On the expressive power of some non-standard

constituents
Richard Zuber
CNRS, Paris

Abstract

A semantic constraint on gapping constructions, based on the type of mono-

tonicity induced by the sequence of NPs, is proposed. This constraint makes it
possible to show that some type h2i functions are not denotable by gapping con-
structions and thus excludes a whole class of complex expressions as possibly denot-
ing some type h2i quantifiers. This result contradicts some claims that semantics
of gapped constructions requires the full class of type h2i quantifiers.

The use of generalised quantifiers theory has given rise to various new results con-
cerning expressive power of NLs. Thus we know (Keenan & Westerstahl 1997) that over
a fixed finite universe each type h1i quantifier is denotable by an English noun phrase.
Similarly Keenan (2007) shows that the expressive power of English would be less that it
is if the only NPs we need were ones interpretable as subjects of main clause intransitive
verbs. A sufficient reason is that anaphors like himself must be interpreted by functions
from relations to sets which lie outside the class of generalised quantifiers as classically
defined.
A related result is announced in Nam (2005). He claims to have shown that natural
languages require for their interpretation, in finite universes, the full power of type h2i
quantifiers, i.e. the full set of functions from binary relations to truth-values. The reason
Nam gives is that finite meets and joins of sequences of two type h1i quantifiers (considered
as type h2i quantifiers) lie outside of the set of reducible type h2i quantifiers.
In this note I contest Nam’s result basically on empirical grounds. I show that
his result is based on overgeneralised empirical data and does not take into account some
probably universal constraints, on the grammaticality of gapped constructions which play
an essential role in the argument for the claim. I propose a specific semantic constraint
on gapping and show how it and one of its more specific symmetric versions can be used
to show that some type h2i quantifiers are not denotable by gapped constructions.
Very roughly Nam proves his claim as follows. He considers type h2i quantifiers
induced by a sequence of two type h1i quantifiers. Indeed, given the notion of a case
extension, any sequence of two type h1i quantifiers Q1 Q2 , where Q1 corresponds to subject
NP and Q2 to direct object NP, induces a type h2i quantifier F in the following way:
F (R) = Q1 Q2 (R) = Q1 (Qacc (R)), where Qacc is the accusative case extension of Q2
defined as : Qacc (R) = {x : Q2 (xR) = 1} (where xR = {y : xRy}).
Obviously such type h2i quantifiers induced by case extensions are (Fregean) re-
ducible. However, as observed in Keenan (1992), meets and joins of such quantifiers need
not be reducible. Given that natural languages exhibit constructions whose interpreta-
tions necessitate the use of meets or joins of quantifiers induced by sequences of two type
h1i quantifiers (this happens in the case of gapped sentences as in Leo kissed Lea and Bill
Sue) we need finite meets and joins of such quantifiers. Then Nam shows that meet/join

⊲ LoLa 10/Richard Zuber: Expressive power 5

 closure of such reducible quantifiers forms an atomic Boolean algebra identical to the
algebra of all type h2i quantifiers.
It is essential for Nam’s proof to consider the closure of pairs Qi Qj under arbitrary
joins and meets since only such arbitrary closure is closed under Boolean complements.
In other words the closure of meets and joins of reducible type h2i quantifiers satisfying
some specific conditions may not result in the full algebra of type h2i quantifiers.
In order to understand Nam’s claim the following precisions are necessary. I take it
that Nam claims that the full compositional semantics of a NL must have recourse to all
(reducible and unreducible) type h2i quantifiers only if it is supposed to interpret directly
expressions of this language. For indeed, sentence (1a) can be interpreted by (1b):
(1) a. Leo kissed Lea and Bill Sue.
b. Leo kissed Lea and Bill kissed Sue.
It is obvious that to interpret (1b) we do not need unreducible type h2i quantifiers.
However, in this case the interpretation is not direct and is done at the cost of some more
complex syntax since it supposes a syntactic relationship between (1a) and (1b).
Furthermore, I take it that Nam’s claim is essentially based on the existence of
gapped constructions in NLs. This means that I am not going to show that NLs do not
need the full power of type h2i quantifiers for their interpretation but only to show that
such a necessity does not follow from the existence of gapped constructions.
Finally I take it that Nam’s claim is related to the existence of the conjunctions
or disjuncions of pairs of NPs denoting type h1i quantifiers. Semantically this means
that he considers meets or joins of reducible type h2i quantifiers. There are, however,
constructions which, syntactically, are also results of gapping but whose semantics is
probably different. Consider for instance the following example of an anaphoric quantifier:
(2) a. Leo kissed Lea and nobody else kissed anybody else.
b. Leo kissed Lea and/but nobody else anybody else.
c. Leo kissed Lea and nobody but Leo kissed noone but Lea.
Semantics of these constructions does not involve just meets or joins of sequences of two
type h1i quantifiers. In fact it is far from obvious that for instance nobody else denotes an
extension of a type h1i quantifier or that the sequence nobody else anybody else denotes
a type h2i quantifier or any of its “higher order” (case) extensions. For instance they
cannot be replaced by any “exceptional” type h1i quantifier since there is no equivalence
between sentences like (2c) and (2a). At the same time we know that these constructions
contribute to the formation of non reducible type h2i quantifiers as well (cf. Keenan 1992)
though the irreducible quantifiers to which they give rise may be of different nature.
Technically else-else constructions should be probably considered as “atomic modifiers”
of some expressions denoting type h2i quantifiers: they “chose” a specific atom from the
reducible type h2i quantifier found in the first conjunct making the whole true of just
one binary relation. Such atomic quantifiers are in general also irreducible, except when
they are true of a binary relation which is a product-relation of a particular form (Keenan
1992). Of course the additional expressive power of such constructions has nothing to do
with the possibility of gapping since (2a) and (2b) are very likely equivalent.
The last precision concerns the main logical connective linking a full sentence and
a gapped constituent. Since we are interested in meets and joins we will consider that
all gapped constructions refer to Boolean coordinations which can always be expressed as

⊲ LoLa 10/Richard Zuber: Expressive power 6

 conjunctions or disjunctions of a sentence with a gapped constituent. In particular nei-
ther . . . nor . . . coordination will be considered as a conjunction of negated arguments.
This precision concerns only the main connective of the gapped construction. Since the
gapped constituents have the Boolean structure, their complex Boolean compound can
explicitly contain another connective as in (3a) and (3b):
(3) a. Leo kissed Lea and either Bill Sue or Sam Mary.
b. Leo kissed Lea and neither Bill Sue nor Sam Mary.
In (3a) and (3b) the main connective is the conjunction and. The gapped constituent
in (3a) is composed of sequences of NPs conjoined by either . . . or . . . and in (3b) the
gapped constutuent contains sequences of NPs conjoined by neither . . . nor . . .
Given above remarks Nam’s claim that I am going to criticise can be formulated
as follows: any type h2i quantifier (which is a meet or a join of type h1i quantifiers)
over a finite domain can be denoted by a complex expression containing gapped, possibly
Booleanly complex, constituents. We can even add, given the possibility of “transferring”
the sentential negations to quantifiers and to the corresponding gapped constituent, that
only proper names, conjunctions and disjunctions, and their negations, should be involved.
My disagreement with Nam concerns the question of whether all the meets and joins
of all two-member sequences of type h1i quantifiers (having as argument a binary relation)
interpret (directly) some gapped sentences. In order to show that this is not the case it is
enough to exhibit one particular constraint on possible NPs which can form non-standard
constituents in gapped sentences and which implies the impossibility for some type h2i
quantifier to be denoted by a gapped construction.
Gapping is a delicate phenomenon and judgements on it are subtle and fluctuating.
It has been early observed, however, that negations have a particular status in gapped
sentences and that their use is strongly constrained. Since in the context of quantifiers it
is preferably to use the notion of monotonicity instead of negation, we can state roughly a
semantic constraint on the NPs which can be used in gapped constructions in the following
way: if we have a meet or a join of two sequences of type h1i quantifiers which interpret
a gapped sentence then the second member of the meet (or join) has to take into account
the polarity or the type of monotonicity of the first member. Consider from this point of
view the following examples:
(4) a. Nobody/no student kissed Lea but Bill/some teachers kissed Sue.
b. *Nobody/*no student kissed Lea but Bill/some teachers Sue.
(5) a. Leo kissed no student and Bill no teacher.
b. *Leo kissed no student and Bill Sue.
(6) a. Every student kissed Leo and every teacher Lea.
b. *No student kissed Lea and/but every teacher Sue.
(7) a. Every student kissed Lea but not every teacher kissed Sue.
b. *Every student kissed Lea but not every teacher Sue.
(8) Neither (did) no teacher kissed no student nor (did) Bill Sue.
The above examples indicate dependence of gapping on monotonicity of quantifiers
denoted by various NPs. Given the fact that the negation of the argument (that is of
the binary relation) of a reducible type h2i quantifier is related to the postnegation of
the second quantifier in the sequence it may happen that monotonicity of quantifiers is

⊲ LoLa 10/Richard Zuber: Expressive power 7

 decisive in the choice of the connector in the resulting gapped construction. This fact can
be easier illustrated with sentences where only proper names as NPs are used:
(9) Leo did not kiss Lea and/but Bill kissed Sue.
(10) Leo did not kiss Lea and Bill did not kiss Sue.
(11) a. *Leo did not kiss Lea and Bill Sue.
b. Neither Leo kissed Lea nor Bill Sue.
One observes that (9) does not have a gapped counterpart and the preferable gapped form
of (10) is (11b) and not (11a).
Given the above examples in which gapping is not possible we can formulate the
following more precise constraint on gapping: if the first member (a sequence of two type
h1i quantifiers) of the meet or join induces a monotone decreasing type h2i quantifier then
the second sequence must also induce a monotone decreasing quantifier.
Since this constraint involves monotonicity of reducible type h2i quantifiers we have
to see how the monotonicity of a reducible type h2i quantifier is determined by the mono-
tonicities of type h1i quantifiers whose iteration composes it. This dependence is expressed
by the following propositions concerning conditions for a reducible type h2i quantifier
F = Q2 Q1 to be monotone (from now on the indices of specific case extensions are
omitted). Thus we have (Zuber forthcoming):
Proposition 1: If Q1 and Q2 are monotone increasing then Q1 Q2 is monotone
increasing.

Proposition 2: If Q1 is monotone decreasing and Q2 is monotone increasing

then Q1 Q2 and Q2 Q1 are monotone decreasing.

Proposition 3: If Q1 and Q2 are monotone decreasing then Q1 Q2 is monotone

increasing.

Proposition 4: If Q1 Q2 is monotone increasing and Q1 and Q2 are not trivial

(not constant) then Q1 and Q2 are both monotone increasing or are both monotone
decreasing.

Proposition 5:
(i) Q1 Q2 is monotone increasing iff Q1 and Q2 are both monotone increasing or
are both monotone decreasing.
(ii) Q1 Q2 is monotone decreasing iff one of Q1 , Q2 is monotone increasing and
the other monotone decreasing.
(iii) Q1 Q2 is not monotonic iff one of Q1 , Q2 is not monotonic.

Proposition 5 can be still generalised in the sense that Q2 can be just a particular function
from binary relations to sets without being a quantifier case extension. We will use such a
generalised instance of Proposition 5 in which Q2 is replaced by the function SELF where
SELF(R) = {x : hx, xi ∈ R}. It is monotone increasing and (cf. Keenan 2007), it is not a
case extension of any type h1i quantifier.
We assume also that if F 1 and F 2 are monotone increasing (decreasing) then F 1 ∨F 2
and F 1 ∧F 2 are monotone increasing (decreasing) and that a meet or a join of a monotone
increasing quantifier and of a monotone decreasing quantifier may be not monotonic.

⊲ LoLa 10/Richard Zuber: Expressive power 8

 Given the above propositions it can now be easily checked that the proposed con-
straint applies to various examples we have seen. For instance, given that nobody and
no student denote monotone decreasing quantifiers and proper nouns and the NP some
teachers denote monotone increasing quantifiers, in (4a) the first sequence of NPs denotes
a decreasing type h2i quantifier and the second sequence denotes an increasing quantifier.
Hence, given the constraint, the impossibility of (4b).
In addition if we interpret conditionals by material implication then our constraint
also explains the impossibility of gapping in conditionals: if (12a) is (“partially”) inter-
preted by (12b) then our constraint explains the marked ungrammaticality of (12c):
(12) a. If Leo kissed Lea (then) Bill kissed Sue.
b. Either Leo did not kiss Lea or Bill kissed Sue.
c. *If Leo kissed Lea (then) Bill Sue.
The proposed constraint is very weak. In particular it does not say anything about
gapped constructions in which corresponding quantifiers are neither monotone increasing
nor monotone decreasing as in (13):
(13) a. Only Leo kissed Lea and only Bill Sue.
b. *Only Leo kissed Lea and Bill Sue.
In (13a) the quantifiers denoted by Only Leo and Only Bill are not monotone (increasing
or decreasing) and consequently their iterations do not form monotonic quantifiers and
still the sequence is grammatical.
Our purpose here is not to fully specify constraints on gapping. I want to indicate
that some constraints are enough to show that not all type h2i quantifiers are denotable
in NLs. The proposed constraint and one of its symmetric version that I am going to
discuss briefly now are sufficient for that purpose.
One question that can be asked is whether the above constraint can be symmetrically
inversed: do we have gapped sentences in which the first quantifier is monotone increasing
and the second is monotone decreasing? The following example shows that such symmetric
constraint does not hold in general:
(14) a. Leo kissed somebody/some student and/but Bill nobody/no teacher.
b. *Bill kissed nobody and Leo somebody.
So the the constraint saying that if in a gapped construction the first induced type h2i
quantifier is monotone increasing then the second should be also monotone increasing does
not hold in general. If we consider, however, gapped constructions in which only proper
nouns occur and in which the gapped constituent is simple, that is consisting of one pair
of NPs which both are proper nouns, then this symmetric version of the constraint holds:
(15) a. *Leo kissed Lea and/but not Bill Sue.
b. *Every student kissed Lea and/but not Bill Sue.
Thus we cannot negate a simple gapped constituent if the first conjunct of the gapped
construction expresses a monotone increasing quantifier.
The above examples should be distinguished from the one in (16a,b):
(16) a. Leo kissed Lea and/but neither Bill Sue nor Sam Pat.
b. *Leo kissed Lea and Bill Sue or/and nobody Pat.

⊲ LoLa 10/Richard Zuber: Expressive power 9

 In (16a) the second conjunct of the gapped construction is a complex gapped constituent.
This complex constituent corresponds to a monotone decreasing type h2i quantifier since
it is a meet of two decreasing quantifiers. This means that when in a gapped construction
the first conjunct induces an increasing quantifier the gapped constituent cannot be syn-
tactically atomic and express a decreasing quantifier. We observe in addition that when
the gapped constituent is syntactically complex, that is a coordination of many sequences
of two NPs then any such sequence must induce a quantifier of the same monotonicity.
This fact is illustrated by example (16b).
So the specific version of the constraint that I am going to use to show the falsity of
Nam’s claim can be formulated as follows: if the first conjunct of a gapped construction
expresses a monotone increasing reducible type h2i quantifier then the second conjunct,
the gapped constituent, cannot be atomic and denote monotone decreasing quantifier.
I am now going to indicate how the above constraint allows us to show that some
type h2i quantifiers are not denotable by gapped constructions. I will consider two types
of such quantifiers: (1) specific atoms of the algebra of all type h2i quantifiers and (2)
quantifiers which interprete sentences in which the reflexive pronoun self occurs.
Atoms of the algebra of all type h2i quantifiers are functions which are true of just
one relation are noted F R , where R is the only relation of which F R is true. The relation
R in F R is called the index of the atomic quantifier F R . We will not consider all atomic
quantifiers but only those which are indexed by co-atomic relations, that is relations
which contain all but one ordered pair. Thus we will consider atomic quantifiers of the
form F {ha,bi}′ , for any a, b ∈ E (where E is a finite universe and {ha, bi}′ denotes the set
composed of all elements of E × E except the pair ha, bi). Such atomic quantifiers are
not reducible (Keenan 1992). It will be shown that they are not denotable by gapped
constructions (from which the transitive verb has been deleted).
Any effability claim concerning NPs implies that it is possible to name in some way
various objects in the universe. In our case we suppose for simplicity that we can name
any object in the finite universe, that is for any finite set of objects we can find a set of
individual denoting NPs which refer to these objects. Consider now the atomic quantifier
F {ha,bi}′ . This quantifier is true of just the relation which contains all pairs except the pair
ha, bi. As any atomic quantifier it can be represented as a meet of pairs consisting of two
ultrafilters or pairs consisting of a complement of an ultrafilter and an ultrafilter (cf. Nam
2005). Ultrafilters are denotations of individual denoting NPs or of proper nouns. More
specifically the atomic quantifier F {ha,bi}′ is defined as in (17) (where I x is the ultrafilter
generated by x):
V
(17) F {ha,bi}′ = ( x6=a,y6=b I x I y ) ∧ (¬I a I b )
It follows from (17) that F {ha,bi}′ (R) = 1 iff R = {ha, bi}′ .
The representation in (17) indicates how to construct a sentence corresponding to
the function F {ha,bi}′ applied to a given relation R denoted by a transitive verb. Its non-
gapped counterpart will be just a “long” conjunction composed of sentences of the form
“NPi TV NPj ”. More precisely suppose that the universe E contains n + 2 elements and
let the expression PrN(a) denote a, PrN(b) denote b and PrN(x), with varying x, denote
all other elements of universe and the TV denote the relation R. The form of the non-
gapped counterpart corresponding to the quantifier given in (17) applied to the relation
R, denoted by TV, is given in (18):
(18) PrN(a) not TV PrN(b) and PrN(1) TV PrN(1) . . . and PrN(n) TV PrN(n)

⊲ LoLa 10/Richard Zuber: Expressive power 10

 It follows from the discussion of constraints on gapping that (18) does not have a gapped
counterpart. The reason is of course the only “negative” conjunct it contains. We know
that it cannot be placed at the beginning of the gapped constructions since the remaining
part induces a monotone increasing quantifier. Similarly it cannot constitute the gapped
constituent (after the deletion of the verb) since unique (“atomic”) sequences of proper
nouns inducing a monotone decreasing quantifier are not possible in gapped construc-
tions. Finally, it cannot be a part of a gapped constituent since all other parts of such a
constituent should be “positive”.
What has just been said is not quite a proof that the quantifier represented in (17)
is not denoted by any expression in a given natural language (English). It just shows that
gapping expressions of some natural form cannot denote it. This is enough to consider that
the effability claim announced by Nam is not correct: natural languages, at least English,
do not need all type h2i quantifiers for their semantics. Consequently, the existence of
gapped constructions does not force us to type h2i effability: not all type h2i quantifiers
are necessary to interpret gapped constructions in natural languages.
It might be objected that the above argument is essentially based on the non-
grammaticality of (15a), (15b) or of (19). However, for some speakers (15a) and (15b) are
not completely out and (19) is sufficiently cumbersome to make judgments overwhelming:
(19) *Leo likes Lea, Bill Mary, Lea Lea. . . but not Bill Sue.
Concerning the grammatical status of sequences in (15) I want to stress that corre-
sponding sequences in other languages are generally judged as clearly non-grammatical.
This is in particular the case in Polish which has relatively free word order and a rich case
marking system. This means that in Polish there are at least three versions of gapped
constructions corresponding to (15a): in addition to the form corresponding to the En-
glish sequence in (15a) one has a version where the order between the subject and the
object NP in the gapped constituent is reversed and a version in which the pairs of NPs
are coordinated before the verb and the verb takes the final position. What is important
is the fact that no of these versions of gapped constructions is grammatical in Polish.
The example in (19) can be used to construct another type h2i quantifier not de-
notable by a gapping construction. Observe that conforming to the form given in (17)
we have in (19) sequences of two identical NPs. In other words the transitive verb also
has a “reflexive” use in (19). Now reflexive function SELF may give rise, when com-
bined with some quantifiers denoted by the subject NP, to irreducible type h2i quanti-
fiers (Keenan 1992). Consider from this point of view the (complex) type h2i quantifier
ONLY LEO . . . SELF needed for the interpretation of (20a) and (20b):
(20) a. Only Leo likes himself.
b. Only Leo does not like himself.
This quantifier, contrary to the quantifier LEO . . . SELF, is not reducible (Zuber 2003).
It is, however, a (finite) meet of reducible type h2i quantifiers: (20a) can be represented
by (21a) and then by (21b). In addition, if one wants to use the anaphoric expression
nobody else . . . himself, it can be represented by (21c):
(21) a. Leo likes himself and neither does Lea like herself . . . nor does Sue like herself.
b. Leo likes Leo and neither does Lea like Lea . . . nor does Sue like Sue.
c. Leo likes himself and nobody else likes himself.

⊲ LoLa 10/Richard Zuber: Expressive power 11

 We know that gapping can also apply to transitive verbs with a reflexive pronoun. The
following examples are not completely unacceptable:
(22) a. Leo washed himself and Lea Bill.
b. Leo washed Lea and Bill himself.
One observes, however, that gapped conjunctions corresponding to (20a) or (20b) are not
grammatical. Thus neither (23a) nor (23b) have gapped counterparts since (24a) and
(24b) are not in general considered as grammatical:
(23) a. Leo washed Lea and/but Bill did not washed himself.
b. Leo washed himself and/but Bill did not wash Lea.
(24) a. *Leo washed Lea and/but not Bill himself.
b. *Leo washed himself and/but not Bill Lea.
Similarly gapped constructions whose gapped constituents contain the anaphor everybody/
nobody else . . . himself are excluded. For instance the gapped counterpart of (21c) given
in (25) is clearly not grammatical:
(25) *Leo likes himself and nobody else himself.
It is clear now that the above examples allow us to construct another example of a
type h2i quantifier which cannot be denoted by a sequence of NPs occurring in gapped
constructions. I will not present the argument fully here since it is very similar to the one
discussed above in connection with co-atomic functions.
Let me conclude this note by a more general, negatively oriented remark, related
to some ideas underlying Nam’s project. We know that gapping can occur not only in
sentences with transitive verbs but also those containing di-transitive verbs and even in
sentences where verb phrases take more than three arguments. In addition when the verb
phrase has more than two arguments one can have “partial gapping” where only a part
of the VP has been deleted. This is partially illustrated in the following examples:
(26) Leo gave a chocolate to Lea and a candy to Sue.
(27) Leo gave a chocolate to Lea and Bill a candy to Sue.
In (26) we have partial gapping and in (27) gapping with a three argument verb. Con-
cerning various quantifiers to which these constructions give rise one observes that in (26)
we have a meet of two reducible type h2i quantifiers (denoted by a sequence of two NPs)
which apply to a relation of arity three giving as a result a property. In (27) we have two
sequences of three NPs which give rise to a meet of two reducible type h3i quantifiers. I
mention these examples because as far as I can see the move used by Nam to show his
effability claim for type h2i could also be used to show the effability of quantifiers of other
types. Thus we could show, given the gapping construction like the one in (27), that the
existence of gapping with di-transitive verbs NLs necessitate the full power of type h3i
quantifiers. Such a claim would be very strong not to say strange.

references
Keenan, E. L. 1992. Beyond the Frege Boundary. Linguistics and Philosophy 15: 199–221.
Keenan, E. L. 2007. On the denotations of anaphors. Research on Language and Computation 5: 5–17.
Keenan, E. L. and D. Westerstahl. 1997. Generalized quantifiers in linguistics and logic. In: J. van
Benthem and A. ter Meulen (eds.). Handbook of logic and language. Amsterdam: Elsevier. 837–
893.

⊲ LoLa 10/Richard Zuber: Expressive power 12

 Nam, Seungho. 2005. n-ary quantifiers and the expressive power of DP-compositions. Research on
Language and Computation 3: 411–428.
Zuber, R. 2003. Quantificateurs faiblement rÃľducibles. Logique et Analyse 184–185: 441–446.
Zuber, R. forthcoming. A note on monotonicity of reducible quantifiers. To appear in Journal of Logic,
Language and Information.

⊲ LoLa 10/Richard Zuber: Expressive power 13