Introduction

The aim of this paper is to show that the `bound-variable' reading of pronouns, exemplified in (1), follows automatically from independently motivated assumptions about (i) the existence of expletive elements of category D (determiner/pronoun) and (ii) a strict mapping between meaning types and structural positions in the noun phrase.

(1) a. [Every driver](i) knows that he(i) is better than average
     b. Nobody(i) thinks that the lost-and-found office has his(i) stuff.

Specifically, I propose that bound-variable (BV) pronouns are nothing but expletive pronouns, i.e. pronouns void of any semantic content, except for phi-features, and that their `variable' interpretation comes entirely from the position [Head,DP] in which they reside. Therefore, there is no need for a theory of BV pronouns, which is entirely subsumed under a theory of expletives and `weak' DPs (in the sense of Milsark, 1974).

The next two sections will not talk about binding theory per se, but they will introduce a few necessary preliminary assumptions. Some predictions of the analysis will be discussed in section 3.

1. Preliminary assumptions: the role of D in the noun phrase.

To begin with, I adopt the idea that pronouns and determiners are the transitive and intransitive heads of the same category DP (Sommerstein 1977, Abney 1987), which can take NP as its complement. I disregard for the time being the role of various intermediate functional projections which might appear between D and N (Cinque, 1992, Crisma, 1991). As many have pointed out in defense of the `D' status of pronouns, determiners and pronouns are often close in morphological shape, and 1rst and 2nd person pronouns can even accompany nouns or other modifiers.

(2) a. We linguists should go on strike.
     b. You with red hair have a bad temper.

Second, I take NPs to denote properties that are predicated of D (cf. Heim 1982, Higginbotham 1987, Hudson, 1989, Longobardi, 1994, Zamparelli 1995). In this view, D is the only true entity-denoting portion of the noun phrase, while the noun plus any nominal modifier merely specifies the nature of this entity.

As an entity-denoting element, D can be either a variable or a constant; let's consider the two cases in turn, focusing on determiners first. To work as a variable, [Head,DP] must be bound by an appropriate quantificational operator. Typically, this function is performed by the class of `strong' lexical determiners, including "every", "most", "each", "the", "that", plus "many", "few" and "some" in their `proportional' interpretation. Notice that at spell-out these determiners `stand' on the very [Head,DP] position they are supposed to bind. Following Heim (1982), we may assume that at LF they undergo head movement and then unselectively bind their own trace; other technical solutions are conceivable, but unrelevant here.

Determiners in the `strong' class cannot be interpreted as `pure' existentials, and are in fact out to various degrees in Existential Sentences ("There are/is {every / each / the / most / many of the} soldiers ready"); moreover, they all induce presuppositions of existence for the entities they range over (cf. Barwise&Cooper, 1981, Zucchi, 1993). For instance, (3)a presupposes the existence of unicorns.

(3) a. If you find {every / each / the / most / many (of the)} unicorns(s),
        please let me know.
     b. If you find {unicorns / a unicorn / two unicorns / many unicorns},
        please let me know.

`Weak' determiners such as the indefinite "a", cardinal numerals ("one", "two"...) and the null determiner of `bare plurals' ("D0 Dogs" in "Dogs arrived from the yard") are not presuppositional ((3)b is not committed to the existence of unicorns), receive an existential interpretation and are fine in existential contexts. Given these differences, it has been proposed since Heim (1982) that determiners in the `weak' class are a type of modifiers without quantificational force, which leave the [Head,DP] position free, to be bound by external operators such as a `default' operation of Existential Closure (Kratzer,1989, Diesing 1992, etc.).

Using data from Italian and English, Zamparelli (1995) proposes that, qua modifiers, weak determiners are in fact located in a position lower than DP, in the predicative part of noun phrases (see Hoeksema, 1983, Hudson, 1989 for similar ideas). If this is correct, DPs with strong determiners are headed by the determiner (4)a, which binds D at LF, while DPs with weak determiners are headed by an empty D0, to get bound externally (4)b.

(4) a. [DP every [ boy]]
     b. [DP D0 [ a boy]]

So far, we have only seen cases in which [Head,DP] behaves as a variable. I will assume that to function as a constant [Head,DP] must be occupied by a non-quantificational (`referential') interpretable element at LF. Natural candidates are demonstratives, personal pronouns (presumably generated in D), and proper names, which---following Longobardi (1994)---I take to be generated in [Head,NP] and raised to the D position by LF (or earlier), so as to function as rigid designators (Kripke, 1972).

(5) a. [DP He ] arrived.
     b. [DP John(i) [NP t(i) ]] arrived.

Notice now that the DPs in (5) are presuppositional, just like the DPs with strong determiners in (3)a above. (6) requires the existence of, respectively, some salient individual "him" refers to, and a guy named "John".

(6) If you find {him / John}, please let me know.

Given the similarity between the structures in (4) and (5), we can capture this fact by linking the presuppositions of existence to the presence of an interpretable element in D. Informally:

(7) Generalization: Presuppositions of existence are triggered
    whenever the D projection contains an interpretable
    element at LF.

To sum up, I have sketched a view of the noun phrase which regards pronouns and determiners as the same category, and assigns to the structural position [Head,DP] a very special function, that of denoting an entity, constant or variable. If a variable, D must be bound externally or internally (by an overt determiner); if a constant, D must be occupied by a `referential' element at LF. (7) links existence presuppositions to the presence of an interpretable element in D.

2. Expletive articles and expletive pronouns.

In Northern Italian and in Catalan, a definite determiner precedes proper names even in their use as rigid designators.

(8) a. Il Gianni N. It.
        the John
     b. En Pere Cat.
        the Peter

This seems at odds with the idea that, by LF, the proper name must occupy the D head, the only suitable position for entity-denoting elements. Longobardi (1994) argues that these apparently `definite' determiners are best regarded as non-quantificational expletives, to match well-known cases of expletive pronouns. At LF, the proper name moves to D replacing (or adjoining to) the expletive and triggering the interpretation of D as a constant.

In some varieties of Northerns Italian, expletive determiners depart from their meaningful relatives in showing a degraded possibility for appositive adjectives to appear between D and N[1]: thus (9)a tends to be the definite descriptions of a particular Gianni-named guy who is enigmatic, or of an enigmatic Carlsonian `stage' of Gianni. It can hardly mean: "Gianni, who is enigmatic", or "That enigmatic Gianni", a possibility open to (9)b, where "signor Calderan" is a description and the definite is interpretable.

(9) a. Ho incontrato l' enigmatico Gianni
       `I have met the enigmatic Gianni'
     b. Ho incontrato l' enigmatico signor Calderan
       `I have met that enigmatic mister Calderan'

In English (or Italian) the weak usage of "the" with relational nouns like "daughter" or "student", in (10) (Poesio 1994) can be plausibly construed as an expletive. The `definite' in (10) lacks uniqueness and familiarity presuppositions, and it is quantificationally dependent on the innermost quantifier. Thus, the most natural paraphrase of (11)a is not (11)b (with "the" rendered as a simple existential), but (11)c, where "any" binds both "linguist" and "student".

(10) Yesterday I met [the daughter of [a linguist]]

(11) a. [The student of [ANY linguist]] should be able to solve this!
      b. Take any linguist; there is at least one of his students
        (possibly his best one) who should be able to solve this.
      c. Take any linguist and any student of that linguist: that student
        should be able to solve this.

In Zamparelli (1995) I have argued that weak "the" is licensed by the complement raising at LF through [Spec,"the"] and binding D.

Let's pull together the results so far: (i) pronouns and determiners belong to the same category D; (ii) the D position can denote a variable or a constant; (iii) there exist expletive determiners which let the D position be interpreted as a constant (by proper-name raising) (iv) there exist expletive pronouns which are (vacuously) interpreted as constants (e.g. "it" in "it rains").

The missing conceptual possibility is that there exist expletive pronouns which let the D position be interpreted as a variable;[2] I claim that this is precisely what bound variable pronouns are.

3. Some predictions

The analysis proposed has the advantage of requiring no additional machinery, or stipulations about an additional special `meaning' of pronouns. What remains specific to pronouns is the usual binding conditions A and B, which enforce constraints on the coindexation between the expletive and its operator.

On the empirical side, this idea automatically derives the fact that BV pronouns are always 3rd person (there are no 1rst and 2nd person expletives), plus two aspects of the BV reading of pronouns.

3.1. Montalbetti's Overt Pronoun Constraint.

Montalbetti (1984) observes that overt pronouns cannot have a bound reading when a corresponding non-overt form is available. In Spanish (and Italian) this excludes overt BV pronouns in subject position; in Japanese (Saito&Hoji, 1983) the same restriction bans overt BV pronouns throughout.

In the present approach this fact is linked to a tendency of realizing expletive pronouns non-overtly. For instance, in Italian, where normally pronouns can be overt or non-overt, no alternation is possible with weather expletives:

(14) a. {Esso / pro} esplose.
         {it / pro} exploded.
      b. {*Esso / pro} piove.
         { it / pro} rains.

In a minimalist framework, this suggests that whenever possible, expletives are realized as bundles of formal features with no phonetic content.

3.2 Lack of Existence Presuppositions.

When a personal pronoun is interpreted as a constant, it carries a presupposition of existence of its referent, be it in the discourse model, as in (5)a, or in the same sentence, as in "John said that he would leave". The same is not generally true of BV pronouns, as illustrated in (1)b, and below:

(12) No serious bug in the Shuttle software would leave NASA
      time enough to fix it.

(12) does not presuppose that there is a bug.[3]

Recall that we have linked the presuppositions to the presence of a interpretable element in D by LF (7). But a BV pronoun is not contentful: all its semantic function can be ascribed to the D position itself. Therefore, the lack of presuppositions is to be expected.[4]

A confirmation comes from the behaviour of other types of expletives. Obviously, "it" in "it rains" carries no commitment on the existence of an "it" which rains. Less obviously, weak "the" doesn't carry presuppositions per se. Imagine a context in which a new disease affects the sons of the women in a certain population. We do not know whether any woman has had a son since the outbreak, yet we can easily utter:

(13) b. Go out, and if you find [the son of a woman who has had the
       disease], report to us.

(weak-"the" inherits presuppositions: "the son of the woman ..." does imply that one such son exists, see Zamparelli, 1995, ch.5).

To sum up, the reasons for lack of presuppositions with weather expletives, BV pronouns and weak-"the" can be unified and traced to the lack of an interpretable element in D.

In conclusion, the idea that the BV meaning is carried by a property of D quite independent of the pronoun (which remains subject to the constraints of binding theory) captures empirical facts with no additional machinery.