The status of referential indices in syntactic theory is unclear. On one hand, theories of coreference and binding have been devised which avoid referential indices altogether (e.g. Safir 2004; Reuland 2011; Rooryck & Vanden Wyngaerd 2011). The development of these theories was motivated by the Inclusiveness Condition of Chomsky 1995 (p. 228), which posited that symbols such as indices could not be added to syntactic representations during the computational procedure.

On the other hand, several streams of research in syntax and semantics continue to assume that indices are part of grammatical representations. For example, the rigidly referential properties of demonstrative noun phrases suggests that they include a referential index, as in (Nunberg 1993; Elbourne 2008; Simonenko 2014; Šimík 2016). The semantic effect of this index is to anchor certain noun phrases to particular referents in the discourse or context. Similarly, Schwarz (2009) and Jenks (2018) argue that anaphoric definites have an overt index while unique definites do not.

Indices have also proven indispensable in research on the grammar of focus. The widely adopted formalism of Kratzer (1991) relies on a binding relationship between a focus operator which evaluates alternatives and a distinguished focus index on the focused constituent which generates those alternatives, necessitated by the island-insensitivity of this relationship (see also Wold 1996; Beck 2016; Bassi & Longenbaugh 2020). In addition to providing an account of island-insensitivity, index-based theories of focus account for intervention effects with multiple foci (Beck 2006). Overt syntactic correlates of focus indices have been proposed in Basaá focus constructions (Leffel et al. 2014) and Sinhala wh-particles de (Romero & Meertens 2022).

Outside of focus and reference, indices have been essential in the analysis of switch-reference morphology, which has been recently been analyzed as index-sensitive agreement in Washo (Arregi & Hanink 2018) and Amahuaca (Clem 2019a; b; To appear). Leading analyses of sentence-external same and different also make crucial use of indices Brasoveanu (2011); Hardt & Mikkelsen (2015); Hanink (2018). Finally, reference to indices are needed to account for the distribution of null versus overt pronouns in Mayan languages such as Popti’ (Aissen 2000) and Chuj (Royer 2021): if two overt third person pronouns occur in the same clause and neither binds the other, they must have disjoint reference. If one of these pronouns is null, it must be coindexed with the other. Royer (2021) argues that this effect arises at PF, requiring the second of two coindexed DPs to be elided, hence necessarily represented in the syntax. Summing up, then, there is strong evidence from a variety of sources pointing to the fact that indices are a part of the syntax and semantics of natural languages.

It seems that there is strong support for the idea that referential indices may be present in syntax. The only way to include indices in the syntactic representation without violating inclusiveness is to provide them with a featural counterpart from the start. As is the case with any feature, a featural counterpart of indices would be expected to be part of some but not all syntactic expressions. Building on the above work, this paper argues that incorporating indices into syntactic theory illuminates the differences between unique definite descriptions and a class of expressions which are labeled indexed definites. Indexed definites, or D^xPs, are proposed as a natural class of rigidly referential noun phrases which include demonstratives, anaphoric definites, and pronominal definites (such as we linguists). The main idea is that pronouns, anaphoric indices, and demonstratives share a syntactic feature idx, interpreted as a variable. In the general theory of idx sketched in Section 3, idx is category-neutral and combines with different syntactic categories to create anaphoric or indexical counterparts for those categories. The class of indices which refer to individuals are of category D/idxPs; when they occur in indexed definites they occupy [Spec, D^xP]. Plain unique definite DPs, on the other hand, lack an index in their specifier. This contrast is illustrated in Figure 1, where the president is a unique definite DP while that president is D^xP. In this example, the demonstrative determiner that is the realization of idx in [Spec, D^xP].

Semantically, plain or unique definites DPs are headed by a definite determiner interpreted as an ι operator. Once composed with the NP complement of D, plain definite DPs denote the unique individual or maximal plurality with property P in situation s, e.g., the unique president in s where s= ‘somewhere in France in the year 2022’. Indexed definite DPs, on the other hand, are headed by a D^x head. D^xPs denote the unique individual or maximal plurality with property P in situation s who is identical to the denotation of the D/idxP, namely, the individual located at the index on idx_x, namely the discourse referent g(x), e.g., ‘the unique president who is g(1),’ where g(1)=Emmanuel Macron in some context.

The idea that demonstratives, pronouns, and anaphoric indices occur in the specifier of some DPs has been assumed in some form in much work on DP syntax (Giusti 2002; 2015), anaphoric definiteness (Schwarz 2009; Jenks 2015; 2018), demonstratives (Šimík 2016; 2021b; Ahn 2019), and strong or demonstrative pronouns (Patel-Grosz & Grosz 2017; Clem 2017; Bi & Jenks 2019). An alternative perspective on syntactic indices is pursued by Simonenko (2014) and Hanink (2018; 2021), in which an equivalent of idxP lives below DP; this paper will build on this work in pursuing a general theory of syntactic idxPs, but will argue that the syntax in Figure 1 is correct, that is, that indices generally surface as a specifier or perhaps adjunct of the DP projection.

The first part of this paper motivates and sets out the distinction between unique vs. indexed definites. Section 2 reviews arguments that plain definites are characterized only by uniqueness, and shows how this conclusion accounts for the ability of plain definites to function predicationally and to covary based on situations. Section 3 sketches a theory of syntactic indices, in particular the idea that there is a general-purpose syntactic idx feature, realized as what linguists usually call pronouns and demonstratives, which is interpreted as a variable which picks out a discourse referent. Section 4 argues for the existence of a dedicated syntax and semantics for indexed definites. The arguments for the proposed syntax focus on the phrasal status of demonstratives and pronouns and their ability to co-occur with definite articles in many languages. The argument for the proposed semantics come from the rigidly referential properties of indexed definites and the observation that a trio of interpretations characterize pronouns, demonstratives, and indexed definites: direct reference, anaphoric reference, and covariation via dynamic binding.

The second part of the paper (Section 5) builds further support for the proposal with a case-study of Marka-Dafing, a Mande language spoken in Burkina Faso. Like many Mande languages, Marka-Dafing has an enclitic definite article, shown in (1a). We show that this enclitic is a plain or unique definite marker. However, Marka-Dafing has a dedicated construction for anaphoric definites which includes an additional syntactic marker, the prenominal element wó, glossed idx in (1b).

The distribution and interpretation of wó in (1b) and elsewhere provides direct support for the conclusion that it introduces a discourse-anaphoric index, as in the analysis of German strong definites of Schwarz (2009). The anaphoric index wó is shown to be in complementary distribution with possessors, both of which are in [Spec, DP].

Finally, section 6 centers around the observation that wó is incompatible with the exophoric demonstrative in Marka-Dafing, despite the fact that they occupy syntactically distinct positions. The proposed explanation is the following general constraint:

This constraint follows if the proposed structure for indexed definiteness is a general property of human language, i.e., if indexed DPs more generally only rely on a single referential index to refer.

Many languages make use of a morphosyntactic feature [±definite], typically associated with a functional element within DPs (Lyons 1999: p. 16). In this section I review existing arguments which show that uniqueness/maximality is both necessary and sufficient to account for the distribution of [+definite], and hence to characterize the distribution of plain definite DPs. One consequence of this claim is that semantic functions such as referentiality which have often been associated with definite articles are not necessarily associated with definiteness as a notional category per se (cf. Coppock & Beaver 2012).

One compelling argument that definiteness can be reduced to uniqueness/maximality is made by Coppock & Beaver (2012; 2015), who point out that in some predicative uses of the definite article, particularly under negation, the existence presuppositions which are textbook properties of definite articles fail to surface (Coppock & Beaver 2015: p. 380). Such contexts are illustrated below:

In the positive example in (3a), the definite article contributes the inference that there is just a single author of Waverly. This same uniqueness inference holds in (3b) as well, but the sentence explicitly negates the existence of such an author, so the existence of a unique author of Waverly cannot be presupposed.

More generally, definite articles are freely available in predicative contexts where uniqueness holds but there is no reference, for example with superlatives like Fido is the biggest dog in the pound, including cases without a real-world denotation such as the largest real number. Such uses are unique, but they are not referential and in the latter case cannot be said to presuppose existence.

Coppock and Beaver propose that in predicative contexts, the is interpreted as an identity function which checks for uniqueness.

Coppock & Beaver (2015) propose that the existence presuppositions characteristic of argumental uses of definites arise due to an additional ι type-shift (Partee 1987; Chierchia 1998; Dayal 2004). However, the only consistent semantic notion associated with [+definite] in predicative and argumental contexts is uniqueness.

Another central argument for purely uniqueness-based accounts of definiteness come from cases of situation-based covariation, i.e., contexts where a definite is not referential but covaries due to quantification over situations:

What is important about this sentence is that different presidents pardon turkeys on different Thanksgivings, so the definite description the president fails to denote a single individual, though here it functions as an argument.

Elbourne (2005; 2013) and Schwarz (2009) shows how cases like (5) follow from a theory where the uniqueness presupposition characteristic of definiteness is relativized to a situation s, part of the semantic metalanguage. Given a predicate P and a situation s, predicative uses of the definite article simply returns the set of individuals for whom P is true in the situation s:

In (5), the situation variable of the definite is bound by the quantifier every Thanksgiving, deriving covarying readings of presidents with Thanksgivings (cf. Berman 1987; Kadmon 1987; Heim 1990). Informally, the situation-dependent denotation of the president can be paraphrased as ‘the unique president in sʹ, where sʹ is a situation variable bound by the quantificational topic, which can be paraphrased ‘Every sʹ which is a Thanksgiving in the United States.’

The main alternative to uniqueness-based views of definiteness comes from discourse-oriented approaches to meaning, where definite NPs are associated with familiar discourse referents (Heim 1982; Kamp & Reyle 1993; Kamp et al. 2011; Roberts 2002). However, some of the most striking evidence that definiteness should be characterized primarily by uniqueness rather than discourse familiarity comes from languages with a morphosyntactic contrast between unique and anaphoric definiteness (Schwarz 2013). Such contrasts have been described in some detail in at least German and Fering (Schwarz 2009), Icelandic (Ingason 2016), Swedish (Simonenko 2014), Akan (Arkoh & Matthewson 2013),1 Thai (Jenks 2015), Korean (Ahn 2017), Mandarin (Jenks 2018), Lithuanian (Šereikaitė 2019), Cuevas Mixtec (Cisneros 2019), ASL (Irani 2019), and Chuj (Royer 2019; To appear).2 If anaphoric definites are distinct from plain definites in these languages, then plain definites must introduce a requirement distinct from anaphoricity; uniqueness is the only plausible option. As predicted, the simpler plain definites in such languages are always those which mark uniqueness and occur in situation-dependent covarying readings (e.g. Schwarz 2009; Jenks 2018), while the more complex expressions are are directly referential and anaphoric (Schwarz 2009; Simonenko 2014; Jenks 2018).

If plain definites only presuppose uniqueness, we can then ask why expressions such as demonstratives and pronouns can be used in non-unique contexts to refer, either to contextually salient individuals in the immediate context or to a discourse referent, and moreover, why they must refer. It has been proposed in recent work that there is a syntactic feature for a class of syntactic heads that introduce a referential index (cf. Simonenko 2014; Hanink 2018; 2021; Ahn 2019); we will adopt the label idx, from Hanink 2021. The idx is the core of demonstrative and pronoun projections, expressions whose commonalities have long been recognized in the semantics literature (Kaplan 1977; Nunberg 1993). Hanink (2018; 2021) provides striking evidence for this conclusion from Washo (Isolate; Lake Tahoe, California), where third person pronouns and demonstratives share a common morpheme gi:

Hanink builds an analysis of noun phrases in Washo based on the idea that gi realizes an idx feature, which is proposed to realize either variable binders or variables themselves.

For Hanink, however, idx occurs in the extended projection of nouns above NP and below DP. The ability of demonstratives to occur with adjectives (that tall) and adverbs (that slowly) suggest their distribution is not tied to a particular syntactic category. Usually functional projections such as number or aspect only associate with a particular lexical category.

Suppose instead that idx is not part of the extended projection of DP, but instead a feature which freely combines with any lexical or functional category, resulting in an indexed counterpart of that category. The resulting complex head, [idx, F]—F a category feature—is a phrasal projection with the distribution of F. This immediately accounts for the fact that pro-forms are generally distributed across various categories in natural languages. For example, the idx counterparts of various categories are suggested below for English:

English demonstratives this, that are underspecified for category, explaining their ability to occur as anaphora for a wide range of categories, including for predicates Partee (1986; 1987).3

The observation that this and that have a largely category neutral distribution provides an independent argument for the existence of a category-neutral feature like idx. More specifically, if lexical items are the realization of abstract bundles of syntactic features, then there must be some feature like idx which serves as a shared feature in the category neutral realization of functional morphemes like English so, this, or that.4 When idxPs are associated with a particular category F we will label the resulting phrase F/idxP; this paper will focus on D/idxP.

D/idxP are internally complex. This complexity includes ϕ-features and features indicating spatial deixis. These features serve as part of the articulated superstructure of idx, and are restricted to occur with idx features at least in part by their semantics, as specified below.

The idea that pronouns are featurally different from other noun phrases has been pursued extensively elsewhere (Bejar 2003; Harley & Ritter 2002; Harbour 2016; Sichel & Wiltschko 2021; Sichel & Toosarvandani To appear), including the idea that they bear an index feature which restricts their binding behavior (Rezac 2004; Adger & Ramchand 2005). The contribution of the D feature specifically is to restrict the variables which idx allows to be included in the domain of individuals. I follow Ahn (2019) in introducing the particular index as type n (numeral) argument of idx.

Now suppose that the features associated with an index—e.g. gender, number, or place—are partial identity functions on the variable (of type ⟨e,e⟩) which ‘add in’ presuppositional content, restricting the referential potential of the index (Cooper 1979; Heim & Kratzer 1998; Sauerland 2009). First and second person features, for example, would require that the variable be restricted to individuals which have as a subpart the contextually supplied speaker or hearer. Inherently spatial demonstratives, on the other hand, might restrict the potential denotation of the variable to individuals at a contextually defined location.5

The variable introduced by D/idx is what Groenendijk & Stokhof (1990) call a discourse marker, an element of D_e, formally distinct from the regular metalanguage variables used by the compositional procedure. The idea that discourse markers constitute a distinct set of variables from regular metalanguage variables has been used to account for a variety of phenomena including donkey anaphora (Groenendijk & Stokhof 1990; Chierchia 1995), the distinction between null and overt pronouns (Kurafuji 1998), and the semantic behavior of anaphoric definites (Schwarz 2009), and recently, weak crossover (Chierchia 2020).

A well-known problem for the idea that pronouns are (always) interpreted as variables is the availability of descriptive uses of pronouns, also called E-type pronouns, pronouns of laziness, paycheck pronouns, or neontological pronouns.7 The following example based on Cooper (1979) illustrates the problem:

In the most natural reading of this sentence, the president in the first sentence and he in the second are two different people, the latter perhaps the president-elect who we know identifies as a man. Such data pose problems for the idea that pronouns are variables over discourse referents, because the pronoun he should pick up its discourse antecedent, which would be the current president. This reading is available if we imagine a context where a president changes parties, but the president could also be a different person.

One widely adopted solution to this problem is to view he in (12) as a unique definite description whose referent can covary by situation. As such, a plausible analysis of descriptive pronouns is as plain definite descriptions with a deleted NP (Karttunen 1969; Partee 1970; Heim 1990; Elbourne 2001; 2005). If this proposal is correct, the pronoun itself could not bear a D/idx feature, as the index is introduced by the deleted NP.8

However, there is a way of reconciling an ellipsis-based analyses of descriptive pronouns with the idea that pronouns always are the morphological realization of D/idx. Suppose that ellipsis is licensed by an ‘E’ feature, as proposed by Merchant (2001: 61), who defines E as an identity function which introduces a presupposition that the elided constituent is e-given, which, as defined by Merchant, entails the presence of a “salient antecedent” (p. 26), a plausible idx. So E seems to entail the presence of [idx], perhaps in addition to a feature such as +anaph, which introduces the requirement for an explicit discourse antecedent (Hankamer & Sag 1976), not unlike the Marka-Dafing anaphoric definite pronoun discussed below. Descriptive pronouns then such as it in (12) would then be cases of a plain definite D (ι) which occur with complements containing [N, idx, +anaph]:

The [+anaph] feature requires that the property anaphor of N/idxP be an established discourse referent. Interpretively, the N/idxP is interpreted as the property of being the president in a particular situation.

Combined with the definite determiner and a situation variable introduced at D, we achieve the descriptive reading.

On the morphological side, suppose that third person pronouns in English have lexical entries like the following one:

These lexical entries could either realize a D/idxP with ϕ-feature restrictions, which could function as DP arguments on their own, or as plain definite DPs with an N/idxP complement. Such an approach could explain why third person pronouns do not allow overt NP complements in standardized English: it is precisely the ellipsis of N which corresponds to the availability of the idx. Another way of putting this is that English third person pronouns are ambiguous between something like ‘the one’, where ‘one’ is N/idx, and referential pronouns, D/idxPs. A twist on this proposal would be to adopt the view that gender features always originate on N (Kramer 2015; 2016; Conrod 2019), composing with either D or D^x to derive descriptive vs. referential readings of pronouns.

This ambiguity seems to be realized overtly in languages which have phonologically strong pronouns, which are referential, and phonologically weak or null pronouns. Typically if one set of pronouns allow descriptive or sloppy readings it will be with weak or null pronouns (e.g. Kurafuji 1998; 1999; Oku 1998; Saito et al. 2008; Runic 2014; Sakamoto 2017; Patel-Grosz & Grosz 2017; Bošković 2018; Barbosa 2019; Bi & Jenks 2019).9 This finding is consistent with the idea that weak and null pronouns realize NP anaphora of some kind rather than DP anaphora (see e.g. Tomioka 2003; Runic 2014); null pronouns can be analyzed as N/idx, at least in cases of certain sloppy readings, while strong pronouns are always referential D/idxP.

An indexed definite DP is a DP with an D/idxP in its specifier position. In support of this theory, this section presents evidence that adnominal pronouns and demonstratives occupy [Spec, DP] in many languages. We will then examine the semantic effects of the D/idxP specifier in indexed definites, namely the unavailability of situation-based covariation and three semantic functions which are shared with strong pronouns: exophoric uses, anaphoric uses, and dynamically bound uses. This section focuses primarily on complex demonstratives and pronominal definites of the we linguists variety; the following section shows that the theory here extends naturally to account for the overt anaphoric D/idxP in Marka-Dafing.

This section turns to a case-study on the distinction between plain and anaphoric definiteness in Marka-Dafing, which provides empirical support for the conception of indexed definites described above. In Marka-Dafing, anaphoric definites are shown to require an overt anaphoric index, which also functions as a pronoun, this anaphoric index can be clearly shown to occupy [Spec, DP] a position distinct from the definite article. Furthermore, the expected semantic contrasts between plain and indexed definiteness arise when this element is present.

We will zoom back out now to try and understand the kinds of referential indices in indexed definites in a more general sense. We have seen evidence for the following three kinds of restrictions provided by idxPs:

All three restrictions point to some element of the context. Local personal pronouns point directly to the speaker or hearer in a particular context (Sudo 2012), demonstrative pronouns point to different parts of the context: location, in the case of spatial demonstratives, and reference in time in the case of anaphoric demonstratives.25

When we try to combine these restrictions, we find that it is impossible. Instead, languages must select only one kind of deictic restriction per indexed definite, even when there is good reason to think that multiple kinds of deixis should be possible. More generally, this section suggests that there evidence for the following generalization:

This principle can be derived from the requirement that indexed definite DPs have only a single D/idxP specifier, and each D/idxP is always restricted along a single deictic dimension, whether by referring directly to the conversational participants, to the location of an individual in space, or in time. This constraint accounts for the generalization by Hsu & Syed (2020) that markers of identifiability (roughly, anaphoric definiteness) and deixis (roughly, exophoric demonstratives) never seem to co-occur in the same DP. We will review several cases below where this constraint is in effect.26

First, regarding the co-occurrence of local pronominal D/idxPs and demonstrative D/idxPs, there simply do not seem to be any such cases, which would look like the following:

Apparent counterexamples come from cases of demonstrative-pronoun co-occurrence in (23) and (24). But if the demonstratives in these languages are D^x heads, then the pronoun can be analyzed as the D/idxP.

The evidence for the incompatibility of exophoric and anaphoric reference is clearer, particularly when we look at Marka-Dafing. In exophoric contexts, Marka-Dafing makes use of a distinct demonstrative determiner míi which occurs in the DP-final head position and is restricted to pointing, or exophoric, reference:

There is good evidence that míi is a D head, as not only is it DP-final but it is in complementary distribution with definite = ú:

The puzzle is just that, despite occurring in different syntactic positions, deictic míi and anaphoric wó cannot co-occur:

Simple syntactic and semantic explanations for the ungrammaticality of (69) fail. Syntactically, wó and míi occupy different positions: wó is in [Spec, DP], while míi is a D head. In both a general pragmatic as well as in purely logical terms, the meanings of the two demonstrative elements are compatible: there is no reason that a speaker could not be pointing at some dog, satisfying the exophoric requirement of míi, and that the same dog could also be an established discourse referent, satisfying the anaphoric requirement of wó.

The simple theory of indexed definiteness described above, however, provides a natural account for this restriction. Suppose exophoric indexed definite míi always contains a phonologically null idxP in its specifier position, restricted to the individual being pointed at by the speaker. Call this specifier idx:exo. Because indexed definites only have a single specifier, wó is blocked by idx:exo, which is required in the context of míi.

Other languages in the literature where exophoric and anaphoric demonstratives both exist show the same restriction. For example, in Limbum (Grassfields Bantu, Becker 2018, which has an anaphoric article and an exophoric demonstrative pronoun (the opposite of Dafing), the two elements also cannot co-occur:

Another illustration of the Single Index Constraint comes from Korean. Korean has three demonstrative determiners, two of which are strictly exophoric and one of which is strictly anaphoric Ahn (2017). Unlike in Marka-Dafing and Limbum, such demonstratives are in paradigmatic opposition as D heads (Chang 2009), and again are blocked from co-occurring.27

It is possible to demonstrate the effect of the Single Index Constraint in English, lending support to the idea that it is a universal constraint. However, its effect is apparent not in terms of the incompatibility of distinct demonstratives, but rather in restricting that, which generally is capable of both anaphoric and exophoric reference, to pick just one of these deictic restrictions. To see that this restriction holds, one needs to set up an context where anaphoric reference is felicitous, and then to point at an referent with an exophoric demonstrative. Assuming a bias towards the immediate context, i.e., pointing, it should be impossible for the individual being pointed at and the one being referred to anaphorically to be the same individual. This kind of setup is illustrated in the following example, adapted from Wolter (2007: ex. 11, p. 612):

In this discourse the two dogs—Barky and the one Mary believes to have ruined her garden—cannot be the same, although context would seem to strongly prefer such an interpretation. Somehow the final sentence still feels like a non sequitur; to get the intended interpretation, the presupposition cleft that is the dog that ruined her garden would be necessary. The restriction observed above follows from the Single Index Constraint, exactly as in Marka-Dafing: the exophoric use of the demonstrative in the final sentence blocks an anaphoric interpretation.28

In the following example the effect of the Single Index Constraint is perhaps even clearer:

(72a) can felicitously be followed by (72b), where the subject is an anaphoric definite and the object is exophoric, each making use of separate (restrictions on) indices. The demonstrative subject of (72c) attempts to utilize simultaneous anaphoric and exophoric reference, but as long as the speaker is pointing to the man behind the pot, the sentence again feels non sequitur in the context of (72a): there must be two creepy men in the room. This effect follows in English if simultaneous anaphoric and exophoric features indices are not allowed. This restriction likely follows because exophoric indices serve to establish reference, hence are new, whereas anaphoric indices must be part of the context. But crucially, only one index is allowed per DP.

This paper has argued that indexed definites are definite noun phrases which contain a specifier which is a projection of an idx head, interpreted as a variable. This category defines the class of pronominal elements in language, including demonstratives, which can occur in [Spec, DP], or more generally occur as anaphoric elements. Plain definiteness in contrast only conveys uniqueness, and need not be referential, allowing covariation by situation as well as predicative readings.

It seems likely that correlates of indexed definiteness should occur in every language, particularly as demonstratives have been claimed to be a universal of human language (Diessel 1999; Diessel & Coventry 2020). On the other hand, it seems unlikely that plain definiteness is a universal of human language. That is, not all languages have ways of marking uniqueness. This is particularly clear for languages where putative definiteness correspond to bare nouns, such as Russian or Mandarin. Recently, Šimík & Demian (2020) have shown that uniqueness and maximality effects are not detectable in Russian. In addition, indices might also be extended to referential uses of demonstratives in some indefinite contexts (Ionin 2006; Šimík 2016). So the class of referential expressions this paper has attempted to characterize as indexed definiteness might actually just be a slightly more general category of indexed referentiality, which would stand in opposition to a larger class of non-indexed, not-necessarily-referential noun phrases, which include plain definites and bare nouns.

Finally, returning to the theme of the introduction, referential indices do seem to play a central role in the structure and meaning of certain noun phrases contrary to the expectations of Chomsky (1995). However, the fact that indices occur only some of the time, and in particular that they seem to be absent on plain definites, indicates that the standard generative approach to indexing all potentially referential noun phrases is likely not the right model of grammar. Yet it remains to be seen what role, if any, indices may play in the traditional domain of (co-)indexation and binding.