1

First draft: SEPTEMBER 2004. Second Draft: December 2004.

Third Draft March 2006, to appear in Linguistics and Philosophy

The Universal Density of Measurement
1

Danny Fox and Martin Hackl
MIT and Pomona College

Abstract
The notion of measurement plays a central role in human cognition. We measure peoples height,
the weight of physical objects, the length of stretches of time, or the size of various collections of
individuals. Measurements of height, weight, and the like are commonly thought of as mappings
between objects and dense scales, while measurements of collections of individuals, as
implemented for instance in counting, are assumed to involve discrete scales. It is also
commonly assumed that natural language makes use of both types of scales and subsequently
distinguishes between two types of measurements. This paper argues against the latter
assumption. It argues that natural language semantics treats all measurements uniformly as
mappings from objects (individuals or collections of individuals) to dense scales, hence the
Universal Density of Measurement. If the arguments are successful, there are a variety of
consequences for semantics and pragmatics, and more generally for the place of the linguistic
system within an overall architecture of cognition.


1. Introduction

It is common to assume that the semantics of natural language makes reference to a notion of
degree or quantity. For example, a standard formulation of the claim that Mary is taller than John
involves reference to two degrees corresponding to Marys height and to Johns. Such a
formulation motivates a domain of degrees that can be compared to each other to determine
which is greater, i.e., a (totally) ordered domain, or a scale.
A great deal of literature in natural language syntax and semantics argues that this is the
correct formulation and attempts to determine the properties of the relevant scales. The goal of
this paper is to contribute to this literature by arguing that scales are always dense. More

1
This paper benefited greatly from audiences at MIT, Pomona College, Universidad del Pas Vasco (Basque
Country), USC, Tel-Aviv University, ZAS (Berlin), and UCLA. We've benefited in particular from comments by
Hagit Borer, Gennaro Chierchia, Noam Chomsky, Kai von Fintel, Irene Heim, Jon Gajewski, Elena Guerzoni,
Sabine Iatridou, Polly Jacobson, Ezra Keshet, Jeff King, Nathan Klinedinst, Giorgio Magri, Barry Schein, Philippe
Schlenker, Roger Schwarzschild, Jesse Snedeker, Benjamin Spector, Bob Stalnaker, Tamina Stephenson, Jean-
Roger Vergnaud, Steve Yablo, Maria Luisa Zubizarreta, and an anonymous reviewer for Linguistics and
Philosophy.
2
specifically, this paper has two goals. The first is to argue that scales of height, size, speed, and
the like are dense. This, although not universally accepted, seems rather intuitive. If Mary is
taller than Bill, there has to be a degree of height that is somewhere between Marys height and
Bills. Degrees of height correspond to our spatial intuitions, which likely involve the notion of a
continuous substance. Nevertheless, this assumption is not necessary. We will argue that various
puzzles relating to the theory of exhaustivity, scalar implicatures, and the semantics of questions
and definite descriptions can be resolved in a principled way under our intuitive claim that the
relevant domains are in fact dense.
The second goal is to argue for a radical extension of this claim to all degree domains. In
particular, we will defend the unintuitive claim that these domains are dense even when they are
commonly thought to be discrete. In other words, when we say that John has 3 kids or that he has
more kids than Mary, the presupposed scale (according to this unintuitive claim) is not the
ordered set of natural numbers or anything like it. Instead, it is the same domain of
measurements that is needed to capture our intuitions of space and time, something closer to the
rational or real numbers.
The intuitive claim has already been discussed in the literature, in the context of the
semantics of sentences such as John is taller than Bill is. Various proposals building on von
Stechow (1984) assume that the semantics of this statement makes crucial appeal to the maximal
degree of height that Bill possesses (see Heim 2001). In particular, the sentence is claimed to
assert that Johns height is above that maximal degree. Pinkal (1989) challenges this assumption
on the basis of the fact that sentences such as John is taller than he has ever been before are
acceptable and have a coherent interpretation. (See also Artstein 1998.) He argues that this is
unexpected under von Stechows view.
2
According to von Stechow, if the sentence is to be true,
Johns height at the present has to be greater than the maximal degree d such that John has been
d tall at a point in time prior to the present. But there couldnt be such a maximal degree, at least
not if Johns growth is assumed to be continuous. Therefore, the sentence shouldnt receive a
coherent interpretation.
Pinkals argument is based on the assumption that the domain of degrees of height (and the
domain of times) is dense. This assumption, while intuitive, is not independently supported in

2
Pinkal argues therefore for an alternative analysis, one that involves universal quantification. See Heim (2003) for
the comparison of various alternatives.
3
Pinkal (1989). Von Stechow might therefore simply respond to this challenge by denying the
density assumption.
3
We will not engage in this particular debate. Our goal, instead, will be to
provide a related, though independent, argument that the relevant domain of degrees is formally
dense.
4

Suppose that natural language has operators that take degree properties and return their
maximal member (maximality operators). Pinkal expects that the system will fail when such
operators combine with properties of the sort that he identified (e.g. d. John has never been d
tall before). This expectation which leads to the conclusion that the comparative construction
does not make use of a maximality operator is justified only if the relevant degree domains are
dense and Pinkal is right in assuming continuity of growth.
A more direct way to test the expectation will be available if we identify various
constructions that can be argued to involve a maximality operator on independent grounds. This
will be our strategy. We will start with properties that, like Pinkals, apply to members of
intuitively dense scales. We will construct the properties in a way that will ensure that they have
no maximal member if and only if intuitively dense domains are in fact dense. The crucial
observation will be that the combination of such properties with maximality operators yields
systematic failure, an observation that will provide strong support, we hope, for our claim about
intuitively dense domains. Secondly, we will look at parallel constructions that are traditionally
analyzed as involving the non-dense domain of cardinalities. We will show that these
constructions display a virtually identically pattern, which is explained under our radical
extension of the density assumption, and is mysterious otherwise.
We will argue for our two claims in various ways. We will start in section 2 with a puzzle
pertaining to implicatures and the semantics of only in intuitively discrete domains and explain
why the puzzle disappears once we think of the domains as dense. We discuss additional
empirical consequences studying in tandem both the intuitively discrete and dense domains. We
then move in section 3 to a discussion of apparent constraints on wh-movements which,
following Rullmann (1995) and von Stechow (1984), we analyze as syntactically well-formed
but semantically incoherent (due to a violation of a maximality requirement). In contrast to these

3
Kai von Fintel has pointed out to us the potential relevance of a debate between Stalnaker and Lewis about the
semantics of counterfactuals. A similar claim to Pinkals was made by Lewis, and a similar response to the one that
seems available for von Stechow is made by Stalnaker. For detailed discussion see Bennett (2003).
4
Interestingly, the consequences for the von Stechow Pinkal debate will be inconclusive given the modularity claim
that we will end up making. See section 5.4.
4
authors, however, we argue that incoherence is a result of density along the lines outlined above.
The two sets of problems (from section 2 and 3) are shown in section 4 to have a common
logical structure. This leads us to new predictions that we argue are corroborated in a variety of
contexts. Finally, we draw various consequences from our proposal for the architecture of the
linguistic system and the nature of interactions between semantics and pragmatics.
If our arguments are correct, they have potential consequences that go beyond the core
linguistic system. In particular, a possible connection is suggested with work in cognitive science
which aims to discover the origins of human reasoning about quantities and numerosities. One
recurring hypothesis that arises in that context is that prior to the development of adult arithmetic
there is a core system that allows the measurement (or at least the estimation) of quantities, but
crucially does not have access to anything like the notion of a natural number. If this hypothesis
is correct, it will be interesting to investigate whether the postulated core system could possibly
be the system at the heart of the semantics of degree constructions. Needless to say, an
investigation of this question goes way beyond the scope of this paper.


2. Implicatures and Only

When the sentence in ( 1) is uttered, a belief that John doesnt have 4 children is normally
attributed to the speaker.

(1) John has 3 children.

This attribution is standardly accounted for by a mechanism that generates pragmatic inferences,
in this case the scalar implicature that (the speaker believes) that John doesnt have 4 children.
One might expect that a similar implicature will arise when the sentence in ( 2) is uttered, i.e.
the scalar implicature that John doesnt have more than 4 children.

(2) John has more than 3 children.

5
However, this expectation is not borne out. As discussed in Krifka (1999), a sentence with a
numeral determiner n generates a scalar implicature, but a sentence with the comparative
determiner more than n does not.
5

In responding to this puzzle one might hope to capitalize on the relative complexity of
comparative constructions. Specifically, one might suggest that the use of a complex
comparative invites the inference that the use of an alternative simpler construction was
impossible (maxim of brevity).
6
In particular, the idea might be that ( 2) doesnt have the
implicature that John has exactly 4 children because there is a simpler (briefer) way to convey
that information. However, consider an utterance of the following sentences in a context in
which it is known that Bill has exactly three children.

(3) a. John has more children than Bill does. (Context: Bill has 3 children.)
*Implicature: John has exactly four children.
b. John has two more children than Bill does.
Implicature: John has exactly 5 children. (Context: Bill has 3 children.)

By parity of reasoning (maxim of brevity), if it is known that Bill has exactly 3 children, the
assertion in ( 3)b should not trigger the implicature that it in fact does. Similarly, although it is
somewhat artificial to say that John has 2 more children than 3, to the extent that it is possible, it
can convey the information that John has exactly 5 children.
We therefore think that an alternative account is needed. We suggest that the account should
follow from the principles that explain the contrast between ( 4) and ( 5).


5
Krifka discusses another problem namely that the corresponding at least sentence (John has at least 3 children.)
lacks an implicature as well. In fact, on initial inspection the following constructions seem to be quite similar:

(i) a. John has more than 3 children.
b. John has at least 4 children.
c. John has 4 or more children.
d. John has no fewer than 3 children.

Our account does not extend to (b,c), though with a modification it does extend to (d), see Fox (in progress). This is
of course a challenge for the account. However, we think there are good arguments in favor of independent accounts
of (b) and (c), see Fox (2004: class 5, in progress), and for additional arguments pertaining to (b), see Geurts and
Nouwen 2005, as well as Hackl 2000. For an alternative perspective, see Spector (2006).
6
Though see Matsumoto (1995) on the relative scarcity of inferences based on the maxim of brevity.
6
(4) John has very few children. He only has 3
F
(children).
7

(5) *John has very few children. He only has more than 3
F
(children).
8


The sentence in ( 4) provides us with a particular paraphrase of the scalar implicature that the
sentence in ( 1) has. This exemplifies a fairly general fact, pointed out in Fox (2004), building on
earlier observations and proposals made by Groenendijk and Stokhof (1984) and Krifka (1995).
It is generally true that the scalar implicature of a sentence can be stated explicitly with the use of
a focus particle only that associates with the relevant scalar item:

(6) The only implicature generalization (OIG): Utterance of a sentence, S, as a default,
licenses the inference/implicature that (the speaker believes) onlyS', where S' is (a
minimal modification of) S with focus on scalar items.

The sentence in ( 5) is a failed attempt to state a parallel implicature for the sentence in ( 2). We
suggest that an account of the failure of ( 5) should yield an explanation of the missing
implicature in ( 2).
But what might account for the unacceptability of ( 5)? A possible answer is suggested when
we look at the sentences in ( 7). The sentence in ( 7)a, just like ( 1), generates a scalar implicature,
in this case the implicature that John only weighs 120 pounds. The comparative construction in
( 7)b, just like ( 2), lacks a scalar implicature, and again, an attempt to describe the would-be-
implicature with the focus particle only yields an unacceptable sentence, ( 7)c.


7
F-subscript represents the presence of focus, which is, in turn, associated with prosodic prominence.
8
All of the effects discussed in this paper do not arise when the context specifies a discrete set of relevant
alternatives. For example:

(i) A: How many points did Iverson score last night?
B: I dont know.
A: Was it more than 10, more than 20, more than 30, or more than 40.
B: He (only) scored more than 20
F
(points)

The version of Bs final reply which contains only is acceptable (in contrast to ( 5)) and the version without only
yields an implicature (in contrast to ( 2)).
This is reminiscent of Kroch (1989), where it is demonstrated that such context specification (D-linking)
circumvents negative islands. This will be directly relevant to the discussion in section 3 and will be addressed in
section 5.5. The basic idea, as one might expect, will be that contextual specification allows for a richer syntactic
representation in which a pronoun refers to a non-dense domain of alternatives. Cf. Westersthal (1984), von Fintel
(1994), and much subsequent work.
7
(7) a. John weighs 120 pounds.
b. John weighs more than 120 pounds.
c. *John only weighs more than 120
F
pounds.

But in this case, the explanation of the facts is readily available: ( 7)c presupposes that John
weighs more than 120 pounds, 120 + pounds;
9
John, therefore weighs more than 120 + /2
pounds, and, hence, there is a degree, d, greater than 120 such that John weighs more than d
pounds. If this degree is relevant for the meaning of ( 7)c, if it is in the domain of quantification
for only, the unacceptability of the sentence is expected. In other words, if we make the intuitive
assumption that the set of degrees is dense in this case (from which it follows that 120 + /2 is a
member of the set) the presupposition of ( 7)b ensures that the sentence will never be true, and
this, we suggest, is a plausible explanation for unacceptability.
10

It is also obvious why ( 7)b lacks an implicature. The implicature, had there been one, would
have been the implicature that for any degree, d, greater than 120, the speaker believes that the
claim that John weighs more than d pounds is false. In other words, ( 7)c would itself be the
implicature of ( 7)a. However, the speaker asserted that John weighs more than 120 pounds, and
this assertion, given the density of degrees, is incompatible with the implicature.
But how could we get this explanation to carry over to the facts in ( 5) and ( 2)? In ( 7), it is
quite intuitive to assume that the set of degrees relevant for the evaluation of the sentence is

9
For convenience, we assume the treatment of only due to Horn (1969), who argues that a sentence with only
presupposes the truth of the corresponding sentence without only. However, this assumption is not crucial for
anything were saying. If the presupposition is taken to be part of the assertive content, a similar type of incoherence
is predicted. We also talk here as if only quantifies over degrees, rather than alternatives, though translating back and
forth is quite trivial. See note 35.
Furthermore, we are ignoring interesting questions about the semantics of only. In particular, the lexical-entries
we are considering in this paper are all too strong for a variety of constructions, and it is important to ensure that the
correct weakening will be consistent with our account. We think that there are a variety of possibilities, but since the
issue is quite complex, we will not enter it here. The interested reader could verify that the following lexical entry
from Spector (2006), based on Groenendijk and Stokhof (1984) and van Rooy and Schultz (2004), would be
consistent with our account:

(1) [[Exh/only]] (A
<st,t>
)(p
st
) p(w) & w'A[p(w') & (w'<
A
w)]
w'<
A
w {pA:p(w')=1}{pA:p(w)=1}
10
See Stalnaker (1974), and much subsequent work, for a model of communication that prohibits such conflicts.
It is a recurring hypothesis that syntactically well-formed sentences should be unacceptable when they are
contradictory. However, since clearly not all contradictions are judge unacceptable, a semantic explanation of the
unacceptability of sentences like ( 7)c that appeals to its contradictory nature requires also a way of distinguishing
between contradictions that are ruled out by the grammar and those that are not. See section 5 for further discussion
drawing on Gajewski (2002, 2003), and Chierchia (1984).
8
dense. But a similar assumption seems quite radical for ( 5) and ( 2). Nevertheless, our goal is to
argue that it is correct. I.e. we will argue for the following hypothesis about degree scales:

(8) The Universal Density of Measurements (UDM): measurement scales needed for natural
language semantics are always dense.

Let us start with a discussion of the implications of the UDM for ( 5). Suppose that, despite
appearances, the set of degrees that only quantifies over is dense even in this case. ( 5), thus,
asserts that for any degree d greater than 3, John doesnt have more than d children (see Hackl
2000). Without the UDM, there would be no problem. The set of degrees relevant for evaluation
would be, as is standardly assumed, possible cardinalities of children (i.e., 1, 2, 3,). The
sentence would then assert that John doesnt have more than 4 children. Since it presupposes that
John has more than 3 children, it would end up conveying that he has exactly 4. If density is
assumed, however, we get exactly the same contradiction that weve seen in ( 7)c. The
presupposition of course remains the presupposition that John has more than 3 children.
However, the assertion would now not just exclude 4 as a degree exceeded by the number of
Johns children. It would also exclude any degree between 3 and 4. This is obviously in
contradiction with the presupposition.
What about ( 2)? To account for the lack of an implicature, we will assume that the would-be
implicature must be derived in the syntax with a covert exhaustive operator (exh) akin to only in
its interpretation. (This operator is in principle optional and its presence yields implicatures.)
What only and exh have in common, is the requirement that a particular proposition be stronger
than a set of alternatives, and its employment would, thus, be impossible in ( 2) for the same
reason that ( 5) is unacceptable.
The suggestion that implicatures are due to exh was advanced in Groenendijk and Stokhof
(1984), Krifka (1995), and van Rooy and Schulz (2004). It has been defended in Fox (2004)
based on evidence that the OIG (in ( 6)) is a better predictor of the distribution of scalar
implicatures than the traditional/neo-Gricean account (but see note 40).
The data in ( 9)-( 11) seem to provide additional support for our line of reasoning. In ( 9) and
( 10) (in contrast to ( 5) and ( 7)b) only is able to associate with the numeral of the comparative
more than n.
9

(9) a. I can only say with certainty that John weighs more than 120
F
pounds.
b. I can only say with certainty that John has more than 3
F
children.

(10) a. I was only able to demonstrate that this refrigerator weighs more than 120
F
pounds.
b. I was only able to demonstrate that this candidate received more than 500
F
votes.

Similarly, in ( 11) and ( 12), in contrast to ( 2), the comparative more than n can generate an
implicature.

(11) a. I can say with certainty that John weighs more than 120 pounds.
Implicature: I can only say with certainty that John weighs more than 120
F
pounds.
b. I can say with certainty that John has more than 3 children.
Implicature: I can only say with certainty that John has more than 3
F
children.

(12) a. I was able to demonstrate that this refrigerator weighs more than 120 pounds.
Implicature: I was only able to demonstrate that this refrigerator weighs more than 120
F
pounds.
b. I was able to demonstrate that this candidate received more than 500 votes.
Implicature: I was only able to demonstrate that this candidate received more than 500
F
votes.

All these facts are expected under the UDM given the presence of the modal operator.
Specifically, even if more than n, and only, quantify over a dense domain, contradictions of the
sort weve seen in ( 2), ( 5), and ( 7) do not arise in the present case.
To see this, lets focus on ( 9)a. One can presuppose it to have been demonstrated that x
weighs more than 120 pounds, and subsequently assert consistently that there is no degree d,
greater than 120, such that it has been demonstrated that x weighs more than d pounds. The
reason for this is obvious: a demonstration that x weighs more than d pounds doesnt entail a
demonstration that x weighs d + pounds for some specific degree . In possible world
semantics, this fact would be captured by assuming a dense set of worlds corresponding to the
degrees. Once this is assumed, there could be for every a world consistent with the
demonstration such that in that world John weighs less than d + pounds. The fact that the (a)
and the (b) sentences continue to behave identically lends further support to the universality of
10
the UDM, i.e. to the claim that, contrary to traditional assumptions, the linguistic system treats
the (a) and the (b) sentences on a par. However, there are further issues to discuss pertaining to
the (b) cases to which we return in section 5.
Weve learned that the effect we are looking at cannot be attributed to a general inability of
numerals inside comparatives to associate with only or to generate implicatures, since this effect
can be ameliorated in the presence of a modal operator. However, it is also important to show
that the effect is not ameliorated by any modal operator, but only by those operators that can
eliminate the incoherence that is otherwise predicted under the UDM. An argument to this effect
is provided by the following contrast:

(13) a. You're required to read more than 30 books.
Implicature: There is no degree greater than 30, d, s.t. you are required to read more than d books.
b. You're only required to read more than 30
F
books.

(14) a. You're allowed to smoke more than 30 cigarettes.
*Implicature: There is no degree greater than 30, d, s.t. you are allowed to smoke more than d
cigarettes.
b. *You're only allowed to smoke more than 30
F
cigarettes.

In ( 13)a, the presence of a modal operator licenses an implicature and in ( 13)b it allows only to
associate with the numeral 30. In ( 14), by contrast, an implicature is absent, ( 14)a, and
association with only yields an incoherent interpretation, ( 14)b.
11
The difference between ( 13)
and ( 14) can be traced to the difference between the semantic properties of the two modal
operators required and allowed. If you are allowed to smoke more than 30 cigarettes, it follows
that youre allowed to smoke 30 + cigarettes, for some degree . This consequence would
contradict the potential implicature, which is equivalent to the sentence in ( 14)b. A similar
consequence does not follow when you are required to read more than 30 books. If you are
required to read more than 30 books, there need not be a degree , such that you are required to

11
The facts change for some speakers in a context in which it is presupposed that smoking a certain number of
cigarettes is required, i.e., that there is a minimal amount that needs to be smoked. For the relevant speakers, ( 14)b
can convey the information that 30 is the minimal number such that one is allowed to smoke any amount greater
than that number. Such a reading is not in conflict with the UDM. See our discussion of ( 34) and ( 35).
11
read 30 + books, and incoherence is avoided in exactly the manner weve discussed when we
accounted for ( 9)-( 11).
This intuitive difference can be traced to the fact that required is a universal modal and
allowed is existential. If the presupposition of ( 14)b is met, i.e., if you are allowed to smoke
more than 30 cigarettes, we can say that there is a world, w, (compatible with your requirements)
and there is a degree , such that you smoke 30 + cigarettes in w. From this it follows (by the
UDM) that in w, there is a degree greater than 30, 30 + /2, such that you smoke more than that
degree of cigarettes in w. Two existential quantifiers can commute without affecting truth
conditions. Hence there is a degree, d, greater than 30 such that you are allowed to smoke more
than d cigarettes, and this is the negation of ( 14)b.
When you are required to read more than 30 books, we say that for every world, w,
(compatible with your requirements) there is a degree , such that you read 30 + books in w.
From this it follows (by the UDM) that in w, there is a degree greater than 30 (30 + /2) such that
you read more than that degree of books in w. However, an existential and a universal quantifier
cannot commute without effecting truth conditions. Hence, we cannot conclude that there is a
degree, d, greater than 30 such that you are required to read more than d books. There could be a
dense set of worlds corresponding to the density of the set of degrees (given the UDM), and the
contradiction is avoided along the lines discussed above for the epistemic universal modals (can-
say-with-certainty, demonstrate).


3. Negative Islands: Definite Description and Questions
12


In the previous section we have seen that the UDM accounts for the impossibility of embedding
a comparative sentence under the operator only (when only associates with the degree
expression). We have seen that the same reasoning can be extended to account for the lack of
implicatures in comparatives under the assumption that implicatures are derived from
constructions that contain a covert exhaustive operator. In this section, we would like to study
the predictions of the UDM for additional operators with a similar interpretation.

12
This section is concerned with degree constructions. The new facts we report are problematic for existing
accounts of the basic paradigm. However, it is not clear whether our account extends to other negative islands, and
whether we are loosing important generalizations. In the future, we hope to investigate this in greater detail, in
particular in comparison with Szabolcsi and Zwart (1993).
12


3.1 Rullmann (1995)

Our starting point is the well-known puzzle that degree questions cant be formed by wh-
movement across negation.
13
This is exemplified by the contrast between the question in ( 15)
and in ( 16).

(15) John didnt read many of these books?
Question: Which books did John not read?
(16) John doesnt weigh 190 pounds.
Question: *How much does John not weigh?
14


Rullmann (1995) suggests a very interesting explanation for this contrast. He suggests that there
is nothing in syntax that rules out the question in ( 16). The sentence is generated by the syntax
but ruled out by the semantic component as a question that cannot have an answer.
To understand Rullmanns proposal consider a run-of-the-mill degree question, such as ( 17).
If John weighs exactly 150 pounds, then for any degree d smaller than 150, it is true to say that
John weighs at least d pounds. Nevertheless 150 pounds would be the only acceptable answer to
( 17).

(17) How much does John weigh?

One could imagine pragmatic accounts of this fact along Gricean lines. Rullmann suggests,
instead, that it should be captured in the semantics. Specifically, he suggests that degree
questions ask for the largest degree that satisfies a certain property:

(18) How much/many ?
What is the maximal degree d st. (d)?

13
Cf. Obenauer (1984), Rizzi (1990), etc.
14
As discussed in Kroch (1989) and indicated already in footnote 8, questions of this sort are acceptable when the
context provides a discrete set of alternatives. See section 5.5 for discussion.
13

The question in ( 17), for example, asks for the maximal degree d such that John weighs (at least)
d:

( 17)' How much does John weigh?
What is the maximal degree d such that John weighs (at least) d pounds?

Under Rullmanns proposal, this question has only one true answer in the scenario considered
above, namely 150 pounds.
15

This, as Rullmann points out (developing a claim that von Stechow (1984) has made in a
different context), could be the source of the unacceptability of the question in ( 16). If John
weighs exactly 150 pounds then any degree above 150 is a degree d such that John doesnt weigh
d pounds. Hence, as long as the set of degrees corresponding to weights has no upper bound,
there can be no maximal degree d such that John doesnt weigh d pounds, and the question,
therefore, cannot have an answer.
16
This, Rullmann suggests, is sufficient grounds for
unacceptability.
17

Our goal is to argue for a modification of Rullmanns account that depends on the UDM. But
before we get there, we would like to point out the contrast in ( 19), which provides further
support for the basic line of reasoning.

(19) a. *I have the amount of water that you don't.
cf. I have the amount of water that you do.
b. I have an amount of water that you don't.


15
Obviously the unacceptability of less informative answers is related to emergence of scalar implicatures.
Consequently, the decision to account for the former in the semantics is related to the decision to account for the
latter within the grammar.
16
The assumption that the weight scale has no upper bound seems natural. However, it is not essential for the
account. If there is an upper bound, Rullmann could say that a question such as the one in ( 16) cannot play the
communicative role that questions are designed for; the answer is known in advance to be the upper bound, and the
question, therefore, cannot serve to seek information. Note, however, that a little more would have to be said to
account for the unacceptability of the definite description in ( 19)a.
17
Of course, some questions that don't have an answer are, nevertheless, acceptable, e.g. What is the largest natural
number? Rullman's account, as well as our modification, implies that the grammar distinguishes between two types
of unanswerable questions (parallel to its distinction among two types of contradiction). See section 5 for discussion
as well as appendix 3.
14
The unacceptability of ( 19)a can receive the same explanation that Rullmann gives for the
unacceptability of the question in ( 16). Definite descriptions presuppose that the predicate they
combine with has a maximal element in its denotation (Link 1983):

(20) The is defined only if there is a maximal object x st. (x).
When defined, the refers to the maximal object x st. (x).

This presupposition is exactly identical to the requirement that needs to be satisfied for a degree
question to have an answer. The presupposition of ( 19)a, thus, cannot be met for the same reason
that the question in ( 16) cannot be answered. The fact that ( 19)b is acceptable suggests that the
syntactic object is not itself the source of unacceptability. Extraction of a degree argument across
negation yields unacceptability only when combined with a maximality requirement. Since an
indefinite comes with no maximality requirement, ( 19)b is acceptable.


3.2 Beck and Rullmanns Challenge

Rullmanns (1995) account for the fact in ( 16) is unique in predicting the contrast in ( 19) and is
therefore empirically preferable to competing accounts. More importantly, perhaps, it is a
principled account that does not rely on assumptions that are not needed on independent grounds.
It relies on the meanings of lexical entries, but those need to be specified anyhow.
However, Beck and Rullmann (1999) made an observation that casts serious doubt on
Rullmanns assumptions. In particular, Beck and Rullmann pointed out that if Rullmanns (1995)
assumptions about the meaning of degree questions are maintained, ( 21), just like the question in
( 16), should be unacceptable.

(21) How much flour is sufficient to bake a cake?

If exactly 3 pounds of flour are sufficient to bake a cake then any degree d above 3 pounds will
suffice as well. Hence, there cannot be a maximal degree d such that d-much flour is sufficient to
bake a cake. This is true for precisely the same reason that there cannot be maximal degree d
15
such that John does not weigh d pounds. However, ( 21) is acceptable. Moreover, the correct
answer to ( 21) would specify the minimal amount of flour sufficient to bake a cake, rather than
the maximal.
Beck and Rullmann suggest the following characterization as a unifying statement of what
constitutes a good answer to a (degree) question: A (degree) question of the form How
many/much d (d) should be answered by the most informative degree d that satisfies (where d
is the most informative degree satisfying if d satisfies and for every d' that satisfies the
proposition that d satisfies is more informative than the proposition that d' satisfies ).
If, following Rullmann (1995), this statement is to be captured in the semantics of degree
questions, we would have to modify ( 18), repeated in ( 22)a. One modification that suggests itself
is given in ( 22)b.
18


(22) How much/many ?
a. Rullmann (1995):
What is the maximal degree d st. (d)?
b. modification based on Beck and Rullmann (1999):
What is the degree d that yields the most informative among the true propositions of
the form (d)?
19


The modification accounts for the meaning of ( 21). However, this comes at the expense of
loosing the principled account of the contrast in ( 19) and of the unacceptability of the question in
( 16). For example, at the moment one would expect ( 16) to be an acceptable question requesting
the identification of the minimal degree d such that John doesnt weigh d pounds.
In the next section, we will argue that once the consequences of the UDM are factored in, a
principled account very much in the spirit of Rullmann (1995) is possible after all. But before we
get there, we would like to point out an observation due to von Fintel, Fox and Iatridou (in

18
Another possible modification is already in present in Groenendijk and Stokhof (1984). Such a modification,
however, is not going to allow us to develop our Rullmann-like account of the constraints on degree questions we
are concerned with.
19
This modification would follow from Dayal's (1996) more general approach to the semantics of questions. See
appendix 3 for complications and a possible amendment.
16
progress). This observation suggests that if such an account were possible for the question in
( 16), it would carry over to the definite description in ( 19).

Consider the definite description in ( 23). The standard semantics of the definite article due to
Link (1983) predicts that this definite description would be unacceptable for the same reason that
Rullmann predicts that ( 21) should be unacceptable.

(23) I have the amount of flour sufficient to bake a cake.

The definite description in ( 23) will always yield a presupposition failure since there could never
be a maximal amount of flour sufficient to bake a cake. In order to deal with this and similar
observations, von Fintel, Fox and Iatridou suggest a semantics for the definite article very much
like the modification of Rullmann for questions that we presented in ( 22)b. From this semantics
it follows that a definite description should have the following meaning:
20


(24) The is defined only if there is a unique individual x such (x) is a maximally
informative proposition among the true propositions of the form (x).
When defined the refers to the individual x st. (x) is the maximally informative true
proposition of the form (x).

This semantics is of course in exactly the same predicament as the modification of Rullmann
presented in ( 22)b. Both have the advantage of providing a unified account for the variation
between a maximality and a minimality effect. This variation is determined in a predictable way
from the overall semantics of the relevant property (in particular its monotonicity).
21
Both,

20
The required lexical entry is the following:
[[the]]= P
<e,st>
.w:x[P(x)(w)=1 and yx (P(y)(w)=1 P(x) asymmetrically entails P(y))].
(x)[P(x)(w)=1 and yx (P(y)(w)=1 P(x) asymmetrically entails P(y))].
Where :(). is a function defined only for objects of which is true (convention from Heim and Kratzer
1998).
21
If is a property of degrees or individuals that is upward monotone (in the following sense: x > y iff (x) is more
informative than (y)), then the according to ( 24) would pick up the maximal individual x such that (x), when
defined. Similarly, how many/much ? would ask for the identity of this maximal individual. Conversely for
downward monotone properties (x < y iff (x) is more informative than (y)): if is such a property, the would
refer to the minimal individual x such that (x), and how many/much ? would ask for the identity of this minimal
17
however, render Rullmanns original explanation of the negative island effects in definites and
questions [( 16), ( 19)a] unavailable.
The goal of the remainder of this section is to argue that the UDM allows us to capture
Rullmanns original insight while incorporating the necessary changes in the semantics of
questions and definite descriptions. We will do this in two stages (following the logic outlined in
the introduction). We will start in 2.3 by focusing on domains where the account is fairly
intuitive because the domains are intuitively dense. We will then move to domains where the
account is unintuitive in the sense that our account of the lack of an implicature in ( 2) and the
unacceptability of ( 5) was initially unintuitive. Once again, we argue that the facts motivate the
assumption that despite initial appearances all degree domains are formally dense (i.e. treated
as dense by the grammar).


3.3 Negative Islands and Dense Domains

Rullmann (1995) suggested that degree questions (and definite descriptions) are sensitive to
negative islands because negation stands in the way of satisfying a maximality requirement.
However, we have seen that the maximality requirement cannot be right in exactly the way stated
by Rullmann. This appeared to be a problem for Rullmanns original account. Nevertheless, we
will now see that, once the density of the relevant degree domains is taken into account, an
alternative emerges. Also in this alternative maximality is the major player, but since the notion
of maximality has changed, there will be quite a few empirical differences.
Consider again the question in ( 16) repeated below.

(25) *How much does John not weigh?

If the maximality requirement is modified as in ( 22)b, ( 25) should ask for the maximally
informative proposition among the true propositions of the form John does not weight d-pounds.
If John does not weigh 190 pounds then it follows that for any degree d bigger than 190, John

individual. Von Fintel, Fox, and Iatridou argue for their semantics based on paradigms in which the monotonicity of
properties is varied. For these paradigms, Links lexical entry and the alternative that we provided in footnote 20
make different predictions.
18
does not weigh d pounds. In other words, propositions of the form John does not weigh d-pounds
become more informative the smaller d is. Hence the most informative proposition of this form is
the proposition that John does not weigh d pounds where d is the minimal degree d for which this
proposition is true.
Given the discussion in section 2, it is probably clear where we are going. There is no
minimal degree that yields a true proposition of the relevant form, since the relevant set of
degrees is dense. To see this, suppose that John weighs exactly 150 pounds. This means that for
any degree d in the set of degrees greater than 150, John doesnt weigh d pounds. From density it
follows that there is no minimal member in this set. Hence there is no most informative answer
to the question, and Rullmanns account can be maintained.
Exactly the same reasoning accounts for the unacceptability of ( 19)a. However, there are
further predictions. One useful way to think of these predictions is to compare our maximality
account to an account that assumes a syntactic ban on extraction across negation. As we will see,
there are cases where a question or a definite description meets the relevant maximality
requirement despite the fact that negation is crossed by the wh-operator. These cases involve a
negative island, [+NI], but do not violate the maximality requirement, [-MV]). Conversely, there
are cases where the maximality requirement is violated despite the fact that negation is not
crossed by the wh-operator, [-NI, + MV]. Because maximal informativeness is the driving force
in our account, we predict that cases of the first kind should be acceptable (despite the fact that a
wh-operator crosses negation) while cases of the second kind shouldn't be (even though a wh-
operator does not cross negation). In short, we can distinguish our account from the syntactic
alternative by the following predictions:

(26) a. Prediction 1: [+NI,-MV] cases should be acceptable.
b. Prediction 2: [-NI,+MV] cases should be unacceptable.

We will attempt to corroborate both predictions.

Lets start with prediction 1. We have already seen one piece of evidence that the prediction is
correct. The indefinite in ( 19)b involves a negative island without a maximality violation and is
acceptable as predicted. But we are now in a position to conduct additional tests.
19
Consider the data in ( 27) originally noted in Kuno and Takami (1997).
22
Despite the fact that
a wh-operator is extracted over negation in each case ([+NI]), all examples are acceptable.
23


(27) a. How much are you sure that this vessel wont weigh?
b. How much radiation are we not allowed to expose our workers to?
c. The amount of radiation that we are not allowed to expose our workers to is greater
than we had thought
d. The amount of money that you are sure that this stock will never sell for is quite high.
(Are you sure that your estimation is correct.)

Notice further that in all four examples it is the minimal degree that plays a prominent role; it is
the referent of the definite descriptions in ( 27)c-d and the object whose identity is to be specified
if the questions in ( 27)a or b are to be answered. Consider ( 27)a. With this question the addressee
is asked to specify an upper bound on possible weights of the vessel. But among the various
upper bounds that she could specify, which one does she have to specify in order to satisfy the
speaker? Of course it is the minimal upper bound: the minimal degree d such that the addressee
is sure that the vessel will not weigh d pounds.
These are all [-MV] cases even when density is assumed and their acceptability is therefore
predicted. To see this, lets continue with our discussion of ( 27)a. Even if the domain of degrees
is dense, as we are assuming, there could be a minimal degree d such that the addressee is sure
that the vessel wont weigh d pounds. Even though there can be no minimal degree d such that
the vessel doesnt weigh d pounds, there could be a minimal upper bound to possible weights.
We can make sense of this fact using possible world semantics as we did in section 2. We
take it to be necessary that there is a degree d such that the vessel weighs exactly d pounds. In
other words, the set of degrees d such that the vessel weighs at least d pounds is necessarily a
closed interval. Consequently, given density, the complement set is necessarily an open interval,
which cannot have a minimal member. However, the set of degrees d such that there is certainty

22
Thanks to Maria Luisa Zubizarreta for reminding us of this paper.
23
To see that ( 27)a is acceptable, imagine a context in which it is obvious that the addressee does not know how
much the vessel will weigh but she can put an upper bound on possible weights. (Imagine for example a freighter
with an amount of cargo that varies over time and cannot be totally determined in advance.) The question in ( 27)a
could be presented to a logistics officer in order to figure out the cheapest way to protect the vessel from various
hazards (assuming that the cost of protection is a function of weight).
20
(on someones part) that the vessel doesnt weigh at least d could be a closed interval with a
minimal member. To see this, consider the fact that this set could also be described as the set of
degrees d for which there is no possibility that the vessel weighs d pounds. This set is the
complement set of the set of degrees for which there is a possibility that the vessel weighs d
pounds and this set in turn could be an open interval. For example, it will be an open interval if
there is a degree d such that the vessel cant weigh d pounds, but for any smaller degree d', there
is a world (in the modal base) in which the vessel weighs exactly d' pounds. We leave it to the
reader to see that the same reasoning applies to ( 27)b-d.
Clearly the prediction we make for the examples in ( 27) depends on the introduction of a
further operator in addition to negation, which allows the examples to be of the [-MV] type. But
as we have seen in section 2, not any operator can do the job. To see that the same holds here,
consider the following pairs.

(28) a. How much radiation is the company not allowed to expose its workers to?
b. *How much food is the company not required to give its workers?
24


(29) a. How much radiation is the company required not to expose its workers to?
b. *How much food is the company allowed not to give its workers?

We have already seen why the a-questions are [-MV] despite the density of the degree domains.
We will now see that this is not the case for the corresponding b-questions.
25
Notice that ( 28)b
and ( 29)b are equivalent. So it is sufficient to show that one of them cannot satisfy our
maximality requirement. Suppose that there is an answer to ( 29)b. Lets say the answer is 3
bowls of rice. Assume that it is true that the company is allowed not to give its workers three
bowls of rice. In other words, assume that there is an allowed world, w, in which the company
doesnt give its workers 3 bowls of rice. Lets say that in w the company gives its workers
exactly 3- bowls of rice (0< 3). But given density there is a more informative answer. In w


24
The (b) sentences are acceptable on a reading which does not result from degree extraction across negation: what
is the amount of food, d, such that there is food x in amount d and the company is not required to give x to its
workers?. To avoid this confound, we can compare the following:
(i) (When you enter the country) How much money are you not allowed to have?
(ii) *(When you enter the country) How much money are you not required to have?
25
As we will see in section 4, this contrast follows from exactly the same logic that was employed in our account of
( 13) and ( 14).
21
the company doesnt give its workers 3-/2 bowls of rice, and the proposition that this is allowed
is more informative. We see the same contrast with definite description and the account is of
course the same.

(30) a. The amount of radiation that the company is not allowed to expose its workers to is
very high.
b. *The amount of food that the company is not required to give its workers is quite
high.

(31) a. ?The amount of radiation that the company is required not to expose its workers to
b. * The amount of food that the company is allowed not to give its workers

We can now move to test the second prediction in ( 26). If there are cases where a question or
definite description doesnt meet the relevant maximality requirement despite the fact that
negation is not crossed by the wh-operator ([-NI,+MV]), our account makes a clear prediction:
such cases should be unacceptable. We will show that one such case exists under a semantics
that postulates a dense domain of time and that it is in fact unacceptable. Furthermore, we will
employ a universal modal operator in the same way we did above to argue that density is indeed
the crucial factor that determines the acceptability status of the relevant examples. Finally, we
will use, once again, the definite/indefinite contrast to argue that maximality is the determining
factor. In section 4.2 we will discuss another ([-NI,+MV]) case, which directly relates to our
discussion in section 2.
Consider the fact in ( 32) noted in von Fintel and Iatridou (2003). One could imagine that
there is a syntactic constraint that blocks extraction of a before-phrase. However, this will not be
a negative island. Barry Schein (p.c.) suggests that the question is ruled out as a maximality
violation and hence corroborates our prediction that [-NI,+MV] cases are unacceptable.
26


(32) *Before when did John arrive?


26
Schein made his suggestion during the question period of a talk presented by Kai von Fintel and Sabine Iatridou at
MIT. At the time, we were just beginning our work on the UDM.
22
If the domain of time is dense, just like the domain of degrees, and if when-questions require a
similar semantics to how-many-questions, we have a straightforward explanation for the
unacceptability of ( 32). To see this, assume that John arrived at 10 am.
27
It follows that every
time after 10 am is a time t such that John arrived before t. The proposition that John arrived
before 11 am is more informative than the proposition that John arrived before 11:30 am. Hence,
under the assumption that maximal informativeness plays a role in the semantics of when-
questions, the appropriate answer for ( 32) would specify the earliest time t such that John arrived
before t. I.e. the earliest time after 10 am. However, if density is assumed, there is no such
time.
28

This account makes a further prediction, which is by now quite familiar. If we add a
universal modal operator (in the appropriate place) to sentences such as ( 32), they will no longer
constitute [+MV] cases. They will, therefore, be predicted to improve in status, a prediction
verified in ( 33).

(33) Before when do you have to arrive?

To understand why there is no maximality violation in ( 33) it is sufficient to realize that even if
the time line is dense, there could be a time t which is earliest in the set of times t' such that you
have to arrive before t'. To see that there could be such a time, imagine that the rules which
constrain your arrival time specify that it has to be before a certain time, say 10 am, but dont
specify anything other than that. Under such a scenario, there is a most informative time such
that you have to arrive before that time.
29

The question in ( 34), which employs an existential modal in place of a universal, appears to
be problematic for our generalization that existential modals are not capable of obviating a
maximality violation. Consider ( 34), when addressed to a person who is staying at a Youth
Hostel which is locked up over night, say at 10 pm. To get in, one has to arrive before 10pm.
( 34) seems to be a natural way to ask for the time at which the entrance doors are locked.

27
If John did not arrive at all, then of course the question does not have an answer.
28
Notice that the expression we use in the text, the earliest time after 10 am, is not illformed, although at first
sight it should be, if density is assumed. We thank Philippe Schlenker for discussion of this issue, which we address
in section 5.4.
29
In possible world semantics we would say that the set of worlds that conform to the rules is the set of worlds in
which you arrive before 10 am. This is a set that corresponds to the dense set of times: for every time t before 10 am
there will be a world w such that you arrive at (exactly) t in w.
23

(34) Before when are you allowed to arrive?

On closer examination, however, this is not problematic for our proposal. The presence of a
modal operator makes a richer contribution to the semantics of ( 34) than that of simple
existential quantification over worlds. Once the full contribution is taken into account, it turns
out that an obviation of a maximality violation is predicted after all.
To see what is involved, consider the true answer to ( 34):

(35) You are allowed to arrive before 10 pm

This answer is much stronger than the statement that there is a world (compatible with the
relevant rules) in which there is a time before 10 at which you arrive. The later statement would
be true if you were allowed to arrive at 7 and no later (from which it follows that there is a time
before 10 at which you are allowed to arrive). ( 35), however, would not be true in such a
situation. ( 35) requires the rules to allow you to arrive any time before 10 pm. The source of this
free choice interpretation has been debated in the literature quite extensively. However, it is
most likely not pertinent for our purposes. What is important is that the existence of the free
choice interpretation allows the question to receive a (most informative) answer.
In our example, 9 pm is a time t such that you can arrive any time before t. 10 pm is such a
time as well, and the proposition that this is the case is more informative. However, for any time
t after 10 pm it is no longer true that you are allowed to arrive (any time) before it. 10pm is
therefore the latest such time and would furnish the maximally informative answer to ( 34).
Finally, the definite/indefinite contrast that we saw with degree extraction, ( 19), emerges in
this case, as well, thereby providing additional support for our claim that the unacceptable cases
with before are instantiations of [-NI,+MV] :
30


30
We can use definites with resumptive pronouns to constructs examples with differentials, similar to those
discussed in section 2:
(i) a. *I can predict that you will arrive at the time such that John will arrive before it.
b. I can predict that you will arrive at the time such that John will arrive exactly 20 seconds before it.

There are further tests that we might conduct, and quite a few questions that arise when density is assumed in the
domain of time, which we hope to investigate in the future. Our goal here is to understand the consequences of the
density assumption for the domain of degrees (the UDM). The domain of time serves a limited function from this
24

(36) a. *I arrived at the time before noon.
b. ?I arrived at a time before noon.


3.4 On the Universality of the UDM

In 2.2 we have seen evidence supporting our claim that intuitively dense domains are in fact
dense. Specifically, the claim enabled us to account for what initially seemed to be a syntactic
constraint against extraction over a negative expression. The constraint turned out to be an
artifact of a maximality requirement along the lines of Rullmann (1995). However, the source of
the maximality violation for us was very different. This allowed us to derive a host of new
predictions concerning the types of modal expressions that should obviate the effect.
In this section, we would like to argue that this line of reasoning should be extended to all
degree domains, even those that seem at first sight to be based on the non-dense notion of
cardinality. This will serve as an additional argument (to the one presented in section 2) in favor
of the universality of the UDM.
It is well-known that the problem of extracting a degree wh-phrase across negation is not
limited to intuitively dense domains. This can be illustrated by the unacceptability of ( 37).
31


(37) *How many kids do you not have?

If we are right in our account of the parallel fact in dense domains, we are forced either to
claim that the two types of facts have a different source or that (despite initial appearances) all
degree domains are dense. Just as we did in section 2, we argue for the second alternative based
on the observation that the two types of constructions (corresponding to the intuitively dense and

perspective: it lends plausibility to the maximality account of negative islands, which constitutes one of our central
arguments for the UDM.
31
We use the definiteness effect in possessive have constructions to insure that a degree expression is indeed
extracted out of the scope of negation, i.e., in order to exclude a representation in which the degree variable starts
out outside the scope of negation and an individual variable is present in the theta position (see footnote 24, as well
as Heim 1987, and Frampton 1991). In the French examples below we use a similar technique based on a particular
construction that the language makes available.
25
the intuitively non-dense domains) behave identically on a variety of tests, and that this
otherwise mysterious behavior is explained under the UDM.
The facts in ( 38)-( 41) are a direct consequence of the maximality requirement and density if
the UDM is assumed, but are quite mysterious otherwise. The examples in ( 38) show that the
constraint against extraction across negation exemplified in ( 37) is obviated when certain modals
are introduced in the appropriate location, specifically the combination of modals and locations
that can circumvent a maximality violation.

(38) a. If you live in China, how many children are you not allowed to have?
b. How many days a week are you not allowed to work (according to the union
regulations)?
c. How many soldiers is it (absolutely) certain that the enemy doesnt have?

Under the UDM, this receives an identical explanation to the one weve given for parallel
examples involving intuitively dense domains, e.g., ( 27), ( 28)a, and ( 29)a.
32
Parallel facts in
French involving the much discussed split combien construction are provided in ( 39) and ( 40).
(See Obenauer 1984 and much subsequent work.)

(39) a. ???Combien Jean n'a-t-il pas lu de livres?
How many John nhas-he not read of books
b. ? Combien peux-tu me dire avec (absolue) certitude que Jean n'a pas lu de livres?
How many can-you me tell with (absolute) certainty that John has not read of books
(Benjamin Spector, p.c.)
(40) a. *Combien Jean n'a-t-il (pas) denfants?
How many John nhas-he not of children
b.? Combien les chinois ne peuvent ils (pas) avoir denfants?
How many the chineese nallowed-them not have of-children?
(Valentine Hacquard, p.c.)


32
The reader might have noticed a problem pertaining to the precise interpretation of ( 38)c. We will argue in section
5, that this problem is identical to the one alluded to on page 9 pertaining to the truth conditions of ( 11)b and ( 12)b.

26
Once again, the overall logic of the account is corroborated by the observation that only the
modals that can circumvent a maximality violation will allow the relevant degree operator to
cross negation:

(41) a. *If you live in Sweden, how many children are you not required to have?
b. *How many days a week are you not required to work (even according to the
companys regulations)?
c. *How many soldiers is it possible that the enemy doesnt have?


4. Exhaustivity and Density: a Unified Account
33


In the previous two sections we have seen 4 pieces of evidence for the UDM coming from the
study of implicatures, only, questions and definite descriptions. In this section we will see that
there is a level of description under which all of these exemplify the same generalization. Once
we see that, we will be able to draw additional predictions.
Lets start with a more formal characterization of the results from section 2. Let be an
upward monotone property of degrees (of type <d,st>), where is upward monotone if for every
d
1
,d
2
: d
1
>d
2
iff (d
1
) is more informative than (d
2
) (asymmetrically entails it). Assume also that
necessarily describes an open interval, i.e. for every world w, d.(d)(w) is an open interval.
(There is a degree d such that (d)(w)=0 but for every degree d' smaller than d, (d')(w)=1.)
34

The following drawing represents what holds of in every world:


is false
is false

is true



33
We thank an anonymous reviewer for useful suggestions on how to make this section more reader friendly.
34
For the sake of discussion, we assume that degrees are points on a scale rather than intervals. If degrees were
intervals (as assumed in Kennedy 2001), our predicate would describe an open interval of intervals. We also
ignore the pathological case where is an empty interval in some world.
27
For every world w, it will be false to say that there is some degree d such that it is the maximal
degree such that is true of that degree in w. Equivalently (given upward monotonicity), it will
be false to say that there is a degree d such that (d)(w) is true and (d) is more informative than
(d') for every d' such that (d')(w) is true.
These logical facts are at the heart of the account of our observations in section 2. A simple
property of degrees based on comparatives, e.g. d. John has more than d children, is upward
monotone. Assuming the UDM it also has to describe an open interval. Hence, it follows that a
statement that there is a maximal (or maximally informative) degree that satisfies it would be
infelicitous (whether it results from only or a covert exhaustive operator).
Furthermore, we can observe that if we append a universal modal operator, the modified
property, d. (d), would no longer necessarily describe an open interval. The universal modal
could quantify over a dense set of worlds that corresponds to the set of degrees with the result
that the modified property describes a closed interval. For example, given a constant a, the modal
base could consist of all worlds in which is true of some degree bigger than a, i.e. MB = {w:
d>a [(d)(w)=1]}. [A useful aside: MB is precisely the deontic modal base that we would have
in situation in which the only requirement is that (a) hold. Suppose that w MB. Given
upward monotonicity, (a)(w)=1. Conversely, suppose (a)(w)=1. Since necessarily describes
an open interval, a cant be the maximal element in its extension. Hence, there is some d > a such
that (d)(w)=1. Hence w MB.]
Under MB, d.(d) will describe the closed interval I = [0,a]. To see this, suppose first
that d' is in I. Let w MB. Hence, there is a d>a such that (d)(w)=1. Since d>d' and is
upward monotone, (d')(w)=1. This is true for every w MB. Hence, (d') = 1. Suppose next
that d' is not in I (i.e. d' = a + ). Since (d') asymmetrically entails (a+/2), there is a world, w,
in MB, s.t. (d')(w) = 0. Hence (d') = 0. This means that complex predicates over degrees
such as d. you are required to read more than d books do not necessarily describe an open
interval. And this, we suggest, accounts for the fact that such predicates can be the source of an
implicature or a parallel sentence with only.
Appending an existential modal operator, on the other hand, is of no help: the property
d.(d) necessarily describes an open interval. To see this, assume that d. (d) describes a
closed interval (in a world w) and that d' is its maximal member. Of course in w (d')=1. This
28
means that there is a world w' in the modal base such that (d')(w')=1. But since necessarily
describes an open interval, there is a degree d'' bigger than d' such that (d'')(w')=1. Hence, in w,
(d'') =1 and d' is not the maximal member. This, of course, explains the fact that complex
predicates over degrees such as d. you are allowed to smoke more than d cigarettes cannot
support an implicature and are unacceptable with only.
Lets now move to a similar characterization of the results from section 3. Let now be a
downward monotone property of degrees (of type <d,st>), where is downward monotone if for
every d
1
,d
2
: d
1
>d
2
iff (d
2
) is more informative than (d
1
). Assume also that necessarily
describes an open interval, i.e. for every world w there is a degree d such that (d)(w)=0 but for
every degree d' bigger than d (d')(w)=1. The following drawing represents what holds of in
every world:



is true is false


is false


For every world w it will be false to say that there is some degree d such that it is the minimal
degree such that is true of that degree in w. Equivalently (given downward monotonicity), it
will be false to say that there is a degree d such that (d)(w) is true and (d) is more informative
than (d') for every d' such that (d')(w) is true.
These logical facts are once again at the heart of our account. In section 3, we talked about
downward monotone predicates of degrees that (given the UDM) necessarily describe open
intervals such as d. I dont have d children. We pointed out that a statement that there is a
minimal (or maximally informative) degree that satisfies such a property would be infelicitous.
This time, the source of the infelicity was a question or a definite article.
Once again, we can observe that if we append a universal modal operator, the modified
property, d.(d), would no longer necessarily describe an open interval. The universal modal
could quantify over a dense set of worlds that corresponds to the set of degrees with the result
that the modified property describes a closed interval.
29
The example we use to show this is basically the mirror image of the one we constructed for
upward monotone predicates. Given a constant a, we define the modal base as the set of worlds
in which is true of some degree smaller than a, i.e. MB = {w: d<a [(d)(w)=1]}. [Again, a
useful aside: MB is precisely the deontic modal base that we would have in situation in which
the only requirement is that (a) hold. Suppose that w MB. Given downward monotonicity,
(a)(w)=1. Conversely, suppose (a)(w)=1. Since necessarily describes an open interval, a
cant be the minimal element in its extension. Hence, there is some d < a such that (d)(w)=1.
Hence w MB.]
Under MB, d.(d) will describe the closed interval I = [a,). To see this, suppose first
that d' is in I. Let w MB. Hence, there is a d<a such that (d)(w)=1. Since d<d' and is
downward monotone (d')(w)=1. This is true for every w MB. Hence, (d') = 1. Suppose
next that d' is not in I (i.e. d' = a - ). Since (d') asymmetrically entails (a-/2), there is a world,
w, in MB, such that (d')(w) = 0. Hence (d') = 0. This means that complex predicates over
degrees such as d. you are required to read more than d books do not necessarily describe an
open interval. And this, we suggest, accounts for the fact that such predicates can be the source
of an implicature or a parallel sentence with only.
Appending an existential modal operator is again of no help: the property d. (d)
necessarily describes an open interval. To see this, assume that d. (d) describes a closed
interval (in a world w) and that d' is its minimal member. Of course, in w, (d')=1. This means
that there is a world w' in the modal base such that (d')(w')=1. But since necessarily describes
an open interval, there is a degree d'' smaller than d' such that (d'')(w)=1. Hence, (d'') =1 and
d' is not the minimal member. This will then explain the fact that complex predicates over
degrees such as d. you are allowed not to work d days (which is equal to d. you are not
required to work d days) cannot serve as arguments of definite articles or question operators.
Of course there is a generalization here that pertains to all monotone properties of degrees
(whether they are upward or downward monotone). When these properties describe open
intervals in every possible world, we will call them necessarily open properties, or in short N-
open properties. If a property is N-open (in the relevant direction: top for upward monotone and
bottom for downward monotone, i.e. always at the more informative end), maximization should
30
be impossible. This follows if maximization always makes use of a primitive, MAX
inf
, which
selects the most informative member of an interval.

(42) Constraint on Interval Maximization (CIM): N-open monotone properties cannot be
maximized by MAX
inf.

(43) MAX
inf.
(
,st
)(w)= (x)[(x)(w)=1 and y((y)(w)=1 (x) entails (y))].

We take the CIM to be self-evident. Our paper uses it to argue for specific claims about the
nature of degree domains in natural language. The fact that the CIM plays an explanatory role in
accounting for the status of expressions in natural language can be taken as evidence that natural
language has N-open properties, i.e. for our first claim that certain degree domains in natural
language are dense. The fact that the CIM seems to be at work in all degree constructions (even
those that putatively make reference to cardinality) constitutes our argument for the universality
of the UDM.
But we can now see additional predictions. We have isolated four operators that make use of
MAX
inf
: the definite article, the question operator how many/much, only and the exhaustivity
operator, exh (responsible for implicatures):
35


(44) a. [[exh]] ()(d)(w) d = MAX
inf
()(w).
b. [[only]] ()(d)(w) d = MAX
inf
()(w), when defined.
c. [[?]] = w:d[d=MAX
inf
()(w)].{(x): x D

}
d. [[the]] ()(w) = MAX
inf
()(w)

These operators allow us to state four specific consequences of the CIM for three different
environments:

(45) Basic Consequence:

35
Later on in the paper, we talk about only and exh as focus sensitive operators that take a proposition p and a set of
alternatives A (usually the focus value of its sister). This way of talking translates automatically to ( 44)a and b as
long as xD

[p=(x)] & qA[yD

[q=(y)]. Under such circumstances, only

focus-sensitive
(A)(p) = only()(x).
Both ways of talking will, of course, require room for further domain restriction. See appendix 1 for discussion of
domain restriction under a syntax that assumes the focus sensitive only.
31
If expresses an N-open monotone property of degrees, then the following should be
unacceptable
36

a. *exh (d)
b. *only (d
F
)
c. *wh
d
(d)
d. *the d.(d)

(46) Consequence for universal modals:
A universal modal can close an interval, hence even if is an N-open monotone property
of degrees, the following should be acceptable
a. exh (d)
b. only (d
F
)
c. wh
d
(d)
d. the d.(d)

(47) Consequence for existential modals:
An existential modal cannot close an interval. Hence, if is an N-open monotone
property of degrees, then the following should be unacceptable
a. *exh (d)
b. *only (d
F
)
c. *wh
d
(d)
d. *the d.(d)

In section 2 we saw evidence for the UDM under the assumption that a and b hold for
upward monotone degree properties. In section 3, we saw parallel evidence under the assumption
that c and d hold for downward monotone properties. But we can now search for additional
predictions. Specifically, we might find cases in which a and b hold for downward monotone
properties or conversely cases in which c and d hold for upward monotone properties.


36
These are of course sloppy descriptions of syntactic representations which we hope are nevertheless useful.
32

4.1 Implicatures, Only, and Negation

It is well known that scalar implicatures are predicted to reverse in downward entailing
environments. (See Horn 1972, Fauconnier 1975, Levinson 2000, and Chierchia 2004, among
others.) To see this, consider the scalar implicatures of the sentences in ( 48). ( 48)a, with the
existential quantifier some, has the implicature that the corresponding sentence with the universal
quantifier all is false. Similarly, ( 48)b, which contains disjunction, has the implicature that the
corresponding sentence with conjunction is false.

(48) a. John did some of the homework.
Implicature: John didnt do all of the homework
b. John talked to Bill or Sue.
Implicature: John didnt talk to both Bill and Sue.

In downward monotone environments, we get what appears to be the opposite pattern. ( 49)a,
with the universal quantifier all, has the implicature that the corresponding sentence with the
existential quantifier any is false.
37
Similarly, ( 49)b, which contains conjunction, has the
implicature that the corresponding sentence with disjunction is false.

(49) a. John didnt do all of the homework.
Implicature: John did some of the homework.
b. John didnt talk to both Bill and Sue.
Implicature: John talked to one of the two.

This reversal is expected under traditional Neo-Gricean accounts, where an implicatures is a
denial of a stronger alternative that the speaker could have uttered. For us, the reversal is derived
on the basis of a covert exhaustivity operator (exh) which makes use of MAX
inf
. Exh will be
generated with matrix scope and will take the audible sentence as its complement (the prejacent).
As mentioned, this operator asserts that the prejacent is the most informative true member of a
set of alternatives. Assuming that a sentence with a corresponding scalar alternative is a member

37
The corresponding sentence with the Positive Polarity Item some is not grammatical.
33
of the set, the implicature that this alternative sentence is false follows, as long as it is not weaker
than the prejacent.
Furthermore, since an implicature involves the exclusion of strong alternatives, an alternative
that is too weak to be excluded in an upward monotone environment will yield a strong (hence
excludable) alternative in a downward monotone environment. From this perspective, it is
somewhat surprising that the sentences in ( 50)a-b do not give rise to parallel implicatures.
38


(50) a. John didnt smoke 30 cigarettes.
*Implicature: John smoked 29 cigarettes.
b. John didnt read 30 books.
*Implicature: John read 29 books.

We would like to argue for an explanation of this fact based on the UDM. Of course, this fact
could also be explained if the crucial alternatives are somehow disqualified form the alternative
set. More specifically, if it could be ensured that the scalar word 29 is not considered as an
alternative to 30 (when computing the strong meaning of ( 50)), the implicature would not be
predicted (thanks to Polly Jacobson, p.c.). However we dont see an obvious way to justify this
move. It seems to us that if 29 is excluded when the strong meaning of ( 50) is computed, the
same should hold for 31 when computing the strong meaning of John smoked 30 cigarettes. In
other words, we will not understand why John smoked 30 cigarettes has the implicature that it
does, namely that John smoked no more than 30 cigarettes.
39

The UDM provides an immediate account for our puzzle. For the sentences in ( 50) to have an
implicature, exh would have to be employed. The resulting interpretation would be equivalent to
the conjunction of the standard semantics of the prejacent and the assertion that all stronger
alternatives are false. For example ( 50)a would make the assertion in ( 51).

(51) John didnt smoke 30 cigarettes and for all degrees d smaller than 30 John smoked d
cigarettes.

38
The observations in this section grew out of attempts to grapple with facts noted by B. Spector (2004).
39
For some reason that is not completely clear to us, the facts seem to change for some speakers when a non-round
numeral is used. We dont understand this phenomenon, but hope that it can be made consistent with our proposal
once the pragmatic considerations that enter into choosing a level of precision (e.g. Krifka 2002) are taken into
account perhaps along the lines discussed in section 5.5.
34

( 51) is of course contradictory if the UDM is assumed. As is by now familiar, if John didnt
smoke 30 cigarettes, then the exact degree, d, of cigarettes that he smoked is below 30. By the
UDM, there has to be a degree d' between 30 and d. Since John didnt smoke d' cigarettes, ( 51)
cannot be true.
Under the UDM, this is just an instantiation of ( 45)a for downward monotone properties. The
UDM also makes a prediction based on ( 46)a, namely that the problem with adding the
exhaustive operator should disappear with the introduction of a universal modal. Consider ( 52)
and ( 53).

(52) John is not allowed to smoke 30 cigarettes.
Implicature: John is allowed to smoke 29 cigarettes.

(53) John is required not to smoke 30 cigarettes.
Implicature: John is allowed to smoke 29 cigarettes.

These sentences are equivalent given the duality of allow and require, and, as is transparent in
( 53), they involve appending a universal modal to a downward monotone N-open property. Since
the universal modal can close an interval, maximization is predicted to be possible and the
implicature is predicted to be acceptable. These cases serve another purpose, namely they
indicate that the scalar word 29 can be an alternative to 30 in a negative context, contrary to the
competing proposal that we sketched above.
Finally, we make the prediction based on ( 47) that an existential modal will not obviate the
CIM:

(54) John is not required to read 30 books.
*Implicature: John is required to read 29 books.
(55) John is allowed not to read 30 books.
*Implicature: John is required to read 29 books.

35
Parallel facts are predicted for only (instantiating the b cases in ( 45)-( 47)). The predictions
seem to go in the right direction.
40
In ( 56) we see that only can associate with scalar items across
negation. These examples are not perfect (hence our marking %, which reflects considerable
disagreement among speakers), but contrast markedly with the sentences in ( 57), which violate
the CIM under the UDM. This contrast corroborates the predictions of the UDM based on ( 45)b.

(56) a. %John has only not done all
F
of the homework.
b. %John has only not spoken to both
F
Bill and Sue.

(57) a. *John has only not smoked 30
F
cigarettes.
b. *John has only not read 30
F
books.
c. *John only does not weigh 190
F
pounds.

The addition of a universal modal, ( 59), and the move to the equivalent construction with the
existential modal, ( 58), result in sentences that are quite good, by contrast.

(58) a. John is only required not to smoke 30
F
cigarettes.
b. John is only required not to weigh 190
F
pounds.

(59) a. %John is only not allowed to smoke 30
F
cigarettes.
b. %John is only not allowed to weigh 190
F
pounds.

Finally, the addition of an existential modal, ( 60), and the move to the equivalent dual, ( 61), are
of no help.

(60) *John is only allowed not to read 30
F
books.
(61) *John is only not required to read 30
F
books.

40
Not all the data are as clean as we would like them to be. We think that this might be due to an independent
syntactic constraint on the positioning of only. This constraint, we think, disfavors the placement of only
immediately to the left of negation, and is thus responsible for the status of ( 56) and ( 59). See, however, Spector
(2004) for an argument that this constraint might capture additional facts that we derive from the UDM.
As weve seen the exhaustivity operator cannot be restricted by the same constraints. This is a counter-example
to the OIG as it is stated in ( 6), but not to the account that it motivates, which appeals to an exhaustivity operator.
The two operators (although similar) need not share all syntactic properties.
36


4.2 Questions, Definites, and Comparatives

We now want to check the predictions of the UDM for upward monotone properties given ( 45)
(c) and (d), specifically, the prediction that properties that are N-open under the UDM cannot be
maximized by a question operator or a definite article. Furthermore we want to check that the
effect will be obviated by the universal modal operator, ( 46), but not by the existential modal
operator, ( 47).
Our test for this prediction is unfortunately far from perfect, given independent constraints on
the formation of the relevant questions and definite descriptions. Nevertheless, the facts seem to
us to go in the right direction.

(62) a. *More than how many books did John read?
b. ??More than how many books does John have to read?
b. ??/*More than how many cigarettes is John allowed to smoke?

(63) a. *How many feet is John under?
b. ?How many feet do you have to be under to take this ride?
(Steve Yablo, pc)

(64) a. *The amount of food such that you ate more than that (is greater than I had thought).
b. ?The amount of food such that we are required to eat more than that (is greater than
we had thought).
c. *The amount of candy such that we are allowed to eat more than that (is greater than
we had thought).

It is worthwhile to investigate whether there are languages in which the interfering constraints
are not at work, for example, due to the availability of appropriate resumptive pronouns. If there
are, the contrasts ought to be sharper. Hagit Borer (p.c.) suggested the following, as
corroborating data:
37

(65) a. * ?axalti ?et kamut ha-?oxel Se ?ata ?axalta yoter mi-mena.
ate-I ACC amount the-food that you ate more than-from-it
I ate the amount of food, d, such that you ate more than d.
b. ?axalti ?et kamut ha-?oxel Se carix le-?exol yoter mi-mena.
ate-I ACC amount the-food that necessary to-eat more than-from-it
I ate the amount of food, d, such that it is necessary to eat more than d.
c. * ?iSanti ?et kamut ha-tabak Se mutar le-?aSen yoter mi-mena.
smoked-I ACC amount the-tobacco that allowed to-smoke more than-from-it
I smoked the amount of tobacco, d, such that it is allowed to smoke more than d.

5. Arguments for a Deductive System

In the previous sections we have learned that the UDM predicts that certain properties of degrees
will necessarily describe open intervals, and that this, in turn, accounts given the CIM for
various constraints on the formation of questions, definite descriptions, sentences with only, and
corresponding implicatures. In this section, we would like to show that our account can be
maintained only if the CIM is assumed to apply within a formal system which is encapsulated
from various pieces of information that enter into the determination of the truth-conditions of a
sentence. We will call this system the deductive system, DS (following Fox 2000) and we will
provide additional evidence for the postulation of such a system drawing on Chierchia (1984)
and, in particular, Gajewski (2002,2003).


5.1 Cardinality as a Level of Granularity

Assume that the UDM is correct. This means that the measurement scales that are part of the
formal apparatus involved in the interpretation of linguistic expressions are always dense.
However, it is clear that various considerations enter into the interpretation of a sentence that are
not determined by the formal apparatus. It is therefore consistent with the UDM that integers will
end up being relevant for interpretive purposes after all.
38
There is a more general issue here. It is well known that sentences involving degree
expressions are evaluated differently in different contexts. The claim that the distance between
Amsterdam and Vienna is 1000 kilometers will be taken to be true in casual conversation about
travel plans. The fact that the exact distance is closer to 965 kilometers will be irrelevant.
However, the evaluation might be different when the statement is written on an official sign-post
in which other distances are listed with a more fine grained level of precision.
41
Such contextual
dependency illustrates the fact that non-formal considerations enter into the determination of
meaning. This fact can be captured within two dimensional semantics (Kaplan, Stalnaker, and
much subsequent work), under the assumption that among the contextual parameters that are
relevant for the evaluation of the truth conditions of a linguistic expression is a parameter, G (for
granularity), that determines the relevant level of precision.
If the UDM is correct, dense scales (or the rules that characterize them) are part of the formal
component of natural language semantics, but discrete scales are not. However, the interpretation
of a linguistic expression is ultimately determined by factors that are outside the formal
component, and among these factors are the contextual considerations that determine the value of
G. It is therefore possible that a discrete measurement scale even if it is not part of the formal
apparatus will end up being relevant to the evaluation of the truth conditions of a linguistic
expression.
To see how this could come about, we need to be a little bit more specific about the
contextual parameter, G, and how it might determine a level of precision. Consider the sentence
in ( 66).

(66) John is (exactly) 15 years old.

In many contexts, this sentence would be considered true as long as Johns age turns out to be
somewhere between 15 and 16. To capture this fact, we might think of G as an equivalence
relation on the members of the dense measurement scale:

(67) xGy iff there is a natural number n, s.t. x[n, n+1) and y[n, n+1).

41
The example is taken from Krifka (2002) in which it is pointed out that the form of the degree expression affects
the relevant level of precision. See Lasersohn (1999) for a different point of view.

39

This would get us the right result if we work out our semantics so that the truth conditions in ( 68)
are derived for the sentence in ( 66), where the contextual parameter C includes information
about various aspects of the context, among them the value of the granularity level G
C
:

(68) [[John is exactly 15 years old]]
C,w
= 1 iff Johns age (in years) in w, a
J,w
, stands in the G
C

relation to 15.
= 1 iff a
J,w
[15, 16) (given ( 67))

The definition of the level of precision/granularity in ( 67) involves quantification over the set
of natural numbers, and, if the UDM is correct, cannot be part of the formal apparatus relevant
for the semantics of degree constructions. Instead, such quantification must rely on our extra-
linguistic knowledge of arithmetic, and can affect truth conditions only in so far as other aspects
of our (contextual) knowledge can.


5.2 First argument for a Deductive System

Consider from this perspective the way that the UDM is supposed to explain the unacceptability
of sentences such as ( 5) exemplified again below (with a minor change to which we will return).

(69) *John is only more than 15
F
years old.

Weve claimed that this sentence is unacceptable because its truth conditions can never be met.
For concreteness, assume that the sentence receives the logical form in ( 70)a, where the variable
A refers to the focus value of the prejacent John is more than 15
F
years old (see note 35). The
resulting truth conditions that we assumed (ignoring the difference between assertion and
presupposition) are given in ( 70)b.

(70) a. Only[A][John is more than 15
F
years old]]
b. [[Only[A][John is more than 15
F
years old]]
w
= 1 iff
i. John is more than 15 years old and
ii. for every proposition, , in A, if (w)=1, then the proposition that John is
more than 15 years old entails .
40

The truth conditions can never be met since the set of propositions in A is the dense set of
alternatives that corresponds to the dense set of degrees. However, this line of reasoning ignores
contextual parameters and in particular the granularity parameter G.
It turns out that once G is taken into account, it is no longer obvious that the truth-conditions
of ( 69) are contradictory. This, on the face of it, is problematic for our account of the
unacceptability of the sentence. In order to deal with this problem, we will argue that sentences
such as ( 69) are ruled out in a formal system that does not take contextual knowledge into
account. This, in turn, will allow us to solve certain problems that we have left unresolved in
earlier sections. Furthermore, the argument will be reinforced in section 5.4 by a consideration of
other facts of a similar sort reported in the literature.
To understand the problem, consider how we might modify ( 70)b once we introduce the
contextually given level of granularity:

(71) [[Only[A][John is more than 15
F
years old]]
C,w
= 1 iff
a. [[John is more than 15 years old]]
C,w
=1 and
b. for every member of A, , if (C)(w)=1, then w. [[John is more
than 15 years old]]
C,w
entails (C).

Are the truth-conditions in ( 71) still contradictory? In order to determine this, we have to know
what proposition the sentence John is more than 15 years old expresses under C and what the
members of A are. The answer to the first question should fall out from a general perspective on
the role played by granularity in the interpretation of comparatives. It seems that we should
derive the following under the granularity in ( 67):

(72) [[John is more than 15 years old]]
C,w
=1 iff John is at least 16 years old in w.

In appendix 1, we show that this follows from basic assumptions about the semantics of
comparatives and context dependency. More generally, we get the following for every such
that 0 < 1:

(73) [[John is more than n+ years old]]
C,w
=1 iff John is at least n+1 years old in w.

41
The important thing to observe is that the truth conditions given in ( 73) are the same for every
[0,1) (given a choice of n). This means that the set of alternative propositions that we get given
the focus structure of the prejacent is not going to be dense. Rather, we will get the following
answer to our second question about the set of alternatives, A, in ( 71):

(74) [[A]]
C,w
= {w. John is at least n years old in w: n N - {0}}

( 72) and ( 74) allow us to elaborate ( 71) in the following way:

(75) [[Only[A][John is more than 15
F
years old]]
C,w
= 1 iff
i. John is at least 16 years old in w and
ii. for every n N, if n>16 John is not n years old.

These truth conditions are of course not contradictory. Specifically, they will be met as long as
John's age in years is in [16,17). Since this will eliminate the account of the facts described in the
preceding sections, we are forced to conclude that G doesnt enter the picture at the level at
which the CIM is evaluated. Instead we have to assume that it becomes relevant only at a "later"
stage when the properties of a particular context are consulted in order to determine the truth
conditions of a sentence. We thus conclude that the CIM is evaluated in an informationally
encapsulated system that does not take into account the (contextual) knowledge that is appealed
to when determining a level of granularity. If we wanted to have a semantic characterization of
the workings of this system, we could say that it evaluates sentences under the most stringent
granularity relation, namely identity. Under this granularity, a sentence such as ( 69) can never be
true and is therefore ruled out.
42

What emerges from these considerations is that the cognitive system contains an
informationally encapsulated deductive system, DS, (Fox 2000) in which sentences are evaluated
and ruled out if they can be proven to be contradictory.
43
Once a sentence passes DS, it is
evaluated in a particular context, where a level of granularity may affect the interpretation. The
dividing line between the rules of DS and the principles that enter into the derivation of truth

42
A different perspective, perhaps more natural, is that G is an optional contextual parameter, which, when present,
leads to a weakening of the semantic interpretation. The inferences of DS are sound relative to a context independent
(G-less) interpretation.
43
Foxs evidence for DS is also based on Information encapsulation.. See Fox (2000): 66-74.
42
conditions in a particular context is close to the traditional divide between syntax/semantics and
pragmatics. If sentences are to be ruled out because they are contradictory, it seems natural to
assume that this is done based on the properties of a formal level such as DS rather than on the
basis of all contextually available information. This assumption, as we will see in section 5.4, is
supported on independent grounds.


5.3 Back to Universal Modals

In previous sections we have explained why sentences such as those in ( 9) repeated in ( 76) (with
a minor variation in ( 76)b to which we will return shortly) are not contradictory despite the
UDM.

(76) a. I can only say with certainty that John weighs more than 120
F
pounds.
b. I can only say with certainty that John is more than 15
F
years old.

We provided a general proof for this and illustrated it for ( 76)a. Specifically, ( 76)a is true as long
as, on the one hand, the speaker is certain that John weighs more than 120 pounds, while, on the
other hand, for every , the speaker is not certain that John weighs more than 120+ pounds. A
problem, however, arises when we consider ( 76)b. We are able to derive the fact that this
sentence is acceptable. However it is not obvious that we derive the correct truth conditions. In
certain contexts ( 76)b would be equivalent to ( 77) and thus would not entail that there is no in
the dense set of degrees such that the speaker is certain that John is more than 15+ years old.

(77) I can only say with certainty that John is 16
F
years old.

It seems that (in the relevant contexts) the set of degrees that is relevant for the evaluation of
( 76)b is the set of integers. However, in order to account for the unacceptability of the
corresponding sentence without a universal modal operator, it is crucial to make a different
assumption, namely the UDM.
This virtual contradiction is resolved by the division of labor between syntax/semantics and
pragmatics that we postulated in the last section. Specifically, the sentence without the universal
43
operator is ruled out by DS as a contradiction whereas the sentence in ( 76)b is not. ( 76)b thus has
a chance to be evaluated in a particular context while the sentence without a universal modal
does not. In the relevant contexts, the level of granularity would be the one given in ( 67) which
will result in the truth conditions in ( 78) (as we will show in the appendix).

(78) [[Only[A][I can say with certainty that John is more than 15
F
years old]]
C,w
= 1 iff
i. The speaker in C can say in w with certainty that John is at least 16 years old
and
ii. for every n N, if n>16 it is not the case that the speaker in C can say in w
with certainty that John is n years old.

We now move to discuss our original examples in ( 5) and ( 9)b repeated below.

(79) a. *John only has more than 3
F
children.
b. I can only say with certainty that John has more than 3
F
children.

( 79)a will be ruled out within DS, while ( 79)b will not. ( 79)b, however, presents a problem
similar to the one discussed in the context of ( 76)b: it does not entail that for every the speaker
is not certain that John has more than 3+ children. Rather, it seems to entail that the speaker is
certain that John has 4 children. This, again, can be dealt with by postulating a level of
granularity. The following seems to us to be a reasonable assumption about the granularity of
measurement for collections of objects that are indivisible (based on world knowledge), such as
children.

(80) Granularity for the measurement of collections of indivisible objects
44

xGy iff there is a natural number n, s.t. x(n, n+1] and y(n, n+1].

Under this granularity, we get the right results. Specifically, we get the following truth
conditions:


44
To talk about "context dependency" is a bit misleading for these cases. A better term would be "extra-linguistic
dependency" indicating that the extra-linguistic information that enters into the determination of the truth-conditions
might not vary across contexts.
44
(81) [[Only[A][I can say with certainty that John has more than 3
F
children]]
C,w
= 1 iff
i. The speaker in C can say in w with certainty that John has more than 3
children and
ii. for every n N, if n>4 it is not the case that the speaker in C can say in w
with certainty that John has more than n children.
45


These are of course the correct truth-conditions. The assumption that levels of granularity are
contextually specified and that DS is insensitive to this specification provides us with the means
to account for the contrast in acceptability in ( 79) as well as the truth-conditions of ( 79)b.
46



5.4 Further Evidence for a Deductive System

In the previous sections we have seen that the pattern of acceptability of syntactic representations
of the sort we have been looking at can be derived under the UDM. However, this can only be
achieved if levels of granularity, which are needed to derive adequate truth-conditions, are
ignored. This, in turn, would follow if we assume a formal system, DS, that rules out
contradictions independently of a level of granularity. In this section we will review another
argument for DS.
The argument will be stated, once again, in the context of evidence that certain contradictions
ought to be ruled out by the linguistic system. There is quite a bit of evidence, which comes from
a variety of empirical domains distinct from the ones discussed in this paper. However, this
evidence is in conflict with the well-known fact that the linguistic system is able to express many
contradictory propositions. To resolve this conflict, we suggest, following Chierchia (1984) and
Gajewski (2002), that the linguistic system rules out contradictions on the basis of formal
considerations alone. If this claim is correct, it will provide further evidence for DS.
The idea that sentences might be ruled out because they express contradictory propositions
has been advanced in the context of the analysis of a variety of constructions. (See Ladusaw

45
There are other levels of granularity that would produce the same results, e.g. the one in ( 67). ( 80) seems more
natural to us since under this level of granularity the sentence John has n+1/2 children and the sentence John has
n+1 children express the same propositions. An alternative approach to ( 79)b might appeal to a contextual notion of
entailment.
46
Similarly, we can account for the answers available for questions such as How many children are you certain that
John doesn't have? The acceptability of these questions is explained by the fact that universal modals can close
intervals. However, these questions are naturally answered by expressions such as n children, which should be ruled
out by more informative answers of the form, e.g. n-1/2 children. This conflict is resolved by the observation that,
once G is taken into account all answers of the form n- (where 0<1) express the same proposition.
45
1989 for a discussion of a few of the original proposals.) Take for example the contrast in ( 82).
Von Fintel (1993) suggests that this contrast should be explained on the basis of the semantic
import of but, specifically, that the lexical entry needed to account for the truth conditions of
( 82)a ensures that ( 82)b expresses a logical contradiction.

(82) a. Every man but John arrived.
b. *A man but John arrived.

The sentence in ( 82)a entails, on the one hand, that it is not the case that every man arrived and,
on the other hand, that it is the case that every man who is not John arrived, (i.e. that {John} is
the minimal set X that yields a true proposition of the form every man who is not a member of X
arrived). The sentence in ( 82)b if it were acceptable would entail, on the one hand, that it is
not the case that a man arrived and, on the other hand, that it is the case that a man who is not
John arrived. These two entailments are of course contradictory, which, under von Fintel's
proposal, accounts for the unacceptability of the sentence. Von Fintel shows that this type of
reasoning is a good predictor of the distribution of exceptive but based on a variety of empirical
considerations, and Gajewski (2002) shows that, under a particular implementation, it could be
extended to account for certain previously unresolved puzzles.
Similarly, Dowty (1979) argues that the unacceptability of sentences such as ( 83) is to be
explained on the basis of the logical properties of the verb phrase accomplished his mission and
the adverb for an hour. The properties of the adverb yield the result that, if the sentence were
acceptable, it would have expressed the proposition that there is a time interval in the past, T,
lasting for an hour, such that tT, John accomplished his mission in t. The properties of the
verb phrase ensure that this would never hold.

(83) *John accomplished his mission for an hour.
There is a time interval in the past T s.t. Length(T) = one hour and
tT John accomplished his mission in t.

Further examples are discussed in Chierchia (2004, 2005), Dayal (1998), Gajewski (2002)
drawing on Barwise & Cooper (1981), Ladusaw (1989), and Menendez-Benito (2005).
47
This
line of reasoning thus seems to have some support. However, it doesn't appear to be universally

47
Some of these examples require the assumption that tautologies and contradictions are treated on a par.
46
applicable, as there are many contradictions that do not yield unacceptability. Some examples are
given in ( 84).

(84) a. This table is both red and not red.
b. Hes an idiot and he isnt.
c. I have a female (for a) father.
d. No student is sick and yet some student is sick.

One might suggest that the sentences in ( 84) are acceptable because the contradictions can be
eliminated by virtue of a dynamic interpretation of contextual parameters. One might suggest
that in ( 84)a, for example, the context provides a standard for redness which is different in the
first and second conjunct, thereby yielding a contingent proposition. (See, among others,
Chierchia & McConnell-Ginet 1990: chpt. 8 sec. 5, and Kamp and Partee 1995.
48
)
This, however, will not be consistent with what is needed to maintain the proposals weve
made in the previous sections. As we have seen, there is a way to fix the contextual granularity
parameter, G, that would eliminate the contradictions that are responsible (under our account) for
the unacceptability of certain sentences. We would therefore like to adopt an alternative account
of the contrast between sentences such as those in ( 82) and ( 83), on the one hand, and ( 84) on the
other, namely a version of the proposal made in Gajewski (2002).
But before we get there, we would like to mention that there are cases where sentences are
clearly interpreted as contradictions, i.e., cases for which postulating a contradictory
interpretation is needed to account for the communicative function:

(85) What you are saying is clearly wrong. It leads to the conclusion that
a. some student (other than John) arrived and yet that no student arrived.
b. this table is both red and not red.


More importantly, the sentences that are ruled out because they are contradictory can never serve
a similar function, even in cases in which they are predicted to express exactly the same
proposition.

48
Thanks to Jon Gajewski for pointing out this possibility.
47

(86) What you are saying is clearly contradictory. *It leads to the conclusion that some
student but John arrived.

The difference between ( 85) and ( 86) suggests that there is a qualitative distinction between
different types of contradictions, as suggested by Gajewski (2002).
Gajewski proposes to account for the facts described above under the assumption that there is
a formal principle of grammar that rules out contradictory sentences and that this principle
applies at a particular level of representation (LF') which encodes information about the content
of functional/logical words but not about the content of lexical categories.
49
More specifically,
Gajewski suggests that we think of LF' as if it were derived from a traditional logical form by the
replacement of every occurrence of a non-functional word with a new variable.
50

For example, the sentences in ( 84) would have the following representations at LF' on the
basis of which a contradiction cannot be derived.

(87) a. This NP
1
is both AP
1
and not AP
2
.
b. DP
1
is a NP
1
and not a NP
2
.
c. I have a AP for a NP.
d. No NP
1
is AP
1
and yet some NP
2
is AP
2
.

By contrast, the sentence in ( 82)b will have the following representation at LF', which is
sufficient to determine its contradictory status under the assumption that the system can access
various axioms that characterize the meaning of the indefinite quantifier and the exceptive
morpheme but.

(88) A NP but DP VPed.
51,52


49
A suggestion along similar lines was made by Kai von Fintel in a class presentation in 1999. We thank Kai for
helpful discussions of all the issues examined in this section.
An obvious question to ask is what characterizes the set of functional/logical words. Gajewski suggests that
every member of the set is permutation-invariant in Tarskis sense (Sher 1991), building specifically on a
formulation due to Van Benthem (1989). See also von Fintel (1995).
50
If proposals such as that of Marantz (1997) and Borer (2005) are correct, LF' could be identified with LF.
51
See Gajewski 2002 for further details of the nature of the structure.
48

To summarize, we have presented two arguments for DS. One argument was based on our
account of the properties of certain degree expressions: definite descriptions of degrees, degree
questions, sentences with only and corresponding implicatures. This account in terms of the
UDM and the CIM can be maintained only if we assume a division of labor between DS and
context-dependent interpretation, that is, between a formal system that rules out contradictory
expressions independent of levels of granularity, and a later system responsible for the context
dependent computation of truth-conditions (based, among others, on the setting of the granularity
parameter). This allowed us to explain why certain expressions are ruled out by DS despite the
fact that a plausible level of granularity could give them a non-contradictory interpretation. DS is
a formal level whose inferences are independent of contextual interpretations: the inferences are
sound relative to a semantics without a level of granularity (or relative to G=identity) and not
relative to a semantics that considers contextual weakenings.
Our second argument was based on the observation that proposals due to Chierchia, Dowty,
von Fintel, Gajewski and others rely on a similar distinction. Since not all contradictions are
judged unacceptable, but only those that follow from certain formal considerations, a formal
system, DS, needs to be postulated. This system determines for each expression whether it is
contradictory based on the relevant formal properties of the expression.
53

Both arguments are based on a similar logic. They both teach us that the overall
considerations that determine the truth conditions of a sentence (semantics+pragmatics) yield
different results from those needed to derive the relevant acceptability judgments. Our first
argument teaches us that we need to postulate a system that derives contradictions which would
be eliminated if contextual parameters were introduced (a system that is not sound relative to
semantics+pragmatics). Our second argument teaches us that the required system should not be

52
For ( 83) one might assume a more refined syntactic structure in which telic predicates are built from two
independent morphemes one of which is a cause operator, or alternatively one might assume that all primitive
predicates of time are (axiomatically) telic, and that non-telic predicate are syntactically complex.
53
See Heim 1984, Lahiri 1998, Guerzoni 2003, and Zucchi 1995, among others, for arguments that contradictory
presuppositions yield unacceptability. There, too, it seems that not all contradictions are ruled out, only those that
can be plausibly linked to DS:

(i) a. #The man who isnt a man arrived late.
b. #The man is not a man
c. *I read even a single book.

49
able derive all intuitive contradictions (i.e., that it cannot be complete relative to
semantics+pragmatics).
If we are correct, we would like to know which expressions belong to the vocabulary of DS
and what formal mechanisms are involved (e.g. axioms and rules of inference). For the cases
discussed in this paper, one would need to assume that the comparative morpheme er, negation,
EXH, only, and the temporal preposition, before belong to the relevant vocabulary. This
assumption, when coupled with appropriate DS axioms should be able to deliver the required
results. More specifically, it seems that the following would have to be theorems of DS:

(89) a. Universal Density: d
1,
d
2
[d
1
> d
2
d
3
(d
1
> d
3
> d
2
)]
b. Lexical Monotonicity: lexical n-place relations are upward monotone.
c. Lexical Closed Intervals: if R is a lexical n-place relation, whose m
th
argument is a
degree, then for every w, and for every x
1,
,x
n-1
Max
inf
(d.R(x
1
) (d) (x
n-1
))(w) is
defined.
d. Commutativity: Two existential quantifiers can be commuted.
54



Based on this set of assumptions (together with basic logical machinery), DS should be able to
prove that a Gajewski-structure for ( 90)a, provided in ( 90)b, is contradictory. Similar
considerations apply to structures that employ exh.

(90) a. *John is only more than 15
F
years old.
b. only ({w.-er than d (d'.NP d'-AP in w): d a degree})
w.-er than d" (d'.NP d'-AP in w')
55



54
( 89)d is also crucial to deliver the results needed for the proposal in Fox (2000).
55
The LF that corresponds to ( 90)b assumes that the comparative morpheme er is a restricted quantifier over
degrees that adjoins to a clausal node. The restrictor of er is given by than d while the nuclear scope is the matrix
clause, which, given abstraction over a degree position triggered by movement of er than d, denotes a predicate of
degrees (cf. Heim 2001). Only, as before, is analyzed as a clausal operator whose first argument is a set of
alternatives derived by replacing the focused expression in the prejacent with its alternatives. The Gajewski structure
can be derived from the LF by replacing any non-logical item with a different variable except in the calculation of
the alternatives, which requires that the same variable names be used as in the prejacent.

(i) a. John is only more than 15
F
years old.
LF b. only A [[er than 15]
i
[John is d
i
years old]].
A = {w. [[er than d]
i
[John is d
i
years old]]: d a degree}
GS c. only A [[er than d']
i
[NP is d
i
AP]].
A = {w. [[er than d']
i
[NP is d
i
AP]]: d a degree}

50
This division of labor between DS and a more general pragmatic system allows us to address a
variety of problems for what weve said in section 1-4. We start with a problem that we
mentioned briefly in an earlier section (footnote 28). Consider our account of ( 32), repeated
below as ( 91)

(91) *Before when did John arrive?

We explained the unacceptability in the following way: suppose that John arrived at 10, the
answer to ( 91) would have to specify the earliest time after 10, but given the density of time,
there could be no such time. However, this explanation seemed to be contradicted by the
acceptability of ( 92).

(92) the earliest time after 10

We can now give a rational for a distinction between the two cases. For DS to derive the
inference that ( 91) doesnt have an appropriate answer, the preposition before has to belong to
the vocabulary of DS. If it does, and density is assumed, the crucial fact can be derived, i.e. that
there is no maximally informative t such that John's arrival time, t', is before t. To derive a
similar inference for ( 92) requires the same assumption for the preposition after, the adjective
early, and the superlative morpheme est. It seems reasonable to us that this assumption will not
hold for the adjective. If so, we would have a distinction that could derive the contrast. More
specifically, ( 93)b might be an appropriate Gajewski-structure for ( 93)a.
56


(93) a. Mary arrived at the earliest time after 10 pm.

56
We assume for concreteness, following Heim (2000), that in the LF that corresponds to ( 93)b the superlative
morpheme est is a restricted quantifier over degrees that adjoins to a clausal node. Unlike er (but very much like
only), the restrictor of est is a set of focus induced alternatives based on the prejacent of est, which denotes a
property of degrees. The truth-conditional import of est involves the existential claim that there is a degree d such
that the prejacent is true of d but none of the alternatives (that are not entailed by the prejacent) is true of d. (Note
that the definite determiner on the earliest time after 10 pm is translated as an indefinite.)

(i) a. Mary arrived at the earliest time after 10
F
pm.
LF b. [[-est A]
i
[Mary arrived at a d
i
-early time after 10 pm]].
A = {d.w. Mary arrived in w at a d-early time after t]]: t a time}
GS c. -est A [d. NP VP at a d-AP time after t in w].
A = {d.w. [NP VP at a d-AP time after t]]: t a time}.
51
b. -est ({d. w. NP VP at a d-AP time after t in w)}: t a time)
(d'. NP VP at a d-AP time after t' in w)

To see that ( 93)b is not DS-contradictory, consider various instantiations of AP, the variable that
ranges over degree predicates of times such as early, late, opportune, etc. Clearly, a problem
akin to the one in ( 91)a arises if the degree predicate of times that licenses est is based on the
inherent linear ordering among times. Early seems to be such a predicate since t is earlier than t'
iff t is before t'. If early were part of the DS vocabulary, the fact that ordering among times is
that of temporal precedence would be visible to DS and the system would determine that it is
impossible to find a time after t that is earlier than any other time after t. However, if we assume
instead that early is not part of the DS-vocabulary, early would be represented by a variable AP
at DS and this would make it impossible to derive a contradiction. To see this, consider an
instantiation of AP which orders times which on grounds other than temporal precedence, e.g.
opportune in the most opportune time after 10.
In fact, we observe quite generally that sentences that employ only associating with a degree
expression across a comparative operator will be ruled out as DS-contradictions while seemingly
parallel sentence with a superlative operator will not.

(94) a. *John only weighs more than 120 pounds /*John is only heavier than 120 pounds.
b. The smallest weight bigger/above 120 pounds

(95) a. *John only has more than THREE children.
b. The smallest amount/number of children bigger/above 3

This contrast is predicted by our assumptions. Because of the UDM, any degree predicate
suffices to produce a contradiction in structures such as ( 90)b. However, in the superlative cases
a contradiction or presupposition failure would result at DS only if the degree predicate licensing
the superlative, small, belonged to the vocabulary of DS.
Consider next the following contrast.
57
If we are correct, ( 96)a lacks an implicature because
DS can derive a contradiction from a structure that contains exh. In ( 96)b and ( 96)c we see a
contrast that weve observed before with universal and existential modals.

57
Thanks to Nathan Klinedinst, Philippe Schlenker, and Benjamin Spector for insightful comments on this topic.
52

(96) a. John handed in more than 3 assignments.
*Implicature: John handed in exactly 4 assignments.
b. Every student handed in more than 3 assignments.
?Implicature: There is at least one student who handed in exactly 4 assignments.
c. Some student handed in more than 3 assignments.
*Implicature: There is no student who handed in exactly 4 assignments.

More specifically, a universal quantifier over individuals in ( 96)b, just like a universal modal,
can prevent DS from deriving a contradiction, and, thus, allows for the generation of an
implicature. An existential quantifier over individuals, like an existential modal, cannot produce
this effect.
Since the contrasts are parallel, we would like to have a parallel account. Our account in the
modal domain depended crucially on the idea that grammar does not preclude quantification over
an infinite set of worlds. That is a necessary assumption if the domain of degrees can ever be
infinite (given assumptions about asymmetric entailment). However, for the cases in ( 96), it is
not a necessary assumption. Still, it seems rather natural to us.
If DS does not have axioms that rule out an infinite domain of individuals, it will be unable
to prove a contradiction for the structure of ( 96)b that yields the relevant implicature. To show
this, we construct in appendix 3 a world in which the exhaustive interpretation of ( 96)b is true.
The actual interpretation that we derive for the sentence, once it passes the DS filter is based on a
level of granularity. Once the level of granularity is introduced it is as if the domain of degrees
contained only natural numbers, and the representation with exh will entail the proposition that
not every boy handed in more than 4 assignments.
Independently of assumptions about the size of quantification domains, existential quantifiers
over individuals, ( 96)c, are just like existential modals. In both cases a contradiction can be
derived by DS based the fact that two existential quantifiers are commutative.
This type of reasoning will extend to the following contrast pointed out to us by Nathan
Klindinst and Philippe Schlenker. ( 97)b could have a most informative answer: the proposition
that no student handed in d* assignments could be the most informative proposition of the
relevant form (no student handed in d assignments). This is true under the UDM if the domain of
individuals can be infinite.

(97) a. *How many assignments did John not hand in?
b. (?) How many assignments did no student hand in?
53

However, Klindinst and Schlenker (p.c) point yet another contrast:

(98) a. *How many assignments did John not hand in?
b. (?)How many assignments did neither John nor Bill hand in?

In order to account for the (relative) acceptability of ( 98)b, we would have to claim that DS
cannot prove that the property d. neither John nor Bill handed in d many assignments is an
N-open property. However, if the UDM is assumed, it is easy to see that there is no minimal
degree, d, such that neither John nor Bill handed in d many assignments. Assume (with no loss of
generality, of course) that John handed in more assignment than Bill, say d assignments. The set
of degrees above d is an open interval with no minimum. This seems to be a real problem, for
which we dont yet have a satisfactory solution. However, it is worth pointing out that the scope
of the phenomenon is not clear. For example, it is not obvious to us that a parallel puzzle arises
in the domain of conjunction. While universal quantification over individuals is able to
circumvent a UDM effect in ( 96)b, the same is not true of ordinary conjunction in ( 99).

(99) Both John and Bill handed in more than 3 assignments.
*Implicature: Either John or Bill handed in exactly 4 assignments.

At this point, we have to leave the matter as an unresolved puzzle.
58


5.5 Syntactic Contextual Restriction

Kroch (1989) notes that negative islands can be circumvented if the context provides an explicit
set of alternatives. For example, while ( 100)a is clearly unacceptable, the specification of a set of
alternatives in ( 100)b gets rid of the problem.

58
But we can offer some speculations. We might treat neither John nor Bill as a negative existential, with a non-
logical domain restrictor, hence potentially infinite. We are not quite sure how to implement this idea. One
possibility is to de-compose neither into a negative morpheme not and an existential quantifier either, and to assume
that the word or, itself, is not part of the logical vocabulary. If so, the Gajewski representation of ( 98)b would be the
following:

(i) How
d
did not either[x [C [y]]] V d-many NPs?

Since or is not a logical word, C is replaced with a variable of the appropriate type, in this case a function that takes
an individual to functions from individuals to sets. Since the function is not restricted in any way, the range could be
any set, among others, an infinite set, that would allow for a most informative answer. Many questions arise at this
point, and addressing them would involve various speculative moves, which are very likely pre-mature.
54

(100) a. *Tell me how many points Iverson didnt score?
b. Among the following, please tell me how many points Iverson didnt score?
A. 20 B. 30 C. 40 D. 50

On initial inspection, this contrast is directly predicted by our approach to negative islands.
( 100)a is ruled out given our assumption that the set of degrees is dense, from which it follows
that the question cannot be answered: an answer to the question would specify the most
informative true proposition in the Hamblin denotation of the question, but there can be no such
proposition.
In ( 100)b, by contrast, the addressee is requested to focus on a discrete sub-set of the
propositions in the Hamblin set. This contextually specified set of alternative propositions (that
Iverson didn't score 20 points, that he didn't score 30 points, etc.) contains a most informative
true element which renders the question answerable, hence acceptable.
This observation extends to implicatures and only as the data in ( 101) and ( 102) show.

(101) Iverson sometimes scores more than 30 points. But today he only scored more than 20
F
.

(102) A: How many points did Iverson score last night?
B: I dont know.
A: Was it more than 10, more than 20 or more than 30.
B: He scored more than 20 points
Implicature: he didnt score more than 30.

However, a real question arises at this point. While we expect that the existence of a most
informative degree will license the use of Max
inf
, it is at this point puzzling that contextual
information can save the day. Earlier, we argued that levels of granularity, which are
contextually determined, are invisible to DS and therefore cannot resurrect DS contradictions.
Here we see that a contextually provided set of alternatives saves a structure that would
otherwise be ruled out as a DS contradiction.
We can make sense of this apparent paradox if we assume, following e.g. von Fintel (1994),
that the contextual restrictions at work in ( 100)-( 102) are represented in the syntax as variables
that modify the restrictor of the relevant operator, as illustrated in ( 103)a for only. This yields the
55
Gajewski structure in ( 103)b.
59
This structure is not ruled out by DS since a contradiction cannot
be derived given the presence of C.

(103) a. [Only[A C] [Iverson scored more than 20 points]]
where C = {that Iverson scored more than 20 points, that Iverson scored more than 30
points, }]
b. Only [{that NP
1
V more than d NP
2
: d a degree} C] NP
1
V more than d' NP
2
.

If the context does not provide a specific set of alternatives, a structure with a free variable C
will not be licensed. Under such circumstances, the restrictor of only will be determined solely
based on focus induced alternatives to the prejacent, which, as we have shown above, yields a
DS contradiction.
Note the crucial assumption: there are two distinct ways in which the context can affect
meaning. Levels of granularity enter into the determination of the truth-conditions quite
differently from contextually licensed variables, such as C in ( 103)c. Levels of granularity are
among the parameters of the valuation function, and, as such, are not represented in the syntax,
and, consequently, are invisible to DS. The desired result is that levels of granularity, in contrast
to contextually licensed variables, cannot save a structure from a DS violation.


6. Conclusion

In sections 1 through 4, we argued that the semantics of degree constructions does not
presuppose the concept of natural number. This argument was based on an intricate set of
acceptability judgments which follow from an extremely natural assumption (the CIM) only if
degree domains are always dense, i.e., if the set of natural numbers is not treated by the linguistic
system as a separate domain of entities.
We think that our arguments fit well with the fact that not all languages have a vocabulary
that can be used to refer to the set of natural numbers (Gordon 2004, Pica et. al. 2004), and with
the much discussed difficulties in the acquisition of the relevant numerical concepts. These give
reason for skepticism about claims that there is a core knowledge system that directly defines the
set of natural numbers. Therefore, even if various semantic notions are axiomatized within the

59
The structure of the comparatives in ( 103)b,c needs to be further articulated.. See fn. 55 and appendix 1 for one
possibility.
56
linguistic system (for which there is abundant evidence), one would probably not expect the
same to be true of the theory of natural numbers. From this perspective, the conclusions which
we have argued for are fairly natural.
However, our arguments are in conflict with much work in natural language semantics in
which the concept of cardinality plays a prominent role. We suspect that this conflict is not
substantial, and that standard results can be recast in terms of contextually determined levels of
granularity. If this turns out to be correct, there are far reaching consequences for the architecture
of the grammar, which we began to investigate in section 5.

57
Appendix 1: Granularity in the Interpretation of Comparatives

In the following we implement a few of the crucial assumptions about the way a level of
granularity, G, affects the interpretation of degree constructions:

(104) Standard Lexical entries (without G):
a. [[old]]
w
(d)(x) = 1 iff xs age in w is greater or equal to d (a
x,w
d).
b. [[er/more than]] (D
d,t
)(
d,t
) = 1 iff d[d[[]] & d'[[D]] [d>d']
60

c. [[15 years]] = {15}.

d. [[only

S]]
w
= 1 iff [[S]]
w
= 1 & [[S]]
f
(w)=1 [[S]]
w
|=
61

(where [[S]]
f
= {w. [[S']]: S' can be derived from S by appropriate replacement of focus marked constituent})

(105) Two dimensional interpretation:
[[
d,t
]]
G
= {d: d'[d'Gd & d'[[]] ]}
[[
d,et
]]
G
= {d,x: d'[d'Gd & d',x[[]] ]}
(generalizes in an obvious way to n-place predicates)

Some Derivations (building up to an account of ( 76))

(106) [[er/more than D ]]
G
= 1 iff d[d[[]]
G
& d'[[D]]
G
[d>d']]
iff dd'[d'Gd & d'[[]] ] &

d' (d''[d''Gd' & d''[[D]] ] [d>d'])

We will assume that every one of Gs equivalence classes (ECs) is an interval (continuous, i.e.
with no holes). If a member of [[]], d, belongs to EC(d'), and d' is above a set of ECs, d must also
be above the set of ECs. Hence:

iff d[[]] &

d' (d''[d''Gd' & d'' [[D]] ] [d>d'])

(107) [[er/more than x years ]]
G
= 1 iff d[[]] &

d' EC(x) [d>d'] (since EC(x) = {d: d'[d'Gd & d' = x})


Level of Granularity for Age:
G = {d,d': n [ n N & d [n,n+1) & d' [n,n+1) ]}

(108) [[ John is more than 15 years old]]
G,w
=1 (example ( 72))
iff [[ more than 15-years [d.John is d-years old] ]]
G,w
=1 (assumption about LF)
iff d[a
J,w
d] &
d' EC(15) [d>d']
iff d[a
J,w
d] &

60
We are ignoring the difference between characteristic functions and the sets they characterize.
61
In order to avoid intricate syntactic representations, we provide a syncategorematic rule. Furthermore, we ignore
the distinction between assertion and presupposition, again to keep things simple.
58
d' [15, 16) [d>d'] (by definition of G)
iff a
J,w
16


(109) More generally, we obtain for any n N and any [0,1)
[[ more than n+ [d.John is d-years old] ]]
G,w
=1
iff a
J,w
n+1 (since EC(n+) = [n, n+1))

(110) The focus value of John is more than 15
F
years old under G:
[[more than 15 [d. John is d-years old] ]]
f
G
=
{w. [[more than d [d'.John is d' years old] ]]
G,w
: d R} =
{w. [[more than n+ [d'. John is d' years old] ]]
G,w
: nN, (0,1)} =
{w. a
J,w
n+1: n N}

(111) The focus value of I can say with certainty that John is more than 15
F
years old under G:
{w. w' [w' Acc
w
a
J,w'
n+1]: n N} (by the same reasoning)


(112) [[ Only [I can say with certainty that more than 15 d. John is d-years old ]]
G,w
=1
iff w' [w' Acc
w
a
J,w
16] &
[ focus-of-prejacent-under G & (w)=1
w.[[prejacent]]
G,w
|= ].
(where the prejacent is the sister of only)

Since: i. the-focus-of-prejacent-under-G is
{w.w' [w' Acc
w
a
J,w'
n+1] : n N} (by ( 111)), and
ii.
16
|=
n
iff n<17
where
n
= w.w' [w' Acc
w
a
J,w'
n+1]


[[( 112)]]
G,w
= 1 iff w' [w' Acc
w
a
J,w'
16] &
n17[
n
(w) = 0]


Exactly the same computations will produce the desired interpretations for examples ( 2) and
( 81).

Appendix 2: The UDM and Quantification over individuals


In this appendix we show that the consequences of universal and existential modals for the UDM
extend to all universal and existential quantifiers, when no restrictions are imposed on the
59
domain of quantification. To save space, we limit our discussion to upward monotone
functions.
62

Let be an upward monotone function of type <d,<e,st>>, where is upward monotone if
for every d
1
,d
2
: d
1
>d
2
iff for every x, (d
1
)(x) is more informative than (d
2
)(x) (asymmetrically
entails it). Assume also that necessarily describes an open interval, i.e. for every world w, and
individual x, d.(d)(x)(w) is an open interval. The following drawing represents what holds of
in every world, given a choice of an individual x:


(d)(x) is false (d)(x) is false


(d)(x) is true


We can see that if we append a universal quantifier over individuals with restrictor C,
C
(of
type <<e,st>, st>), the modified property, d.
C
[(d)], would no longer necessarily describe an
open interval. The universal modal could quantify over a dense set of individuals that
corresponds to the set of degrees with the result that the modified property describes a closed
interval. To see this, consider a sequence of open intervals I
k
that converges to a closed interval I
= [0,a] from the top (I
k
= [0,a
k
) where a
k
converges to a from the top).

a
1
a
2
a
3
a
k
a


I
1
I
2
I
3
I
k
I

There is a world w in which there is a corresponding sequence of individuals in C, x
k
, such that
for every n, d.(d)(x
n
)(w) = I
n
. In this world, d.
C
(d)(w) denotes the closed interval I. To

62
To see that the same holds for downward monotone function, one needs to follow the same steps, with the
difference that all open intervals are reversed and the sequence of open intervals converges to a closed interval from
the bottom.
60
see this, suppose that d is not in I. This means that d is bigger than a. Hence, starting with some
N, for all n bigger than N, (d)(x
n
)(w)=0 (given convergence from the top). Hence,

C
[(d)](w)=0. Suppose conversely that d is in I. Given that I is a subset of every I
n
,

C
[(d)](w)=1. This means that complex predicates over degrees such as d. every student
handed in more than d assignments do not necessarily denote an open interval. And this, we
suggest, accounts for the fact that such predicates can be the source of an implicature in ( 96).
As in the modal case, appending an existential quantifier is of no help: the property
d.
C
[(d)] necessarily describes an open interval. To see this, assume that d.
C
[(d)] denotes
a closed interval in a world, w, and that d' is its maximal member. Of course
C
[(d')](w)=1.
This means that there is an individual, x, in C such that (d')(x)(w)=1. But since describes an
open interval for every x, there is a degree d'' bigger than d' such that (d'')(x)(w)=1. Hence, d' is
not the maximal member in d.
C
[(d)](w). This, of course, explains the fact that complex
predicates over degrees such as d. some student handed in more than d assignments cannot
support the relevant implicature.

Appendix 3: Remaining Issues in the Semantics of Questions

The results of sections 1-4 were based on the assumption that Maximality (specifically Max
inf
) is
relevant for the semantics of four different linguistic constructions: sentences with only,
sentences with scalar implicatures, definite descriptions, and questions. We think that this
assumption is straightforward for the first three cases. However the role of Maximality

in the
semantics of questions is less straightforward, and various conflicting positions have been argued
for. (See, among others, Groenendijk&Stockhof 1984, Dayal 1996, Higginbotham 1993, Lahiri
2002.) While we cant do justice to the extensive literature on this topic, we feel obliged to deal
with specific challenges to Maximality that have been raised in the context of degree questions.
Consider the question answer sequence in ( 113) and ( 114) discussed by Beck and Rullmann
(1999).

(113) Speaker A: How many people can play soccer?
Speaker B: 6 people (indoor soccer), 8 people (small field) and 11 people (regular)

(114) Speaker A: How many courses are you allowed to take?
Speaker B: Any number between 4 and 6.

61
The answer given by Speaker B in both cases is inconsistent with a maximality presupposition.
There are two possible moves that are available to us at this stage. We can interpret Bs answer
as involving a rejection of the questions presupposition, or alternatively we can try to replace
the maximality presuppositions of degree questions with a weaker presupposition that would still
capture the results described in sections 2 and 3. The first move might be appropriate for the
dialogue in ( 113), which can be replicated with a definite description.

(115) Speaker A: What is the number of people that can play soccer?
Speaker B: 6 people (indoor soccer), 8 people (small field) and 11 people (regular)

It is very likely that in this case Speaker B is indicating that three different questions need to be
asked and is answering each in turn. However the situation seems to be different in the case of
( 114), which is totally appropriate and quite different from its counterpart with a definite
description:

(116) Speaker A: What is the number of courses are you allowed to take?
Speaker B: #Any number between 4 and 6.
Spealer B: Oh, the rules are not that rigid. I can take any number of courses between 4
and 6.

We thus need to investigate the second possibility. The need is quite pertinent for our concerns
as illustrated by the dialogue in ( 117), pointed out to us by Irene Heim (p.c.).

(117) A: How much money are you not allowed to bring into this country?
B: $10,000
C: The maximum allowed is $10,000.
= Youre not allowed to bring in any amount that exceeds $10,000.

Weve claimed that As question is acceptable in contrast to the non modalized variant in ( 118)
because a modal operator can close an N-open interval.

(118) *How much money did you not bring into this country?

Under the assumption that Max
inf
is employed in the semantics of degree questions, ( 118) is bad
because there can be no minimal degree d, such that you did not bring d-much money into this
country. By contrast, there could be a minimal degree of the relevant sort in ( 117). For example,
62
Bs answer conveys the information that $10,000 is the minimal (i.e. maximaly informative)
amount of money that you are not allowed to bring into this country. However, as pointed to us
by Irene Heim, this line of reasoning would force us to analyze Cs answer as contradicting the
presupposition of the question. In his answer, C is claiming that there is no minimal amount of
money that you are not allowed to bring into this country.
Such an analysis doesnt seem plausible to us, and we would therefore like to tentatively
propose an alternative. Specifically, we would like to adopt Hamblins (1973) proposal that a
question denotes a set of propositions, to which we would like to add, as a presupposition, the
requirement that it be possible for a member of the set to be a maximally informative answer:
63


(119) [[how many ]] , when defined = {p: d (p = (d)}.

(120) A question Q is defined, in a world w, iff it is possible that the conjunction of all true
propositions in Q is itself a member of Q.
64


This presupposition will not be met when the Question property, in ( 119), is N-open. However,
its possible for the question property to end up describing an open interval as long as it was also
possible for the property to describe a closed interval (as long as it is not necessarily open).



63
As in the text, we are following the insight of Dayal (1996), with the necessary weakening that comes from
adding the word possible.
64
Note the conjunction of all true propositions to Q is Heims (1994), Answer-1.
63
References
Artstein, R. (1998): Maximality of Quantification? A Critique of Rullmann (1995) Dissertation
Chapter 5, ms. Rutgers University
Barwise, J., Cooper, R. (1981): Generalized Quantifiers and Natural Language, Linguistics and
Philosophy 4: 159-219
Beck, S., Rullmann, H. (1999): A Flexible Approach to Exhaustivity in Questions. Natural
Language Semantics 7(3): 249-297.
Bennet, (2003): A Philosophical Guide to Conditionals, Oxford University Press, Oxford.
Borer, H. (2005): Structuring Sense. Oxford University Press, Oxford.
Chierchia, G. (1984): Topics in the Syntax and Semantics of Infinitives and Gerunds, PhD
dissertation University of Massachusettes Amherst, GLSA.
Chierchia, G. (2004): Scalar Implicatures, Polarity Phenomena, and the Syntax/Pragmatics
Interface, in A. Belletti (ed.): Structures and Beyond. Oxford University Press, Oxford.
Chierchia, G. (2005):
Chierchia, G., McConnell-Ginet, S. (1990): Meaning and Grammar. MIT Press, Cambridge,
MA.
Dayal, V. (1996): Locality in Wh Quantification: Questions and Relative Clauses in Hindi.
Studies in Linguistics and Philosophy, Kluwer Academic Publishers
Dowty, D. (1979): Word Meaning in Montague Grammar. D. Reidel Publishing Co., Dordrecht,
Holland.
Fauconnier, G. (1975): Pragmatic Scales and Logical Structures, in Linguistic Inquiry 6: 353-
375.
von Fintel, K. (1993): Exceptive Constructions. Natural Language Semantics 1.2: 123-148.
von Fintel, K. (1994): Restrictions on quantifier domains, PhD dissertation University of
Massachusetts Amherst, GLSA.
von Fintel, K., Fox, D., Iatridou, S. (in progress): Notes on the Meaning of the Definite Article.
ms. MIT
von Fintel, K., Iatridou, S. (2003): Since Since. handout MIT
Fox, D. (2000): Economy and Semantic Interpretation, Linguistic Inquiry Monographs 35,
MITWPL and MIT Press, Cambridge, MA.
Fox, D. (2003): Implicature Calculation, Only, and Lumping: Another Look at the Puzzle of
Disjunction. handout, Yale University http://web.mit.edu/linguistics/www/fox/Yale.pdf.
Fox, D. (2004): Implicatures and Exhaustivity. Handouts from a class taught at USC,
http://web.mit.edu/linguistics/www/fox/index.html.
Fox, D. (in progress):
Frampton, J. (1991): Relativized Minimality: A review. The Linguistic Review 8: 1-46.
Gajewski, J. (2002): L-Analyticity in Natural Language, ms. MIT
Gajewski, J. (2003): Connected Exceptives and NPI-any. generals paper, MIT
Geurts, B., Nouwen, R. (2005): At least et al.: the semantics of scalar modifiers. Ms University
of Nijmegen.
Gordon, P. (2004): Numerical Cognition without Words: Evidence from Amazonia. Science 306:
496-499
Groenendijk. G., Stokhof, M. (1984): Studies on the Semantics of Questions and the Pragmatics
of Answers. PhD dissertation, University of Amsterdam.
Guerzoni, E. (2003): Even-NPIs in Yes/No Questions. Natural Language Semantics 12: 319-343.
Hackl, M (2000): Comparative Quantifiers. PhD dissertation MIT, MITWPL, Cambridge, MA.
64
Hamblin, C. (1973): Questions in Montague Grammar. Foundations of Language 10:41-53.
Heim, I. (1984): A Note on Negative Polarity and DE-ness. In C. Jones and P. Sells (eds.)
Proceedings of NELS 14, GLSA, Amherst, Mass.:98-107.
Heim, I. (1987): Where does the Definiteness Restriction Apply? In E. Reuland and A. ter
Meulen (eds): The Representation of (In)definiteness. MIT Press, Cambridge, MA.
Heim, I. (1994): Interrogative Complements of Know. In R. Buchalla and A. Mittwoch (eds.),
Proceedings of the 9
th
Annual IATLConference and the 1993 IATL Workshop on
Discourse, Akademon, Jerusalem: 128-144
Heim, I. (2001): Degree Operators and Scope. C. Fry and W. Sternefeld (eds.), In Audiatur Vox
Sapientiae. A Festschrift for Arnim von Stechow, Berlin: Akademie-Verlag: 214 - 239
Heim, I. (2003): Class notes from a seminar on comparatives co-taught with C. Kennedy MIT
Fall 2003
Higginbotham, J. (1993): Interrogatives. In K. Hale and S. Keyser (eds.), The View from
Building 20: Essays in Honor of Sylvain Bromberger, MIT Press, Cambridge, MA: 195-
227.
Horn, L. (1969): A presuppositional analysis of only and even. CLS 5: 97-108
Horn, L. (1972): On the Semantic Properties of Logical Operators in English. UCLA
dissertation. Distributed by IULC, 1976.
Kennedy, C. (2001): Polar Opposition and the Ontology of `Degrees'. Linguistics and Philosophy 24:
33-70.
Krifka, M. (1995): The Semantics and Pragmatics of Polarity Items. Linguistic Analysis 25: 1-49
Krifka, M. (1999): At least some Determiners aren't Determiners, in K. Turner (ed.), The
Semantics/Pragmatics Interface from Different Points of View. (= Current Research in
the Semantics/Pragmatics Interface Vol. 1). Elsevier Science B.V.: 257-291
Krifka, M. (2002): Be Brief and Vague! And how Bidirectional Optimality Theory Allows for
Verbosity and Precision, in D. Restle & D. Zaefferer (eds): Sounds and Systems. Studies
in Structure and Change. A Festschrift for Theo Vennemann, Mouton de Gruyter (=
Trends in Linguistics 141), Berlin: 439-458.
Kroch, A. (1989): Amount Quantification, Referentiality, and Long Wh-movement. Ms.
University of Pennsylvania.
Kuno, S., Takami, K., (1997): Remarks on Negative Islands. Linguistic Inquiry 28: 553-576.
Ladusaw, W. (1986): Principles of Semantic Filtering, WCCFL 1: 129-141
Lahiri, U. (1998): Focus and Negative Polarity in Hindi. Natural Language Semantics 6.1: 57-
123.
Lahiri, U. (2002): Questions and Answers in Embedded Contexts. Oxford University Press,
Oxford.
Lasersohn, (1999): Pragmatics Halos. Language 75.3: 522-551.
Levinson, S. (2000): Presumptive Meanings. MIT Press.
Link. G. (1983): The Logical Analysis of Plurals and Mass Terms: A Lattice-Theoretic
Approach. In R. Buerle, et. al. (eds.): Meaning, Use, and Interpretation. De Gruyter,
Berlin: 302-323.
Marantz, Alec (1994): A Late Note on Late Insertion, in Y. Kim, et. al. (eds.): Explanation in
Generative Grammar: A festschrift for Dong-Whee Yang, Seoul, Hankuk: 396-413
Matsumoto, Y. (1995): The Conversational Condition on Horn Scales, Linguistics and
Philosophy 18: 21-60
Menendez-Benito, P. (2005): The Grammar of Choice. PhD dissertation University of
Massachusetts Amherst.
65
Obenauer, H. (1984): On the Identification of Empty Categories. Linguistic Review4.2: 153-202
Pica, P., Lemer, C., Izard, V., Dehaene, S. (2004): Exact and Approximate Arithmetic in an
Amazonian Indigene Group, Science 306: 499-503
Pinkal, M. (1989): Die Semantik von Satzkomparativen. Zeitschrift fr Sprachwissenschaft 8:
206-256.
Rizzi, L. (1990): Relativized Minimality. MIT Press, Cambridge, MA.
Rullmann, H. (1995): Maximality in the Semantics of Wh-constructions. PhD dissertation
University of Massachusetts Amherst, GLSA.
van Rooy, R., Schulz, K. (2004): Exhaustive Interpretation of Complex Sentences, Journal of
Logic, Language and Information 13: 491-519
Sher, G. (1991): The Bounds of Logi: A Generalized Viewpoint. The MIT Press, Cambridge, MA
Spector, B. (2004): Intervention Effects in How-Many Questions and the Syntax/Semantics
Interface. Abstract for Sinn und Bedeutung 9.
Spector, B. (2006): Aspects de la pragmatique des oprateurs logiques, Thse de doctorat,
Universit Paris 7.
von Stechow, A. (1984): Comparing Semantic Theories of Comparison. Journal of Semantics 3:
1-77.
Stalnaker, R. (1974): Pragmatic Presuppositions. in M. Munitz and P. Unger (eds.): Semantics
and Philosophy. New York University Press: 197-214
Stalnaker, R. (1999): Context and Content. Oxford University Press, New York, NY.
Szabolcsi, A., Zwart, F. (1993): Weak Islands and Algebraic Semantics for Scope Taking,
Natural Language Semantics 2: 1-50
Westerstahl, D. (1984): Determiners and Context Sets. in J. van Benthem and A. ter Meulen
(eds.): Generalized Quantifiers in Natural Language, Dordrecht, Foris: 45-71
Zucchi, A. (1995): The ingredients of definiteness and the definiteness effect. Natural Language
Semantics 3: 3378