2.1 Indices in the DP

Recent work on anaphora engages with the nature of indices, and particularly with the way in which they are formally represented in the syntax of definite descriptions. Broadly speaking, this line of research questions the idea that indices are merely subscripts on D or N, as they are often schematized in syntactic representations (following e.g., the Binding Theory of Chomsky 1981: Chapter 3, according to which each NP must be indexed by a single numerical index value at LF; see Fiengo & May 1994 for an overview).

More specifically, building on Elbourne (2005), Schwarz (2009) argues that German morphology provides evidence that indices occupy structural positions within anaphoric definites to the exclusion of their non-anaphoric (i.e., solely unique) counterparts: In German, the form of the definite article in anaphora differs from its form in non-anaphoric contexts, providing a morphological distinction that makes a cut along Heim’s (1982) notion of familiar vs. unique definites.¹ This is observed through the fact that non-anaphoric definites allow for contraction with a preceding preposition (the ‘weak’ form of the article, in Schwarz’s terminology), as in (9), while ‘familiar’ definites whose reference must be resolved via an antecedent may not participate in this contraction (10), exhibiting the so-called ‘strong’ form instead.

The crucial distinction is that in (9), ‘the mayor’ is a solely unique definite, uttered in a context in which world knowledge dictates that a given town has just one mayor (cp. Hawkins’s (1978) typology), while ‘the politician’ in (10) has already been introduced into the discourse, making it familiar.² Based on this distinction, Schwarz reaches the generalization that the definite article may not contract when a definite description is anaphoric (see also Hartmann 1978, Haberland 1985, and Cieschinger 2006).³

To capture the morphological reflex of this generalization, Schwarz proposes that the definite determiner may not contract with a preceding preposition in case an anaphoric index (notated as ‘1’) intervenes in the structure of the DP, which D selects as its specifier, as in (11) (cf. Elbourne 2005; see also Patel-Grosz & Grosz 2017; Jenks 2018).⁴ The unique form lacks this index, and contraction is therefore possible (12):

Schwarz then attributes the interpretational differences to the meanings for the definite article in (13)–(14), which differ in whether anaphoricity is hardwired in. The denotation in (13) for the unique (weak) form represents a standard Fregean/Strawsonian meaning for the determiner, and results in a DP denoting a unique individual.⁵

The anaphoric (strong) form then differs in that D must take an index as an argument (in addition to its nominal complement), as schematized in (14). This index introduces a free variable that picks out an antecedent via an assignment function according to the Pronouns and Traces rule of Heim & Kratzer (1998).

Going forward, I follow the spirit of Schwarz’s (2009) analysis, but I adopt the revised implementation in Simonenko 2014 and Hanink 2017; 2018, in which syntactic indices head their own phrase (see also Hanink 2021). Hanink (2018) argues in particular that the structure of the anaphoric DP is instead as in (15), in which idx is not a specifier, but a head within the extended projection of N, on a par with other functional layers such as NumP (Ritter 1991; 1993; see Alexiadou et al. 2007 for an overview).⁶

As argued by the above authors, one benefit of this implementation is that it allows for a single denotation for D across both anaphoric and non-anaphoric interpretations (cf. (13)–(14)), as anaphoricity within the DP is divorced from D itself and placed entirely on idx. Idx is not a semantic argument of D but instead a nominal modifier that has undergone the ident type shift (Partee 1986) from type e to the property meaning in (16) (as also in Elbourne 2005, though in a different syntactic position).

This modifier of type ⟨e, t⟩ then undergoes Predicate Modification (Heim & Kratzer 1998) with the NP, and we arrive at precisely the DP meaning proposed by Schwarz (2009), though in this case by different means:

In order to more explicitly explain the contraction contrast, Hanink (2018) adopts an analysis couched in Distributed Morphology (Halle & Marantz 1993). Hanink argues that P-D contraction is the result of postsyntactic Lowering (Embick & Noyer 2001: 561), a type of merger that applies hierarchically to head-complement sequences, after all syntactic operations, but prior to linearization. This process is schematized in (18).

Similar to variation in head movement, whether or not a head triggers Lowering is specified on a head-by-head basis in a given language. For instance, Embick & Noyer (2001) analyze “affix hopping” in English tense marking as a case of T lowering to v (20):

Adopting this idea, Hanink (2018) argues that P-to-D lowering is a specified process in German, resulting in contraction in the general case, as schematized in (20):⁷

Crucially, the structure of anaphoric DPs is more complex than (20) in that it includes idx. In this case, Hanink (2018) argues that idx itself triggers lowering of D. Adopting a bottom-up, cyclic approach to postsyntactic operations (e.g., Embick 2010), D-to-idx lowering (21) precedes P-to-D lowering. This then bleeds the environment for contraction, as D, itself having lowered, is no longer present within the relevant domain to trigger lowering of P (22). (This particular approach will be further motivated in Section 5.1, which evaluates the behavior of same in contraction environments.)⁸

Taken together, the accounts reviewed in this section offer semantic as well as morphosyntactic evidence that indices occupy structural positions in anaphoric DPs, a proposal that has been further supported by cross-linguistic work revealing similar effects (Akan, Arkoh & Matthewson 2013, cf. Bombi 2018; ASL, Schlenker 2017; Cantonese, Jenks 2018; see Schwarz 2019 for a recent overview). Going forward, I adopt the proposal laid out above that the head of idxP is the locus of anaphoricity in the DP, and that P-D contraction is contingent on whether D-to-idx Lowering is able to apply first.

2.2 Anaphoric modification in the DP

We now turn to the relationship between structurally-encoded indices and anaphoric modifiers within the DP. While there has been a good deal of work on anaphoric modifiers such as same and different (i.a. Alrenga 2007; 2009; Dotlačil 2010; Brasoveanu 2011; Bumford & Barker 2013; Charnavel 2015; Hardt & Mikkelsen 2015; see also Dowty 1985; Carlson 1987; Moltmann 1992; Beck 2000), these works focus by and large on various aspects of interpretation, leaving structural questions unaddressed (though see Oxford 2010). As a testing ground for investigation, I focus on the modifier same, as in (23):

On what can be taken as the null hypothesis in the current view, same is a modifier within an idxP-hosting DP structure, just as in any unmodified anaphoric definite (e.g., (15)). If this is correct, then the morphosyntactic behaviors of anaphoric DPs with and without same should be identical. I argue however that this is not the case: Instead, same has a distinct syntax and encodes anaphora in its own way. On this view, we expect to see morphosyntactic differences between definite descriptions with and without same. I briefly introduce data supporting this prediction below before offering an analysis that accounts for the observed differences in Section 3.

Our starting clue to the structural difference between anaphoric definites with and without same comes from an apparent anomaly in P-D contraction behavior in German. Recall Schwarz’s generalization that the definite article may not contract with a preceding preposition in anaphora. As Schwarz points out however, anaphoric same appears to contradict his generalization, a puzzle that he leaves for future work.⁹ The puzzle is exemplified in (24)–(25): While the definite descriptions are anaphoric in these examples, the contracted form of the article is nevertheless preferred.¹⁰ (N.B. der/die/dasselb- is written as a single orthographic word):¹¹

These data constitute evidence against the null hypothesis: If same appeared in an otherwise unchanged DP structure, contraction should be blocked: idx should trigger lowering of D to idx (see (21)) on Hanink’s account (or 1 would intervene between P and D as in (11), on Schwarz’s account), thus bleeding the environment for P-D contraction. This prediction is crucially not borne out, and we therefore find evidence that distinct structures underlie anaphoric definites with and without same.

Beyond P-D contraction in German, another more general and well-established clue to this proposed structural distinctness is that only definite descriptions with same license an as-relative (Carlson 1987) as in (26a) (cf. 26b), a characteristic that has long been attributed to same’s selectional properties (Heim 1985).

While this is by no means a new observation, I argue in what follows that understanding the ability of same to introduce an as-relative can shed light both on the structure in which it occurs as well as the way in which it gives rise to an anaphoric interpretation.

Looking forward, the core of the analysis relates the availability of both anaphora and as-relatives to the selectional properties of equative heads more broadly, in a way that can explain the seemingly unrelated puzzle from German contraction. As previewed in Section 1, I contend that, unlike anaphoric definites without same, anaphoric definites with same have an underlying syntax in which same is an equative head that selects for idx as an implicit argument (Baker et al. 1989, see Williams 2015 Chapter 5 for overviews). Crucially, I propose, building on Matushansky (2010a; b) and Charnavel (2015), that idx and the as-relative compete for the same position ((27)–(28)) – as the complement of Deg – explaining why they alternate with one another.

3.1 Degree equatives

Beginning with degree equatives, the examples in (31) demonstrate that as alternates between an anaphoric reading and one in which an as-relative is present instead. While anaphoric uses of as are not well-studied in the literature (though see Umbach 2007 and Hohaus 2015), I propose that they can be explained along the lines of the proposal for anaphoric comparatives made by Alrenga et al. (2012) and Hohaus (2015) (see also Beck et al. (2012) on so-called ‘contextual comparatives’). In particular, Alrenga et al. demonstrate that comparatives alternate between an anaphoric use and an overt standard of comparison in the form of a than-clause:

Alrenga et al. account for this contrast by proposing that an implicit degree argument competes for the same syntactic slot as the than-clause, that is, in addition to a gradable adjective, Deg selects either for an overt standard, or for an implicit degree argument, giving rise to an anaphoric interpretation. The parallel to equatives is then as follows: While in (34) the as-relative provides the overt standard of equation, as on its anaphoric use selects for an implicit degree argument (35) instead.¹⁵

Both the implicit degree argument and the as-relative are of the same type – d – making sense of this syntactic alternation. Adopting the view that gradable predicates denote relations between degrees and individuals (Creswell 1976; von Stechow 1984; Heim 1985; Kennedy 1999), a standard semantics for degree equatives (with an overt as-relative) is as in (36) (Horn 1972; Seuren 1984; von Stechow 1984; Schwarzschild & Wilkinson 2002). As imposes a weak linear ordering between maximal degrees, in this case contributed by the matrix and embedded clauses, respectively:

In anaphoric equatives, where an as-relative is lacking (35), the standard is instead supplied by the implicit degree argument (see Umbach 2007 for a similar, but slightly different semantic analysis of the German counterpart to as, ‘so’):

3.2 Kind equatives

Beyond degree equatives, a similar state of affairs has been observed moreover by Anderson & Morzycki (2015) for kind equatives, who treat such as a degree head in its own right. Building on Landman & Morzycki 2003, Anderson & Morzycki argue for a view of degrees according to which they are sets of (Davidsonian) state-kinds (cf. Parsons 1970; Baglini 2015; Wellwood 2015; 2019), with one key result being that such is analyzed syntactically as a degree head.

These authors present evidence for the unification of degrees and kinds from the fact that such has a degree use in addition to its better-studied kind use. For instance, (38a) shows such as a degree exclamative (Bolinger 1972; Bresnan 1973; Carlson 1977a); (38b) shows such on an extent reading (Meier 2003; Castroviejo-Miró 2011).¹⁶

Independently from Alrenga et al. (2012), Anderson & Morzycki (2015) account for this alternation with the now familiar proposal that such selects either for an overt standard of equation in the form of an as-relative (39), or an implicit kind variable instead, giving rise to an anaphoric interpretation (40) (see also Carlson (1977b), who argues that the as-relative itself supplies the anaphoric relation):¹⁷

The semantics that Anderson & Morzycki propose is also such that both complements of such are of the same type – k. Such itself has the meaning in (41), which takes a kind k as its first argument (where ^∪ is Chierchia’s (1985) up operator, shifting kinds to properties) and returns a property of individuals instantiating that kind.

In cases where the standard is provided by an as-relative (39), the embedded clause is property-denoting (following λ-abstraction) (42a), but then undergoes the existential closure type shift (Partee 1986) (42b), as previously proposed for free relatives by Caponigro (2003; 2004).¹⁸

The embedded clause then undergoes QR, rendering the meaning of the matrix clause that in (43), with abstraction over a kind variable of type k:

The property in (43) then composes with (42b), resulting in (44), which states that the kind of dog that Floyd instantiates is also instantiated by the dog that barked:

In cases of anaphora on the other hand (e.g., (32a)), such selects syntactically not for a CP, but for an implicit kind argument instead, on a par with anaphoric comparatives à la Alrenga et al. 2012 and the current proposal for degree equatives. The resulting meaning may then undergo another type-shift depending on its argument position, again resulting in a similar derivation in both cases.¹⁹

3.3 Interim summary

Taken together, a parallel emerges between same, as, and such on the view that they are all equative degree heads that show a common selectional alternation: one between an implicit argument and an as-relative. Same makes a natural addition to the paradigm: while as equates degrees and such equates kinds, same equates individuals. The proposal that all of these morphemes take as-relatives to establish identity meanings is moreover in line with Bacskai Atkari’s (2016) observation that the general purpose of as-relatives is to establish identity, regardless of the type of matrix material they compose with.

4.1 Previous work on same as Deg

The claim that same is a degree head goes back to Heim (1985), who notes that degree constructions do not only target positions on a dense scale, but can also target binary values on a scale of [0,1], i.e., the scale of (non-)identity. More recently, Alrenga (2007) has argued that same is a degree head based on its gradable behavior in what he calls similarity comparatives. Alrenga shows that the gradability of same is observed in that both degree equatives and similarity comparatives may be modified by the same set of closed-scale approximating modifiers (Huddleston & Pullum 2002), as in (46a-b):

To capture this distributional equivalence, Alrenga argues that – beyond its identity reading – same may make reference to degrees on a measurement scale of similarity, where same encodes a relation expressing the grade of similarity between two objects. This is what unites similarity and degree equatives, explaining why both are able to occur with the set of modifiers in (46). Crucially, on Alrenga’s (2007) account, same is not an adjective, but a degree expression on a par with matrix as in degree equatives.

As Alrenga points out however, the modifiers in (46) are not possible in identity cases where the equative establishes a relation between individuals, rather than degrees, as illustrated with their badness in (47) where token identity holds between dogs:²⁰

On this basis, Alrenga draws the conclusion that same is polysemous between a gradable and non-gradable use; similarity same encodes an ordering between degrees in a fashion similar to as in degree equatives, while identity same encodes identity between individuals. In both cases, the purpose of this head is to express the absence of dissimilarity. Notably however, beyond the modifiers that Alrenga discusses in (46), we find a similarity between degree equatives and even non-gradable same. For instance, both may be modified by exactly, which in Lasersohn’s (1999) terms effectuates an identity mapping:

I therefore take Alrenga’s work one step further and argue that individual-equating same is also syntactically an equative head, not an adjective.²¹ While Oxford (2010) reaches the same conclusion based on same’s ability to take an as-relative, I show that such cases can be unified with the anaphoric use of same on the present approach.

Before moving on, I clarify that I am using the term degree head purely syntactically, making reference to the shared selectional properties of this head across equatives and comparatives. Another possible label for DegP in this case is for example ‘EquatP’, proposed by Brandner (2016) and discussed by Bacskai-Atkari (2016). I note that while I do not assign identity-same a degree semantics here (see below), it might well be the case that this head may be analyzed with reference to degrees, related to its gradable use (see e.g., Alrenga 2007: Chapter 4 for discussion).²² I must leave this matter aside, focusing instead on the syntactic properties of degree heads related to their selectional behavior.

4.2 Basic proposal

If same has the same equative degree syntax as both as and such, it immediately explains why same shows the same selectional alternation between an implicit argument and an overt standard as those heads do. I therefore propose that identity same may select either for an overt standard of equation in the form of an as-relative (49), or for the implicit individual argument idx (50), resulting in anaphora in a related but distinct way from definite descriptions without this modifier.

Turning to interpretation, while the precise meaning of same has received a good deal of attention in recent literature (Alrenga 2006; 2007; 2009; Barker 2007; Charnavel 2010; 2015; Hardt et al. 2012; Hardt & Mikkelsen 2015), going forward, I adopt the modest proposal that it relates two individuals in a relationship of equation (Heim 1985; Alrenga 2009; Matushansky 2010a; b):²³

Adopting this meaning along with the structure in (49), composition with an as-relative proceeds as follows. (I return to a detailed analysis of the internal make-up as-relatives in Section 6.) The as-relative and implicit variable are both individuals of type e; while this is a straightforward enough meaning for the implicit variable, I treat as-relatives as properties of individuals that undergo the iota type-shift (Partee 1986) to achieve the individual meaning, on a par with the treatment of definite free relatives proposed by Caponigro (2003; 2004):

The as-relative is then of the right type to saturate the first argument of same, as in (53):

The resulting DegP composes with the matrix NP recipe through Predicate Modification, before composing with the definite article (cp. the structure in (49)):

On its anaphoric use on the other hand (e.g., (50)), same selects instead for idx as its first argument, resulting in what is essentially the property of being anaphoric, as in (55):²⁴

With this meaning for DegP in place, the interpretation of same on its anaphoric use is then as in (56), once the property denoted by DegP undergoes Predicate Modification with the property denoted by the noun it modifies (cp. the structure in (50)):

The result is a definite description containing a free variable that may refer back to an antecedent, much in the same way as described for unmodified anaphoric definite descriptions in Section 2.1. Crucially, while the result is the same – (56) has the same meaning as individual anaphora without same (see also Brasoveanu 2011) – the underlying structures are distinct.²⁵ Again, while this is most likely a simplification of the semantic contribution of same, it derives two distinct structural paths to anaphora, supporting the idea that indices may enter the derivation in different ways.²⁶

4.3 A comparison to related proposals

Before moving on, I offer a brief comparison to previous analyses that the present proposal builds on, found in Matushansky 2010a; b and Charnavel 2015. The crucial similarity with these proposals is that they likewise aim to capture the fact that same is compatible with either an anaphoric interpretation or a comparative-type clause instead.

First, Matushansky (2010a; b) offers an analysis of same very similar to the structures proposed in (49)–(50). In particular, she argues that same’s internal argument is either a contextually supplied antecedent (on its anaphoric reading), or a CP (in cases with an overt standard):

Crucially, as shown in (57), Matushansky assigns same the category of an adjective and argues that it is the lexicalization of Partee’s (1986) ident type-shift (with the same semantics adopted in the previous section, from Heim 1985). My proposal therefore differs from Matushansky’s in the treatment of same as the equative head of an NP-modifying DegP, which offers a unified analysis with other equative degree heads such as as and such.²⁷ (I return to Matushansky’s (2010a) analysis of as-relatives and empirical challenges for it in Section 6).

Second, Charnavel 2015 proposes a related account of same and particularly different in French according to which they are relational adjectives (see also Safir 1996). The meaning she proposes for même ‘same’ is shown in (58):

Similarly, Charnavel proposes that same may select a ‘comparative clause’ (introduced by que in French) or occur with what she calls an implicit DP. While Charnavel does not offer a detailed syntactic analysis, she writes that anaphoric same constructions involve an implicit argument x, in which this x is underspecified and interpreted pragmatically as an anaphoric pronoun (see also Dowty 1985), as in (59) (Charnavel 2015: 151). (Charnavel is not explicit about the way in which same composes with an as-relative.)

In sum, while the present accounts builds on these proposals, it diverges specifically in its explicit treatment of same as an equative head (building here instead on Heim 1985; Alrenga 2007; 2009; Oxford 2010) rather than an adjective, in a way that makes sense of the relationship between identity-same and similarity comparatives, as well as the morphosyntactic differences between same and adjectival modifiers (see Section 5). The proposal also offers an explicit representation of same’s implicit argument as idx, and unifies the selectional properties of same with those of other equative heads.

5.1 The German contraction facts revisited

The present proposal contends that anaphoric definites with and without same are structurally distinct, differing crucially in the position in which idx is introduced. Naturally, this claim leads to the prediction that we should find morphosyntactic reflexes of this difference. Section 2.2 introduced two morphosyntactic behaviors of same that appeared to support this prediction: i) the anomalous behavior of same in German contraction environments, and ii) the availability of as-relatives only in definite descriptions with same. Having argued in Section 4 that the selectional properties of same explain the alternation between anaphora and as-relatives, I now return to the contraction behavior of same in German and show how the present account accounts for this unsolved puzzle.

The anomaly presented in Section 2.2 is that P-D contraction is licensed in anaphoric environments just in case same is present, shown in the repeated example in (60). This is unexpected, as the presence of idx in anaphoric definites otherwise blocks contraction.

Before presenting the analysis, I briefly rule out two alternative explanations for this behavior. The first is the idea that the presence of any modifier licenses contraction, and that this behavior is therefore not particular to same. This is crucially not the case however, as seen for example in (61), in which same is replaced by the adjective blue. In this case, the definite article may not contract – much in accordance with Schwarz’s generalization:

Another possibility is that contraction is preferred because definites with same are not in fact anaphoric, but are unique uses of the definite article that are independently licensed in anaphoric contexts. Evidence against this idea comes from the observation that same may appear in covarying contexts, argued to be the sine qua non environment for diagnosing anaphoricity in languages with different definiteness-marking strategies (Schwarz 2009; Jenks 2018).²⁸ A covarying context is shown in (62) (adapted from a similar example without same in Schwarz 2009: 242): (62a) shows that, in definites without same, the strong article is preferred in such contexts, consistent with the idea that they involve anaphora. (62b) shows two things: First, same is likewise felicitous in anaphoric covarying contexts, and second, that contraction is observed in this environment as well (as in non-covarying contexts).

From the above data we can conclude not only that same is exceptional in inhibiting contraction, but also that definites with same in cross-sentential anaphora are truly anaphoric. The emerging puzzle is therefore that P-D contraction is licensed in anaphoric definites just in case same is present, which is unexplained if same is simply an adjective, like blue, that co-occurs with the anaphoric index as on Schwarz’s (2009) or Hanink’s (2018) account. These contraction data are therefore a crucial clue that anaphoric definites with same involve a different underlying structure from those without this modifier.

On the current proposal, we can now make sense of this contraction anomaly as a result of the fact that DegP is present in the structure. One crucial characteristic of Lowering is that it may not result in the displacement of a head into an adjunct: A head X⁰ may only lower to Y⁰ iff Y⁰ is the head of X⁰’s complement (Embick & Noyer 2001: 561, see also Bobaljik 1994). For example, returning to the case of affix hopping in English from Embick & Noyer, an adjunct such as loudly is not a suitable host for Lowering:

In the case of anaphoric same, idx is housed in an adjunct (DegP) and is likewise not a suitable host for lowering. The correct environment for D-to-idx Lowering is therefore not met (64), allowing P-to-D lowering to go ahead, resulting in P-D contraction.

Before moving on, I note that nothing in the present account rules out the presence of idxP in the extended projection of N (cp. (15)), such that idx is introduced in two positions in the structure (as the complement of Deg, and in its typical position in anaphoric definites). This state of affairs predicts that non-contraction should in fact be possible in anaphoric definites with same, resulting in a structure containing two indices. To this end, it is worth pointing out that non-contraction in this case is not categorically ungrammatical.²⁹ Consider for example (66), in which im selben is replaced with in demselben.

One way of interpreting this is that the presence of two coreferential indices is redundant and therefore pragmatically dispreferred (see also footnote 21), though I leave the precise discourse conditions for each use to future work.

In summary, a more detailed look at German supports the proposal that definite descriptions with and without same are built from different underlying structures whose morphosyntactic differences are conditioned by the part of the structure in which idx is introduced; this proposed difference is consistent moreover with the ability of same to introduce an as-relative. In the next subsection, I offer a comparison between same and relational adjectives that have been argued to be anaphoric. I show that such terms pattern differently from same with respect to how their anaphoric interpretation comes about, offering further support for the idea that anaphora with same is syntactically encoded, and that same does not pattern with adjectives in this regard.

5.2 Same, bridging, and relational terms

I argued above that the lack of P-D contraction with same is explained if idx in such cases is selected by Deg, rather than by D as in simple anaphoric definites. One question that emerges from this claim is how same compares to other relational terms that have likewise been argued to involve an implicit argument (Mitchell 1986; Partee 1989; see also Condoravdi & Gawron 1996, a.o.). For example, relational adjectives such as local are known to have a discourse anaphoric use, as in (67):

While Partee (1989) points out that the presence of a null pronoun might be responsible for the anaphoric interpretation in such cases (ultimately rejecting this approach, cf. Bhatt & Pancheva 2006), another possibility is that local imposes restrictions on the context in a more global sense, a view adopted by Condoravdi & Gawron 1996. Condoravdi & Gawron argue that implicit arguments in relational terms are interpreted along the lines of definite descriptions in Heim’s (1982) sense, which is to say that they impose particular familiarity conditions on the context in which they are uttered. On this view, the use of local requires some kind of accommodation (see Condoravdi & Gawron 1996: 15), in this case of a new variable that is assigned to a particular location relative to the context, even if that location has not been explicitly introduced by an antecedent.

Importantly, Condoravdi & Gawron (1996) do not argue for a syntactic representation of this variable. Relevant for our purposes is that Condoravdi & Gawron draw in their analysis a connection between relational terms such as local and definite descriptions that may also be accommodated without an explicit antecedent, as in (68):

The use of the definite the engine in (68) constitutes a case of bridging (or associative anaphora, Clark 1975; Hawkins 1978). The availability of local in similar contexts (in that no explicit location is introduced as an antecedent in (67a)) leads Condoravdi & Gawron (1996) to conclude that implicit arguments behave like bridged definite descriptions, and not like pronouns, as the latter require an overt antecedent to be used anaphorically.

Importantly, Schwarz (2009) investigates bridging contexts with unique and anaphoric definites in German, and shows that not all bridging contexts are alike. Instead, different types of bridging relations result in different forms of the definite article. In particular, Schwarz shows a distinction between what he calls producer-product bridging and part-whole bridging. In the former, the anaphoric form of the article is used (69), while in the latter, the non-anaphoric form is (70):

Schwarz argues that cases of producer-product bridging are special in that they involve two-place relational terms such as author, whose meanings necessarily encode an internal argument as in (71), differentiating them from one-place nouns such as crisper:

Crucially, as Schwarz points out, the internal argument of author in an example such as (69) can then be understood as the index, schematized in (72) within the specific implementation of the present proposal. The presence of idx in the structure blocks P-D contraction, resulting in the strong form of the article just as in non-bridged anaphoric definites, explaining why producer-product bridging makes use of the strong form.

This structure does not obtain for part-whole bridging on the other hand, as such cases do not involve relational nouns. Schwarz (2009) therefore proposes that, unlike cases of producer-product bridging, part-whole bridging definites are licensed by the same type of situational uniqueness as other weak form definites. I take this result to suggest that certain dependencies are structurally encoded, while others come about via domain restriction instead (though I do not comment on how the latter is achieved).

While Schwarz does not discuss the German data or offer a concrete analysis, he suggests that terms such as local be treated as part-whole relationships (p. 226), interpreted via situational uniqueness (cp. Condoravdi & Gawron’s (1996) accommodation). This suggestion is consistent with the observation that, with local, the weak form of the article is preferred:

I submit that while both same and anaphoric adjectives like local involve the weak form of the article, the structural conditions in each case are distinct, that is, contraction results for different reasons. In the case of same, anaphora is structurally encoded, but the locus of idx differs from that in unmodified anaphoric definites in such a way that contraction is blocked (Section 5.1). In the case of local however, there is no implicit variable in the structure, and the context-dependent relation is achieved via more global means (cf. Martí 2003); P-D contraction applies in this case as the default, as idx is not present in the structure.

Supporting this view, we in fact find evidence for the structural nature of anaphora with same from further investigation of data from bridging. The novel observation here is that same is felicitous in the same contexts in which the strong, anaphoric form of the article is, namely in producer-product bridging (74), but not in part-whole bridging (75):

The same is true for covarying uses. Schwarz reports that the contrast observed in contraction also holds in cases of covariance (from Schwarz 2009: 63), such that contraction is still dispreferred with producer-product pairs (76a), but preferred with part-whole pairs (76b):

I show with the contrast in (77) that, in cases of covariance, same is similarly licensed only with producer-product – not part-whole – pairs:

What these data reveal is that same, just like the anaphoric form of the definite article, requires the presence of idx, explaining why it patterns with the strong, anaphoric form. It is not available in part-whole bridging, in which anaphoricity is not the result of extra structure, but accommodation of some kind.

In sum, while the current contention is that anaphora has a structural source in anaphoric definites, the context-dependency of non-anaphoric definites in certain bridging contexts, along with that of adjectival, relational terms such as local, is derived differently, pointing to structural differences between these superficially similar types of expressions. While the full range of behaviors of anaphoric modifiers in P-D contraction environments is not addressed here, it poses an interesting avenue for future work.

5.3 Interim summary

The treatment of same as an equative head unifies several seemingly unrelated characteristics. Specifically, the present account unifies the selectional alternation between an implicit argument and an as-relative with the selectional requirements of other equative heads, simultaneously offering a solution to the initially puzzling data from German contraction. Evidence from relational terms offers further support that anaphora with same is encoded in the structure of the DP, though in a different way from anaphoric definites without same. Shifting gears, I turn next to the constituents that compete for the same syntactic slot as idx: as-relatives selected by same.

6.1 Overview of as-relatives

As-relatives have received considerably less attention (Carlson 1977b; Stowell 1987; Potts 2002; Matushansky 2010a; b; Lee-Goldman 2012) than relative clauses of other types (see Bianchi 2002; Grosu 2002; 2012; de Vries 2002 for overviews). For instance, in most existing literature on as-relatives introduced by same (e.g. Alrenga 2007; Charnavel 2015), no explicit syntactic analysis of the as-relative or its association with the matrix clause is provided. An exception to this is Matushansky (2010a), whose account informs the present proposal (though see Section 6.4 for an empirical challenge). As-relatives introduced by same are nevertheless on a par with those found in a variety of equative constructions (78), calling for a comparison and an explicit analysis:³⁰

Previous works largely propose a semantics of as-relatives similar to that of other relative clauses. For example, Rett (2013) argues that degree equatives are property-denoting, and that as performs the function of a wh-operator. More specifically, Rett builds on Stowell (1987) and argues that the gap in the as-relative is simply a variable that remains unbound. In her analysis, the role of as is to serve as an unselective binder, binding this free variable and thereby turning the as-relative into a property-denoting expression (see also Potts 2002; Lee-Goldman 2012).

While I treat as-relatives as a type of relative clause, any analysis of as-relatives without Ā-movement faces a major challenge: A lack of movement would predict a lack of island sensitivity. However, as-relatives introduced by same are in fact subject to a variety of island effects (Ross 1967). These facts are discussed in Landman 2006 for kind equatives, and for same-equatives in Matushansky 2010b and at greater length for French in Charnavel 2015. Some examples of island violations are shown below:

Moving forward, my analysis therefore reflects Ā-movement (Chomsky 1977) of the relativized argument (as in Landman 2006; Matushansky 2010b; Charnavel 2015).

Another relevant question concerns the category of as. Some authors treat as as a preposition (e.g., Emonds 1985; Potts 2002; Rett 2013), while others treat it as a realization of C (e.g., Matushansky 2010b; Lee-Goldman 2012; Charnavel 2015). Going forward, as is represented as C, but nothing crucial hinges on this assumption.³¹

6.2 The nature of the gap

One immediate question concerning as-relatives is the nature of the gap. For instance, Landman (2006) argues that the gap in such-as (kind) equatives consists not in a definite variable as in restrictive relatives, but rather in a noun phrase which she terms a k-like NP. Evidence for this is the fact that the as-relative can contain an existential there-construction:

Landman therefore takes the gap to be a ‘k-like NP’ rather than a definite variable, as definites are barred from participating in such constructions according to the Definiteness Restriction in (81) (Milsark 1974):

As Landman points out, such-as relatives differ from restrictive relatives in this regard, whose gaps may not be definite arguments (Carlson 1977b; Heim 1987):

Importantly, existential there-constructions are likewise not licensed in as-relatives embedded by same, as shown in (83), which pattern with restrictive relatives in this regard:

Note that Landman’s proposal differs also from treatments of degree equatives involving gradable adjectives, which largely assume that the gap contains a deleted adjective in addition to a degree variable, as seen in (84) (modified from Rett 2013: 1109):

In many analyses of degree equatives, the embedded adjective is deleted after moving to Spec, CP (Bresnan 1973), as described in the following formulation by (Kennedy 2002b: 556) (see also Kennedy 1999 and Kennedy & Merchant 2000):

[Comparative deletion] involves overt movement of the compared constituent to the specifier of a clausal complement of than/as, plus deletion under identity with the head of the comparative.

In comparatives, an adjective must be present in the embedded clause for semantic reasons: The degree morphemes more/-er and as impose an ordering on two sets of degrees, and a gradable adjective is required to specify the scale in both sets.

Generally speaking, evidence for comparative deletion in the domain of degree expressions comes from the existence of comparative sub-deletion (Bresnan 1973; 1975), in which some (but not the entire) constituent is missing from the surface presentation of the complement of an equative or comparative. We observe this sub-deletion in degree equatives as well (Bresnan 1975; Kennedy & Merchant 2000):

Comparative sub-deletion is licensed when the two adjectives denote different properties (e.g., unusual and noble in (85)). Importantly, in the case of as-relatives embedded by same, sub-deletion is also not (generally) available, as in (86):³²

As the two objects being equated are necessarily the same individual, the general contrastive conditions licensing sub-deletion are not met; the man that was seen in (86) can never be identical to the woman that was seen. These facts lead to the conclusion that same is not covertly present in the gap in the way that a gradable adjective is in degree equatives or comparatives – following on this analysis from the fact that it is not an adjective, but itself a degree head, and that individuals are being equated, rather than maximal degrees. Again here, as-relatives embedded by same pattern with restrictive relatives in this regard.

6.3 As-relatives as matching relatives

Preserving an observed similarity to comparative deletion, I propose that as-relatives selected by same involve a matching analysis along the lines of that argued for restrictive relative clauses by Sauerland (1998; 2003) and Hulsey & Sauerland (2006):

In (87), the external and internal heads of the relative clause are not derivationally related, but must be identical in order for the lower head to be deleted. Sauerland (2003) draws a comparison between this type of deletion and the type of obligatory deletion observed in comparatives, proposing that relative clauses involve so-called relative deletion of their internal head:³³

A major piece of evidence for the matching analysis comes from the lack of a Condition C violation in relative clauses, as shown in (89) (Munn 1994: 402):

The lack of a Condition C violation between matrix John and he in (89) can be explained if John need not reconstruct inside the embedded clause, which holds under the matching analysis: In Sauerland’s (2003) account, the matching requirement imposed on the internal and external heads can be satisfied if the internal head is itself a pronoun (rather than a proper name), which would not incur a Condition C violation (90a). Contrarily, on a raising analysis of relative clauses (90b) (Brame 1968; Schachter 1973; Vergnaud 1974; Kayne 1994; Bhatt 2002), according to which the internal head undergoes an additional movement out of the relative clause into a matrix position, the possibility of reconstruction is unavoidable.

Notably, the lack of Condition C effects is paralleled in as-relatives embedded by same:

Taking these effects as well as the observed similarity with comparative deletion and degree expressions more generally into consideration, I propose a matching analysis of as-relatives introduced by same.³⁴ The derivation of the as-relative is as follows. First, the internal head noun ‘recipe’, selected by Op, undergoes Ā-movement to Spec, CP:

(92)	a.	Today I used the same pancake recipe [as-rc as I used yesterday ].
	b.

Here I follow Fox (2002) by proposing that the lower copy of recipe undergoes Trace Conversion (adopted also in Hulsey & Sauerland 2006), as schematized in (93):

The composition of the clause embedded by as is then as in (94) following trace conversion and λ-abstraction triggered by movement (Heim & Kratzer 1998), assuming that the relative operator does not make a semantic contribution:³⁵

As discussed in Section 4, in order for the embedded clause to compose with same (95) (repeated from (51)), it needs to denote an individual, rather than a property.

I therefore propose that the meaning of the as-clause in (94) undergoes the iota type-shift (Partee 1986), on a par with the proposal for definite free relatives by Caponigro (2003; 2004) and for degree equatives by Anderson & Morzycki (2015) (cp. Charnavel (2015), and cf. Matushansky (2010a), who makes use of a max-operator instead; her account also differs in several respects, and she is not explicit about adopting either a matching or raising analysis).

The meaning of the embedded clause is now of the correct type to saturate the first argument of same, and composition with the matrix NP proceeds as in (97):³⁶

(97)

The resulting meaning of the matrix DP containing the as-relative is then the unique recipe that is identical to the unique recipe that the speaker used, as in (98):

6.4 As-relatives vs. that-relatives

In this final subsection, I focus on the generalization that, like as-relatives, that-relatives license the presence of same, even though there is no antecedent in such contexts.

There are in principle two possible sources for that-relatives with same. The first is that same is simply an optional modifier in relative clauses, and that the relative clause adjoins to the NP in the typical way. In this approach, same selects for idx in the same way it does in anaphora, since no as-relative is present to saturate its first argument:

This approach however faces a problem, given the anaphoric nature of same when it does not select for a relative clause: If same first composes with the idx and then the relative clause, we will arrive at a meaning for the entire DP as in (101), which results in an anaphoric meaning for a DP that is not necessarily anaphoric.

As with as-relatives, the puzzle is again however that same is licensed in relative clauses even when no antecedent is present, and thus anaphoricity should not be built in.³⁷

The second option, proposed by Matushansky (2010b), is that same selects for that-relatives in the same way it does for as-relatives. I adopt this option here and provide novel empirical evidence in support of it. On the present account, this means that the embedded CP is selected for by Deg, as in (102) (and undergoes extraposition just like the as-relative does). On this view, the relative clause is not a nominal modifier as normal restrictive relatives are, but is instead an argument of Deg (102), just like an as-relative is.³⁸ This precludes the selection of idx and avoids an unnecessary anaphoric interpretation.

Because the that-relative, like the as-relative, is property denoting, it saturates the first argument of same in exactly the same way that the as-relative does after undergoing an iota type shift. In the present proposal, the meaning of (102) is then as in (103):

Note the parallel here to anaphora with and without same. Restrictive relative clauses with and without same appear to be almost identical, differing only in the presence of a modifier, they in fact arise from different underlying structures; while it may seem that same is simply an optional modifier, its presence is in fact indicative of the presence of a degree structure.

Empirical support for this proposal comes from languages that make use of the same marker in equatives and relative clauses. Spanish is one such language, which uses the invariant relativizer que in both relatives with same (104) and those without (105).³⁹

Importantly, relative clauses with and without same can nevertheless be distinguished based on the deletion they allow. In case same is present, the standard may be phrasal (106). If same is omitted however, the sentence becomes ungrammatical (107).

This contrast is not expected if same is simply an adjective that modifies an NP independently from the relative clause. The grammaticality of (106) can be explained however if same is a degree head that selects for the relative clause in such cases. What we observe then in Spanish is optionality of the size of the standard, which is expected: As a result of deletion processes restricted to comparatives and equatives, phrasal standards are allowed in degree clauses, but not in relative clauses, and can be independently accounted for (with either the direct or reduction analyses in the terminology of Lechner 2004; Bhatt & Takahashi 2011, see also Merchant 2009).

In sum, by adopting the structure of relative clauses with same as in (102), we can explain the same obviation of anaphoricity that is observed in as-relatives, as well as the size of deletion effects in a language like Spanish. In cases where same is present, a degree syntax is present rather than a typical modifying restrictive relative clause.

I conclude briefly by comparing the account here to the one proposed by Matushansky (2010a), which my account builds on. A crucial component of Matushansky’s analysis is that as-relative formation involves the movement of a null operator – housed in an AP modifying the NP – to Spec CP, to the exclusion of the gapped NP, as shown in (108):

The motivation for this aspect of her analysis is to derive what she describes as obligatory Verb Phrase Ellipsis (VPE) as in (109), which is not required in restrictive relatives (110).

In (108), the movement of the operator to Spec CP to the exclusion of the NP it modifies constitutes a Left Branch violation (Ross 1967; Corver 1990). Matushansky therefore aims to derive the obligatory deletion observed in (109) through Kennedy and Merchant’s (2002) notion of salvation by deletion (Lasnik 1995): In order to rescue the violation incurred, all material following the moved operator must be deleted, in line with the proposal that ellipsis may serve as a rescue operation for PF violations. That-relatives then differ from as-relatives in the amount of material that undergoes movement: In relative clauses, the entire gapped DP moves to Spec, CP, rather than just the operator.

Matushansky’s analysis however faces an empirical problem: The generalization that as-relatives obligatorily involve Verb Phrase Ellipsis is inaccurate, as attested by native speaker judgments as well as the following naturally-occurring example (111); Matushansky’s analysis erroneously predicts obligatory VPE in all as-relatives.⁴⁰

In the present proposal, the optional ellipsis found in as-relatives is instead on a par with VPE observed in this type of antecedent-contained deletion (Sag 1976; May 1985; Larson & May 1990; Wold 1995; Merchant 2000; Fox 2002; Kennedy 2002b; Bhatt & Pancheva 2004).

To summarize, I have argued in this section for a matching analysis of as-relatives that closely relates them to that-relatives, differing only in whether the relative clause is an NP modifier, or selected by Deg inside of a DegP modifier. I have also argued that that-relatives with same arise from the same equative syntax as as-relatives, and have shown that if such relatives follow a degree syntax, we can understand why we observe deletion therein on a par with comparative clauses. Unlike English, Spanish does not differentiate morphologically between as-relatives and that-relatives introduced by same, but still offers clues that that-relatives may have two distinct structural sources. One of these sources is facilitated by an underlying degree syntax, supporting the view that same is not an adjective, but an equative head.