cambridge:
Cambridge University Press.
Heine, Bernd, Claudi, Ulrike and Hunnemeyer, F., 1991. Grammaticalization: A
Conceptual Framework, Chicago: Chicago University Press.
Hopper, Paul J. and Traugott, Elizabeth C., 1993. Grammaticalization, Cambridge: Cambridge University Press.
Hornstein, Norbert and Lightfoot, David W., 1981. `Introduction', in N. Hornstein and D. W. Lightfoot (eds), Explanation in Linguistics: The Logical Problem of
Language Acquisition, London: Longman, 931.
Huddleston, Rodney, 1984. Introduction to the Grammar of English, Cambridge:
Cambridge University Press.
Hudson, Richard A., 1987. `Zwicky on heads', Journal of Linguistics 23, 109132.
Jackendoff, Ray S., 1972. Semantic Interpretation in Generative Grammar, Cambridge, MA: MIT Press.

d:/96-1/aarts.3d 31/3/98 17:43 disk/mp

aarts binominal nps

157

Jespersen, Otto, 19091949. A Modern English Grammar on Historical Principles,


London: George Allen & Unwin / Copenhagen: Munksgaard.
Kajita, M., 1977 `Towards a dynamic model of syntax', Studies in English Linguistics.
5, 4477.
Kayne, Richard, 1994. The Antisymmetry of Syntax, Cambridge MA: MIT Press.
Klein, M., 1977. Appositionele Constructies in het Nederlands, Unpublished doctoral
dissertation, University of Nijmegen.
Kuhn, S. M. (ed.), in progress. Middle English Dictionary, Ann Arbor: The University
of Michigan Press.
Littre, E., 195658. Dictionnaire de la Langue Francaise, Edition integrale, Paris:
Jean-Jacques Pauvert editeur.
Mateus, H. M., Brito, A. M., Duarte, I. and Faria, I. H., 1989. Gramatica da
Lngua Portuguesa, Lisbon: Caminho.
McCawley, James, 1968. `The role of semantics in grammar', in E. Bach and T.
Harms (eds), Universals in Linguistic Theory, New York: Holt, Rinehart and
Winston, 125169.
McCawley, James, 1987. `A case of syntactic mimicry', in Rene Dirven and Vilem
Fried (eds), Functionalism in Linguistics, Amsterdam: John Benjamins, 459470.
McCawley, James, 1988. The Syntactic Phenomena of English, 2 volumes, Chicago:
Chicago University Press.
Milner, J-C., 1973. Arguments linguistiques, Paris: Mame.
Milner, J-C., 1978. De la Syntaxe a l'interpretation, Paris: Editions du Seuil.
Napoli, Donna Jo, 1989. Predication Theory: A Case Study for Indexing Theory,
Cambridge: Cambridge University Press.
Paardekooper, Piet C., 1956. `Een schat van een kind', De Nieuwe Taalgids 49, 93
99.
Paardekooper, Piet C., 1984. Beknopte ABN Syntaksis, Eindhoven: private publication.
Payne, John, 1993. `The headedness of noun phrases: slaying the nominal Hydra', in
Greville G. Corbett, Norman M. Fraser and Scott McGlashan (eds), Heads in
Grammatical Theory, Cambridge: Cambridge University Press, 114139.
Pekelder, J., n.d. `Wat doen taalkundigen met voorzetsels?', ms. Louvain-la-Neuve.
Pekelder, J., n.d. `Een droom van een professor', ms. Louvain-la-Neuve.
Quirk, Randolph, Greenbaum, Sidney, Leech, Geoffrey and Svartvik, Jan, 1985.
A Comprehensive Grammar of the English Language, London: Longman.
Radford, Andrew, 1993. `Head-hunting: on the trail of the nominal Janus', in
Greville G. Corbett, Norman M. Fraser and Scott McGlashan (eds), Heads in
Grammatical Theory, Cambridge: Cambridge University Press, 73113.
Ross, John R., 1973. `Nouniness', in O. Fujimura (ed.), Three Dimensions of
Linguistic Theory, Tokyo: TEC, 137257.
Ramsey, M. M., 1962. A Textbook of Modern Spanish (revised by R. K. Spaulding),
New York: Holt, Rinehart and Winston.
Royen, Gerlach, 194754. Buigingsverschijnselen in het Nederlands, 4 vols, Amsterdam: North Holland Publishing Company.
Ruwet, Nicolas, 1982. Grammaire des insultes et autres etudes, Paris: Editions du
Seuil.
Selkirk, Elizabeth, 1977. `Some remarks on noun phrase structure', in Peter W.
Culicover, Thomas Wasow and Adrian Akmajian (eds), Formal Syntax, Orlando:
Academic Press, 285316.

d:/96-1/aarts.3d 31/3/98 17:43 disk/mp

158

transactions of the philological society 96, 1998

Toorn, Maarten C. van den, 1966. `Bedoelen en verstaan; de aard van het
syntactisch verband', De Nieuwe Taalgids 59, 2935.
Williams, Edwin S., 1980. `Predication', Linguistic Inquiry 11.1, 203238.
Woordenboek der Nederlandsche Taal, volume XVIII, F. de Tollenaere (ed), 's-Gravenhage: Martinus Nijhoff/Leiden: A. W Sijthoff.
Zwicky, Arnold, 1985. `Heads', Journal of Linguistics 21, 130.
Zwicky, Arnold, 1993. `Heads, bases and functors', in Greville G. Corbett, Norman
M. Fraser and Scott McGlashan (eds), Heads in Grammatical Theory, Cambridge:
Cambridge University Press, 292315.

You might also like