2.1. Comparison to agreement, clitics, and gestures

As we noted above, one prominent view of verbs locationally modified to indicate person information is that they reflect verb agreement with the subject and/or object (Fischer & Gough Reference Fischer and Gough1978, Janis Reference Janis, Emmorey and Reilly1995, Lillo-Martin & Meier Reference Lillo-Martin and Meier2011, Meir Reference Meir1998, Reference Meir2002, Padden Reference Padden1983, Pfau et al. Reference Pfau, Salzmann and Steinbach2018). Descriptively, agreement can be said to involve a systematic covariance between a semantic or formal property of one element and a formal property of another (Steele Reference Steele, Greenberg, Ferguson and Moravcsik1978). English shows verb agreement with an inflection -s when the subject of a verb is third-person singular, as in 5. Spanish shows subject agreement, with, for example, the verb hablar ‘to speak’ ending with o for a first-person singular subject and with a for a third-person singular subject, as in 6.

Under an analysis of locational modification as agreement, the modification of the initial and/or the final location within the movement of a verbal sign in ASL and other signed languages is equated to person inflections on the verb. The agreement analysis is supported by the observation that locational modification exhibits typologically common effects of verb agreement, such as argument drop and noncanonical word order. Specifically, overt modification of the verb allows word order to be noncanonical (Fischer Reference Fischer and Li1975, Quadros Reference Quadros1999) and licenses null arguments (Lillo-Martin Reference Lillo-Martin1986).

There are two alternatives to analyzing locational modification as agreement. Both are more semantically contentful than syntactic agreement, which is taken to be a reflection of features associated with syntactic positions, such as the subject or object position. The first alternative treats locational modification as personal (e.g. they/that) and/or locational (e.g. there) pronouns, which have simply been cliticized onto the verb (Koulidobrova Reference Koulidobrova2012, Nevins Reference Nevins2011b). The second alternative analyzes locational modification as gestures, that is, as incorporation of a pointing gesture into the form of the sign to signal an association with the referent, in the same way that a pointing gesture by a nonsigner would (Liddell Reference Liddell and Coulter1993, Reference Liddell2003). Specifically, Liddell (Reference Liddell, Emmorey and Reilly1995) argues that referents marked by locationally modified verbs should be seen as being present in some mental spatial representation (called ‘surrogate space’ or ‘token space’), where they are assumed to have topographic representation. The only difference between a token space and a mental representation of the real space of discourse is that the tokens are invisible. Thus, pointing in the token space should be considered to be identical to pointing to actual entities in the discourse context.

Motivations for these alternative analyses come from what Nevins (Reference Nevins2011b) calls ‘significant departures from agreement patterns’ in spoken languages (see Koulidobrova Reference Koulidobrova2012, Lillo-Martin & Meier Reference Lillo-Martin and Meier2011, Mathur & Rathmann Reference Mathur, Rathmann, Pfau, Steinbach and Woll2012, and Sandler & Lillo-Martin Reference Sandler and Lillo-Martin2006 for extensive discussion). Below, we focus in more closely on two of these differences.

2.2. Optionality and its solution

Another important difference between verb agreement in spoken languages and locational modification in sign languages is that agreement is by definition obligatory in every instance of a particular verb, while we see in sign languages the same verb sometimes appearing with locational modification and sometimes without. In other words, while Lourenço and Wilbur (Reference Lourenço and Wilbur2018) argue that all verbs in Libras ‘can show co-location’ as an argument in favor of analogizing it to agreement, we note that verbs in a verb-agreement language must show agreement, irrespective of semantic/pragmatic factors.

Large-scale corpus studies have shown that directional modification in sign languages is far from obligatory. De Beuzeville et al. (Reference de Beuzeville, Johnston and Schembri2009) report that only 41% of inflecting verbs in Auslan (Australian Sign Language) were clearly modified, even for marking undergoer arguments. Similarly, Fenlon et al. (Reference Fenlon, Schembri and Cormier2018) examine 1,436 tokens of eighty-one inflecting verb types in British Sign Language (BSL) and, even under a generous coding of locational modification, find modification in only about 65% of tokens, with stricter assumptions yielding lower rates.Footnote ⁴ They identify numerous factors that condition modification, including person, thematic role, animacy, verb position, and constructed action. In particular, constructed action—where the signer adopts the perspective of a character—correlates with higher rates of patient modification, likely because the signer deictically points to imagined referents. They also note that verbs are more often modified following null arguments, suggesting that modification may serve as a communication strategy to maintain reference and ensure transparency of meaning. Beyond such global patterns, Schembri et al. (Reference Schembri, Cormier and Fenlon2018) emphasize idiosyncrasies that are argued to be unpredictable: while pay was modified in twenty of twenty-six tokens, push was modified in only five of twelve, supporting Liddell’s (Reference Liddell, Emmorey and Lane2000) proposal that directionality functions more like constructed action or gestural addition than like obligatory agreement.

While such corpus-based work highlights optionality, it often treats modification as unpredictable. By contrast, Kocab et al. (Reference Kocab, Ahn, Lund and Davidson2019) argue for systematic pragmatic conditioning. They suggest that the higher modification rate of pay relative to push arises from differences in argument reversibility: the inanimate patients of pay (e.g. a bank) can plausibly also be agents, whereas the inanimate patients of push (e.g. a wall) cannot. Thus, pay is more likely to be overtly modified to disambiguate the agent and patient roles.

Small-scale experimental evidence from ASL corroborates this view. In one set of contrasts with the verb dance, narratives with a single referent were judged natural when the final verb appeared without modification, as in 9, accepted by all three native ASL signers consulted by Kocab et al. (Reference Kocab, Ahn, Lund and Davidson2019), but narratives with two referents were degraded unless loci were established and the verb was signed at the agent’s locus, as shown by the contrast between 10a, accepted by only one signer, and 10b.

A second set of contrasts with push yielded similar results. When no loci were established, unmodified verbs were rejected, as in 11a; when loci were marked and the verb moved between them, sentences were judged acceptable, as in 11b. Crucially, when the surrounding context rendered the intended agent salient, the same unmodified verb from 11a was judged natural and clear, as in 11c (see also Neidle et al. Reference Neidle, Kegl, Bahan, MacLaughlin and Lee2000).

These results demonstrate that modification is not random but is systematically licensed by pragmatic need: it is required when roles are reversible or referentially ambiguous, but unnecessary when discourse context suffices.

In summary, converging findings show that locational modification on verbs is optional, in sharp contrast to obligatory agreement in spoken languages. At the same time, its distribution is not arbitrary: both corpus studies and experimental data reveal systematic conditioning by pragmatic factors such as disambiguation and discourse salience. This perspective aligns indicating verbs with pronominal elements, whose overt realization is likewise governed by discourse-pragmatic constraints (Ariel Reference Ariel, Barlow and Kemmer2000, Gundel et al. Reference Gundel, Hedberg and Zacharski1993, Nevins Reference Nevins2011b), rather than with morphosyntactically obligatory agreement.

2.3. Analysis: location incorporation

Motivated by optionality facts, we favor an analysis of locational modification that involves the notion of incorporation. We define incorporation in this context as a morphological process of combining different parameters of two signs, such as handshape, location, and movement, to form a new word (Liddell Reference Liddell1997, Valli et al. Reference Valli, Lucas, Mulrooney and Villanueva2011). For example, the ASL sign for six-months involves an incorporation of two signs, six and month, as shown in Figure 5. To incorporate the two signs, the handshape of the sign six is combined with the location, movement, and orientation of the sign month. So instead of the dominant hand forming the indexical handshape generally used for month, the dominant hand takes the shape of the sign six. This incorporation is productive up to ‘nine’, so if the dominant handshape is that of seven, the incorporated word means ‘seven months’, and so on.

Figure 5.

The incorporation of the signs six and month in ASL involves the dominant handshape of six combined with the dominant hand orientation, nondominant handshape, movement, and location of month.

We argue that the locational modification of verbs is a kind of incorporation between a nominal element and the verb, aligned with the view put forward by Koulidobrova (Reference Koulidobrova2012) and Nevins (Reference Nevins2011b) and consistent with the idea in Liddell Reference Liddell, Emmorey and Reilly1995 that a locationally modified verb is composed of a grammatical verb and deixis. This view also maintains the intuition in Liddell Reference Liddell, Emmorey and Reilly1995 and Liddell Reference Liddell, Emmorey and Lane2000 that loci use is uniform across the nominal and verbal domains. The nominal element that gets incorporated into the verb is the indexical point (IX) directed to the relevant loci (such as A and B), represented as IX_A and IX_B.Footnote ⁵ To incorporate these into the verb, the start and end locations of the verb would be replaced with the location parameters of IX_A and IX_B, respectively, with the other parameters of the verb remaining. An example is shown with a locationally modified verb help in 12.

The same analysis can apply to verbs that do not have a path component. A locationally modified verb dance is analyzed in 13 as having an overt IX in the underlying semantics. Here, too, we argue that the locational parameter of IX_A is incorporated into the verb dance in the phonological production of the sentence.

Analyzing locationally modified verbs as verbs with incorporated IX has implications for the tripartite categorization of verbs in sign languages into ‘inflecting’, ‘spatial’, and ‘plain’ seen in 8 above. Because the referential expression (IX) can be incorporated into both an ‘inflecting’ verb and a ‘plain’ verb, the distinction between the two categories is removed.Footnote ⁶

Note that it is possible in ASL for the overt IX to cooccur with locationally modified verbs, as shown in 14.

This, however, does not pose a challenge to the incorporation analysis because clitic doubling—where overt referential expressions such as pronouns and names cooccur with cliticized pronouns—is commonly found across languages, and in fact is often first misanalyzed as agreement (Anagnostopoulou Reference Anagnostopoulou, Everaert and van Riemsdijk2006, Baker & Kramer Reference Baker and Kramer2018, Koulidobrova Reference Koulidobrova2012, Kramer Reference Kramer2014, Nevins Reference Nevins2011a).Footnote ⁷

2.4. Summary

In this section, we have examined how ASL verbs may be locationally modified to encode person information. While traditionally analyzed as agreement morphology, we have argued that such modification is better understood as the incorporation of a nominal element into the verb. This analysis captures the observed optionality of modification and its sensitivity to pragmatic factors, distinguishing it from obligatory agreement in spoken languages. Taken together, the data suggest that the typologically distinctive patterns of sign language verbs reduce to a general process of incorporation.

3.1. Optionality and pragmatic conditioning of IX

As with loci in the verbal domain, one crucial but often overlooked property of IX_LOC is its pragmatically conditioned optionality and its overall low frequency in natural ASL discourse. Production studies of ASL show that choices among referential expressions—null arguments, classifiers, IX, or overt nominals—reveal how discourse structure is organized (Czubek Reference Czubek2017, Frederiksen & Mayberry Reference Frederiksen and Mayberry2016). In Frederiksen & Mayberry Reference Frederiksen and Mayberry2016, twelve native signers were presented with a simple, six-panel picture depicting a story and were asked to retell the story. A number of different anaphoric expressions available in ASL—such as the null argument, classifiers, and IX—were counted in the collected corpus, focusing on those appearing in the subject position. In order to determine how referent status affects the choice of the anaphoric expression, Frederiksen and Mayberry adopt from Gullberg Reference Gullberg2006 a three-way distinction of referent status, shown in 17. These represent different statuses a referent can have in a narrative and have been shown to affect the kinds of referential expressions that spoken and sign languages use (Debreslioska et al. Reference Debreslioska, Özyürek, Gullberg and Perniss2013, Perniss & Özyürek Reference Perniss and Özyürek2015, among others).

In order to see how frequently a locus is used to refer to a familiar entity, we report Frederiksen and Mayberry’s (Reference Frederiksen and Mayberry2016) counts from the latter two categories in Table 1.

Table 1.

Kinds of anaphoric expressions (adapted from Frederiksen & Mayberry Reference Frederiksen and Mayberry2016:59, table 3).

The null argument is dominant in both categories of anaphoric reference, with around 70% of all anaphoric expressions being null. Overt nominals can be used in anaphora as well, especially when the definition of anaphora is broadened to include reintroduced referents. Within the nominal category, bare nouns are most frequent, as Table 2 shows.

Table 2.

Nominal anaphoric expressions (adapted from Frederiksen & Mayberry Reference Frederiksen and Mayberry2016:60, table 4).

Critically, IX rarely appears in this kind of production study. The prenominal and postnominal uses of IX occur only one time each in the whole set of data.

Czubek (Reference Czubek2017) replicates the investigation of referential expressions in Frederiksen & Mayberry Reference Frederiksen and Mayberry2016 and compares the simple production data to more elaborate, professionally produced ASL narratives. He notes that the more complex the narrative, the more frequent the use of IX. It is, however, not possible to determine whether the use of locus was increased systematically in more complex discourse because the count of locus use was separated into pronominal uses and adnominal uses in this study and conflated with bare nouns occurring without overt loci under the ‘definite description’ category. What is clear from these studies is that the use of IX is not especially frequent for anaphoric reference.

Elicitation data further show that IX_LOC is not only optional but also systematically conditioned by pragmatic need, paralleling findings in the verbal domain. Ahn, Kocab, and Davidson (Reference Ahn, Kocab, Davidson and Cho2019) elicited judgments from three native ASL signers and found that IX_LOC was preferred when disambiguation between potential antecedents was required, but was otherwise redundant. For example, with only one referent, both null arguments and bare nouns were judged acceptable, as in 18.Footnote ⁸

Participants also accepted the locus-marked version in 19, but considered it redundant and thus dispreferred compared to the null.

However, the preference for null arguments disappears when the context calls for disambiguation, as in 20. In 20, two referents are introduced, a boy and a girl. Two of the three signers rejected the sentences with a null argument, while all three accepted the bare noun boy.

All three signers readily accepted 21, which established the boy and the girl in different loci. Here there was no dispreference for marking the loci.

These findings suggest that IX_LOC behaves like marked referential expressions found across languages: it is licensed when disambiguation is necessary and dispreferred when the referent is unambiguous (Schlenker Reference Schlenker2005). This pattern parallels demonstratives (Ahn Reference Ahn2022, Hinterwimmer & Bosch Reference Hinterwimmer, Bosch, Patel-Grosz, Grosz and Zobel2018, Nowak Reference Nowak2019, Roberts Reference Roberts, van Deemter and Kibble2002, Wolter Reference Wolter2006), which Koulidobrova and Lillo-Martin (Reference Koulidobrova, Lillo-Martin, Grosz and Patel-Grosz2016) propose as the right comparison for IX_LOC. Irani (Reference Irani2016) compares IX_LOC to another ‘heavy’ referential expression, so-called ‘strong’ definites of the sort found in German (Schwarz Reference Schwarz2009). We argue that, minimally, IX_LOC needs to be heavier in phonological and semantic content than other available anaphoric expressions (Ariel Reference Ariel, Sanders, Schilperoord and Spooren2001), and we propose such an analysis in the next section.

3.2. Analysis: IX_LOC as a spatial modifier

We propose that indexical points (IX) to locations are predicates that locate entities in signing space, a function from <individual, location> pairs to truth values. We formalize this as an operator IX_LOC which contributes a semantic restriction that relates an entity to a location. Here, locations are considered to be semantic primitives like entities. We could formalize them as entities, but entities that correspond to specific points or areas in the physical spatial layout of the discourse context. Just as the pronoun she picks out a single entity relevant in a given domain, locational demonstratives like here pick out a single location relevant in the domain. The location does not have to be a single point and can range over a larger area, as in a sentence such as We take cards here, where here refers to a store. While we do not cover loci use that ranges over an area, we refer interested readers to works such as Padden Reference Padden and Newmeyer1988 for a more comprehensive overview of how loci are used. For now, it is sufficient to think of location variables as picking out a specific point in the spatial layout. In this respect, we follow Liddell Reference Liddell, Emmorey and Reilly1995 and assume that locations should not be part of the grammar; that is, we are not arguing that different points in the spatial layout in front of the signer are represented in language. Rather, we are able to make use of the space around us readily and to identify different points and areas of the spatial layout, and language is sensitive to these distinctions in the same way it is sensitive to distinctions in entities. We also follow Padden Reference Padden and Newmeyer1988:234, where some uses of loci are analyzed as predicates marking locations of their arguments. Extending and formalizing these intuitions, we analyze IX_A as a simple predicate taking an entity x as its argument and returning ‘true’ if and only if some relation holds between x and some location A. More generally, the function IX_LOC takes a location variable o and an entity x as arguments, and returns ‘true’ if and only if some relation holds between x and o. The lexical entry for IX_LOC is provided in 22a, and IX pointing specifically to a locus (A in this case; IX_A) is shown in 22b. R(α,β) is true if some relation R holds between α and β. We are intentionally allowing for the relation expressed by R to be one that is entirely arbitrary, somewhat motivated, or completely iconically transparent, as in direct deixis.

3.2.1. Combining with a null demonstrative

This spatial predicate then composes with a null demonstrative to form a nominal argument. Consider Figure 6 as an example, which is translated as ‘she/he dances’ or ‘they(singular) dance’. Because in sign languages we see the indexical point used so frequently where we expect to see pronominal expressions, it is often taken to be equated to a pronoun or a demonstrative such as she or this person in English. However, we argue that all we see on the surface is the spatial predicate that accompanies a demonstrative.

Figure 6.

‘They(singular)_A dance’ (ASL).

We capitalize on the strong similarity between indexical pointing in both spoken and sign languages to motivate our analysis of the point to a location not as a nominal argument itself, but as a descriptive modifier that can accompany a (null) nominal head and a determiner to form a nominal argument, following the proposal in Ahn Reference Ahn2020. The idea is that in the underlying semantic representation of Figure 6, there is a covert demonstrative that takes IX_LOC along with some available grammatical restriction, such as an NP, and returns an entity.

This is analogous to the composition of indexical pointing with a demonstrative description in English, as in 23. In 23, ‘+’ in ‘that+IX_A’ represents a co-speech pointing to A that cooccurs with the utterance of ‘that’. Demonstratives are often analyzed as taking two arguments and returning the unique entity that meets both restrictions (Elbourne Reference Elbourne2008, King Reference King2001, Robinson Reference Robinson2005). We specifically follow Ahn Reference Ahn2022 in assuming that the demonstrative takes a spatial modifier like pointing and a nominal restriction and returns the unique entity that meets both properties. This is shown in 24. We represent it in our semantics as ι taking two restrictions, with ι giving rise to a presupposition that there exists a unique entity that meets the two restrictions.

(23)

⟦[that+IXA] girl⟧ = ιx. girl(x) & R(x,a)

(English)

(24)	⟦that⟧ = λR. λP. ιx. P(x) & R(x)

We further assume that the nominal restriction slot can contain not just full NPs but also ϕ-features (Ahn Reference Ahn2022). This allows pronominal demonstratives and pronouns to be analyzed in parallel, as shown in 26 for the pronoun she. Here, instead of a full NP restriction of being a girl, the nominal restriction contains only the ϕ-feature requirement of being female and returns the unique entity that is female and is associated with location a.

(25)

⟦she+IXA⟧ = ιx. fem(x) & R(x,a)

(English)

In Mandarin, where the pronoun ta can be used to refer to any animate or inanimate entity (see Sun Reference Sun2006), it simply refers to the unique entity that is associated with location a, as in 26. ASL IX_A is analyzed in a parallel manner, except that the pronoun is null, as in 27.

(26)

⟦ta+IXA⟧ = ιx. entity(x) & R(x,a)

(Mandarin)

(27)

⟦∅+IXA⟧ = ιx. entity(x) & R(x,a)

(ASL)

We see motivation for this kind of analysis within sign languages, with deverbal anaphors identified in Coppola & Senghas Reference Coppola, Senghas and Brentari2010 and Senghas Reference Senghas1995. Senghas (Reference Senghas1995:139) observes that signers of Lengua de Señas Nicaragüense (LSN) produce what she calls a deverbal anaphor, ‘a reduced, truncated form of a recently-signed verb … to refer back to the referent in the narrative that last served as the most salient argument of that verb’.Footnote ⁹ An example of a deverbal anaphor from her work is shown in 28, where collect in the last line refers to ‘the one who collected’ (glossed as ‘the collector’ in Senghas Reference Senghas1995).

(28)	man fall-down-[iterative].
	‘The man falls down head-over-heels.’
	feather-pl float-down, man collect-pl.
	‘Feathers float down and the man collects them.’
	bird laugh.
	‘The bird laughs.’
	[collect] N look up.
	‘The collector looks up.’	(NSL; Senghas Reference Senghas1995, ex. 16)

Kocab et al. (Reference Kocab, Senghas, Coppola and Snedeker2023) show that the modifiers used this way in LSN show prosodic reduction, which may suggest their status as a relative clause or a smaller clausal structure that acts as a modifier. The rough meaning of the deverbal anaphor in 28 could be ‘the unique entity that did the collecting’. The man was described in the previous discourse with the property of collecting, and that same property is being used to point out the intended referent. If this is analyzed as an anaphoric expression that has a null head and a restrictive relative clause, deverbal anaphors can be seen as parallel to the analysis of IX_LOC presented here. Just as the deverbal anaphor refers to ‘the one who collected’, IX_A refers to ‘the one associated with location A’.

3.3. Advantages of analyzing IX_LOC as a modifier

3.4. Extending to deictic uses

We analyze IX_LOC as a spatial modifier that takes a location o and an entity x and returns ‘true’ if x holds some relation to o. We assume that this relation is essentially spatial in nature, meaning that the location variable o inherently comes with the mereological properties that all spatial locations have and that loci have also been noted to have (Schlenker et al. Reference Schlenker, Lamberton and Santoro2013). When the relation R is interpreted in its most direct, transparent form, it can represent a true spatial layout, where the referent is physically in the indicated location at the time of the discourse. This use would account for deictic uses of IX_LOC where the referent is present in the signing context, as in 38.

Note that the association of actual entities in the signing context extends to the verbal domain, so that the verb seems to agree with the locations of the relevant entities in the actual context of discourse. Incorporation of transparent deixis is generally not found in any existing verb agreement system that we are aware of in spoken languages, but is naturally predicted under an account in which instances of IX_LOC involving an actual location variable o are incorporated into the verbs, as we proposed in Section 2.

The relation between a referent and a location can be less deictic and more metaphorical, given that the location simply has to be associated with the intended referent. For example, a signer can point to an office and describe the occupant of that office as reading, as in 39.

Finally, an abstract locus use can be thought of as a further extension of this inherently deictic use. Instead of pointing directly to the referent or to an entity that is associated with the referent, the signer associates an arbitrary location in signing space with that referent and points to that location, as in Figure 7, represented in 40. Thus, the different uses of IX_LOC fall onto a continuum, from a fully deictic reference to a real entity to an abstract association of a referent with an arbitrary space.

Figure 7.

‘He reads’ (ASL). See example 40.

Note that this continuum from deixis to arbitrary association is found in spoken languages too. For example, a speaker can point to an actual entity in the discourse context, as in 41a, in which case R would simply be a relation of physical location. A speaker can also point to a location that is contextually associated with the intended referent, like the person’s office as in 41b, or to an arbitrary location they have assigned for the referent in an anaphoric use, as in 41c.

Our drawing a direct parallel between deictic and arbitrary uses of IX_A may be surprising given that deictic and anaphoric expressions have long been assumed to differ fundamentally in the way they interact with the rest of the linguist content. Perhaps most famously, it has been known since Kaplan Reference Kaplan, Davidson and Hintikka1969 that deictic expressions like demonstratives have a rigid reference that is not compatible with variable-bound interpretations. Rigidity is defined in Kaplan Reference Kaplan, Almog, Perry and Wettstein1989 as having a fixed reference across different contexts of evaluation. For example, indexical expressions such as actually, here, and now in 42 refer only to a specific group of people who are actually in the utterance context, regardless of the temporal and locative displacement.

Rigidity of deictic demonstratives is illustrated in 43: regardless of the fact that the speaker is supposing a hypothetical situation where Jin and Sol switched places, the use of a demonstrative description that person while pointing to Sol rigidly picks out Sol in the actual speech context. This is contrasted with a definite description, as in 44, where the referent of the person on the right shifts following the hypothetical context.

IX_LOC across languages is known to allow bound-variable uses where it covaries in the scope of a quantifier (Kuhn Reference Kuhn2016, Schlenker Reference Schlenker2011, Steinbach & Onea Reference Steinbach and Onea2015, among others). For example, in 45 from Schlenker Reference Schlenker2011, IX_A and IX_B do not refer to specific entities, but covary with the context of evaluation.

If the abstract use of IX_LOC is equated with deictically transparent uses, how can the bound-variable interpretation be derived? While it is true that bound-variable interpretations are generally not compatible with deictic expressions, we also observe that deixis can, in contrast to the classic generalization, be contextually restricted. Consider 46, where a speaker is describing a scene at a theme park, pointing to the addressee’s left arm (Paul Marty, p.c.). Here, despite the use of the demonstrative description and accompanying deictic gesture, the referent of this side is not fixed rigidly: the unrealistic interpretation where everyone got a stamp on the addressee’s left arm is not available. Instead, the relative side and location covaries with the individuals being quantified over.Footnote ¹¹

The interpretation of this side _→left-arm as ‘the relevant person’s respective side’ can be applied to examples like 45, where the interpretation of IX_A is not strictly ‘the person at location A in the discourse context’ but ‘the relevant person associated with A’.

Approaching the problem of deixis and loci from a more gestural/spatial angle, the overwhelming similarity between abstract and deictic uses of space in both the nominal and verbal domains in sign languages has not gone unnoticed, and in fact a continuum between these has been a large motivating factor for cognitive approaches to loci, such as Liddell Reference Liddell2003. We suspect that precisely such an approach to map space via analogy between mental representations is underlying the possible mappings for our relation R. In this, we agree completely with the goal as stated by Schlenker et al. (Reference Schlenker, Lamberton and Santoro2013), to marry a formal analysis of abstract loci with a cognitive linguistic approach to iconicity. We differ with that account on some of the details: the ‘iconic variables’ as modeled by Schlenker et al. (Reference Schlenker, Lamberton and Santoro2013) are essentially independent in the semantics they impose in being iconic and in being loci. In other words, in their system, presupposition failure in iconic loci is disjunctive: it can either arise via failure of the dynamic semantic assignment function to assign a referent to the variable (where the variable is made visible via the locus), or it can arise if there is not an available mapping between (aspects of) the form of the locus (say, its height) and (aspects of) the referent. By contrast, in our view there is a single relation in the semantics R between a location and an entity, and whether that relation itself is entirely arbitrary, somewhat motivated, or fully deictic is something that can vary continuously. One can imagine how gradience/continuity in iconicity could be derived in the system in Schlenker et al. Reference Schlenker, Lamberton and Santoro2013 too to some extent; that is, one can essentially have a very weak or nonexistent constraint on some mapping that might exist, but there the locus as variable still exists as an independent restriction. In our system where the locus instead contributes a predicative restriction via IX_LOC, there is no independent restriction of being a locus, but rather an abstract locus is a special case of a particularly arbitrary relation, a relation that in other cases may be far more motivated.

3.5. Extending to locational loci and spatial verbs

So far we have looked at the use of a personal locus, namely IX_LOC, that associates an entity with a location. This predicate combines with a null demonstrative to refer to an entity, and they get incorporated into ‘agreement’ and ‘plain’ verbs to mark person information. In this section, we show that this analysis extends to the third category of verbs identified in Padden Reference Padden1983, the ‘spatial’ verbs, which incorporate spatial information rather than person information.

The analysis of locational loci that mark spatial information makes use of the same analysis for IX_LOC, where it is a predicate relating some entity to some location. This denotation from 22 is repeated in 47.

The only difference between personal loci and locational loci is what entity is associated with that location. Crucially, we assume that entities for which the spatial predicate is defined do not always have to be individuals but can also be locations. Individuals, events, and locations can all be involved in canonical semantic operations such as quantification, anaphora, and reference (and Schlenker (Reference Schlenker2013) observes that sign languages in particular quantify over these in similar ways). For example, just as we can anaphorically refer to entities (I met a person and liked her ), we can refer back to events (I danced yesterday and was happy about that ) or locations (I went to Boston yesterday and want to go there again). Given this parallel, we assume that x in 47 can pick out a location rather than an individual entity. Then, the contextually salient R variable that holds between x and o can be that of identity, physical containment, or some link between a mental representation and the actual physical location. Combining with a null demonstrative, the resulting meaning would correspond to expressions referring to locations, such as there or that location there.

Finally, we assume that spatial verbs differ from ‘agreement’ verbs only in their selectional properties, indicating locations such as source and goal as arguments (instead of e.g. agent/patient). Thus, a verb like move involves locational expressions such as here and that location, so that _Amove _B corresponds to an English sentence such as (I) moved from there to there. By extending our analysis to allow incorporation of locational expressions such as there and that location in addition to referential expressions such as the person at A, we can account for verbs that mark person information and spatial information uniformly in terms of argument incorporation. Note that this analysis aligns with the main intuition in Meir Reference Meir2002, where agreeing and spatial verbs are uniformly analyzed as carrying a directional morpheme that agrees with the source and the goal arguments. For spatial verbs, the source and the goal are locations, while for agreeing verbs, the source and the goal are entities. Our analysis is similar in that we can uniformly treat person and spatial modification, though our analysis relies on incorporated nominals that refer to entities or locations.

How is a correct interpretation of loci possible when personal and locational nominal heads are phonologically null and indistinguishable? We argue that verbs are lexically specified for the type of loci they incorporate: for example, help is specified to take personal loci, walk-to is specified to take locational loci, and put is specified take a personal locus for the subject and a locational one for the object. This lexical specification determines which nominal a verb selects and underlies the correct interpretation of loci.

The verb’s selectional properties can be formalized either syntactically (e.g. put selects for an NP and a PP) or semantically in terms of thematic roles (e.g. put selects for an agent, a patient, and a theme). Learning the argument structure of different verbs is a challenge in signed languages just as in spoken languages. Correct interpretation of loci is analogous to pro-drop in spoken languages: even when overt pronouns are absent, participants can identify the relevant arguments (animate, inanimate, or locational) based on the verb’s lexical specification. Additional, language-specific factors, such as agenthood and focus, may also influence which pronouns are marked overtly. Our analysis predicts parallel and expected patterns in how verbs select arguments, and how signers and speakers mark some pronouns overtly depending on context.

Under this proposal, we reduce the sign-specific stipulations of the three-way distinction argued for verbs in sign languages to a language-general problem of learning lexical-specific argument properties of verbs. Under this view, inflecting verbs select for animate entities, and spatial verbs select for animate and locational entities. Some plain verbs can also mark person or spatial information, as we do not restrict modification to verbs with a directional movement. This makes possible the generalization that all verbs in principle may allow incorporation of nominal arguments, which may be locational or personal. There are, of course, various constraints—including the phonology of the sign, the specific selectional properties, and the interpretation of the verbs—that may affect whether incorporation is possible.Footnote ¹² Such idiosyncratic restrictions are common across lexical items, and we are not able to lay out the details in this article. However, we highlight that through this analysis, the unique three-way distinction of verbs in sign languages is reduced to the process of incorporation conditioned by selectional, pragmatic, and phonetic factors. Seen this way, verbs in sign language do not seem very different from verbs in spoken languages.

3.6. Nonmanual marking of loci

While we have focused on manual marking of loci with IX, loci can also be signaled through nonmanual signals such as eye gaze, head tilt, and body shifts (Aarons et al. Reference Aarons, Bahan, Kegl and Neidle1992, Bahan Reference Bahan1996, Neidle et al. Reference Neidle, Kegl, Bahan, MacLaughlin and Lee2000, Quadros Reference Quadros1999, among others). These interact with verb modification and word order (Quadros Reference Quadros2003). For example, in Brazilian Sign Language (Libras), nonmanual marking is optional with verb modification in SVO sentences, but becomes necessary in noncanonical orders. Quadros (Reference Quadros2003) shows that eye gaze (marked as ‘eg’) is optional with SVO (48a), while OSV requires eye gaze (48b) and OSV without eye gaze is degraded (48c).

(48)

Under our analysis, IX_LOC is a modifier requiring an entity to be associated with a location. While this is often achieved via pointing, it extends naturally to nonmanuals such as eye gaze or head tilts, through which locations can be made salient. These location-associating nonmanual markers can also be analyzed as spatial predicates associating the given entity to a location.

This explains both the optionality of nonmanuals and their role in flexible word order. Note that in all of the examples in 48, directional modification of the verb is present, thus indicating the subject and object clearly. However, with noncanonical word order, there might be more need to mark the relevant entities overtly because the systematic correspondence between syntactic position and thematic roles is lacking. Thus, the need for disambiguation that licenses overt modification of verbs in ASL can be argued to play the same role in licensing eye gaze in Libras. This would also be compatible with the view that verb modification is a syntactic phenomenon that marks obligatory feature agreement in Libras (Lourenço & Wilbur Reference Lourenço and Wilbur2018, Quadros Reference Quadros1999) and that nonmanual markers are used to disambiguate potential referents in a pragmatically driven way.

3.7. Comparison with other accounts: indices and features

Although sharing many features of various analyses of sign language loci, our proposal differs from existing accounts on a few key points. We highlight these contrasts here and explain why we think a spatial modifier view has advantages. However, we see several ways in which the same insights involving pragmatic contrasts we have focused on here can be extended to other accounts, in many ways making them closer to each other and to the analysis that we have described.

First, our proposal contrasts in important ways with the view of loci in sign languages as equivalent to the indices used in assignment functions to keep track of variables in dynamic semantic accounts. If loci carry descriptive semantic content that is optional, then they can be subject to pragmatic considerations. This optionality and the markedness of loci are not easy to account for under an account where a direct parallel is drawn between sign language loci and covert indices assumed for anaphoric resolution in spoken languages. In semantic frameworks that make use of indices, every discourse referent is assumed to be assigned an index, and the index is what enables an anaphoric link between a pronominal element and its antecedent. If loci are overt manifestations of anaphoric indices, this would mean that every anaphoric relation will require loci to be present. While it is true that loci can be used in anaphoric contexts, as shown by many examples in this line of work, the optionality suggests that loci are not necessary in anaphoric relations.

This does not mean that we must rule out the index analysis altogether. In fact, the original proposal relating loci to indices in Lillo-Martin & Klima Reference Lillo-Martin, Klima, Fischer and Siple1990 builds this pragmatic optionality into the grammatical model. Unlike a traditional Y-model with Syntax feeding (separately) logical form (LF) and phonological form (PF), they argue that the level of discourse representation must take place in a third dimension that connects both logical representation and phonological representation, as well as syntactic and pragmatic representation. In our view, allowing the level of discourse representation realized by loci to have access to pragmatic representation makes Lillo-Martin and Klima’s index view quite different from some of the more recent adaptations of their work, such as Barberà & Zwets Reference Barberà and Zwets2013 and Schlenker Reference Schlenker2011, where loci are analyzed as strict counterparts to semantic indices.

In subsequent works, the notion of ‘loci as indices’ from Lillo-Martin & Klima Reference Lillo-Martin, Klima, Fischer and Siple1990 has been couched in semantic frameworks with assumptions that they are indeed strictly part of the LF. For example, Schlenker (Reference Schlenker2011, Reference Schlenker2016, Reference Schlenker2018) argues that the semantics of loci is given by an assignment function like logical indices in spoken languages and that loci are indices that can be bound dynamically as well. In analyzing IX used in Sign Language of the Netherlands (Nederlandse Gebarentaal; NGT), Barberà and Zwets (Reference Barberà and Zwets2013) adopt the File Change Semantics in Heim Reference Heim1982 and argue that IX occurring with a nominal that refers to a referent not previously mentioned introduces a discourse referent, like indefinite nouns do in spoken languages. In all of these studies, the underlying assumption is that loci are logical indices that are assumed in semantic frameworks that account for anaphora resolution in spoken languages. In other words, loci are assumed to play the critical role in anaphorically linking a pronominal element to its antecedent. We call this line of works the ‘strict index analysis’.

Steinbach and Onea (Reference Steinbach and Onea2015) preserve the pragmatical optionality built into Lillo-Martin & Klima Reference Lillo-Martin, Klima, Fischer and Siple1990 by separating the pragmatics of the phonological production of pointing from the semantic notion of indexing discourse referents. Semantically, every referent is assigned a locus through a default feature-assignment mechanism. When pragmatically appropriate, this locus can be phonologically marked with an overt IX_LOC. Thus, their account differs from that of other works discussed here in that the overt use of loci is not directly associated with a logical, semantic indexing mechanism, but is used to restrict coreference.

Our proposal preserves the critical intuition in Lillo-Martin & Klima Reference Lillo-Martin, Klima, Fischer and Siple1990 and Steinbach & Onea Reference Steinbach and Onea2015 that the overt use of loci should be pragmatically conditioned, but with a slightly different implementation. Instead of arguing that loci are semantic indices that can be overtly produced when pragmatically licensed, we argue instead that they are inherently locational restrictions, and thus different from semantic indices.

However, if one is committed to the advantages of dynamic semantics and finds the use of loci to be a good candidate for an overt index in that system, there is a way to reconcile these views, namely, to treat the overt information as descriptive. In this case, one may prefer to phrase this content not so much as indicating a location but rather as equating with an index, and as long as this is considered descriptive content, our same reasoning will hold. Note that this alternative analysis may necessitate a different approach to indices in those views that do not treat indices as descriptive. For instance, in Schlenker Reference Schlenker2005, indices are ranked as being the least descriptive: DPs that carry only indices—pronouns in Schlenker’s account—are pragmatically less redundant than DPs that carry descriptions. This analysis of loci as indices predicts that the use of loci would be preferred over not using them, and that a DP without a locus would be deemed more redundant than a DP with one. This prediction seems less compatible with the empirical picture we have seen.

The second category of disadvantages of a treatment of loci as indices concerns their semantics. There are several places where loci do not parallel semantic indices as they are typically conceived. Consider, for example, the observation in Abner & Graf Reference Abner and Graf2012 and Graf & Abner Reference Graf and Abner2012 that loci do not participate in syntactic binding, based on their inability to appear under negative quantifiers, among other things. Here we see evidence against the use of loci for variable traces, which is one of the properties that makes indices potentially distinct from any other feature one would assign that would restrict coreference to pronouns.

Ultimately, we think that a semantics that involves actual location and gets the anaphoric use by metaphorical extension is to be preferred over a semantic content that simply predicates of a referent that it is associated to a locus A, but the two will be formally the same in terms of contrast and ability to restrict coreference. One may choose the more abstract or the more locative option as a preference or as other empirical or theoretical considerations come into play. Regarding the abstract anaphoric use of signing space as a metaphorical extension of topological space, we think this fits naturally with the idea that contextually given information is signed in the ‘nearby’/default space, to contrast with expanded contexts in more distal/higher space, another use of space motivated by metaphor that has become grammaticalized for anaphora in multiple sign languages (Barberà Reference Barberà2016, Davidson & Gagne Reference Davidson and Gagne2022), including here the metaphorical use of space via distance-as-difference.

An alternative proposal in formal linguistic approaches to loci is to take them to be (ϕ‑)features of the sort that trigger agreement. This naturally accounts for their behavior in both the nominal and verbal domains and for their ability to resolve co-reference. It also raises the question of the ‘interpretability’ of such features. While variable indices are by definition always arbitrary, features may convey (not-at-issue) semantic information: for example, gender in English personal pronouns contributes the information that the referent is of a particular gender (e.g. female), and proximal features on English demonstratives attribute a locational relation with respect to the speaker or some other center of evaluation, yet both are ignored/uninterpreted in focus-sensitive contexts such as under the focus operator only. For example, Kuhn (Reference Kuhn2016) points out that we can see in 49–50 that the alternatives negated by only do not consider the gender feature on the personal pronoun or the proximal feature on the demonstrative pronoun in constructing alternatives—they exclude everyone, not just the mothers of females in the first sentence, and not just the proximal sweaters in the second sentence—and that the same pattern holds in ASL loci as well, where the alternatives include entities that are not associated to the location a.

How do we square the fact that we ignore the gender restriction in 49, the proximal restriction in 50, and the spatial restriction in 51 with our emphasis throughout this article on the pragmatic contrasts that loci induce? The idea is simply that like any modifier, a locus can become ‘at issue’ and thus interpreted under focus, but that it is also able to be not-at-issue, as discussed at length across modalities by Esipova (Reference Esipova2018). We further note that ‘featural’ restrictions do not seem qualitatively different from any other content in anaphoric definite descriptions. For example, in 52, a speaker may choose to use a pronoun her or a full description the linguist to refer back to the antecedent. Neither the property of being female nor the property of being a linguist is interpreted under the focus operator ‘only’: 52 says that no one else, other than the said linguist, liked the book that was recommended to them by the philosopher.

In this respect, the restriction of a definite description does not seem to differ from the gender feature of pronouns, so we see analyzing IX_LOC as a restriction as still compatible with examples like 51.

3.8. Summary

In this section, we have shown that the use of loci in the nominal domain is similar to that in the verbal domain in its sensitivity to pragmatic need for disambiguation: loci are dispreferred when reference is unambiguous, but licensed when disambiguation is required. Noting that these properties align IX_LOC with marked referential expressions crosslinguistically, such as demonstratives and anaphoric definites, we analyze IX_LOC as a spatial modifier that accompanies a covert demonstrative. The predicate locates entities in the signing space, just as pointing that accompanies demonstratives and pronouns in spoken languages does. We show that analyzing IX_LOC as a spatial modifier allows us to (i) give a uniform account of IX_LOC in its introducing and anaphoric uses, (ii) draw a parallel between the use of IX_LOC and the use of abstract and deictic labels in spoken languages, (iii) provide a natural account of predicative uses of loci observed in descriptions of a spatial layout, and (iv) derive their optionality of use without additional stipulations.

We extend this analysis to locational loci, showing that the only difference between locational and personal loci is the type of entity the spatial predicate is defined for: a location and an entity, respectively. These form anaphoric expressions referring to locations (‘that location at A’) or people (‘that person at A’), which then can be selected by verbs. We argue that the selectional property of a verb is determined in the lexicon, so that a verb like help selects for two person entities, while a verb like move selects for two location entities. This allows for a consolidation of Padden’s (Reference Padden1983) three-way classification of verb types, at least with respect to the treatment of loci: agreement, spatial, and plain verbs can all be analyzed in terms of how they incorporate nominal material, with their differences stemming from their lexical, selectional properties.

4.1. Study design

The study used (i) three sentence conditions, each with two levels ([±context]), and (ii) manipulation of locus presence. A Latin square design created four lists of twelve sentences each, with one sentence per verb. For example, list A contained four sentences for the number-of-referents condition, each with a different verb: [1 referent, −locus, verb 1], [1 referent, +locus, verb 2], [2 referents, −locus, verb 3], [2 referents, +locus, verb 4], and so on for the other two conditions. Participants were randomly assigned to one list.

On each trial, participants viewed a short signed video of the target sentence, followed by a GIF with a question of the form ‘who [verb]?’, where the verb matched the trial. They answered by selecting one of two GIFs showing possible NP answers. The GIFs contained only bare, singular NPs. Afterward, participants rated the discourse’s naturalness on a slider from a sad face (left) to a smiley face (right).Footnote ¹⁴

4.2. Results

We tested the effects of Number, Narrative, Animacy, presence of locus ([±locus]), and contextual support ([±context]). Figure 8 shows response accuracy, that is, the percentage of choosing the intended antecedent of the agent argument. Overall, we found a marginal main effect of condition (p = 0.097), significant main effects of locus (p < 0.001) and contextual support (p < 0.001), and a significant interaction between the two (p < 0.001). Participants were more likely to choose the intended referent when loci were present and when context made roles clear.

Figure 8.

Percentage of intended referent choice by locus presence and context support. [−locus] (gray) and [+locus] (blue) represent the absence and presence of locus marking on nouns and verbs. ‘0’ and ‘1’ for context support represents the absence and presence of context support (one vs. two referents in ‘Number’; no narrative vs. narrative in ‘Narrative’; and two animate referents vs. one animate referent in ‘Animacy’).

In the Number condition, there were significant main effects of locus (p < 0.001) and contextual support (p < 0.001), plus a significant interaction (p < 0.001). Correct responses were more likely when loci indicated roles or when only one entity was present. In the Narrative condition, locus had a significant effect (p = 0.002) and interacted with contextual support (p = 0.001), but contextual support itself was not significant (p = 0.318). Thus, loci increased accuracy regardless of narrative. In the Animacy condition, locus (p = 0.017) and contextual support (p < 0.001) both had significant main effects, as did their interaction (p < 0.001). Accuracy was higher with loci present or when the competing referent was inanimate.

At-ceiling responses in [clear, +locus] conditions for Animacy and Number reflect our screening procedure. Still, in both conditions participants were significantly less accurate when loci were absent or when context did not clarify roles; performance was lowest when both cues were missing. In the Narrative condition, overall accuracy was lower (31–81%), but the presence of loci still improved accuracy.

Figure 9 shows naturalness ratings with standard error bars (twenty-six responses per condition). A linear mixed-effects model (LMM) with participant and item as random effects revealed a marginal effect of contextual support (p = 0.026), a significant effect of locus (p < 0.001), and a significant interaction (p < 0.001). Sentences were rated more natural when loci were present or when context clarified roles.

Figure 9.

Average rating of sentences by locus presence and context support. [−locus] (gray) and [+locus] (blue) represent the absence and presence of locus marking on nouns and verbs. ‘0’ and ‘1’ for context support represents the absence and presence of context support (one vs. two referents in ‘Number’; no narrative vs. narrative in ‘Narrative’; and two animate referents vs. one animate referent in ‘Animacy’).

In the Number condition, contextual support (p = 0.034), locus (p < 0.001), and their interaction (p < 0.001) were significant; in [−locus] conditions, contextual support marginally increased ratings (p = 0.069). In the Narrative condition, locus (p < 0.001) and its interaction with contextual support (p < 0.001) were significant, but contextual support alone was not (p = 0.470). In the Animacy condition, locus had a significant effect (p = 0.028) and a marginal interaction with contextual support (p = 0.067), while contextual support was not significant (p = 0.476). In [−locus] conditions, ratings trended higher with contextual support, though not significantly (p = 0.350).

4.3. Discussion

Several observations emerge from the sentence ratings. First, [−locus] sentences were not rated at 0, indicating that signers accept sentences without locus marking in both nominal and verbal domains as natural ASL. This supports our claim that loci are optional modifiers used to disambiguate antecedents, not obligatory agreement markers. By contrast, an agreement-based account would predict rejection of sentences without a locus.

Second, ratings of [−locus] sentences improved with contextual support, most clearly in the number condition (significant) and marginally in the animacy condition (not significant). This aligns with our pragmatic account, where loci are licensed for disambiguation rather than agreement or anaphoric linkage.

Third, we do not see the pragmatic effect in the narrative condition. We hypothesize that this is because the Narrative condition did not in fact have the intended effect of disambiguation. This is indicated by the lower percentage of correct referent choice in the Narrative conditions.

Finally, Animacy sentences showed slightly lower ratings in [+locus, +context] (p = 0.906), possibly reflecting avoidance of loci on inanimates. Prior work (Ahn et al. Reference Ahn, Kocab, Davidson and Cho2019) reports degraded judgments for inanimate loci. While this could reflect redundancy, we do not conclude so for two reasons: (i) redundancy effects are absent in Number sentences, where locus use is also redundant, and (ii) redundancy is typically detected in on-line measures (e.g. the Repeated Noun Penalty: Gordon et al. Reference Gordon, Grosz and Gilliom1993, Gelormini-Lezama & Almor Reference Gelormini-Lezama and Almor2011, Song & Lee Reference Song and Lee2009, Van Gompel et al. Reference Van Gompel, Liversedge, Pearson, Carreiras and Clifton2004, Yang et al. Reference Yang, Gordon, Hendrick and Wu1999) rather than off-line ratings. Off-line tasks may mask processing costs (Law & Syrett Reference Law and Syrett2017, Syrett & Brasoveanu Reference Syrett and Brasoveanu2019). Thus, detecting degradation of [+locus, +context] cases likely requires on-line methods.

To summarize our results, we go back to the three subhypotheses presented in 54: (i) sentences without locus use are accepted if there is contextual support to indicate who did what, (ii) sentences without loci are rated lower if contextual support is not available, and (iii) potential redundancy of loci in unambiguous contexts was not detectable here and may require a different methodological approach. Overall, the findings converge with prior work and support the view that locus use is pragmatically licensed for disambiguation.