Demonstratives As Individual Concepts

Demonstratives as Individual Concepts
Paul Elbourne (p.d.elbourne@qmul.ac.uk)

Queen Mary, University of London
Abstract. Using a version of situation semantics, this article argues that bare and
complex demonstratives are interpreted as individual concepts.
Keywords: Demonstratives, individual concepts, pronouns, situation semantics
1. Introduction
This article attempts to give a unified semantics for the English demonstrative determiners that and this. In particular, it will argue that they
are both definite articles of a certain kind, where definite articles will be
assumed to introduce existence and uniqueness presuppositions in the
manner described by Frege (1893) and Strawson (1950). DPs headed by
demonstrative determiners will be interpreted as individual concepts.
By a unified semantics I mean an account that is capable of handling all the uses to which this and that seem to be put. The classic
theory of Kaplan 1989a can handle only a proper subset of the uses
of these words. As is well-known, Kaplan focused on occurrences of
demonstratives like the one in (1), that is to say on utterances in which
the speaker demonstrates and intends to talk about some perceptually
salient object.
(1) That [gesture at Venus] is a planet.
The resulting doctrine was that demonstratives in these uses are directly referential, where a singular term is directly referential if and
For valuable discussion of this topic I would like to thank Sigrid Beck, Ray
Buchanan, Michael Glanzberg, Irene Heim, Richard Kayne, Lisa Levinson, Stephen
Neale, Bernhard Nickel, Gary Ostertag, Francois Recanati, Stephen Schiffer, Susanna Siegel, Jason Stanley, Peter Staudacher and Anna Szabolcsi. I am particularly
grateful to Daniel B
uring and two anonymous reviewers for Linguistics and Philosophy for detailed and perceptive commentary on a first draft. I should emphasize
that the basic view defended in this article (that bare and complex demonstratives
are individual concepts) is not mine alone. In particular, Irene Heim and I both
argued for it strongly at a 2002 Harvard-MIT seminar on reference and complex
demonstratives organized by Glanzberg and Siegel, apparently having come to the
same conclusion independently; Ray Buchanan came to the same conclusion, again
independently, in a 2002 paper; and the same idea occurred independently to Jason Stanley, who speaks favorably of it in a review of King 2001 (Stanley 2002).
This research was partly funded by the Deutsche Forschungsgemeinschaft as part of
Sonderforschungsbereich 632 (Information Structure).
c 2007 Kluwer Academic Publishers. Printed in the Netherlands.

demoniacal.tex; 21/08/2007; 19:21; p.1
Paul Elbourne
only if the only contribution it makes to the content of an utterance

containing it is its referent. The referent is determined by the context
of utterance and the expressions character, an aspect of its meaning
consisting of a function from contexts of utterance to contents. Since
only single referents are able to be contributed, there is no way that this
theory is able to handle occasions where demonstratives appear to have
covarying uses, as Kaplan himself freely acknowledges (1989a: 489490,
1989b: 592). It cannot handle (2), for example, where on one natural
reading that senator is a covarying expression, giving the approximate
truth conditions in (3).
(2) Mary talked to no senator before that senator was lobbied.
(3) There is no individual x such that x is a senator and Mary talked
to x before x was lobbied.
A survey of the basic data, including various kinds of covarying uses
of demonstratives, will be undertaken in section 3. But for now, since
Kaplans theory and close variants of it cannot handle all the data
thrown up when we look at how demonstrative determiners are used,
there is a strong case for saying that this approach is inadequate, as
King (2001: 147161) has already emphasized. In order to account for
(2) and the examples we will see in section 3, Kaplan has to say either
that these words are systematically ambiguous, or, still more drastically, that they have homonyms in which he is not interested (Kaplan
1989a: 489). But since we are dealing with cases that plausibly have
some intuitive common core (unlike, say, the case of bank and bank ),
Occams Razor dictates that theories that have all the uses follow from
one lexical entry per word are superior to those which posit polysemy
or homonymy.
I know of only two theories that attempt to give a unified semantics
for the English demonstrative determiners, those of King (2001), who
argues that DPs introduced by this and that are quantifier phrases, and
Roberts (2002), who treats them as definites in a dynamic semantics.
Accordingly, the present paper will have the following structure: section
2 introduces the semantic framework in which I will be couching my
discussion; section 3 sets out my theory of demonstrative determiners,
including a survey of what I take to be the basic data in this field;
section 4 summarizes the theories of King (2001) and Roberts (2002)
and compares them to the one advocated here; and section 5 concludes.
demoniacal.tex; 21/08/2007; 19:21; p.2
2. Framework
2.1. Situation Semantics
This paper will use a version of situation semantics (Barwise and Perry
1983). The particular version I will use is based closely on the system in Elbourne 2005b, which I will here summarize for the sake of
convenience.
The system in Elbourne 2005b is based most directly on the work
of Kratzer (1989), Berman (1987), Heim (1990), von Fintel (1994) and
Heim and Kratzer (1998). It is based on the notion of a situation, where,
as in Barwise and Perry 1983, a situation is a part of a possible world.
(The part-of relation is understood reflexively, so that possible worlds
count as big situations.) A situation consists of one or more individuals
having one or more properties or standing in one or more relations at
a particular spatiotemporal location (Barwise and Perry 1983: 7).
Unlike in some versions of situation semantics, lexical items in Elbourne 2005b are not accompanied by situation variables in the syntax. So for the simple sentence Mary laughs we would just have the
(simplified) structure in (4).
(4) [Mary laughs]
For the purposes of this section, I will make the simplifying assumption that proper names are syntactically simplex lexical items whose
denotations are individual concepts of a certain kind, those that map
situations directly to individuals, as it were, without the mediation of
a descriptive condition. (This view will be revised slightly in section
2.2.) So we will have the lexical entry in (5) for Mary.
(5) [[Mary]] = s.Mary
Intransitive verbs will be functions that take individual concepts and
map them to functions from situations to truth values. For example,
we have (6) for laughs.
(6) [[laughs]] = uhs,ei . s. u(s) laughs in s
I follow a notational convention from Heim and Kratzer 1998. For any
expressions M and N and any variable x, let [N/x]M be the result of
substituting N for every free occurrence of x in M , and changing bound
variables to avoid clashes.1 A -term consisting of a and a variable
x and a period followed by a constituent M is a function mapping
1
For the exact definition, see Hindley and Seldin 1986:7, from which the summary
in the text is also taken.
demoniacal.tex; 21/08/2007; 19:21; p.3
Paul Elbourne
an argument N to [N/x]M if M is not an expression of propositional

type in the metalanguage; but if M is an expression of propositional
type in the metalanguage, the whole expression is a function mapping
an argument N to 1 if [N/x]M is true and to 0 otherwise. Thus the
denotation of laughs in (6) takes an individual concept u and maps it
to a function that takes a situation s and maps it to 1 if u(s) laughs in
s and to 0 otherwise.
We will need the following basic rule (Heim and Kratzer 1998):
(7) Functional Application
If is a branching node and {, } the set of its daughters, then,
for any assignment g, is in the domain of [[ ]]g if both and
are, and [[]]g is a function whose domain contains [[]]g . In that
case, [[]]g = [[]]g ([[]]g ).
I assume that all nodes in syntactic trees are at most binary branching
(Kayne 1984).
Given this simple setup, the denotation of the sentence in (4) turns
out to be (8).
(8) s. Mary laughs in s
Similarly, the sentence in (9), given the treatment of names just described and the lexical entry in (10), will have the denotation in (11).
(9) [Mary [greets John]]
(10) [[greets]] = uhs,ei . vhs,ei . s. v(s) greets u(s) in s
(11) s. Mary greets John in s
I take the content of a speech act of assertion to be an Austinian proposition, that is a pair of a topic situation about which the speaker intends
to say something and a set of situations like the ones characterized in
(8) and (11). If the topic situation is a member of the set, the speaker
has spoken truly (Austin 1961, Barwise and Perry 1983: 160, Kratzer
2004, 2006).2
2
An anonymous Linguistics and Philosophy reviewer asks how the framework

used here can be modified to yield structured propositions, following the attempt
by Soames (1987) to demonstrate that no sets of circumstances of any kind are adequate to play the role of propositions. But Soamess argument relies crucially on the
premise that certain singular terms, such as names, pronouns and demonstratives,
must be directly referential in the examples he uses, which are of the Hesperus is
Phosphorus type. This premise I take to be undermined by considerations like the
ones given in the current article, and also by arguments in Elbourne 2005b. Edelberg
1994 contains independent criticism of Soames 1987. But if structured propositions
do turn out to be necessary, for example as the objects of propositional attitudes,
demoniacal.tex; 21/08/2007; 19:21; p.4
2.2. Definite Descriptions and Names

In this article, I will be arguing that bare and complex demonstratives
are very similar to definite descriptions. In Elbourne 2005b, Chapter
3, I argued that definite descriptions are individual concepts. For the
sake of comparison, I will briefly recapitulate this position here.
Based on some remarks in Frege 1893, the meaning in (12) has been
suggested for sentences involving definite descriptions (Heim 1991: 495
496, Elbourne 2001b, 2005b, von Fintel 2004: 318).
(12) Regardless of the utterance context, [[the ] ] expresses that
proposition which is
true at an index i, if there is exactly one at i, and it is
at i,
false at an index i, if there is exactly one at i, and it is
not at i,
truth-valueless at an index i, if there is not exactly one at
i.
We can capture this by giving the definite article the meaning in (13).
(13) [[the]] = fhse,sti .s. xf (s0 .x)(s) = 1
I will be using the metalanguage operator pq as follows: for any situation s and function f of type hse, sti, the denotation of an expression
px f (s0 .x)(s) = 1q will be of type e, if it is defined; if there is
exactly one entity x such that f (s0 .x)(s) = 1, the denotation of
px f (s0 .x)(s) = 1q will be that very individual; if there is no such
individual, or more than one such individual, the whole expression
will have no semantic value. (So the expression in effect introduces
a presupposition that there is exactly one such individual, since an
utterance containing it will not be felicitous otherwise.) The individual
that is the value of the expression will naturally vary from situation to
situation. For example, let f be [uhs,ei .s.u(s) has been President of
the USA in s]. Then px f (s0 .x)(s) = 1q is equivalent to px x has been
President of the USA in sq. Assume that we are talking about situations
that are part of the real world at the beginning of 2007, when George H.
W. Bush and Bill Clinton have been President of the USA, but Henry
Kissinger, Arnold Schwarzenegger and Al Gore have not. If s contains
they could easily be created from intensional semantic systems like the present one.
A natural starting point would be interpreted syntactic trees, that is syntactic tree
structures with the meanings of the lexical items replacing the lexical items at each
node. Further structures could be created on the basis of these if necessary. See
Cresswell 1985 for a system of structured propositions along these lines.
demoniacal.tex; 21/08/2007; 19:21; p.5
Paul Elbourne
only Bill Clinton, Henry Kissinger and Arnold Schwarzenegger, then

the denotation of px x has been President of the USA in sq is Clinton;
if s contains only George H. W. Bush, Henry Kissinger and Arnold
Schwarzenegger, the value of the same expression will be Bush; and if
s contains only Al Gore, Henry Kissinger and Arnold Schwarzenegger,
the expression will have no value. This, simply put, is how definite
descriptions differ from constants.
(14) will have the simplified LF in (15).
(14) The cat laughs.
(15) [[the cat] laughs]
Our lexical entry for the, in conjunction with the lexical entries in (16)
and the rule of functional application, gives us the truth conditions in
(17) for (14).
(16) a. [[cat]] = uhs,ei . s. u(s) is a cat in s
b. [[laughs]] = uhs,ei . s. u(s) laughs in s
(17) s.x x is a cat in s laughs in s
Imagine that a speech act is performed whereby a topic situation s is
claimed to be a member of the set of which (17) is the characteristic
function. If s contains exactly one cat, then the -term in the metalanguage will have a denotation, and the utterance will be true or false
according as that cat laughs or not. But if s does not contain exactly
one cat, the -term will not have a denotation, and the proposition will
be unable to be evaluated for truth or falsity. So we have produced the
kind of proposition described in (12). In effect, then, use of a definite
description introduces a presupposition that there is exactly one entity
that satisfies the NP content, just as Strawson (1950) claimed.3
All this, of course, contrasts with Russells well-known (1905) theory
of definite descriptions, according to which the existence and uniqueness of the thing satisfying the NP descriptive content are asserted, not
presupposed. According to Russell (1905), a sentence of the form the F
is G has the truth conditions in (18). This implies that the would have
the lexical entry in (19) in an extensional semantics.
(18) x(F x & y(F y y = x) & Gx)
(19) fhe,ti . ghe,ti . x(f (x) = 1 & y(f (y) = 1 y = x) &
g(x) = 1)
3
I wish to remain neutral on the issue of how to deal with improper or incomplete definite descriptions. It is possible that the answer lies in situation semantics,
including the judicious choice of topic situations, as claimed by Recanati (2004) and
Kratzer (2004). But I will not attempt to adjudicate on this matter here.
demoniacal.tex; 21/08/2007; 19:21; p.6
In my opinion, there are considerations favoring the Frege-Strawson

theory of the definite article over the Russellian theory, but I will not
go into them here, since they have been detailed elsewhere: see Heim
1991, von Fintel 2004, and Elbourne 2005b.
We are now in a position to contemplate a slight revision to the view
of names set out in the last section. As described in Elbourne 2005b,
Chapter 6, I believe that names should quite generally be analyzed as
syntactic and semantic definite descriptions of a certain kind (Burge
1973, Larson and Segal 1995, Elugardo 2002). We see this overtly in
languages such as German (der Hans, dialectally), Italian (la Maria)
and Classical and Modern Greek. But the descriptive condition will
generally involve identity to a particular object, thus ensuring that
names in general behave as rigid designators (Kripke 1972). I will
assume that names come with a phonologically null definite article,
as in (20):
(20) the Mary
the will have the same meaning as overt the. For the purposes of this
article, I will assume that names just have meanings like (21):
(21) [[Mary]] = uhs,ei . s. u(s) = Mary
So the meaning of the complex in (20) will still be (5), repeated here
as (22).
(22) [[the Mary]] = s.Mary
Most of what I say will not depend on this particular theory of names,
and I will not always bother to write the null definite article shown
in (20); but this theory will be relevant in the discussion of referential
pronouns in section 2.4.8.
2.3. De re and de dicto readings
I will now describe the method by which the current theory handles de
re and de dicto readings of definite descriptions (Elbourne 2005b: 99
106), since it introduces a technical device that will be useful later on.
The Russellian theory of definite descriptions is supposed to be able to
handle the distinction between de re and de dicto readings by means of
scope (Russell 1905: 52). To take an example involving a propositional
attitude predicate, we can analyze the de re and de dicto readings of
(23) as in (24) and (25) (Neale 1990: 121):
(23) Mary believes that the Pope is an alien.
demoniacal.tex; 21/08/2007; 19:21; p.7
Paul Elbourne
(24) De dicto
Mary believes that there is an individual x such that x is a
Pope and there are no Popes but x and x is an alien.
(25) De re
There is an individual x such that x is a Pope and there are no
Popes but x and Mary believes that x is an alien.
This result, in fact, is sometimes claimed to be one of the main virtues
of the Russellian theory of descriptions.
Semantic systems based on situations can reproduce this result by
directly manipulating the situations with respect to which certain predicates are evaluated (Bauerle 1983, Heim 1991, Farkas 1997, Heim,
Kratzer and von Fintel 1998, Percus 2000, Elbourne 2005b). The idea
is that we can account for the two readings in question by means of the
following paraphrases, where w0 is the actual world:
(26) De dicto
All worlds w compatible with Marys beliefs in w0 are such that
the Pope in w is an alien in w.
(27) De re
All worlds w compatible with Marys beliefs in w0 are such that
the Pope in w0 is an alien in w.
Imagine that Mary has no particular beliefs about who the Pope is; she
just thinks that the present Pope, whoever he is, is an alien. We could
describe this state of affairs by means of (26). Now imagine that she has
seen a figure dressed in white behaving suspiciously in Vatican City;
the person she saw was in fact the current Pope, Benedict XVI, but
she does not know this; she just forms a belief concerning the person
she saw to the effect that he is an alien. We could describe this state
of affairs by means of (27). But this, of course, is just the distinction
between de dicto and de re readings.
Some theorists have advocated systems in which every predicate
takes a situation variable as its first argument (Heim 1990, Percus 2000,
B
uring 2004). This would allow us to say that the situation variable
associated with the noun Pope in (23) refers to w0 in the case of the
de re reading, and is bound by the propositional attitude verb in the
case of the de dicto reading. See Elbourne 2005b: 100102 for further
details.
It is not necessary to introduce so many situation variables into the
syntax, however, in order to deal with the data at hand. In Elbourne
2005b: 103104, I propose the following theory, which is in a certain respect more economical. We posit one additional operator in the syntax,
which is shown in (28).
demoniacal.tex; 21/08/2007; 19:21; p.8
(28) [[s0 ]]g = fhse,sti .uhs,ei .s.f (u)(g(0)) = 1

This is designed to take a noun or NP as argument4 and give back the
denotation of this nominal modified so as to take as its second argument
a contextually salient situation. Here, for example, is the meaning of
the noun Pope, with and without an accompanying s0 operator, with
the operator introducing the actual world when it appears:
(29) [[Pope]] = uhs,ei . s. u(s) is a Pope in s
(30) [[Pope s0 ]][0w0 ] = uhs,ei . s. u(w0 ) is a Pope in w0
I take it that only actual situations can be salient.
Given the lexical entry in (13) for the definite article, repeated here
as (31), we arrive at two possible meanings for the Pope, shown in (32)
and (33).
(31) [[the]] = fhse,sti .s. xf (s0 .x)(s) = 1
(32) [[the Pope]] = s. x x is a Pope in s
(33) [[the [Pope s0 ]]][0w0 ] = s. x x is a Pope in w0
For the sake of convenience, we can assume that is an alien has the
denotation in (34), like any other predicate.
(34) [[is an alien]] = uhs,ei .s.u(s) is an alien in s
This means that the Pope is an alien, in our example, will have the
denotation in (35) if s0 is used and the denotation in (36) if it is not.
(35) s.x x is a Pope in w0 is an alien in s
(36) s.x x is a Pope in s is an alien in s
By means of the lexical entry in (37) for believes, we arrive at the
following LFs and meanings for the de dicto and de re readings of our
example.
(37) [[believes]] = phs,ti .uhs,ei .s.all worlds w compatible with the
beliefs of u(s) in s are such that p(w) = 1
(38) De dicto:
a. [Mary believes [the Pope is an alien]]
b. The proposition true of world w0 iff all worlds w compatible
with Marys beliefs in w0 are such that the unique x such
that x is a Pope w is an alien in w.
4
It must not be able to be associated with Verb Phrases, however, as Percus 2000 makes clear. We can deal with this restriction by means of syntactic
subcategorization.
demoniacal.tex; 21/08/2007; 19:21; p.9
10
Paul Elbourne
(39) De re:
a. [Mary believes [the Pope s0 is an alien]]
b. The proposition true of world w0 iff all worlds w compatible
with Marys beliefs in w0 are such that the unique x such
that x is the Pope in w0 is an alien in w.
It can be seen that we have produced intuitively adequate renderings
of the de dicto and de re readings of a representative example.
Moreover, there are powerful arguments to the effect that the treatment of de dicto and de re readings just described is actually empirically
superior to the theory that uses scope. See Bauerle 1983 and Elbourne
2005b: 104106 for details.
2.4. Pronouns and Binding
It will be useful at this point to examine the treatment of third-person
pronouns in our fragment, since they will form a useful point of comparison to demonstratives.
2.4.1. Pronouns as definite descriptions
There is a substantial amount of evidence to indicate that pronouns are
interpreted as definite descriptions.5 I will here mention three pieces of
evidence. The first relevant phenomenon is that of donkey anaphora; I
have argued at length elsewhere (Elbourne 2005b) that the approach to
donkey anaphora that treats donkey pronouns as definite descriptions
(Cooper 1979, Heim 1990, Neale 1990, Elbourne 2001a) is the correct
one.
Secondly, there is an argument by Heim (1993) to the effect that
treating pronouns as descriptions helps to explain some otherwise puzzling apparent violations of Chomskys (1981) Binding Theory. (See
also B
uring 2005:156157 for further discussion.) Imagine a scenario in
which two people are looking at a political candidate speaking on a
soapbox. They cannot see the person clearly and are wondering who it
is. They might speak as follows (Heim 1993):
(40) A: Is that Zelda?
B: Shes praising her to the skies. It must be Zelda.
5
To be exact, all of the phenomena I cite here constitute evidence to the effect
that pronouns are sometimes interpreted as definite descriptions; the final step of
concluding that they are always definite descriptions, with referential uses employing
descriptive conditions like identical to John (Elbourne 2001b, 2005b), is made on
the basis of Occams Razor. There seems to be something more than an accidental
homonymy going on here, especially when we look at other languages and see that
their pronouns too seem to be ambiguous, for example, between referential, bound
and donkey anaphoric uses.
demoniacal.tex; 21/08/2007; 19:21; p.10
11
B is convinced that the speaker is Zelda. We can also suppose, if we

think it makes a difference, that B is correct in this conviction. We
might expect, therefore, that the first sentence of Bs utterance would
be a violation of Chomskys (1981) Principle B, which (among other
things) prevents non-reflexive pronouns in object position from having
the same referent as a subject singular term. Compare the last sentence
of (41), which cannot mean Zelda is praising herself to the skies.
(41) A: Whats Zelda doing?
B: *Shei is praising heri to the skies.
The last sentence of (41), on the given interpretation, is correctly ruled
out by Principle B. There is no Principle B violation in the same
sentence when it occurs in Bs utterance in (40), however. Intuitively,
what seems to be going on is the following (Heim 1993): even though
the two pronouns in this utterance end up referring to the same person,
they get there by different routes, as it were; the first one uses the
property of being the speaker on the soapbox over there, whose identity
is in question, whereas the second one refers directly to Zelda. Heim
(1993) suggests that we spell out this intuition by means of individual
concepts. The meaning of the first pronoun will be something like (42a)
(where SB is a name of the soapbox in question), and the meaning of
the second will be something like (42b).
(42) a. s.x x is a speaker on SB in s
b. s. Zelda
We can now explain the lack of a Principle B violation in (40) by saying
that Principle B is computed over individual concepts. We might say
that a Principle B violation will occur when a pronoun and another
singular term in the relevant configuration are interpreted by means
of the same individual concept, but will not occur when they stand
for different individual concepts. The two pronouns stand for different
individual concepts in (40) but this is plausibly not the case in (41),
where [s. Zelda] will presumably be the value each time. Hence the
difference in status. One could of course object that one could avoid a
Principle B violation even in a context like (41) by imagining different
individual concepts that would yield Zelda in the current evaluation
situation and employing them for the two pronouns in question. In order
to avoid this, we can add a natural prohibition against two singular
terms standing for distinct individual concepts that are presupposed
in the common ground to be coreferential when evaluated with respect
to the current evaluation situation (Heim 1982, Heim 1993, B
uring
2005:157). Note that in (40) there is a debate about the identity of the
demoniacal.tex; 21/08/2007; 19:21; p.11
12
Paul Elbourne
political candidate, and so the common ground does not entail that
she and her end up picking out the same person, given that her refers
to Zelda. Interestingly, it has been argued that some of the difficulties
that small children display with Binding Theory are due not to any
lack of knowledge of the Binding Principles but to a lack of knowledge
of this supplementary principle (Thornton and Wexler 1999).6
The third reason for thinking that pronouns are definite descriptions
is the existence of the cases sometimes called descriptive indexicals
(Nunberg 1993, Recanati 1993, Nunberg 2004, Recanati 2005the term
is Recanatis, but the pioneering work in this area was done by Nunberg). To adapt an example from Recanati 2005, credited to Nunberg,
suppose I point at Benedict XVI and say (43):
(43) He is usually an Italian.
It is obvious that I will interpreted as saying that the Pope is usually
an Italian; I will not be interpreted as saying that Benedict XVI is usually an Italian. Note that we cannot explain this example by claiming
that he has Benedict XVI as its semantic value and that the meaning
The Pope is usually an Italian is derived from this by some kind of
pragmatic process; if that were the case, we would expect an utterance
of (44) to have the same reading:
(44) Benedict XVI is usually an Italian.
The name Benedict XVI will presumably introduce Benedict XVI into
the proposition expressed, on which basis the putative pragmatic process that is supposed to produce the meaning of (43) should surely be
able to operate; but (44) cannot have the meaning of (43), causing a
grave difficulty for any attempt to explain away descriptive indexicals
on the basis of direct reference and pragmatics (Nunberg 1993, 2004).
2.4.2. The theory of Nunberg 1993
My strategy in this section will be to produce a formalization of the theory of Nunberg 1993 in the current version of situation semantics, and
to show that it can account for the three phenomena just introduced,
as well as the referential and bound variable uses of pronouns. Nunberg (1993) claims that the semantics of pronouns and other indexicals
involves the following four elements:
1. A deictic component, which picks out a contextually salient object
called an index, on the basis of which the actual interpretation of
the indexical will be computed.
6
But see Elbourne 2005a and Takahashi et al. 2006 for critical discussion.
demoniacal.tex; 21/08/2007; 19:21; p.12
13
2. A relational component, which constrains the relation that must

hold between the index and the interpretation.
3. A classificatory component, including things like -features, which
adds further information about the interpretation.
4. An interpretation, which is an individual or definite description7
contributed to the proposition expressed.
Following Nunbergs strategy in his 1993 article, I will first illustrate the
theory with the example of the first-person plural pronoun we, before
going on to the third-person pronouns that are my main concern.
In the case of we, then, the deictic component picks out the speaker.
(The deictic component of we is identical to the deictic component of I.)
The classificatory component ensures that the interpretation is plural
and animate. (I cannot refer to myself and my laptop as we, unless I
wish to personify the computer.) The relational component specifies
that the interpretation must be either a plural individual of which the
index is a part or a definite description whose actual instantiation is
such that the index is a part of it.8 For example, if Nunberg says (45),
referring to himself and his wife, we presumably refers to the plural
individual whose parts and Nunberg and his wife (Nunberg 1993: 12,
Link 1983):
(45) We cant come to dinner that night.
On the other hand, suppose that John has just been elected to the
Kansas State Board of Education, a body known for changing hands
between advocates of Intelligent Design and advocates of Darwinian
evolution. It so happens that after this election all of the members of
the Board are advocates of Intelligent Design. If John says (46) to the
other members of the Board, his utterance will be ambiguous.
(46) We might have been evolutionists.
On one reading, he is saying of the actual members of the Board that
they might have been evolutionists. On another reading, however, he is
7
Nunberg (1993) frequently talks about indexical contributing either individuals

or properties, but the properties generally turn up as part of definite descriptions,
as in example (43) where we have the Pope. Nunberg also points out (1993: 12)
that pronouns sometimes seem to have the interpretation of bare plurals, as when
a woman says We are less likely to contract the disease than men are, meaning
Women are less likely to contract the disease than men are. I follow Carlson 1977
in supposing that bare plurals refer to kinds, which are a complex sort of individual.
8
This is my own summary of the relational component of we, based on examples
given by Nunberg (1993). I follow Link 1983 in supposing that plural pronouns often
refer to plural individuals, that is individuals that have other individuals as parts.
demoniacal.tex; 21/08/2007; 19:21; p.13
14
Paul Elbourne
saying that there might have been other, more Darwin-friendly, people
elected to the Board. This last reading, at least, seems to require we
to be synonymous with the definite description the current members of
the Kansas State Board of Education (Nunberg 1993: 1314). Compare
the two readings of (47):
(47) The members of the Board might have been evolutionists.
There is evidence, then, in favor of the contention that we can have
readings equivalent to those of definite descriptions.
Third-person pronouns work similarly to we in Nunbergs (1993)
theory. The differences are that the index can be any salient object,
instead of having to be the speaker, and that the relational component
can fix on any salient relation, within certain limits (Nunberg 1979,
1993). If I point to Benedict XVI and say he, then the index is going to
be Benedict XVI.9 If I mean to refer to Benedict XVI, that is if I wish
Benedict XVI to be the interpretation of the pronoun, the relation in
play is going to be identity. If I wish to contribute to the proposition
expressed a definite description synonymous with the Pope, as in example (43), the relation is going to be something like the relation that
maps people to the offices they hold.
2.4.3. A formalization of Nunberg 1993
I will formalize this theory in a straightforward way, positing items in
the syntax corresponding to the deictic component and the relational
component. The pronoun it, for example, will be as in (48):
(48) [it [R1 i2 ]]
In this structure, i is a variable of type e constituting the deictic component; its value will be the index, in Nunbergs terms. R is a variable
of type he, hse, stii constituting the relational component. These will be
the only types of variable in the current system. The value of R will, of
course, map the value of i to a function of type hse, sti; note that this
is the type of Noun Phrases in the current framework. The pronoun it,
then, will be a definite article (Postal 1966, Stockwell, Schachter and
Partee 1973, Abney 1987, Longobardi 1994, Uriagereka 1995, Elbourne
2001a, 2001b, 2005b, Neale 2005b). Abstracting away from -features,
which would be written into the denotation as presuppositions in complete treatment, perhaps in the manner advocated in Heim and Kratzer
9
We can suppose, with Neale 2005b, that the speakers intentions are ultimately
what determine the index and the relation selected by the relational component, but
that cooperative speakers will not intend to pick out entities whose identities they
think their audience would not be able to work out.
demoniacal.tex; 21/08/2007; 19:21; p.14
15
1998, it and other third-person singular pronouns will mean the same
as the:
(49) [[it]] = fhse,sti .s. xf (s0 .x)(s) = 1
Pronouns with their accompanying variables, then, will be definite
descriptions.
2.4.4. Descriptive Indexicals
Let us return to our descriptive indexical example (43), repeated here
as (50):
(50) He is usually an Italian.
We can now deal with (50) as follows. The pronoun he will take indexed
variables as shown in (51); these variables will be interpreted by means
of the rule in (52).
(51) [he [R1 i2 ]]
(52) Variable Interpretation
For all natural numbers n and assignment functions g, if in is a
variable with subscript n, then
[[in ]]g = g(n)
provided n is in the domain of g; [[in ]]g is undefined otherwise.
The individual variable i will have Benedict XVI as its semantic value.
Suppose the utterance is made in world w0 at time t0 . The relation
variable R would then have as its value the relation in (53):
(53) x.uhs,ei .s.u(s) holds in s the office held in w0 at t0 by x
The word he, as mentioned, will be interpreted as a definite article.
This means that the semantic value of the DP as a whole will be as
follows:
(54) s. x x holds in s the office held in w0 at t0 by Benedict XVI
Since Benedict XVI is, we can assume, the Pope in w0 at t0 , this formula
is equivalent to the following:
(55) s. x x is the Pope in s
The sentence He is an Italian, with the pronoun interpreted in the way
just indicated, will have the semantic value in (56):
(56) s. x x is the Pope in s is an Italian in s
demoniacal.tex; 21/08/2007; 19:21; p.15
16
Paul Elbourne
We can now follow Berman 1987, Heim 1990 and Elbourne 2005b in
supposing that quantificational adverbs like usually quantify over situations. The idea is that usually says that most situations of a certain
kind are members of the set of situations that is the denotation of
the matrix clause. We can further assume that, when no restrictor is
given explicitly by means of an if -clause or when-clause, the restrictor
is tacitly formulated by the speaker and constructed on the fly by the
audience, in their best guess at the speakers intentions. Let us assume
for the present example that the speaker intends to quantify over papal
reigns. The speaker of (43) would be claiming, then, that for most
situations s such that s is a papal reign, the unique Pope in s is an
Italian in s. This seems to be intuitively adequate.
2.4.5. Referential pronouns
Let us now turn to the referential case, in which I point to Benedict
XVI and mean to contribute him, not the property of being Pope, to
the proposition expressed. The variables i and R will have the values in
(57). This gives the value in (58) for the combination of pronoun and
variables.
(57) a. [[i]] = Benedict XVI
b. [[R]] = x.uhs,ei .s.u(s) = x
(58) s. Benedict XVI
We could take it that wishing to talk about the actual index is so
universal and natural that the relation of identity is generally salient
in such cases. There is also, in fact, a certain amount of evidence to
the effect that the language faculty contains a type-shifting procedure
that takes individuals and returns, for each individual, the property
of being identical to it (Partee and Rooth 1983, von Fintel 1993, van
Benthem 1995); in other words, it may well be the case that the function
postulated as the value of R in (57b) is actually a built-in part of the
language faculty and hence always available. See section 3.5 for further
discussion.
2.4.6. Bound pronouns and traces
In order to analyze bound pronouns, we will need to have a theory
about the basics of binding, which I will introduce with an example
that does not use pronouns. Consider (59):
(59) John, Mary likes.
And suppose, as is plausible, that this sentence results from John
moving from object position and leaving a trace. Then the LF of this
example will be approximately as in (60):
demoniacal.tex; 21/08/2007; 19:21; p.16
17
(60) [John 2 [Mary [likes [the R2 i1 ]]]]

I follow Heim 1993 and Heim and Kratzer 1998 in positing a -abstractor
in the syntax. The idea is that every movement inserts a -abstractor
just below the landing-site of the moved item and obligatorily coindexes it with the trace. In the current theory, traces are complexes
consisting of a phonologically null definite article, represented by the,
and indexed variables of type we have seen before. the has the same
meaning as overt the.10
We now need some help in interpreting this structure. The following
rule, based distantly on the corresponding one in Heim and Kratzer
1998, will do the job:
(61) Predicate Abstraction
For all indices i and assignments g,
[[i ]]g = uhs,ei . [[]]g
ident(u)/i
(62) ident(u) =def x.vhs,ei .s.v(s) = u(s)

g ident(u)/i is the variable assignment that is exactly like g except that
it maps i to ident(u).
Note that the relation variable R is bound, not the individual variable. (The reasons for this will be given below.) The individual variable,
in fact, will never be bound in this system. Predicate Abstraction will
be the only rule that allows anything to be bound. When there is no
obvious index to be the semantic value of the deictic component i, we
could assume that it takes on a default value , where is a specially
designated individual whose only role in life is to be manipulated by
external forces and then quietly vanish. Another possibility, however, is
that, when R is bound by a binder , i refers to the individual referred
to (or the group of individuals quantified over) by the DP that has
moved and created . See section 3.7 for evidence in favor of this latter
possibility.
The combination of the and [R i] in the trace will have the following
meaning:
(63) [fhse,sti .s. xf (s0 .x)(s) = 1](v.s.v(s) = u(s))
This reduces down to (64), which in turn is equivalent to (65):
10
This treatment of traces is in fact only a simplified version of what research

has shown to be advisable. If Chomskys (1993) copy theory of movement is correct,
we will ultimately want something like Foxs (2002) theory of trace conversion.
In the terms used above, this would mean that a copy of the moved item is left in
the trace position, its determiner is replaced by the and [R i] is adjoined to the
predicate material.
demoniacal.tex; 21/08/2007; 19:21; p.17
18
Paul Elbourne
(64) s.x x = u(s)

(65) u
So, informally speaking, the combination of -abstractor and coindexed
trace sets up a structure of the form pu . . . uq in the semantics. The
somewhat baroque way of achieving this is motivated by the advisability of having the same types of variables in pronouns and traces, and
by the desire for ultimate compatibility with an empirically adequate
theory of traces, as described in note 10.
Given the above rule and the lexical entry in (66), we achieve the
result in (67), as desired.
(66) [[likes]] = uhs,ei . vhs,ei . s. v(s) likes u(s) in s
(67) [[John 2 [Mary [likes the R2 i1 ]]]] = s. Mary likes John in s
is the empty set, the assignment that does not map anything to
anything.
Let us move on now to binding involving pronouns. For the purposes of this article, I will follow the hypothesis advanced in Heim and
Kratzer 1998, to the effect that all binding is accomplished by means
of -abstractors in the syntax and that -abstractors in the syntax are
created only by movement.11 Arguably, there is nothing to prevent us
coindexing the variable in a pronoun with the variable in a trace left
by movement. This means that for an example like (68) we might have
an LF something like (69):
(68) Every diver attacked the shark next to him.
(69) [every diver 2 [[the R2 i3 ] attacked the shark next to [him R2
i3 ]]]
In (69), the constituent [the R2 i3 ] is the trace left by the raised subject.
The subject could have been raised either by Quantifier Raising (May
1977, 1985) or by movement to the canonical subject position from a
VP- (or vP-)internal subject position (Kitagawa 1986). Be that as it
may, the constituent below the moved subject will end up with the
denotation in (70) by means of the rules of Variable Interpretation and
Predicate Abstraction given in (52) and (61).
(70) u.s. u(s) attacked the shark next to u(s)
Thus the pronoun ends up as a bound individual concept variable, as
desired.
11
I do not, however, mean to be seen as opposing proposals like that in B

uring
2004, where there are separate syntactic binders for pronouns and traces.
demoniacal.tex; 21/08/2007; 19:21; p.18
19
2.4.7. Donkey pronouns

Let us move on to analyze donkey anaphora in the current framework. The classic example (71), will involve a pronoun and variable
assignment as in (72) and (73). The semantic value of it with its
accompanying variables will be (74).
(71) Every man who owns a donkey beats it.
(72) [it [R2 i1 ]]
(73)
1 donkey
2 [[ ]]
(74) s. x x is a donkey in s
In (73), donkey is either an occurrence of the word donkey or the word
donkey itself, whatever it is that the interpretation function works on.
[[ ]] is the normal interpretation function, which, in the case of nouns,
takes individuals (nouns or their occurrences) and maps them to functions of type hse, sti; it is of type he, sesti, then, as required. Since it
is hard-wired into the language faculty, the interpretation function will
always be available for use in resolving anaphora.12
The expression in (74) is also the denotation of it (in a donkey
sentence) in Elbourne 2005b. The rest of the current account of donkey
sentences is identical to the one in this earlier work, then; I will summarize it here briefly for the sake of convenience.13 How does (71) come
to have a covarying reading? The answer is by situation variables being
bound. Situations are viewed as being composed of individuals and of
properties and relations predicated of them; not all of an individuals
properties need enter into a situation of which that individual is a part
(Kratzer 1989). We introduce the notion of a minimal situation of a
certain kind, which, informally, is a situation in which certain specified
12
This use of the interpretation function is reminiscent of the discourse deixis of

Roberts 2002, 2003, where provision is made for something very similar. See section
4.2.2 for details. But note that the current use of the interpretation function is
situated within a larger framework justifying the use of relations at this point in the
theory, whereas in Roberts 2002, 2003, it is more like a separate stipulation.
13
In fact all the argumentation in Elbourne 2005b that does not relate to the
details of how (74) is arrived at still stands, as far as I concerned. In Elbourne
2001a, 2005b, I show that the recovery conditions on the descriptive content in
donkey anaphora are the same as those on NP-deletion and suggest that donkey
anaphora actually involves NP-deletion; but I now think that the similar recovery
conditions might come about not because there has literally been NP-deletion in
cases of donkey anaphora, but because in both NP-deletion and donkey anaphora
the task is basically to recover descriptive material of the semantic type of a Noun
Phrase. Note how the deictic and relational components in the current theory lead
one inexorably to functions of type hse, sti, the type of nouns. So the current theory
is in a way only a minor variant of my previous one.
demoniacal.tex; 21/08/2007; 19:21; p.19
20
Paul Elbourne
conditions are fulfilled, but in which nothing else goes on (Kratzer

1989). And we introduce an ordering on the set of situations, such that
a situation s is part of a situation s0 (s s0 ) if and only if s0 contains
all the individuals and predications of properties and relations that are
in s, and possibly some others too; we also say in such circumstances
that s0 is an extension of s (Kratzer 1989). We then define the word
every as in (75):
(75) [[every]] = fhhs,ei,hs,tii . ghhs,ei,hs,tii . s. for every individual x:
for every minimal situation s0 such that s0 s and
f (s.x)(s0 ) = 1, there is a situation s00 such that
s00 s and s00 is a minimal situation such that s0 s00
and g(s.x)(s00 ) = 1
(In this definition, every on the right hand side of the equation is to be
understood as logical or the subset relation, so there is no circularity.)
Skipping the compositional details, which are given in Elbourne 2005b,
Chapter 2, we posit (76) and (77) as the meanings of man who owns a
donkey and beats the donkey respectively:
(76) uhs,ei .s.u(s) is a man who owns a donkey in s
(77) uhs,ei .s.u(s) beats in s the unique donkey in s
Given the definition of every in (75), this means that (71) has the truth
conditions in (78):
(78) s. for every individual x: for every minimal situation s0 such
that s0 s and x is a man who owns a donkey in s0 , there is a
situation s00 such that s00 s and s00 is a minimal situation such
that s0 s00 and x beats in s00 the unique donkey in s00
These truth conditions are intuitively adequate. In a less formal paraphrase, they state that our example is true if and only if, for every pair
of an individual x and a minimal situation s0 in which x is a man who
owns a donkey, there is an extended situation s00 in which x beats the
donkey in s00 . Since the situations s00 are extensions of the situations
s0 , which already contain donkeys, the donkey in s00 in each case
must be identical to the donkey in the smaller situation s0 , since if it
were different there would be two donkeys in s00 and the uniqueness
presuppositions of the definite article (or the -operator) would not be
fulfilled. So each man beats his own donkey, as desired. See Elbourne
2005b, Chapter 2, for extensive discussion.
This completes the discussion of the basics of the theory of pronouns
and binding that I will adopt in this article. Demonstrative determiners,
to look ahead briefly, will turn out to be extremely similar to pronouns;
demoniacal.tex; 21/08/2007; 19:21; p.20
21
in fact one might call them pronouns that also take Noun Phrases as
arguments.
2.4.8. Some loose ends
But there are a couple of loose ends to tie up. In the remainder of this
section, I will describe the following: an additional mechanism that
might be at work sometimes in what we would normally analyze as referential pronouns; and the reason why the rule of Predicate Abstraction
binds the relation variable and not the individual variable.
An observation underlies the first suggestion just referred to. It involves the traditional notion of anaphora. Let us use the term anaphora
for cases where a pronoun or demonstrative is associated with a linguistic antecedent that plays a semantic role in its interpretation.14 By
this definition, bound variable pronouns and donkey pronouns involve
anaphora on the theory just sketched. In the bound variable case, the
most directly involved antecedent is the -abstractor that binds the pronoun; the DP that moves and creates the -abstractor can also be seen
as the binder or antecedent in an extended sense (Heim and Kratzer
1998). In donkey anaphora, on the current theory, the antecedent would
be the noun donkey that directly provides the interpretation of the donkey pronoun. Traditionally, referential pronouns too have been said to
involve anaphora. In a case like (79), he would be said to be anaphoric
on Bill.
(79) Everyone who has met Billi thinks hei is smart.
It is not certain that such a case involves anaphora, however. It seems
perfectly possible to say that the mention of Bill makes Bill salient,
and that he then just refers directly to Bill, without any further role
for the word Bill. We might say in a case like this that Bill plays a
pragmatic rather than a semantic role in the interpretation of he.
Now for the observation. The observation is that the literature on
the processing of anaphora, using lexical decision tasks, eye-tracking
experiments and other methods, makes an overwhelming case for the
mental representation of the antecedent becoming activated during the
parsing of the anaphoric element (Shillcock 1982, McElree and Beaver
1989, Garnham et al. 1995, Cacciari et al. 1997, Carminati et al. 2002,
van Gompel et al. 2004). For example, processing a pronoun anaphoric
to a DP significantly quickens reaction times on the task of deciding
whether or not an adjective that had appeared in the DP is a word
(McElree and Beaver 1989). The reactivation effect obtains in cases
14
Antecedents need not literally come before the pronouns with which they are
associated. I am not, that is to say, making the traditional distinction between
anaphora and cataphora. A linguistic antecedent is just some item in the syntax.
demoniacal.tex; 21/08/2007; 19:21; p.21
22
Paul Elbourne
where the pronoun could be referential as well as in cases that are

certainly bound variable anaphora.15
For the case of pronouns that we would normally analyze as referential, this suggests the following. Recall the theory of proper names
outlined in (22), repeated here as (80).
(80) [[the Mary]] = s.Mary
If proper names in fact involve a definite description structure, as in
(80), an example like (79), where a name is followed by a coreferential
pronoun, could possibly involve the same mechanism as that which
obtains in donkey anaphora according to the above theory. In (79),
that is, we could have the pronoun in (81) and the variable assignment
in (82). This would give the value in (83) for pronoun plus variables.
(81) the Bill. . . [he [R2 i1 ]]

(82) 1 Bill
2 [[ ]]
(83) s. Bill
Since the mechanism in (83) is already available for donkey anaphora,
it would take an undesirable stipulation to prevent it being available
for pronouns coreferential with proper names, given the independently
fruitful theory that names are definite descriptions. It will naturally
also be available for pronouns anaphoric to overt definite descriptions.
It is also one way of making sense of the finding in the processing
literature that people pay renewed attention to the antecedent even
when processing a pronoun that might be referential.
I will now turn to the reasons why the Predicate Abstraction rule
in (61) binds the relation variable in pronouns and not the individual
variable. This might strike some as a strange way to go about things.
After all, one might think that one could have a neater theory by
having the rule of Predicate Abstraction bind the individual variable
and having the relation in such cases be identity, as in the case of the
referential pronoun illustrated in (57). Why not do this?
Firstly, we should note a technical difficulty. The current system is
set up in such a way that predicates take individual concepts as their
arguments, not individuals.16 This means that Predicate Abstraction
15
Most of the processing literature seems to deal with cases that are assumed to
be referential. Carminati et al. 2002 is the only study known to me that sets out to
test bound variable anaphora.
16
This in turn is necessitated by the need to have a situation structure rich enough
to support a lexical entry like (75) for every, which in turn is necessary for analyzing
donkey anaphora in terms of situations and definite descriptions, which in turn is
arguably necessary because of the reasons reviewed in Elbourne 2005b.
demoniacal.tex; 21/08/2007; 19:21; p.22
23
has to bind individual concept variables, not individual variables. So

altering the current system so as to bind the individual variable but
leave everything else the same would not work. The only alternative
that suggests itself is to alter the system to make the deictic component
pick out an individual concept. But then we would be faced with the
unnatural scenario of the index in a scenario where a speaker refers
to Benedict XVI being not Benedict XVI but [s. Benedict XVI]. We
might even need [s. ]. Such a system would not be impossible, but
any alleged gain in naturalness that would come from directly binding
the individual concept variable associated with the deictic component
would be counterbalanced by the new system of individual concept
indexes.
Secondly, there are also more substantive reasons to avoid a system
in which pronouns contain bindable individual variables (or individual
concept variables) paired with relation variables. Such a system would
presumably be one in which it was possible to have the individual
variable bound and the relation variable free. In fact Cooper (1979)
advocates a system like this for the analysis of donkey anaphora. And,
as discussed in detail in Elbourne 2005b, Chapter 2, such a system
encounters difficulties with examples like (84) and (85).
(84) a. In this town, every farmer who owns a donkey beats it, and
the priest beats it too.
b. In this town, every farmer who owns a donkey beats the
donkey he owns, and the priest beats the donkey he owns too.
(85) a. Every man who has a wife is sitting next to her.
b. ??Every married man is sitting next to her.
A system like the one we are considering would allow it in (84a) to
be analyzed the donkey-owned-by x, with x an individual variable
bound by the subject and donkey-owned by the value, informally
speaking, of the free relation variable (Cooper 1979, Heim and Kratzer
1998). In idiomatic English, the nearest thing would be the donkey he
owns, with he bound, as in (84b). But then we would predict that (84a)
would have a sloppy reading, in which the priest beats his own donkey,
as (84b) does, and this is emphatically not the case (Elbourne 2001a,
2005b). In (85), meanwhile, a system in which we have to look for a
salient relation has difficulty explaining the difference in acceptability
between (85a) and (85b), since the potential antecedents wife and married are about equally far from providing a suitable relation (Kadmon
1987, Heim 1990): they are arguably both two-place relations that are
being forced by syntactic circumstances in our examples into behaving
as one-place predicates.
demoniacal.tex; 21/08/2007; 19:21; p.23
24
Paul Elbourne
So a system that would allow a bindable individual variable combined with a free relation variable encounters empirical difficulties with
some examples of donkey anaphora. The system advocated in this article, however, does better with these examples. The advocated analysis
of donkey anaphora brings it about that we end up with descriptive
material of the type of a Noun Phrase in the donkey pronoun. For (85a)
we arrive at the following paraphrase: for every pair of an individual x
and a minimal situation s0 such that x is a man in s0 and there is an
individual y such that y is a wife of x in s0 (treating a as an existential
quantifier, as is traditional), there is an extended situation s00 in which
x sits next to the unique wife in s00 (i.e. the unique wife in s0 , since the
situations s0 already contain wives and the ones in the situations s00
have to be unique there). For (85b), however, the best we could arrive
at would be the following paraphrase: for every pair of an individual x
and a minimal situation s0 such that x is a married man in s0 , there is an
extended situation s00 in which x sits next to the unique married person
in s00 (i.e. in s0 ). The married men end up sitting next to themselves
then, and also being presupposed to be female.
As for (84), (74), the denotation of a donkey pronoun on the current
theory, repeated here as (86), is also the denotation of the overt definite
description the donkey. The prediction, then, is that the basic donkey
sentence (71) will mean the same as (87) and that (84a) will behave
the same as (88):
(87) Every man who owns a donkey beats the donkey.
(88) In this town, every farmer who owns a donkey beats the donkey,
and the priest beats donkey too.
Both of these predictions are fulfilled. (87) seems to have the same truth
conditions as (71), and (88) lacks a sloppy reading, just like (84a). See
Elbourne 2005b, Chapter 2, for more detailed discussion.
This might seem like a lot of material on pronouns for an article ostensibly about demonstratives. But I hope to show that, once pronouns
are correctly understood, demonstratives are just pronouns provided
with some extra descriptive content in the form of Noun Phrases.17
17
More precisely, demonstratives in English are pronouns with the addition of
NPs and proximal or distal features. But some demonstratives in other languages,
like French ce, do not have proximal or distal features, and are hence plausibly just
pronouns with nouns stuck on them.
demoniacal.tex; 21/08/2007; 19:21; p.24
25
3. A Theory of Demonstrative Determiners

3.1. Lexical Entries for the Demonstrative Determiners
Without further ado, let me set out the lexical entries that I assume
for the English demonstrative determiners:
(89) [[that]]w,h,a,t = x.fhe,sesti .ghse,sti .s.z(f (x)(s0 .z)(s) = 1 &
g(s0 .z)(s) = 1 & distal(x, w, a, t))
(90) [[this]]w,h,a,t = x.fhe,sesti .ghse,sti .s.z(f (x)(s0 .z)(s) = 1 &
g(s0 .z)(s) = 1 & proximal(x, w, a, t))
Since these are slightly complicated formulas, I will explain them in
a number of steps. First note that the first and second arguments of
each of these determiners are to be understood to be individual and
relation variables of the kind that we have already seen in connection
with pronounsNunbergs (1993) deictic and relational components.
The third argument is the following NP. So we will have configurations
like (91).
(91) [DP [[that i] R] NP]
We can assume that that and this syntactically subcategorize for the
relevant variables and NPs in these positions. Note that R and i do not
form a constituent in the case of demonstratives, unlike in the case of
pronouns; this is because it will be useful to have direct access to the
index in order to deal with proximal and distal features, as described
below.
LFs in this system are officially interpreted with respect to a possible
world w (the world in which the speech act in question occurs), an
agent a (the originator of the speech act in question), a time t (the
time at which the speech act occurs) and a variable assignment h.18
At a first pass, then, we see that the determiner this, for example,
takes as arguments an individual (the index), a relation (the relational
component) and a property (the value of the NP), and maps them to
an individual concept of a particular kind, which will be the value of
the whole DP. To be precise, the individual concept is the smallest
function that takes a situation s and maps it to the unique individual
z (if there is such an individual) such that z satisfies the NP-property
18
I follow Kripke 1972 in assuming, at least for the purposes of linguistic analysis,
that entities in the actual world can show up in other possible worlds. On the other
hand, if Lewiss (1968) counterpart theory is correct for semantics, identifying an
agent is ipso facto identifying a world, and we would be able to dispense with the
parameter w.
demoniacal.tex; 21/08/2007; 19:21; p.25
26
Paul Elbourne
in s and also satisfies in s the property obtained by composing the

relational component with the index, and which is also such that a yet
to be explained condition about being proximal is fulfilled; if there is
no such unique individual at the circumstance of evaluation, the whole
DP will receive no denotation there, as explained earlier in section 2.2.
Overall, then, complex demonstratives are individual concepts, just like
definite descriptions and pronouns. In fact demonstratives turn out to
be combinations, as it were, of definite descriptions and pronouns, in
that they take both an NP, like a definite description, and a pair of an
index and a relational component, like a pronoun.
This leaves the treatment of proximal and distal features to be
explained. To a first approximation, adopted from traditional grammar, the proximal and distal features of this and that are the features
that indicate that the referent of this is near the speaker and the
referent of that is, at least comparatively, distant from the speaker.
This is only a first approximation for a couple of reasons. If bare and
complex demonstratives are indeed individual concepts (as I maintain)
or quantifiers (as King (2001) maintains), then there will not always
be anything that can be called the referent of these phrases anyway.
(Strictly speaking, in fact, there will not ever be such a thing.) But in
the current system it will frequently happen (to look ahead slightly)
that the property derived from i and R will be the property of being
identical to a particular person or thing, so that the same person or
thing will be the value of the demonstrative (when it has a value)
whatever evaluation situation is considered. In these circumstances, we
could speak loosely of the referent of the demonstrative, meaning the
entity so contributed. For example, if I point at Flossy the donkey I can
refer to her with the phrase that donkey, where, if the current theory is
on the right lines, the variables i and R jointly contribute the property
of being identical to Flossy and the noun contributes the property of
being a donkey; simplifying slightly, the whole phrase will then have as
its semantic value the function from situations to the unique entity in
them that is identical to Flossy and is a donkey. Whenever this function
is given an evaluation situation that allows it to return an individual
at all, that individual is going to be Flossy. In a case like this, then
we might be tempted to say that Flossy has a privileged status, just
explicated, that we can dignify with the term referent, and that this
or that will be chosen by speakers according as they conceive of Flossy
as being near to them or far from them, respectively. In other words,
we might be tempted to link the choice of proximal and distal features
with the status of the referent, when there is such a thing.
This is on the wrong track, however, as Nunberg (1993) has shown.
In his terms, it is not the interpretation (the referent, in special cases)
demoniacal.tex; 21/08/2007; 19:21; p.26
27
but the index that determines the choice of proximal and distal features. Imagine the following scenario. Immediately in front of us is Field
A, and beyond that, at some considerable distance, is Field B. We know
that one donkey is kept in A and another donkey is kept in B, but it so
happens that neither donkey is in its field today. Perhaps they are at
the vet. Nevertheless, despite the total absence of donkeys, I can point
at Fields A and B in turn and say (92):
(92) This donkey [gesture at Field A] is healthier than that
donkey [gesture at Field B].
The choice of this versus that seems to be determined by the fact that
Field A is close to us and Field B is far away. The donkeys themselves
do not come into it; we do not have to have any idea about their
relative distances from us in order for (92) to be felicitous. Field A is
the index, in Nunbergs terminology, that brings to mind the donkey
that resides in it; and likewise Field B for its regular inhabitant. It is
even possible to engineer scenarios where the positions of indexes and
interpretations impose conflicting requirements on proximal and distal
features. In these cases, as Nunberg (1993: 23) says, the index always
wins. In Nunbergs own example, we are to imagine him pointing in
turn at two sample plates in his china shop. The first one is right in
front of him, but the second is across the room. He says the following:
(93) These [gesture at the nearby plate] are over at the warehouse, but those [gesture at the distant plate] I have in
stock here.
As Nunberg says, if he had really been pointing at the referents of
these and those, it would have made more sense to have reversed them
(Nunberg 1993: 24).
For this reason, the demonstrative determiners in (89) and (90) take
the index as a separate argument, and it appears in the final conditions
distal(x, a, t, w) and proximal(x, a, t, w). These are to be read x
is distal with respect to a at t in w and x is proximal with respect
to a at t in w, where distal and proximal boil down to distant and
near. Some work remains to be done on the explication of the terms
distal and proximal. It is evident that what a speaker counts as being
near to them, in the relevant sense, can be influenced by all kinds of
things, including their emotional attitudes to the objects in question
and how recently the objects were mentioned. I will offer occasional
commentary on this question in the discussion of examples below, but
I will not attempt to come up with a comprehensive theory.
Before we leave the subject of proximal and distal features, however,
I would like to comment on one more characteristic of them, which is
demoniacal.tex; 21/08/2007; 19:21; p.27
28
Paul Elbourne
theoretically quite interesting. An interesting quirk of these features is

that they seem to be justified not only with respect to the point of view
of the speaker but also with respect to the point of view of subjects of
propositional attitude reports. For example, we have the contrasts in
(94).
(94) a. Mary talked to no senator without declaring afterwards that
that senator (?this senator) was the one who would cosponsor her bill.
b. Mary talked to no senator without thinking at the time
that this senator (?that senator) was the one who would
cosponsor her bill.
One might have expected distal features to be justified only by distance
of the relevant entities from the speaker, that is to say the utterer of
the whole token that contains the demonstrative and the propositional
attitude report or reported speech act (and similarly with proximal
features). But (94) surely shows us that these features can refer back
to the point of view of an agent of a speech act reported by the speaker
of the whole token. Marys discussions with senators can be equally
far away from the speaker in (94a) and (94b); instead, that and this
indicate relative proximity (in temporal terms, in these examples) to
Mary, the reported thinker. Of course, that senator and this senator
in these examples can also receive readings in which the distal and
proximal features are evaluated with respect to the speaker of the whole
token; the sentences are ambiguous in this respect.19 These examples,
then, are reminiscent of the following example from Schlenker 2003
(slightly adapted from his (44b)):
(95) John has told me repeatedly that he was sick two days ago.
In (95), two days ago can be understood either with respect to the
context of the utterance of the whole token, which would perhaps
require John to have repeated his message several times within the last
few hours, or with respect to various speech acts of Johns, possibly
spread out over a number of years. The parallel with the interpretation
of the distal and proximal features in (94) is striking. Schlenker uses
(95) and similar examples to argue for the existence of monsters, that is
operators which operate on contexts, making indexicals be interpreted
with respect to contexts introduced within the sentence; and he argues
19
In (94b), the question mark in front of that senator refers only to the reading
whereby the distal feature is justified from Marys point of view. The index in these
examples is evidently the group of senators that Mary spoke to, as suggested in
section 2.4.6. See section 3.7 for further discussion.
demoniacal.tex; 21/08/2007; 19:21; p.28
29
that propositional attitude verbs quite generally are monsters. Note

that it is much harder to use this in a similar example in which the
demonstrative is not in the scope of a propositional attitude verb:
(96) a. ??Mary talked to no senator without this senator thinking
she was crazy.
b. Mary talked to no senator without that senator thinking
she was crazy.
The that in (96b) is evidently justified from the perspective of the
speaker of the whole token. It is plausible, then, that (94) is another
example of Schlenkers monstrousness.
Having explained the elements that appear in the lexical entries of
the demonstrative determiners, I will now review the data thrown up
when we ask how demonstratives are actually used and show how these
lexical entries deal with them.
3.2. Canonically Referring Complex Demonstratives
Let us start with sentences in which the determiners take overt NP
complements, and the DPs in question can be taken to refer (loosely
speaking) to a perceptible referent in the environment. I will refer to
DPs like this as canonically referring complex demonstratives. Some
simple examples are in (97).
(97) a. That cat [gesture at Felix] laughs.
b. This cat [gesture at Felix] laughs.
On the current theory, one simplified LF available for (97a) is that in
(98):
(98) [[[[that i1 ] R2 ] cat] laughs]
In this case, the index will be Felix himself, and the relational component will be identity, as illustrated in (99).
(99) a. [[i]] = Felix
b. [[R]] = x.uhs,ei .s.u(s) = x
This is exactly as in (57), the case of the referential pronoun.
Given this value for i and R, and the rules and lexical entries we
have previously introduced (see (100) for a reminder of some relevant
lexical entries), it can be seen that the denotation of the complex
demonstrative in (98) is predicted to be (101), when it is interpreted
with respect to a world w, an agent a and a time t; and the denotation
of (98) as a whole is (102).
demoniacal.tex; 21/08/2007; 19:21; p.29
30
Paul Elbourne
(100) a. [[cat]] = uhs,ei . s. u(s) is a cat in s

b. [[laughs]] = uhs,ei . s. u(s) laughs in s
(101) s.z(z = Felix & z is a cat in s & distal(Felix, a, t, w))
(102) s.z(z = Felix & z is a cat in s & distal(Felix,a, t, w)) laughs in
s
Given the iota notation, this latter function divides up the set of possible situations into three. Given a topic situation s , the utterance will
be true if there is exactly one distal cat identical to Felix in s and
this cat laughs in s ; it will be false if there is exactly one distal cat
identical to Felix in s and this cat does not laugh in s ; and it will
receive no truth value if there is not exactly one distal cat identical to
Felix in s . Recall the discussion in section 2.2.
In (102), the predicates are all relativized to the same situation
variable s, the one abstracted over at the start of the formula. This
meant that Felix had to be a cat in the evaluation situation s . No
problems were caused by this, since we took s to be a situation in
which Felix was, in fact, a cat. But it is conceivable that there should
be examples in which we would not want the descriptive content in the
NP to have to characterize the referent in the evaluation situation. One
good candidate, suggested by an anonymous reviewer, is the following.
Suppose I point at a man crossing the street, Glenn, and say:
(103) That man crossing the street is six feet tall.
Now suppose that we want to evaluate the proposition I expressed with
respect to a situation s that contains Glenn doing something else
having dinner in a nice restaurant, for example. It seems quite possible
to do so, and the result is pretty clearly that what I said is true with
respect to s if and only if Glenn is six feet tall in s . Glenn does not
have to be crossing the street in the evaluation situation in order to
be picked out by the complex demonstrative in that situation.20 This
means that it must not be the case that the semantic value of (103)
has to be (104).
(104) s. x(x = Glenn & x is a man crossing the street in s &
distal(Glenn, a, t, w)) is crossing the street in s
For in s there is no unique appropriately distal individual identical to
Glenn who is a man crossing the street. So it seems that the complex
demonstrative should not be able to pick out anything when we evaluate
20
I take this claim to be supported by the fact that I could say (103), wait a
while, ask someone Is what I said true now that Glenn is in a restaurant? and
receive a positive response.
demoniacal.tex; 21/08/2007; 19:21; p.30
31
it at s , and that the proposition expressed should be neither true nor

false at s . But this is not the case, as we have seen.
This sounds like a job for the operator s0 that we have already met
in section 2.3 in connection with de re and de dicto readings of definite
descriptions. Recall the semantics of the operator:
(105) [[s0 ]]h = fhse,sti .uhs,ei .s.f (u)(h(0)) = 1
One LF available for (103) will be the following, in which s0 is adjoined
to the NP man crossing the street:
(106) [[[that i1 R2 ] [[man crossing the street] s0 ]] [is six feet tall]]
Suppose that I say (103) in a spatially and temporally delimited situation s that contains me speaking and Glenn crossing the road and not
much else. Then I can mean something like man crossing the street in
s by the NP man crossing the street.21 In current theoretical terms,
the index 0 is mapped to s in the variable assignment h, and the
operator s0 modifies the semantic value of man crossing the street so
as to produce the desired result. Let us allow ourselves the results in
(107):
(107) a. [[man crossing the street]]w,h,a,t = uhs,ei .s.u(s) is a man
crossing the street in s
w,h,a,t
b. [[is six feet tall]]
= uhs,ei .s.u(s) is six feet tall in s
The variable assignment h will be as follows:
(108)
0 s
1 Glenn
2 [x.uhs,ei .s.u(s) = x]
Given this assignment, the value of the whole NP with s0 adjoined is

as follows:
(109) [[[man crossing the street] s0 ]]w,h,a,t = uhs,ei .s.u(s ) is a man
crossing the street in s
This means that the semantic value of the whole sentence on this
occasion is the following:
(110) s.x(x = Glenn & x is a man crossing the street in s &
distal(Glenn,a, t, w)) is six feet tall in s
21
There will of course be a certain amount of vagueness about what exactly I take
s to encompass. I do not think this is problematic. See Buchanan and Ostertag
2005 for one attractive way of dealing with such cases.
demoniacal.tex; 21/08/2007; 19:21; p.31
32
Paul Elbourne
This is the result we needed. If we evaluate this proposition at a situation s in which Glenn is having dinner in a nice restaurant we will
not have to find him crossing the street in s in order to get a verdict
of truth or falsity, since the property of being a man crossing the street
is now relativized to s . The proposition will be true or false at s
according as Glenn is six feet tall or not in s .
The current analysis of canonically referring complex demonstratives is also relevant to the question of whether or not a proposition is
expressed if the object demonstrated does not satisfy the NP descriptive content. For example, is a proposition expressed by the utterance
described in (111)?
(111) That knife [gesture at the speakers fork] is dirty.
Braun (1994), Borg (2000) and Glanzberg and Siegel (2006) have argued that no proposition is expressed in cases like this, while Larson
and Segal (1995: 210213) have argued that a proposition is in fact
expressed. The basic issue, of course, is that in these cases, although
the object demonstrated is misdescribed, we do nevertheless pick up
propositional information quite easily, to the effect that the object
demonstrated has the property contributed by the Verb Phrase. Should
we say that therefore the sentence has a semantic value that is a proposition, in the normal way? Or should we rather say that the sentence
does not have a proposition as a semantic value and we manage to
extract propositional information from the utterance in a different way?
It is a consequence of the view I advocate that a proposition is indeed
expressed in cases like this. But the exact details of what takes place
will depend on whether or not the speaker actually thinks the fork in
the scenario in (111) is a knife. If the speaker does think the fork is
a knife (because the tines are covered by a napkin, say, or because of
some defect of vision), then things are comparatively straightforward.
We might have an LF like (112). The variables i and R would contribute
the property of being identical to a, where a is a name of the fork in
question.
(112) [[[that i1 R2 ] knife] [is dirty]]
This LF, in these circumstances, will produce a proposition that can
be characterized as follows:
(113) The function that:
maps situations s to 1 if there is exactly one appropriately
distal knife identical to a in s and this entity is dirty in s;
maps situations s to 0 if there is exactly one appropriately
distal knife identical to a in s and this entity is not dirty in
s;
demoniacal.tex; 21/08/2007; 19:21; p.32
33
does not map situations s to anything if there is not exactly

one appropriately distal knife identical to a in s.
Since a is in fact a fork, we can imagine that on the occasion we are
imagining the proposition would not map the topic situation to any
truth value.
It seems to me that this prediction is entirely compatible with the
facts. When we hear an utterance like (111), in circumstances like those
described, we have a clear intuition that the utterance is defective in
some serious way. Conversation can continue after such an utterance,
if the interlocutor corrects the first speaker or just plays along; but,
as Glanzberg and Siegel (2006) point out, this does not mean that the
original utterance is acceptable after all. The sense of serious deficiency
we feel would be well explained by the current theory, which says,
in effect, that the speaker intended to make a truth-evaluable claim
but failed to do so. This leaves the fact that people who hear such
utterances do extract propositional information from them. But we can
explain this easily by appeal to the manifest intentions of the speakers
in such cases, who do, after all, wish to say something about the object
they demonstrate. The charitable hearer presumably just overlooks the
incorrect NP descriptive content and treats the utterance as if it just
contained a demonstrative with no associated NP.
Let us move on to consider cases in which the speaker does not
actually think the fork in the scenario in (111) is a knife. In this kind of
case, the word knife has been used instead of fork due to a speech error.
It will clarify our thinking on this possibility if we are explicit about
some of the stages of speech production. Abstracting away from many
details, virtually all models of speech production assume that there is
a stage of conceptualization, in which a thought is formulated in some
non-linguistic medium (a conceptual system or language of thought),
followed by a stage of formulation of the thought in language, followed
by the actual utterance of the sentence (Fromkin 1971, 1973; Garrett
1975, 1984; Dell 1986; Levelt 1989). The utterance of the sentence is
accompanied by self-monitoring, a stage in which we check that our
utterance matches our previous intentions (Levelt 1989). Some speech
errors seem to involve a failure to select words in the linguistic formulation stage that accurately express the thought to be verbalized. For
example, some errors involve uttering a semantically related word in
place of the word that would accurately express the thought in question
(Fromkin 1973, 1988; Garrett 1988). Example (111), if it were taken
from real life, would be a typical example of this. We can imagine that
the speaker is trying to talk about forks, but mistakenly chooses the
semantically related word knife in order to do so. The choice of the
demoniacal.tex; 21/08/2007; 19:21; p.33
34
Paul Elbourne
wrong word results from something going wrong in an unconscious

search process that normally takes just a fraction of a second; the
speech production literature just cited proposes that words with similar
meanings are stored in some sense near to each other in the mental
lexicon.
Given all this, should we say that a proposition is expressed by an
utterance of (111) in which the wrong word has been chosen due to a
speech error? And, if so, what proposition is expressed?
In my view, this kind of case points up a certain kind of crudity in the
notion of proposition expressed (or what is said ). We have no difficulty
describing what is going on in this scenario. I have just done so. But we
have difficulty in saying what proposition, if any, was expressed. The
reason, presumably, is that the notion of the proposition expressed was
formulated without recognizing that there might be a discrepancy between the proposition formulated in the conceptual system or language
of thought that someone tries to verbalize and the proposition yielded
(via the same persons language faculty, with suitable supplementation
by speaker intentions) by the LF that is used to verbalize it. But speech
errors show that these two things can come apart.
I propose, therefore, that proposition expressed and what is said
should no longer be terms in our theory, at least when we are considering things at this fine level. We should replace these terms with
a distinction between the proposition formulated in the language of
thought that we try to verbalize and the proposition yielded (via the
same language faculty) by the LF constructed as a result of this effort, with accompanying speaker intentions where necessary, but using
the regular lexical entries for the words used. When these two things
happen to coincide, we can talk loosely of the proposition expressed
or what is said. So the question of whether a proposition is expressed
in the speech error version of examples like (111) cannot receive a
straightforward answer, since it is based on a false presupposition.
Instead, we should say that there is a proposition yielded by the LF uttered, given supplementation by speaker intentions, and that in the case
under consideration it is the proposition in (113); but this proposition,
we should go on to say, does not fulfill the communicative intentions
of the speaker, which were presumably characterized by a different
proposition formulated in the language of thought.
3.3. Canonically Referring Bare Demonstratives

Let us move on to consider cases where a demonstrative is used bare,
without any overt NP complement, to refer (roughly speaking) to some
demoniacal.tex; 21/08/2007; 19:21; p.34
35
perceptually salient entity. I will refer to cases like this as canonically

referring bare demonstratives. They are exemplified in (114).
(114) a. That [gesture at Flossy] is a donkey.
b. This [gesture at Flossy] is a donkey.
The issue raised is naturally that of unifying bare demonstratives with
complex ones.
This unification can be achieved quite naturally if we suppose that
a truth-conditionally trivial property is contributed to the semantics as
the second argument of that in cases of bare demonstratives. A version
of this idea was first suggested, I believe, by Taylor (1980: 197): within
a formal language that translates That F is G as pthat x(F x; Gx)q,
Taylor translates That is G as pthat x(x = x; Gx)q. Since everything
is identical with itself, this formulation does not add anything to the
overt material that is not truth-conditionally trivial. Similarly, King
(2001: 141143) suggests that the NP node might not host lexical
material in these cases, and that a bare NP node might contribute the
property being a thing to the semantics.22 And in Elbourne 2005b I
argue that there is a phonologically null noun ONE with a meaning
is an individual, is of type e which can be taken as an argument by
demonstrative determiners. For present purposes, I will simply assume
that something along these lines is correct, and will not attempt to
distinguish between the views just mentioned.
3.4. Descriptive Indexicals and Deferred Ostension
Nunberg (1993: 28) has already provided examples of demonstratives
being used as descriptive indexicals. Here is one, with slight elaborations in the background scenario. We are in the offices of a very
hierarchical company that prescribes different kinds of office furniture
for employees of different ranks, with low-quality accoutrements being
prescribed for low-ranked workers. Pointing at the desk of a particularly
lowly employee, I say:
(115) According to the regulations, that must always be made of metal.
Here, as Nunberg (1993: 28) says, it seems that that is interpreted as an
attributively used definite description, something like the desk used
by an employee of grade 10 or below.
22
One should note, however, that Kings proposal is not compatible with all
conceptions of phrase structure. In particular, bare phrasal nodes are impossible according to the conception of phrase structure employed in contemporary Minimalist
syntax (Chomsky 1995: 241249).
demoniacal.tex; 21/08/2007; 19:21; p.35
36
Paul Elbourne
Since we have enough to do without being distracted by pile-ups of

intensional operators (must always), I will analyze the slightly simpler
variant in (116):
(116) According to the regulations, that is always made of metal.
Assuming the phonologically null noun ONE from Elbourne 2005b to
handle bare demonstratives, that here will look like this:
(117) [[that i1 R2 ] ONE]
The index will be the desk I point to, and the relational component will
be the function that will map an object to the property of having that
objects function and status in the companys hierarchy of implements.
This is shown in (118), where d is a name of the desk in question.
(118) a. [[i]] = d
b. [[R]] = x.uhs,ei .u(s) is characterized in s by the actual
function and hierarchical status of x
The null noun ONE will contribute the property of being an individual,
henceforth ignored. So the semantic value of the demonstrative in this
case will be (119). Since d is actually a desk for employees of grade 10
or below, this is equivalent to (120).
(119) s.z(z is characterized in s by the actual function and hierarchical status of d & distal(d, a, w, t))
(120) s.z(z is a desk for employees of grade 10 or lower in s &
distal(d, a, w, t))
The quantificational adverb always will, as before, quantify over situations of a certain kind, the exact kind depending on the intentions of the
speaker. Here there is a clue in the phrase according to the regulations;
we can take always to be quantifying over minimal situations that
contain exactly one desk used by an employee of grade 10 or under, and
in which all relevant regulations regarding the choice of the desk have
been observed. In saying (116), then, I am claiming that all situations of
that kind are such that the unique desk used by an employee of grade
10 or below in them is made of metal. This seems to be intuitively
adequate.
This kind of account will suffice to analyze a group of related examples, Kings (2001) cases of demonstratives being used with no
demonstration, no speaker reference. King (2001: 2) introduces the
relevant class of uses as follows:
demoniacal.tex; 21/08/2007; 19:21; p.36
37
there are uses of that phrases in which they are not accompanied by any demonstration, need not be used to talk about something present in the physical context of utterance, and in which
the speaker has no particular individual in mind as the thing she
intends to talk about by means of the that phrase.
Some examples, based closely on Kings, are the following.
(121) a. (We heard a while ago that exactly one student scored 100
on the exam, but we do not know who it is.) That student
who scored 100 on the exam is a genius.
b. (We heard a few seconds ago that exactly one student scored
100 on the exam, but we do not know who it is.) This student
who scored 100 on the exam is a genius.
In my judgment, and according to the judgments of those I have consulted, both this and that are possible in these cases, and are intuitively
justified with reference to how distant the event of our hearing about
the brilliant student was.
This suggests that in each case we can use the utterance announcing
the existence of the student who scored 100 as the index. To take an
example, the complex demonstrative in (121a) would have the following
form at LF:
(122) [[that i1 R2 ] [student who scored 100 on the exam]]
Let b be a name of the utterance in which we heard about this student.
The utterance b, then, will be the index. R will take on as its value the
function mapping utterances to the property of having ones existence
announced in them. So the two variables jointly contribute the property
of having ones existence announced in b. Assume that (121a) is spoken
in w at t. The semantic value of the complex demonstrative is then the
function from situations s to the unique x such that xs existence is
announced in b in s and x is a student who scored 100 on the exam
in s and b is distal with respect to the speaker in w at t. The whole
utterance will be true of an evaluation situation s , such as a relevant
chunk of the real world, if and only if the semantic value of the complex
demonstrative maps s to an individual, and that individual is a genius
in s . This seems intuitively adequate.
The presence of the deictic and relational components also accounts
straightforwardly for cases of deferred ostension (Quine 1969, Nunberg 1979, 1993). These are traditionally defined as cases in which
the speaker demonstrates one thing in order to refer to another; the
speaker does have an object in mind to which the index and relational
component are supposed to lead the hearer. We have already seen an
example of this in (92) in section 3.1, repeated here as (123):
demoniacal.tex; 21/08/2007; 19:21; p.37
38
Paul Elbourne
(123) This donkey [gesture at Field A] is healthier than that

donkey [gesture at Field B].
The indexes are the two fieldsrecall that the donkeys themselves need
not even be presentand the relational component in each case is the
relation is kept in. The first occurrence of this donkey, then, will mean
the unique individual x such that x is kept in Field A and x is a donkey
and Field A is suitably proximal. If we imagine that the speaker and
hearer are personally acquainted with the donkeys in question, and
that the speaker wishes the hearer to identify the donkeys in question
by means of the occurrences of this donkey and that donkey, we would
have a case of deferred ostension. But note that the actual semantics
given by the current system is really no different in this case from
the cases of descriptive indexicals and cases of no demonstration, no
speaker reference. We could imagine a use of (123) in which the speaker
had no personal acquaintance with the two donkeys in question, and
simply deduced their state of health from the state of the fields in which
they were kept. The same definite description would be the immediate
semantic value of the complex demonstrative in each scenario. The
difference, as in the classic essay by Kripke (1979), would be that in
one scenario the speaker has a particular thing in mind that the definite
description is supposed to pick out, whereas in the other scenario this
is not the case.
3.5. Canonically Referring DemonstrativesA Reprise
Kaplan (1989a) in his classic essay used contrasts like the following to
argue that demonstratives were directly referential and that definite
descriptions were not:
(124) (Charles is from Charleston, West Virginia. Paul is from St.
Paul, Minnesota. is a pointing by the speaker in the direction
of Paul, who is seated on a chair in front of the speaker.)
Look over here. [The gesture is initiated now and held
throughout the next sentence.] If Charles and Paul had
changed chairs, then
a. the man being pointed at would be from Charleston.
b. this man being pointed at would be from Charleston.
It is clear, as Kaplan says, that in the circumstances indicated (124a) is
true and (124b) is false. And there is great plausibility in his contention
that this indicates that demonstratives are directly referential while
definite descriptions are not: Paul himself would be contributed by the
complex demonstrative to the proposition expressed by the utterance
demoniacal.tex; 21/08/2007; 19:21; p.38
39
of (124b), and no amount of chair-switching will make it the case that

he is from Charleston.
The doctrine that demonstratives are directly referential cannot,
of course, account for the examples of demonstratives as descriptive
indexicals (section 3.4). Nor can it account for examples where demonstratives have covarying readings (sections 3.63.8). So we should not
feel tempted to revert to it. But on the other hand it does seem that the
devices we have introduced to deal with the cases of descriptive indexicals and deferred ostension might have the unwelcome effect of making
the current theory predict that (124b) will have a true reading in the
given scenario. In the complex demonstrative this man being pointed
at in (124b), the index must presumably be Paul, since he is, after all,
being pointed at. But what is to prevent the relational component being
something like has swapped chairs with, which would presumably be
a salient relation given the antecedent of the conditional? Then the
complex demonstrative would presumably mean something like the
unique x such that x has swapped chairs with Paul and x is a man being
pointed at and Paul is near the speaker. In other words, it looks like
the complex demonstrative should be able to be a definite description
picking out Charles in the counterfactual circumstances introduced in
the antecedent, just like the regular definite description in (124a).
We must ask, then, why the complex demonstrative in (124b) must
be canonically referring and cannot have a deferred use, where by deferred use I mean a use other than canonically referring without being
covarying, such as a use as a descriptive indexical or for purposes of
deferred ostension. As a first step towards answering this question, let
me propose the following descriptive generalization for cases involving
complex demonstratives:
(125) When the index satisfies the NP descriptive content, the index
(or rather the constant individual concept mapping situations to
it) must be the interpretation. Stated otherwise, when the index
satisfies the NP descriptive content, the relational component
must be identity.
Deferred uses of complex demonstratives, like (123), typically involve
demonstrating something that does not satisfy the NP descriptive content. If we try to come up with a minimal variant of (123) where
the indexes do satisfy the NP descriptive content, we find that the
demonstratives in it no longer have deferred uses. Suppose that we both
know that Flossy always shares a field with Esmerelda, and Rupert
always shares a field with Aloysius, and that we can distinguish these
four donkeys infallibly by sight. Suppose further that Esmerelda and
demoniacal.tex; 21/08/2007; 19:21; p.39
40
Paul Elbourne
Aloysius are absent, being attended to by a vet, but we can see Flossy
and Rupert in a farmyard.23 I then say (126):
(126) This donkey [gesture at Flossy] is healthier than that donkey [gesture at Rupert].
As long as it is clear that I am gesturing at Flossy and Rupert, I absolutely cannot mean by this that Esmerelda is healthier than Aloysius.
This is so even if I have just explicitly drawn attention to the necessary
relational component by saying something like You know that Flossy
always shares a field with Esmerelda, and Rupert always shares a field
with Aloysius. I must mean that Flossy is healthier than Rupert, even
though the health of Esmerelda and Aloysius is arguably more salient,
given that they are receiving veterinary care.
I know of only one type of apparent counterexample to the generalization in (125), and the examples in question are in fact exceptions
that prove the rule. Consider (127), a variant of Kaplans (124b):
(127) If Charles and Paul were disguising themselves as each other,
this man [gesture at Paul] would be from Charleston.
Unlike (124b), this has both a true and a false reading. We can account
for the true reading by supposing that the mention of disguise has
made salient what we might think of as the outer shell or appearance of
people. We obtain a true reading, then, by interpreting the index not as
being Paul but as being the outer shell or appearance of Paul, which we
can call a. The complex demonstrative, on the true reading, then means
something like the unique individual x such that x has appearance a
and x is a man and a is near the speaker. Provided that we imagine
the disguises to be good ones, this would indeed pick out Charles in
the counterfactual circumstances introduced by the antecedent. Since
the outer shell or appearance of Paul is not in itself a man, there is no
transgression of the principle in (125).
We can account for Kaplans datum in (124b), then, by pointing out
that the index satisfies the NP descriptive content and hence, by (125),
the relational component has to be identity. But we would ideally like
to go deeper into the matter than this. Why should (125) hold? As
it happens, Nunberg (1979) has already provided a plausible answer.
Using demonstratum for his later term index, and referring function or
RF for the relation between demonstrata and interpretations carried
out partly by the relational component in the current system, he writes
as follows (1979: 160):
23
I stipulate this latter detail in order to prevent the gestures at Flossy and
Rupert in the example being reinterpreted as gestures towards the fields where they
live or the general space that they inhabit.
demoniacal.tex; 21/08/2007; 19:21; p.40
41
Allow me to borrow the psychologists notion of cue-validity to refer to the probability with which a given referent b can be identified
as being the value of a certain function f at a demonstratum a,
or more generally, to the relative usefulness of a given description
for purposes of identification. All things being equal, we will assume
that given an array of possible RFs which take a demonstratum into
a range of reference, a rational speaker will intend that his hearer
should select that function that has the highest cue-validity for its
referent. In other words, when a demonstratum stands uniquely in
several different relations to several members of a range of reference, and there is no reason for assuming that any one of these
members is a more likely candidate for reference than another, we
will assume that the intended referent is that member which is most
easily identified in terms of its relation to the demonstratum. [. . . ]
[T]he reader may have noticed that we did not define the range of
reference above in such a way as to exclude the possibility that the
demonstratum was itself a member of the range. Thus, if I point
at a hat and say, That is a derby, we would normally understand
the range of possible referents as including the hat itself. And under
these circumstances, the hat itself would be the only thing that I
could actually be referring to, if I am being rational. The general
point is this: when the demonstratum could be the referent, it must
be the referent. [. . . ] [T]his principle follows from the more general
observation about cue-validity that we made above. The identity
is a function too, after all, and it is the only function whose value
is trivially computable for all arguments in all domains. So where
the identity could be the referring function, its cue-validity must be
higher than that of any other possible RF, and it must be chosen.
The identity function that Nunberg had in mind was of course [x.x]. In
the system in the current article, the same effect is achieved by having
the relational component be (128), as in (57) and (99):
(128) x.uhs,ei .s.u(s) = x
This combines with the index and an iota-operator, whether in a pronoun or a demonstrative, to give the index back as the interpretation
or, strictly speaking, the constant function mapping situations to the
index. Exactly the same rationale envisaged by Nunberg in 1979, then,
can apply to the system in the present article. Basically, the kind of
constant function just mentioned is always going to be trivially derivable from any index by means of a trivial and universally available
relational component. (This relational component may even be an inbuilt type-shifting mechanism in the language faculty, as I pointed out
below (57).) This kind of constant function is, in a sense, going to be the
demoniacal.tex; 21/08/2007; 19:21; p.41
42
Paul Elbourne
easiest thing to arrive at. Since a cooperative speaker will intend that
the interpretation be the one that is most easily identified in terms
of its relation to the demonstratum, this kind of interpretation will
always win out, unless there are obvious reasons to make it impossible,
such as the NP-descriptive content being incompatible with the index.
Note that Nunbergs (1979) reasoning here has very much the flavor
of Relevance Theory (Sperber and Wilson 1986) and could easily be
recast explicitly in terms of that theory.
3.6. Quantifying In
One of the central planks in Kings (2001) argument against a direct
reference account of complex demonstratives is the existence of examples of quantifying in. These examples involve a pronoun in the matrix
of a complex demonstrative being bound by a quantifier phrase outside,
generally producing covarying readings.24 King (2001: 10, 74, 173) gives
the examples in (129); (129a) occurred first in Neale 1993, where it is
attributed to Jamie Tappenden.
(129) a. Every man eagerly looks forward to that day when he retires.
b. Every father dreads that moment when his eldest child leaves
home.
c. Most avid snow skiers remember that first black diamond
run they attempted to ski.
d. Every professor cherishes that publication of his.
In the case of (129d), which might seem awkward in isolation, King
(2001: 74) suggests that we imagine it to be spoken while the speaker
nods in the direction of a screen showing a scene from a film in which
a professor is fondly perusing his finest piece of published work; it is
supposed to mean something like Every professor cherishes his finest
publication.
I think it is evident that the current theory will face no particular
problems with this kind of sentence. In the case of (129d), for example,
we can imagine that the index is the scene towards which the speaker
gestures, which we can call s, and the relational component is something
like has the qualities depicted in. Translating into extensional terms
24
I believe it is possible to have quantifying in without a covarying reading,
however. Suppose that all the boys in a Scout troupe were told to clean the floor
of a cabin were they were staying, but all of them forgot. We could say Every boy
forgot to clean that floor he was supposed to clean. We have quantifying in, but the
floor that was supposed to be cleaned is the same for every boy, and hence there
is no covariation. The deictic and relational components would work as they do in
canonically referring examples in this case.
demoniacal.tex; 21/08/2007; 19:21; p.42
43
for the sake of simplicity, we obtain the truth conditions in (130) for
(129d):
(130) Every professor x cherishes the unique individual y such that y
is a publication of x and y has the qualities depicted in s and s
is suitably distal.
Since the piece of work in s is depicted as the finest publication of the
professor in question, this is equivalent to saying that every professor
cherishes his finest publication, as desired.
It is notable that the other three examples of quantifying in in King
(2001), (129a)(129c), all involve intensional transitive verbs and all
seem to introduce a presupposition to the effect that the kind of event
or object they talk about (retiring, eldest children leaving home, black
diamond runs skied for the first time) is somehow familiar. I speculate
that the reason for this is as follows. Since we are presented with these
examples out of the blue, as it were, with no scenario provided in which
we are to imagine them being said, the deictic component cannot zero
in on any obvious visual clue, as it does in the case of (129d); and since
we are dealing with covariation, the index cannot be the interpretation,
as it is in the case of referential readings. This leaves the deictic and
relational components with limited room to maneuver, and we hence
settle on something like the concept or idea of retiring, or having ones
eldest child leave home, for the index; the relational component is then
something like exemplified by. The rough truth conditions for (129b),
then, are something like those in (131):
(131) Every father x dreads the unique individual y such that y is a
day when xs eldest child leaves home and y exemplifies the idea
of eldest children leaving home and the idea of eldest children
leaving home is suitably distal.
This explains the feeling we get when we hear examples like (129b)
that the speaker is making reference to a kind of event with which we
are expected to be familiar.
3.7. Bound Demonstratives
As already noted in (94), repeated here as (132), complex demonstratives seem to be able to be bound themselves. Simple ones do too,
although they are perhaps slightly more awkward in bound readings
than complex ones. Witness the following examples.
that senator was the one who would cosponsor her bill.
demoniacal.tex; 21/08/2007; 19:21; p.43
44
Paul Elbourne
b. Mary talked to no senator without thinking at the time that

this senator was the one who would cosponsor her bill.
that was the one who would cosponsor her bill.
b. Mary talked to no senator without thinking at the time that
this was the one who would cosponsor her bill.
On the current system, of course, these demonstratives too will be be
bound into. In particular, their relation variable will be bound, as with
the case of bound variable pronouns in section 2.4.6. So, for example,
(134) will have the simplified LF in (135), in which no senator has
moved by Quantifier Raising and left a trace.25
(135) [no senator] [2 [Mary talked to the R2 i1 before that i1 R2
senator was lobbied]]
On the current theory, parallel to the denotation of every in (75), no
will have meaning in (136):
(136) [[no]] = fhhs,ei,hs,tii . ghhs,ei,hs,tii . s. for no individual x is it the
case that there exists a minimal situation s0 such that
s0 s and f (s.x)(s0 ) = 1, such that there is a situation s00 such that s00 s and s00 is a minimal situation
such that s0 s00 and g(s.x)(s00 ) = 1
If the intuitions reported in (94) reflect what is really going on in cases
of bound demonstratives, then proximal and distal features make reference in these cases to the group of individuals quantified over. Glancing
back to the possibilities aired in section 2.4.6, below (61), this supplies
evidence for the index in cases of binding being this group, as opposed
to a special default individual. As already noted in footnote 19, this
means that in this case the index is the group of senators talked to by
Mary, which we can call g. Using the rule of Predicate Abstraction in
(61), we can work out that the meaning of the second argument of no
in (135) is (137). With the help of the lexical entries in (136) and (138),
we arrive at (139) as the denotation of (135).
(137) uhs,ei .s.Mary talked in s to u(s) before z(z = u(s) and z is a
senator in s and g is suitably distal) was lobbied in s
25
Examples like (134) will presumably have implications for our views on Principle
C of the Binding Theory, since the LF configuration in (135) should be a violation of
this principle. There is a prima facie conflict between (134), which seems to indicate
that Principle C does not apply at LF, and the arguments in Fox 2000 that seem to
indicate that Principle C must apply at LF.
demoniacal.tex; 21/08/2007; 19:21; p.44
45
(138) [[senator]] = uhs,ei .s.u(s) is a senator in s

(139) s. for no individual x is it the case that there exists a minimal
situation s0 such that s0 s and x is a senator in s0 , such that
there is a situation s00 such that s00 s and s00 is a minimal
situation such that s0 s00 and Mary talked in s00 to x before
z(z = x and z is a senator in s00 and g is suitably distal) was
lobbied in s00
This is intuitively satisfactory. Being suitably distal in this case means
being distant from the speaker, since no monsters seem to be involved;
but in the examples in (132) and (133), as noted in section 3.1, the
proximal and distal features can be justified with respect to the point
of view of the agent of the reported propositional attitude.
3.8. Donkey Anaphoric Demonstratives
Both bare and complex demonstratives can be donkey anaphoric, as
we see in the following examples:
(140) Every man who owns a donkey beats that and nothing else.
(141) Every man who owns a donkey beats that donkey.
We have already seen in some detail how the current system analyzes
sentences very like these, back in section 2.4.7. Recall that (71), repeated here as (142), involved the pronoun (143) and the variable
assignment (144), giving the meaning (145) for the pronoun.
(142) Every man who owns a donkey beats it.
(143) [it [R2 i1 ]]
(144)
1 donkey
2 [[ ]]
Similarly, that in (140) will involve a deictic and a relational component that will be interpreted exactly as in (144). Instead of a regular
NP, it will be associated with a null component contributing a truthconditionally trivial property, as described in section 3.3. The only
semantic difference between (140) and (142), then, is that in (140) the
index, the previous occurrence of the word donkey, will be presupposed
to be distal. It is unclear whether this means that the occurrence of
the word itself is being presupposed to be distant, or whether (perhaps more naturally) the occurrence can be distal because the relevant
demoniacal.tex; 21/08/2007; 19:21; p.45
46
Paul Elbourne
entities that fit its descriptive content are being presupposed to be

distant.26
I propose that (141) works in exactly the same way as (140), with
the exception, obviously, that there is an overt NP. That is, the deictic
and relational component will work exactly as in (144) and contribute
the property donkey, redundantly in this case.
4. Two Previous Theories

4.1. A Comparison with King 2001
4.1.1. Introduction
So much for the basic data and their treatment by the theory advocated
in this article. I now move on to the examination of the only two other
theories known to me that also set out to account for a broad range
of data. I will begin with King 2001. Before examining the details of
Kings (2001) theory of that, I will make a few remarks on the more
general argumentation in Chapter 1 of his monograph to the effect that
complex demonstratives must be quantificational. Readers are referred
to Johnson and Lepore 2002 for further discussion of Kings approach.
4.1.2. Arguments that complex demonstratives are quantificational
Chapter 1 of King 2001 is devoted to making it plausible that complex demonstratives are generalized quantifiers. In the first part of the
chapter, King points out uses of complex demonstratives that mean
they cannot possibly be directly referential, such as the NDNS uses
and quantifying-in cases; he then suggests that these cases could be
dealt with by a theory according to which complex demonstratives were
generalized quantifiers. This kind of consideration does not, of course,
rule out the possibility that another kind of theory, such as the one
given in the present article, could account for the data at least as well.
King also, however, points to three constructions that he claims provide more direct evidence that demonstratives are quantificational. He
26
Note that else in (140) is also functioning as a donkey anaphor. That is, nothing
else means something like nothing other than that donkey. If the hypothesis
behind the current paper is correct, the fact that else can be used as a donkey
anaphor means that it should also be able to be used in all the other ways that
pronouns can be used, since the same machinery is put into service for all these
uses. This seems to be correct. Here, for instance, is an example of else being used
as a descriptive indexical: we see Benedict XVI ordaining a cardinal, gesture towards
him, and say No-one else is allowed to do that. This could mean Only Benedict
XVI, at the moment, is allowed to do that. But it could also mean Only the Pope,
in general, is allowed to do that.
demoniacal.tex; 21/08/2007; 19:21; p.46
47
notes, firstly, that complex demonstratives can embed the problematic

pronouns in Bach-Peters sentences, so that parallel to the classic (146)
(Bach 1970) we also have (147).
(146) Every fighter who shot at it hit the MiG that was chasing him.
(147) Every friend of yours who studied for it passed that math exam
she was dreading.
(Note that there have to be different math exams for each friend in
order to ensure that we have a genuine Bach-Peters reading here.) King
says (2001: 12), The most plausible explanations of the acceptability
and semantics of such sentences assume that the phrases containing the
anaphoric pronouns are both quantifier phrases. We are thus supposed
to draw the conclusion that complex demonstratives are quantifier
phrases. Although he does not cite them, it is plausible to suppose that
King here has in mind the accounts of Bach-Peters sentences of Keenan
1972, Higginbotham and May 1981, and May 1989, which involve quantification over pairs, and which do indeed require that the two DPs
embedding the pronouns be quantificational. Note, however, that the
most empirically adequate of these accounts, Higginbotham and May
1981, requires that the two quantifier phrases merge to form a complex
quantifier, and that this requires a special mechanism. There has always
been another tradition in the analysis of Bach-Peters sentences that
does not require this mechanism, and does not involve quantification
over pairs, but which says instead that the first of the pronouns is a Dtype pronoun (Karttunen 1971, Jacobson 1977, 1991, 2000, Elbourne
2001a). This tradition, which is arguably more satisfactory than the
other one since it does not require any special mechanism, does not
require that the DPs embedding the pronouns be quantificational. I do
not think, then, that much weight can be placed on this argument for
complex demonstratives being quantificational.
A second construction that King suggests might afford some evidence for the view that complex demonstratives are quantifier phrases
(QPs) is antecedent-contained deletion (ACD) (King 2001: 17). He cites
May 1985 for the claim that QPs undergo movement at LF and referring
expressions do not. One environment where QR is arguably obligatory
in order to produce a well-formed structure is ACD. So (148) is not
interpretable unless the QP moves to create a structure like (149).
(Angle brackets indicate elided material.)
(148) Tiger birdied every hole that Michael did.
(149) [every hole that 2 Michael did <birdy t2 >] 2 [Tiger did birdy
t2 ]
demoniacal.tex; 21/08/2007; 19:21; p.47
48
Paul Elbourne
Now we note that ACD is possible with that-phrases too:

(150) Tiger birdied that hole that Michael did.
This might give us reason to believe that that-phrases are QPs.
However, as King notes himself in a long endnote (2001: 175, note
16), Mays (1985) claim that referring expressions never undergo QR
has been challenged. In addition to the sources cited by King himself on
this point, I would draw attention to Sags 1976 doctoral thesis, where
it is pointed out that bare argument ellipsis can be naturally analyzed
if we assume that (151) has an LF (152):
(151) I called John, and the teacher too.
(152) [John 2 [I called t2 ]] and [the teacher < 2 [I called t2 ] >]
For further relevant discussion, see Heim 1993, where it is argued convincingly that even bound pronouns must sometimes QR. For present
purposes we can note that ACD provides evidence in favor of a quantificational view of complex demonstratives only if the now controversial
view of May 1985 is correct.
The last construction discussed by King in this regard is weak crossover (WCO) (King 2001: 1819). Following much previous literature on
WCO, he claims that WCO effects arise when the lower DP in sentences
like (153)(158) is a QP but not when it is a referring expression.
He cites the following examples where he alleges we experience WCO
effects:
(153) His mother loves every man.
(154) His mother loves some man.
(155) His mother loves the man with the goatee.
(156) His mother loves no man.
By contrast, King claims we do not experience WCO effects in the
following sentences, where the lower DPs are plausibly referential:
(157) His mother loves John.
(158) His mother loves him [gesture].
We now observe, according to King, that there is WCO with thatphrases:
(159) His mother loves that man with the goatee.
So we conclude that that-phrases must be QPs since they pattern with
QPs according to this test.
demoniacal.tex; 21/08/2007; 19:21; p.48
49
At this point I must raise a question about Kings data. (159) does
not seem ungrammatical to me on the relevant reading, or to any of
several native speakers whom I have asked about it. The same goes
for (155). The judgments I have elicited, then, would tend to make one
suspect that complex demonstratives and definite descriptions were not
quantificational. In the previous and subsequent literature on WCO,
too, it is assumed, as far as I know, that complex demonstratives and
definite descriptions do not cause WCO. Lasnik and Stowell (1991) cite
the following example as involving no WCO violation:
(160) This book I would never ask its author to read, but that book I
would.
We should also note the judgments given on the following example of
Jacobsons (2000: 93):
(161) The man who loves her saw Mary/the woman with red hair/*every
woman with red hair.
Harley (2002: 661) says that (162) involves no WCO violation on the
relevant reading:27
(162) His mother loves the boy that Sue dislikes.
And King himself (2001: 176, note 20) reports judgments on slightly
different examples that are compatible with the ones I have encountered
from informants and in the literature: he reports that (163) and (164)
do not give rise to WCO effects, whereas (165) does.
(163) Someone who liked her asked that woman wearing a red jacket
to the dance.
(164) Someone who liked her asked the woman wearing a red jacket
to the dance.
(165) Someone who liked her asked every woman wearing a red jacket
to the dance.
King does not have an account of why (163) and (164) should be different from (155) and (159). I think the best we can say here, according
full respect to Kings judgments on (155) and (159), is that a confused
27
Harley comments (2002: 661) that proper names and definite descriptions in
object position must not have to undergo QR, for if they did WCO violations would
result. Furthermore, The self-evident reason that they do not undergo QR is that
definite DPs are not, in fact, quantificational, and hence do not need to move to be
appropriately interpreted. As far as I can see, this position is widespread in the
literature on WCO.
demoniacal.tex; 21/08/2007; 19:21; p.49
50
Paul Elbourne
and variable picture is presented by the data regarding possible WCO

caused by definite descriptions and complex demonstratives. But that
would mean that it would be unwise to base any conclusions on these
data concerning the semantic type of these kinds of DPs.28
4.1.3. Kings theory of complex demonstratives
King assigns great importance to speaker intentions in the semantics
of complex demonstratives. He postulates that there are two types of
relevant speaker intentions (2001: 2731):
1. Perceptual intentions. A speaker can intend to talk about a particular object that they are perceiving or have perceived. This will
give rise to supplied properties like being identical to b (abbreviated
=b), where b is the object in question.
2. Descriptive intentions. A speaker can intend to talk about whoever or whatever fits a certain descriptive condition; there is no
perception of the object.
Speakers with these intentions will supply the corresponding properties
at strategic points in the expansion of sentences involving complex
demonstratives.
Kings favored theory of complex demonstratives29 is called T2.
According to T2, the basic contribution of that to the semantics of
a sentence is the following (2001: 43):
(166)
and
are uniquely
in an object x and x is
The first and last slots are to be filled in by the properties denoted by
the NP and VP respectively; that is thus like other quantifiers, in that
it takes the NP and VP sets and contributes a relation between them.
28
Before leaving this topic, however, we should note that it would be theoretically
rather mysterious if definite descriptions and complex demonstratives were to give
rise to WCO effects. Accounts of WCO assume that the pronoun in question is
syntactically or semantically bound by the relevant DP. As Jacobson (2000: 93)
observes in connection with (161), there is no obvious reason why this should be the
case for the relevant reading of The man who loves her saw the woman with red hair,
even assuming that definite descriptions are quantifier phrases, because the pronoun
her could just be a referential pronoun that referred to the red-headed woman in
question. It would not have to be bound, syntactically or semantically, in order for
the relevant interpretation to be obtained: it could have an index separate from that
on, or bound by, the woman with red hair that just happened to pick out the same
woman. Like reasoning applies to complex demonstratives. See Harley 2002: 661,
footnote 2, for further discussion.
29
See section 3.3 for discussion of his views on unifying bare and complex
demonstratives.
demoniacal.tex; 21/08/2007; 19:21; p.50
51
When a speaker at w, t has a perceptual intention whose object is b,

the middle slots are filled in as follows:
(167)
and = b are uniquely jointly instantiated in w, t in an object

x and x is
This semantics is appropriate, then, for traditional examples like (168).

The sentence turns out to mean the same as (169), if it is spoken in w
at t.
(168) That animal [gesture at Flossy] is a donkey.
(169) animal and = Flossy are uniquely jointly instantiated in w, t in
an object x and x is a donkey.
Note how the properties supplied by perceptual intentions on Kings
theory resemble the contributions of the indices R and i on my theory.
When a speaker has a descriptive intention, the intention determines
some property O , and the middle slots are filled in as follows:
(170)
and O are uniquely jointly instantiated in an object x and

x is
For example, T2 deals with the NDNS case (171) as in (172). The
speaker has a descriptive intention, the relevant descriptive condition
being already expressed in the NP; the second slot merely repeats the
first, therefore, in this case.
(171) That hominid who discovered how to start fires was a genius.
(172) Being a hominid who discovered how to start fires and being a
hominid who discovered how to start fires were uniquely jointly
instantiated in an object x and x was a genius.
And here is how T2 handles a case of quantifying in:
(173) Most avid snow skiers remember that first black diamond run
they attempted to ski.
(174) For most avid snow skiers x: first black diamond run x attempted to ski and first black diamond run x attempted to ski
are uniquely jointly instantiated in an object y and y is such
that x remembers y.
T2 can, then, go a long way towards handling cases that are outside
the grasp of the traditional direct reference approaches.
One might ask why the instantiation of properties is limited to being
in the world of the utterance in cases of complex demonstratives used
with perceptual intentions. That is, why are we limited to schemata
like (175), as opposed to (176), especially since the restriction to w, t
does not appear in the cases of descriptive intentions?
demoniacal.tex; 21/08/2007; 19:21; p.51
52
Paul Elbourne
(175)
and = b are uniquely jointly instantiated in w, t in an object

x and x is
(176)
and = b are uniquely jointly instantiated in an object x and

x is
King maintains that this restriction is necessary because of cases like

the following (King 2001: 59):30
(177) (At w, t I look at b and say:)
That senator from California could be a crook.
The argument goes as follows. We expect a reading of (177) that has
reconstruction of the subject, as in the following LF:
(178) [could [that senator from California be a crook]]
T2 gives us the truth conditions in (179), which seem roughly correct.
(179) For some possible world w0 : senator from California and = b
are uniquely jointly instantiated in w, t in an object x and x is
a crook in w0
Note that if the restriction to the first two properties being jointly
instantiated in the world and time of utterance were removed, one
would presumably be able to obtain truth conditions like the following
from a theory in the spirit of Kings:
(180) For some possible world w0 : senator from California and = b
are uniquely jointly instantiated in w0 in an object x and x is a
crook in w0
But this is the wrong result. It incorrectly predicts that the sentence
will share a reading with (181):
(181) b could be a crooked senator from California.
The restriction to the world and time of utterance seems to be necessary, then, in cases of perceptual intentions.
30
I change the example slightly to provide a clearer instance of the kind of scope
ambiguity claimed by King. His own example, It is possible that that senator from
California is a crook, has the complex demonstrative embedded in a position where
it is arguably unable to take scope over possible. Compare It is possible that no
senator from California is a crook, which does not have a reading No senator from
California can be a crook, as it would if the subject of the embedded sentence could
scope out.
demoniacal.tex; 21/08/2007; 19:21; p.52
53
4.1.4. Possible problems for Kings theory

Thus Kings theory. I now move on to raising some possible problems
for this approach.
The first problem concerns occurrences of complex demonstratives
that appear to be bound. How would T2 handle an example like (134),
repeated here as (182)?
It is hard to see how this could be done. A perceptual intention would
not do, since that senator in this example does not refer to any one
senator. And no general descriptive intention of the sort envisaged in
King 2001 will help us out. What we need, sticking as closely as possible
to T2, is something like the following:
(183) There is no senator x such that Mary talked to x before: senator
and = x were uniquely jointly instantiated in w, t in an object
y and y was lobbied.
In other words, contrary to T2 as it stands, we need something like
a bindable index within the semantics and probably the syntax of
complex demonstratives, as provided by the theory advocated in the
current article.
It will presumably be possible to emend T2 so as to incorporate
something like this. Indeed an anonymous reviewer for Linguistics and
Philosophy, commenting on an early draft of the current article, suggested changing (184), the relation expressed by that in T2, to (185):
(184)
and
are uniquely
in an object x and x is
(185)
and
are uniquely
in an object x and x = y and x is
The new variable y, the reviewer suggests, could be assigned the demonstrated object in referential uses and bound in bound uses. In cases of
NDNS and quantifying in, y could be made identical to x, providing a
truth-conditionally trivial property:
(186)
and
are uniquely
in an object x and x = x and x is
There will presumably have to be some special rule or procedure to

accomplish this last effect, leaving this revised theory open to the
charge of being slightly ad hoc, but it would at least deal with bound
demonstratives as well as with the cases dealt with by T2. It might
very well be possible to devise other treatments along these lines.
demoniacal.tex; 21/08/2007; 19:21; p.53
54
Paul Elbourne
A second area in which T2 might require extension or modification

is its treatment of contrasts like the one in Kaplan 1989a given above
as (124), repeated here as (187). I have changed this man being pointed
at to that man being pointed at, since T2 is formulated to deal with
that.
(187) (Charles is from Charleston, West Virginia. Paul is from St.
Paul, Minnesota. is a pointing by the speaker in the direction
of Paul, who is seated on a chair in front of the speaker.)
Look over here. [The gesture is initiated now and held
throughout the next sentence.] If Charles and Paul had
changed chairs, then
a. the man being pointed at would be from Charleston.
b. that man being pointed at would be from Charleston.
As far as I can see, T2 as it stands leaves open the possibility of filling in
the second slot in (166) by means of a descriptive intention that would
contribute a property identical to the one expressed by the NP, as is
the case with the NDNS example in (172). So the following analysis
should be available for (187b):
(188) If Charles and Paul had switched chairs, then being a man
being pointed at and being a man being pointed at would be
uniquely jointly instantiated in an object x and x would be from
Charleston.
In other words, T2 seems to predict that there will be a true reading
of (187b) parallel to the reading available for (187a).
What could be done to overcome this problem? The only way I
can think of to make the correct prediction about (187b) while also
dealing with examples like (123), (126) and (127), is, as I explained in
section 3.5, to make use of Nunbergs (1993) apparatus of deictic and
relational components and his (1979) theory of the permitted relations
between indexes and interpretations. It would be possible, I presume,
to take T2 and insert all of this apparatus, but it would be a rather
radical change. Indeed, it would presumably yield a theory very much
like the one in the current article, the only difference being that in this
new theory complex demonstratives would be generalized quantifiers
and not individual concepts. I would welcome the emergence of such
a theory; the question would then be whether there was any reason
to prefer the version with individual concepts over the version with
generalized quantifiers.
This last question is an extremely difficult one, and I fear I cannot
address it adequately within the confines of the current article. The
demoniacal.tex; 21/08/2007; 19:21; p.54
55
phrase an object in (184) suggests that T2 analyzes complex demonstratives as existential quantifiers of a certain kind. The theory is, then,
very reminiscent of Russells (1905) analysis of the definite article. Now
there are arguments against analyzing definite descriptions as existential quantifiers. If the majority of the judgments on WCO that we
surveyed in section 4.1.2 are to be trusted, the lack of WCO violations
by definite descriptions in object position indicates that they are not
quantifiers at all; but then again, King (2001) disagrees with those
judgments. There is also another argument, made by Heim (1991),
Elbourne (2005b: 109112) and Kripke (2005: 1023), to the effect that
the Russellian theory of definite descriptions predicts that (189) will
have a reading equivalent to (190), which is not the case:
(189) If the ghost in my attic starts to make scary moaning noises, my
boring guests will leave.
(190) If there is exactly one ghost in my attic and it starts to make
scary moaning noises, my boring guests will leave.
(189) can only be read as presupposing that there is exactly one ghost
in my attic, whereas (190) leaves the possibility open. The analogous
criticism of T2 would point out that, on the face of it, T2 predicts that
(191) will have a reading (192), exactly equivalent to (190), and that
this does not seem to be the case.
(191) If that ghost in my attic starts to make scary moaning noises,
my boring guests will leave.
(192) If ghost in my attic and ghost in my attic are uniquely jointly
instantiated in an object x and x starts to make scary moaning
noises, my boring guests will leave.
Neale (2005a: 846) has briefly argued against this objection, but it
is still the focus of ongoing research by people on both sides of the
argument. For current purposes, we should note that these attempted
criticisms of the Russellian theory of definite descriptions transfer over
quite naturally to Kings (2001) theory of complex demonstratives. If
they are ultimately successful in the one area, it is likely that analogous
arguments will be able to be made in the other. The argument is basically the one between the Russellian and Fregean theories of definite
descriptions, and I am not about to resolve it here. See section 2.2
above and, in more detail, Elbourne 2005b: 98112 for my view.
To summarize, Kings (2001) T2 is an elegant theory that deals
with a large amount of the data concerning complex demonstratives.
I hope to have shown, however, that it could profitably be altered in
directions that would assimilate it to the theory defended in the current
demoniacal.tex; 21/08/2007; 19:21; p.55
56
Paul Elbourne
article, in order to deal with cases of bound complex demonstratives

and Kaplans (1989a) classic data concerning direct reference effects.
I have not tried to pronounce on whether a quantificational theory
of complex demonstratives will ultimately be possible, but have noted
that ongoing work on the existence entailments of definite descriptions
will probably shed light on this question.
4.2. A Comparison with Roberts 2002
4.2.1. Introduction
Roberts (2002) has already set out a theory within a dynamic framework that is in some respects similar to the one defended in the present
paper. She too argues that complex demonstratives are basically species
of definite descriptions and come with presuppositions about existence,
uniqueness and proximal/distal features. To a certain extent, then,
the present paper is merely to illustrate that the basic idea can be
cast within a situation-semantic framework at least as well as within
a dynamic one, rounding out the comparison between these theories
begun by Heim 1990 and continued by Elbourne 2005b. That said, I
do think that the theory in the current article is preferable in certain
respects to the theory in Roberts 2002.
4.2.2. Robertss theory of demonstratives
Roberts (2002, 2003) assumes the basic theoretical framework of Heim
1982, which I will assume to be at least broadly familiar to my readers.
Briefly, in Heim 1982 the semantic value of an expression is its Context
Change Potential, a function that takes a state of the context and returns another one. The context is a set of propositions that speaker and
hearer hold in common, the common ground, combined with a domain
of familiar discourse referents. A discourse referent is not an actual
thing in the world, but, roughly speaking, a mental entity, represented
by a natural number, that can be introduced to stand for one. Roberts
(2002: 112, 2003: 297306) introduces a taxonomy of ways in which
discourse referents might be familiar: strong familiarity obtains when
a discourse referent has been introduced by the utterance of a preceding DP, typically an indefinite; and weak familiarity obtains when the
existence of the entity referred to is evident to the participants in the
discourse, for example by direct perception or deduction from things
that have been said, even though it has not been mentioned.
Roberts concentrates on the presuppositions of demonstratives, which
in the framework she is using are the conditions that the context must
fulfill in order for the Context Change Potential to be defined. There
are in fact two sets of presuppositions for demonstratives; if either set
demoniacal.tex; 21/08/2007; 19:21; p.56
57
of conditions applies, the use a demonstrative is licit. The first set, in

an informal version, reads as follows (Roberts 2002: 117):31
(193) Presuppositions of Demonstrative DPs
Given a context C, use of a (non-)proximal demonstrative DPi
presupposes (a) that there is an accompanying demonstration
whose unique demonstratum, correlated with a weakly familiar discourse referent by virtue of being demonstrated, lies in
the direction indicated by the speaker at a (non-)proximal distance to the speaker, and (b) that the weakly familiar discourse
referent for the demonstratum is the unique familiar discourse
referent contextually entailed to satisfy the (possibly liberalized)
descriptive content of DPi .
These, then, are the presuppositions associated with demonstratives
in what I have been calling their canonically referring use. If these
conditions are fulfilled, the semantic contribution of the demonstrative
is just the weakly familiar discourse referent in question, handled in a
Heimian dynamic system. The second set of presuppositions are those
associated with what Roberts (2002: 122123) calls discourse deictic
demonstratives:32
(194) Presuppositions of Discourse Deictic Demonstrative DPs
Given a context C, use of a (non-)proximal demonstrative DPi
presupposes (a) that there is an accompanying linguistic constituent that is (non)-proximal to the occurrence of DPi 33 , and
(b) that the discourse referent introduced into the semantics by
31
I have changed Robertss NP to DP for the sake of consistency with the rest
of the current article.
32
Unlike (193), which is a quotation, (194) is my summary of the passage in
question, adhering as closely as possible to the format of (193). The footnotes in (194)
give the quotations from Robertss text upon which I have based the corresponding
parts of my summary. Roberts does give a formal version of the presuppositions
of discourse deictic demonstratives in her (58d ) (Roberts 2002: 123). But (58d )
seems to be self-contradictory: g(j) is supposed to fit the descriptive content Desc
of the demonstrative NPi , according to condition (ii), which makes it sound like
the ultimate referent of the demonstrative; but then in condition (iii) the definition
of the function discourse-referent makes it necessary for g(j) to be a linguistic
constituent. So the definition would work only in cases where the speaker is trying
to refer to a linguistic constituent. For this reason, I have not tried to base my
summary directly on (58d ), but on the summary and discussion in the preceding
text, which seems clear.
33
Roberts 2002: 122123: The proximity associated with the pronoun may be
helpful in picking out an antecedent NP, based on its relative proximity in the text
to the time of utterance of the demonstrative [. . . ] That is, we use the proximity
presupposition in the demonstrative to pick out (demonstrate) a maximally salient
NP or other constituent.
demoniacal.tex; 21/08/2007; 19:21; p.57
58
Paul Elbourne
is the unique familiar discourse referent contextually entailed

to satisfy the (possibly liberalized) descriptive content of DPi .34
If these conditions are fulfilled, the semantic contribution of the demonstrative is just the discourse referent in question, the discourse referent
contributed to the semantics by the antecedent linguistic constituent .
Although she does not go through examples in detail, Roberts (2002:
122) evidently intends this formulation to deal with all examples where
demonstratives are used anaphorically to a preceding DP or as bound
variables.
4.2.3. Possible problems for Robertss theory
I believe there are two kinds of problems with Robertss theory, those
that stem from details of her particular formulation, and those that
would affect any attempt to treat demonstratives as definites within a
contemporary dynamic semantics of the kind she uses.
As it is formulated, at least, it is difficult to see how Robertss theory
could deal with cases of descriptive indexicals and deferred ostension,
like the ones analyzed in section 3.4. The only case apart from canonically referring demonstratives that Roberts explicitly provides for is
the case of discourse deixis. But descriptive indexicals and deferred
ostension fall into neither of these two categories.35 It is not obvious how
Robertss theory should be changed so as to include a provision for these
cases; adding the whole apparatus of Nunberg 1993, as recommended
in the current article, would be a major revision.
A second interesting quirk of Robertss formulation is that her semantics seems to have difficulty dealing with (195), a variant of the
well-known example (196).
(195) If a bishop meets a bishop, he blesses that bishop.
34
Roberts 2002: 123: there is a slippage of the relationship between the demonstrative and its demonstratum, so that instead of presupposing the equation of the
discourse referent for the demonstrative with that for the demonstratum itself, as in
([193]), it is equated with the discourse referent introduced by the demonstratum,
its referent. Note that in this passage discourse referent for is used in two
different ways: in the first instance, it seems to mean discourse referent that is the
contribution to the semantics of, while in the second instance it seems to mean
discourse referent that evaluated in context yields. For the uniqueness condition,
see clause (ii) of (58d ) (Roberts 2002: 123).
35
An anonymous reviewer for Linguistics and Philosophy says that Robertss
theory of discourse deixis covers cases of deferred ostension, but I cannot see any
textual evidence of this. It is made quite clear, for example in the quotations given
in footnote 33 and in the definition of demonstration in discourse in Roberts
2002: 123, that discourse deixis covers only cases in which a discourse referent of a
distinct linguistic constituent is reused as the discourse referent of a demonstrative.
demoniacal.tex; 21/08/2007; 19:21; p.58
59
(196) If a bishop meets a bishop, he blesses him.

It was noted by Hans Kamp (reported in Heim 1990) that D-type
theories of donkey anaphora faced problems with sentences like (196).
D-type theories, broadly construed, interpret donkey pronouns as definite descriptions; and use of a definite description was supposed by
early D-type theorists such as Cooper (1979) to entail uniqueness of the
referent with respect to the descriptive content of the NP. In (196), however, there is no evident way that the pronouns can be interpreted as
definite descriptions, given this constraint: we cannot have the unique
bishop blesses the unique bishop, because there are two bishops in each
case described; nor can we have the unique bishop who meets a bishop
blesses the unique bishop, since there are two bishops who meet bishops
in each case, meeting being a symmetric relation. This, therefore, seems
like a problem for the D-type analysis of donkey anaphora. Dynamic
theories, on the other hand, generally have no difficulties with sentences
like these, since they just interpret the pronouns as different bound
variables. Robertss theory, however, does seem to run into trouble:
although Roberts assumes a Heimian dynamic semantics, she also reintroduces the uniqueness requirements that are thought to characterize
definite descriptions in the non-dynamic tradition. (See clause (b) of
(193) and (194).) Thus when we come to interpret the occurrence of
that bishop in (195) the current theory predicts that there will be presupposition failure, since the discourse referent used, whichever one we
choose, will not be the unique familiar discourse referent contextually
entailed to be a bishop. This is not the case, however: there is no
presupposition failure and the sentence is quite natural.36
Before we leave the topic of Kamps bishop sentences, we should note
that Roberts (2003: 333) does attempt to deal with (196), the variant
involving only pronouns. Is there any chance that her explanation here
might be extended to deal with (195)? In order to understand her
explanation, we must be aware of the presuppositions that she associates with the use of pronouns (Roberts 2003: 330), which are given
in (197):
(197) Presuppositions of Pronouns
Given a context C, use of a pronoun P roi presupposes that it
has as antecedent a discourse referent xi , which is:
36
The theory advocated in the present article, on the other hand, can avail itself
of the solution to the problem of (195) proposed in Elbourne 2003, 2005b. The
D-type solution proposed in these works makes crucial use of situation semantics
and hence would not generalize to dynamic theories unless they were to introduce
situation semantics in addition to their own dynamic devices, which would surely
be undesirable on grounds of theoretical economy.
demoniacal.tex; 21/08/2007; 19:21; p.59
60
Paul Elbourne
a) weakly familiar in C,
b) salient in C, and
c) unique in being the most salient discourse referent in C
which is contextually entailed to satisfy the descriptive content suggested by the person, number and gender of P roi .
Given this definition, the analysis of (196) in Roberts 2003: 333 goes
as follows. The two occurrences of a bishop in the antecedent will be
salient to different degrees when we come to find a discourse referent to
be the interpretation of he; we can assume, following work in Centering
theory (Grosz et al. 1995), that the subject is more prominent than the
object. Because of clause (c) in (197), this means that the discourse
referent contributed by the subject of the antecedent must be used for
the interpretation of he in the consequent. It would have to be used
for him, too, but the fact that a reflexive pronoun is not used prevents
this; this syntactic fact overrides clause (c) in (197) in the case of the
object of the consequent, and the discourse referent contributed by the
object of the antecedent has to be used instead. As Roberts herself puts
it (2003: 333), Use of him instead of a reflexive pronoun for the direct
object rules out taking the antecedent clause subject as antecedent
for him as well [. . . ], amounting to an additional implicated domain
restriction in the instantiation of [clause (c) of (197)].
What are we to make of this, and would it be possible to extend
this explanation to help out in the analysis of (195)? Note first that
in order to be able to analyze (195) by these means, the uniqueness
presuppositions currently part of the presuppositions of demonstratives
would have to be removed and the salience presuppositions currently
associated only with pronouns would have to be put in their place. This
would be no small change, therefore, and it might have consequences
for other parts of Robertss theorizing. But suppose, for the sake of
argument, that such a change could be made with no ill effects elsewhere. Would we have a viable analysis of bishop sentences? I believe
we would not, for the following reason. As Roberts acknowledges (2003:
333), the saliency conditions on pronouns would seem to force both
pronouns in (196) to have to be interpreted by the discourse referent of
whichever antecedent is most salient. The use of this discourse referent
for the first pronoun would presumably not make it any less salient,
and it would seem to have to be used for the second pronoun too.
Only Binding Theory intervenes and overrides this preference, in the
analysis as it stands. This predicts, then, that in an example where
Binding Theory did not place any constraints on the interpretation of
the second pronoun, the same discourse referent would have to be used
for the interpretation of both. The prediction is, then, that (198) should
demoniacal.tex; 21/08/2007; 19:21; p.60
61
only be able to mean . . . he blesses his own buskins, since Principle B

does not prevent genitive pronouns in the object being covalued with
the subject.
(198) If a bishop meets a bishop, he blesses his buskins.
This is not the case, however. This sentence could equally well describe acts of blessing ones own buskins or acts of blessing ones fellow
bishops buskins. I conclude, therefore, that the theory of bishop sentences in Roberts 2003 does not account for all the relevant data, and
that there is no point, therefore, in extending it to encompass bishop
sentences involving demonstratives.
I now move on to some problems that I think would face any attempt
to deal with demonstratives in a dynamic framework like that adopted
by Roberts. As noted in Elbourne 2005b: 20, dynamic theories like
those of Heim 1982, Groenendijk and Stokhof 1990 and van Eijck and
Kamp 1997 have problems with examples like (199), which is taken
from Jacobson 2000:
(199) (A new faculty member picks up her first paycheck from her
mailbox. Waving it in the air, she says to a colleague:)
Do most faculty members deposit it in the Credit Union?
As Jacobson (2000) notes, it has a covarying reading here. The sentence
means something like, Do most faculty members deposit their paycheck
in the Credit Union? Dynamic theories of the type mentioned deal with
examples of covariation without c-command, such as donkey anaphora,
by supposing that certain antecedents, normally indefinite DPs, can
introduce discourse referents that can function as bound variables in
the interpretation of the covarying expression. (199) poses a problem
for dynamic theories, then, in that we have covariation without ccommand, but there is no antecedent expression that could possibly
introduce into the semantics a discourse referent suitable for the interpretation of it, since it has no linguistic antecedent whatsoever. There
could of course be a weakly familiar discourse referent for the particular
paycheck that is waved in the air ; but this is no good to us, since we
need something that can function as a bound variable. In the present
context, then, we should note that there are examples like (199) that
involve demonstratives instead of pronouns, as we see in (200).
(200) (A new faculty member picks up her first paycheck from her
mailbox. Waving it in the air, she says to a colleague:)
Do most faculty members deposit this in the Credit Union?
Exactly the same problem arises here, and it is difficult to see how a
dynamic theory of the kind under discussion could handle it.
demoniacal.tex; 21/08/2007; 19:21; p.61
62
Paul Elbourne
A similar problem is posed by certain classic paycheck sentences,

such as (201) from Cooper 1979:
(201) John gave his paycheck to his mistress. Everybody else put it in
the bank.
The second sentence here means Everybody else put their paycheck in
the bank, with a bound variable reading. Again we have a pronoun
displaying a covarying reading without being c-commanded at any
linguistic level by anything that could bind it; but once more there is
nothing that normal dynamic semantics methods can do, since there is
no antecedent indefinite to contribute a discourse referent that could be
used in the interpretation of the pronoun. The intuitive antecedent, of
course, is his paycheck ; but this cannot contribute a discourse referent
in the requisite fashion, since it is a definite. Once again, then, we
should note in the present connection that similar examples arise with
covarying demonstratives, as we see in (202), where slight contrastive
stress should be placed on that and flowers:
(202) John gave his paycheck to his mistress. Every other man put
that in the bank and gave flowers to his mistress.
The example is interpreted in the same way as the previous one. It is
difficult to see how a theory of demonstratives as definites embedded
in Heimian dynamic semantics could deal with this.37
I conclude, then, that the theory of demonstratives proposed in the
current article has a greater empirical reach than the theory given in
Roberts 2002.
5. Conclusion
I conclude that a theory that sees demonstratives as individual concepts
in a situation semantics is entirely viable. The particular formulation
suggested in this article, which incorporates a version of the indexical
apparatus of Nunberg 1993, may even have some empirical advantages
over current rival theories.
37
An anonymous reviewer points out that Roberts talks about paycheck sentences
in Roberts 2004: 541. There is only a brief mention of paycheck sentences here,
however, and I do not see how what is said would help to overcome the difficulties
just noted.
demoniacal.tex; 21/08/2007; 19:21; p.62
63
References
Abney, Steven. 1987. The English Noun Phrase in its sentential aspect.
Doctoral dissertation, MIT. Distributed by MITWPL.
Austin, John L. 1961. Truth. In J.O. Urmson and G.J. Warnock, eds.,
Philosophical Papers, 117133. Oxford: Oxford University Press.
Bach, Emmon. 1970. Problominalization. Linguistic Inquiry 1, 1212.
Barwise, Jon, and John Perry. 1983. Situations and Attitudes. Cambridge, Mass.: MIT Press.
Bauerle, Rainer. 1983. Pragmatisch-semantische Aspekte der NP-Interpretation. In Manfred Faust, Roland Harweg, Werner Lehfeldt
and Gotz Wienold, eds., Allgemeine Sprachwissenschaft, Sprachtypologie und Textlinguistik. Festschrift f
ur Peter Hartmann, 121
131. T
ubingen: Gunter Narr Verlag.
van Benthem, Johan. 1995. Language in Action. Categories, Lambdas
and Dynamic Logic. Cambridge, MA: MIT Press.
Berman, Stephen. 1987. Situation-based semantics for adverbs of quantification. In James Blevins and Anne Vainikka, eds., Studies in
semantics, 4668. (University of Massachusetts Occasional Papers in Linguistics, vol. 12.) GLSA, University of Massachusetts,
Amherst.
Borg, Emma. 2000. Complex demonstratives. Philosophical Studies 97:
229249.
Braun, David. 1994. Structured characters and complex demonstratives. Philosophical Studies 74: 193219.
Buchanan, Ray. 2002. Understanding complex demonstratives. Ms.,
New York University.
Buchanan, Ray, and Gary Ostertag. 2005. Has the problem of incompleteness rested on a mistake? Mind 114: 889913.
Burge, Tyler. 1973. Reference and proper names. The Journal of Philosophy 70, 42539.
B
uring, Daniel. 2004. Crossover situations. Natural Language Semantics 12: 2362.
B
uring, Daniel. 2005. Binding Theory. Cambridge: Cambridge University Press.
Cacciari, Cristina, Carreiras, Manuel, and Barbolini Cionini, Cristina.
1997. When words have two genders: anaphor resolution for Italian functionally ambiguous words. Journal of Memory and Language 37: 517532.
demoniacal.tex; 21/08/2007; 19:21; p.63
64
Paul Elbourne
Carlson, Gregory. 1977. Reference to kinds in English. Ph.D. dissertation, University of Massachusetts, Amherst.
Carminati, Maria nella, Lyn Frazier, and Keith Rayner. 2002. Bound
variables and c-command. Journal of Semantics 19: 134.
Chomsky, Noam. 1981. Lectures on Government and Binding. Dordrecht: Foris.
Chomsky, Noam. 1993. A minimalist program for linguistic theory.
In Kenneth Hale and Samuel Jay Keyser, eds., The View from
Building 20. Essays in Linguistics in Honor of Sylvain Bromberger,
152. Cambridge, MA: MIT Press.
Chomsky, Noam. 1995. The Minimalist Program. Cambridge, Mass.:
MIT Press.
Cooper, Robin. 1979. The interpretation of pronouns. In Frank Heny
and Helmut Schnelle, eds., Syntax and Semantics 10: Selections
from the Third Gr
oningen Round Table, 6192. New York: Academic Press.
Cresswell, Max. 1985. Structured Meanings. Cambridge, Mass.: MIT
Press.
Dell, Gary. 1986. A spreading activation theory of retrieval in sentence
production. Psychological Review 93: 283321.
Edelberg, Walter. 1994. Propositions, circumstances, objects. Journal
of Philosophical Logic 23: 134.
van Eijck, Jan, and Hans Kamp. 1997. Representing discourse in context. In Johan van Benthem and Alice ter Meulen, eds., Handbook
of logic and language, 179237. Amsterdam and Cambridge, MA:
Elsevier and MIT Press.
Elbourne, Paul. 2001a. E-type anaphora as NP-deletion. Natural Language Semantics 9: 241288.
Elbourne, Paul. 2001b. On the semantics of pronouns and definite articles. In Leora Bar-el and Karine Megerdoomian, eds., WCCFL
20: Proceedings of the 20th West Coast Conference on Formal
Linguistics, 164177. Somerville, MA: Cascadilla Press.
Elbourne, Paul. 2003. Indistinguishable participants. In Proceedings
of the Fourteenth Amsterdam Colloquium, ed. Paul Dekker and
Robert van Rooij, 105110. Amsterdam: ILLC/Department of
Philosophy, University of Amsterdam.
Elbourne, Paul. 2005a. On the acquisition of Principle B. Linguistic
Inquiry 36(3): 333-365.
demoniacal.tex; 21/08/2007; 19:21; p.64
65
Elbourne, Paul. 2005b. Situations and Individuals. Cambridge, Mass.:

MIT Press.
Elugardo, Reinaldo. 2002. The predicate view of proper names. In Gerhard Preyer and Georg Peter, eds., Logical Form and Language,
467-503. Oxford: Oxford University Press.
Farkas, Donka. 1997. Evaluation indices and scope. In Anna Szabolcsi,
ed., Ways of Scope Taking, 183215. Kluwer: Dordrecht.
von Fintel, Kai. 1993. Exceptive constructions. Natural Language Semantics 1, 123148.
von Fintel, Kai. 1994. Restrictions on Quantifier Domains. Doctoral
dissertation, University of Massachusetts, Amherst.
von Fintel, Kai. 2004. Would you believe it? The King of France is back!
(Presuppositions and truth-value intuitions.) In Anne Bezuidenhout and Marga Reimer, eds., Descriptions and beyond: an interdisciplinary collection of essays on definite and indefinite descriptions and other related phenomena, 315341. Oxford: Oxford University Press.
Fox, Danny. 2000. Economy and Semantic Interpretation. Cambridge,
Mass.: MIT Press.
Fox, Danny. 2002. Antecedent-contained deletion and the copy theory
of movement. Linguistic Inquiry 33: 6396.
Frege, Gottlob. 1893. Uber

Sinn und Bedeutung. Zeitschrift f
ur Philosophie und Philosophische Kritik 100, 2550. [Translated as On
sense and reference, in Peter Geach and Max Black, eds., Translations from the philosophical writings of Gottlob Frege, 5678.
Oxford: Blackwell.]
Fromkin, Victoria. 1971. The nonanomalous nature of anomalous utterances. Language 47: 2752.
Fromkin, Victoria, ed. 1973. Speech errors as linguistic evidence. The
Hague: Mouton.
Fromkin, Victoria. 1988. The grammatical aspects of speech errors. In
F.J. Newmeyer, ed., Linguistics: the Cambridge survey, Volume
II, pages 117138. Cambridge: Cambridge University Press.
Garnham, Alan, Oakhill, Jane, Ehrlich, Marie-France, and Carreiras,
Manuel. 1995. Representation and processes in the interpretation
of pronouns: new evidence from Spanish and French. Journal of
Memory and Language 34: 4162.
demoniacal.tex; 21/08/2007; 19:21; p.65
66
Paul Elbourne
Garrett, Merrill. 1975. The analysis of sentence production. In G. Bower,

ed., Psychology of learning and motivation. Volume 9. New York:
Academic Press.
Garrett, Merrill. 1984. The organization of processing structure for
language production. In D. Caplan, A.R. Lecours and A. Smith,
eds., Biological perspectives on language, pages 172193. Cambridge, MA: MIT Press.
Garrett, Merrill. 1988. Processes in language production. In F.J. Newmeyer,
ed., Linguistics: the Cambridge survey, Volume III, pages 6996.
Cambridge: Cambridge University Press.
Glanzberg, Michael, and Susanna Siegel. 2006. Presupposition and policing in complex demonstratives. No
us 40: 142.
van Gompel, Roger, and Asifa Majid. 2004. Antecedent frequency effects during the processing of pronouns. Cognition 90: 255264.
Groenendijk, Jeroen, and Martin Stokhof. 1990. Dynamic Montague
Grammar. In Laszlo Kalman and Laszlo Polos, eds., Papers from
the Second Symposium on Logic and Language, 348. Budapest:
Akademiai Kiado.
Grosz, Barbara, Aravind Joshi, and Scott Weinstein. 1995. Centering: a framework for modelling the local coherence of discourse.
Computational Linguistics 21:203225.
Harley, Heidi. 2002. WCO, ACD, and QR of DPs. Linguistic Inquiry
33: 659664.
Heim, Irene. 1982. The semantics of definite and indefinite noun phrases.
Doctoral dissertation, University of Massachusetts, Amherst.
Heim, Irene. 1990. E-type pronouns and donkey anaphora. Linguistics
and Philosophy 13: 137177.
Heim, Irene. 1991. Artikel und Definitheit. In Arnim von Stechow and
Dieter Wunderlich, eds., Semantik. Ein internationales Handbuch der zeitgen
ossischen Forschung, 487535. Berlin: Walter de
Gruyter.
Heim, Irene. 1993. Anaphora and semantic interpretation: a reinterpretation of Reinharts approach. University of T
ubingen Seminar
f
ur Sprachwissenschaft working paper series (SfS-Report-07-93).
[Also published in Uli Sauerland and Orin Percus , eds., The
interpretive tract, 205246. (MIT Working Papers in Linguistics
25.) Cambridge, MA: MIT Working Papers in Linguistics, 1998.]
Heim, Irene and Angelika Kratzer. 1998. Semantics in generative grammar. Oxford: Blackwell.
demoniacal.tex; 21/08/2007; 19:21; p.66
67
Heim, Irene, Angelika Kratzer and Kai von Fintel. 1998. Introduction
to intensional semantics. Class notes, MIT.
Higginbotham, James, and Robert May. 1981. Crossing, markedness,
pragmatics. In Proceedings of the 1979 GLOW Conference. Pisa:
Scuola Normale Superiore de Pisa.
Hindley, J. Roger, and Jonathan P. Seldin. 1986. Introduction to combinators and -calculus. Cambridge: Cambridge University Press.
Jacobson, Pauline. 1977. The Syntax of Crossing Coreference Sentences.
PhD dissertation, University of California at Berkeley. (Distributed
by Indiana University Linguistics Club, 1979; published by Garland Press, 1980.)
Jacobson, Pauline. 1991. Bach-Peters sentences in a variable-free semantics. In Paul Dekker and Martin Stokhof (eds.), Proceedings
of the Eighth Amsterdam Colloquium, pp. 283302. University of
Amsterdam ILLC, Amsterdam.
Jacobson, Pauline. 2000. Paycheck pronouns, Bach-Peters sentences,
and variable-free semantics. Natural Language Semantics 8: 77
155.
Johnson, Kent, and Ernie Lepore. 2002. Does syntax reveal semantics?
A case study of complex demonstratives. Philosophical Perspectives 16: 1741.
Kadmon, Nirit. 1987. On unique and non-unique reference and asymmetric quantification. Doctoral dissertation, University of Massachusetts, Amherst.
Kaplan, David. 1989a. Demonstratives. In Joseph Almog, John Perry,
and Howard Wettstein, eds., Themes from Kaplan, 481563. Oxford: Oxford University Press.
Kaplan, David. 1989b. Afterthoughts. In Joseph Almog, John Perry,
and Howard Wettstein, eds., Themes from Kaplan, 565614. Oxford: Oxford University Press.
Karttunen, Lauri. 1971. Definite descriptions with crossing coreference.
Foundations of Language 7: 157-182.
Kayne, Richard. 1984. Connectedness and binary branching. Dordrecht:
Foris.
Keenan, Edward. 1972. On Semantically Based Grammar. Linguistic
Inquiry 3: 413461.
King, Jeffrey. 2001. Complex demonstratives. A quantificational account. Cambridge, MA: MIT Press.
demoniacal.tex; 21/08/2007; 19:21; p.67
68
Paul Elbourne
Kitagawa, Yoshihisa. 1986. Subjects in Japanese and English. Doctoral

dissertation, University of Massachusetts, Amherst.
Kratzer, Angelika. 1989. An investigation of the lumps of thought.
Linguistics and Philosophy 12, 607653.
Kratzer, Angelika. 2004. Covert quantifier restrictions in natural languages. Ms., University of Massachusetts, Amherst.
Kratzer, Angelika. 2006. Situations in natural language semantics. Ms.,
University of Massachusetts, Amherst.
Kripke, Saul. 1972. Naming and necessity. In Donald Davidson and
Gilbert Harman, eds., Semantics of natural languages, 253355.
Dordrecht: Reidel. [Also published as a separate volume with
a new introduction, Naming and necessity, Cambridge, Mass.:
Harvard University Press, 1980.]
Kripke, Saul. 1979. Speakers reference and semantic reference. In Peter
A. French, Theodore Uehling, Jr., and Howard Wettstein, eds.,
Contemporary perspectives in the philosophy of language, 627.
Minneapolis: University of Minnesota Press.
Kripke, Saul. 2005. Russells Notion of Scope. Mind 114: 10051037.
Larson, Richard and Gabriel Segal. 1995. Knowledge of Meaning. An
Introduction to Semantic Theory. Cambridge, MA: MIT Press.
Lasnik, Howard, and Tim Stowell. 1991. Weakest crossover. Linguistic
Inquiry 22: 687720.
Levelt, Willem. 1989. Speaking: from intention to articulation. Cambridge, MA: MIT Press.
Lewis, David. 1968. Counterpart theory and quantified modal logic.
Journal of Philosophy 65: 113126.
Link, Godehard. 1983. The logical analysis of plurals and mass terms: a
lattice-theoretical approach. In Rainer Bauerle, Christoph Schwarze and Arnim von Stechow, eds., Meaning, use and interpretation of language, 302323. Berlin: Walter de Gruyter. [Also
published in Godehard Link, Algebraic semantics in language
and philosophy, 1134, Stanford: CSLI Publications, 1998.]
Longobardi, Giuseppe. 1994. Reference and proper names: a theory of
N-movement in syntax and logical form. Linguistic Inquiry 25,
609665.
May, Robert. 1977. The grammar of quantification. Doctoral dissertation, MIT, Cambridge, MA.
May, Robert. 1985. Logical Form. Its Structure and Derivation. Cambridge, MA: MIT Press.
demoniacal.tex; 21/08/2007; 19:21; p.68
69
May, Robert. 1989. Interpreting Logical Form. Linguistics and Philosophy 12: 387436.
McElree, Brian, and Thomas Beaver. 1989. The psychological reality of
linguistically defined gaps. Journal of Psycholinguistic Research
18: 2135.
Neale, Stephen. 1990. Descriptions. Cambridge, Mass.: MIT Press.
Neale, Stephen. 1993. Term limits. In James E. Tomberlin, ed., Philosophical Perspectives 7: Logic and language, pp. 89123. Atascadero, California: Ridgeview Publishing.
Neale, Stephen. 2005a. A century later. Mind 114: 809871.
Neale, Stephen. 2005b. Pragmatism and binding. In Zoltan Gendler
Szabo, ed., Semantics versus pragmatics, 165285. Oxford: Clarendon Press.
Nunberg, Geoffrey. 1979. The non-uniqueness of semantic solutions:
polysemy. Linguistics and Philosophy 3: 143184.
Nunberg, Geoffrey. 1993. Indexicality and deixis. Linguistics and Philosophy 16: 143.
Nunberg, Geoffrey. 2004. Descriptive indexicals and indexical descriptions. In Anne Bezuidenhout and Marga Reimer, eds., Descriptions and beyond: an interdisciplinary collection of essays on
definite and indefinite descriptions and other related phenomena,
261279. Oxford: Oxford University Press.
Partee, Barbara, and Mats Rooth. 1983. Generalized conjunction and
type ambiguity. In Rainer Bauerle, Christoph Schwarze and Arnim
von Stechow, eds., Meaning, use, and the interpretation of language, 362383. Berlin: Walter de Gruyter.
Percus, Orin. 2000. Constraints on some other variables in syntax.
Natural Language Semantics 8: 173229.
Postal, Paul. 1966. On so-called pronouns in English. In F. Dinneen, ed., Report on the seventeenth annual round table meeting
on linguistics and language studies, 177206. Washington, D.C.:
Georgetown University Press.
Recanati, Francois. 1993. Direct Reference. From Language to Thought.
Oxford: Blackwell.
Recanati, Francois. 2004. Literal meaning. Cambridge: Cambridge University Press.
Recanati, Francois. 2005. Deixis and anaphora. In Zoltan Gendler Szabo,
ed., Semantics versus Pragmatics, 286316. Oxford: Clarendon
Press.
demoniacal.tex; 21/08/2007; 19:21; p.69
70
Paul Elbourne
Roberts, Craige. 2002. Demonstratives as definites. In Kees van Deemter

and Roger Kibble, eds., Information sharing: reference and presupposition in language generation and interpretation, 89-136.
Stanford: CSLI Press.
Roberts, Craige. 2003. Uniqueness in definite Noun Phrases. Linguistics
and Philosophy 26: 287350.
Roberts, Craige. 2004. Pronouns as definites. In Anne Bezuidenhout
and Marga Reimer, eds., Descriptions and Beyond: an interdisciplinary collection of essays on definite and indefinite descriptions and other related phenomena, 503543. Oxford: Oxford
University Press.
Russell, Bertrand. 1905. On denoting. Mind 14: 479493.
Quine, Willard van Orman. 1969. Ontological Relativity and Other
Essays. New York: Columbia University Press.
Sag, Ivan. 1976. Deletion and Logical Form. Doctoral dissertation, MIT,
Cambridge, Mass.
Schlenker, Philippe. 2003. A plea for monsters. Linguistics and Philosophy 26: 29120.
Shillcock, Richard. 1982. The on-line resolution of pronominal anaphora.
Language and Speech 25: 385401.
Soames, Scott. 1987. Direct reference, propositional attitudes, and semantic content. Philosophical Topics 15: 4787.
Sperber, Dan, and Deirdre Wilson. 1986. Relevance: Communication
and Cognition. Oxford: Blackwell.
Stanley, Jason. 2002. Review of King 2001. The Philosophical Review
111: 605609.
Stockwell, Robert, Paul Schachter and Barbara Hall Partee. 1973. The
major syntactic structures of English. New York: Holt, Rinehart
and Winston.
Strawson, Peter. 1950. On referring. Mind 59: 320344.
Takahashi, Eri, Anastasia Conroy, Jeffrey Lidz and Colin Phillips. 2006.
Early mastery of constraints on binding and coreference. Poster
presented at GALANA 2, McGill University, August 1719, 2006.
Taylor, Barry. 1980. Truth-theory for indexical languages. In Mark
Platts, ed., Reference, truth and reality, pp. 182198. London:
Routledge and Kegan Paul.
Thornton, Rosalind, and Kenneth Wexler. 1999. Principle B, VP ellipsis, and interpretation in child grammar. Cambridge, Mass.:
MIT Press.
demoniacal.tex; 21/08/2007; 19:21; p.70
71
Uriagereka, Juan. 1995. Aspects of the syntax of clitic placement in

Western Romance. Linguistic Inquiry 26, 79123.
Address for Offprints:
Department of Linguistics
School of Languages, Linguistics and Film
Queen Mary, University of London
Mile End Road
London E1 4NS
United Kingdom
demoniacal.tex; 21/08/2007; 19:21; p.71
demoniacal.tex; 21/08/2007; 19:21; p.72

Demonstratives As Individual Concepts

Uploaded by

Copyright:

Available Formats

Demonstratives As Individual Concepts

Uploaded by

Document Information

Original Description:

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Demonstratives As Individual Concepts

Uploaded by

Copyright:

Available Formats

Demonstratives as Individual Concepts

Paul Elbourne (p.d.elbourne@qmul.ac.uk)

demoniacal.tex; 21/08/2007; 19:21; p.1

only if the only contribution it makes to the content of an utterance

demoniacal.tex; 21/08/2007; 19:21; p.2