Interaction of the differential object marker pam with other prominence hierarchies in syntax in German Sign Language (DGS)

2.1 Sign order

Sign languages described to date have been shown to exhibit primarily subject-verb-object (SVO) or subject-object-verb (SOV) constituent orders. Some languages have been described as having either SVO or SOV basic underlying order (see Leeson and Saeed 2012 for a review); however, it has been suggested that most sign languages are characterized by both SVO and SOV (Napoli and Sutton-Spence 2014; Napoli et al. 2017). Based on a literature review of constituent order data from more than 40 sign languages, Napoli and Sutton-Spence (2014) argue that constituent order is determined to a large extent by modality-specific phonological considerations, in particular, that O will precede V if it affects the form of the verb. For example, the handshape and position of the hands differ in carrying a box compared to carrying a suitcase, corresponding to their different semantics. The argument also applies to verbs that move through space to indicate arguments: the relevant arguments should already be introduced if the addressee is to make sense of the path of movement of the predicate (i.e., to identify the arguments associated with the beginning and end locations of the verb). Pressures that have to do with avoiding potential ambiguity of grammatical roles, i.e., pressures that are not modality-specific, are also noted by Napoli and Sutton-Spence (2014), notably that SVO is favored in reversible sentences (see also Hall et al. 2013, Schouwstra et al. 2022 for similar evidence from silent gesture). In addition, Napoli et al. (2017) argue that event type affects constituent order, showing for Libras (Brazilian Sign Language) that verbs with extensional objects (i.e., objects that exist; eat an apple) are preferably realized with SOV constituent order, while verbs with intensional objects (i.e., objects that are presupposed (not) to exist; want an apple) are more likely to be realized with SVO order. They also note the relationship between constituent order and definiteness cross-linguistically (Tomlin 1986) and point out the more likely combination of extensional verbs with definite objects and intensional verbs with indefinite objects, respectively, and that more research is needed to understand the relationship between event type, constituent order, and definiteness.

For DGS, the basic sign order is generally assumed to be SOV (Happ and Vorköper 2006, Steinbach and Herrmann 2013, Oomen 2021, Steinbach 2022). However, based on an analysis of DGS corpus data, Oomen (2021) finds that constituent order is influenced by verb type, which is consistent with observations by Napoli and Sutton-Spence (2014). Specifically, Oomen (2021) finds that body-anchored plain verbs favor SVO constituent order, while agreement (or indicating) verbs (i.e., verbs that move between locations in space) as well as neutral plain verbs (i.e., verbs without movement, but produced in the space in front of the signer’s body) favor SOV constituent order. Oomen (2021), nevertheless, assumes that SOV is the basic constituent order in DGS, based on the possibility of being able to syntactically derive SVO order from an underlying SOV order and assumptions in the previous literature (see also Bross and Hole 2017, Bross 2018, Pfau et al. 2018). In the present study, we aim to shed light on the different constituent orders found in naturalistic data by bringing in the notion of prominence and the interaction with other phenomena, in particular, verb type, spatial modification behavior, and the use of pam.

Both of the dominant constituent orders share a subject-first position. The preference for a subject-first order reflects a more general agent-first principle (e.g., Primus 2001) and can be further described in terms of a prominence-based approach to constituent order and its alternations. The prototypical agent is human and the initiator of events. As such, the agent is an inherently prominent element, grounded in both cognitive and evolutionary explanations (Himmelmann and Primus 2015). The prominence of the agent is reflected cross-linguistically by being placed in first position in clauses. In the association between agents, subjects, and humans as the best candidates for the first syntactic position, there is some potential circularity: the subject is inherently prominent because it occupies the first position, and the first syntactic position is inherently prominent because it is occupied by the subject. It is the prominence of the agent that is at the root here: the agent is understood cognitively as the most important element of the event, as the initiator and causer of the event, with volition, control, and autonomous movement.

Building on these assumptions, linguistic contexts can be imagined where these three elements (agent, human, and subject) come into opposition, essentially resulting in argument competition in syntax (Himmelmann and Primus 2015). Argument competition occurs when two or more arguments compete for prominence, i.e., compete to be the center of linguistic (and cognitive) attention. On the one hand, there are inherently prominent elements (associated with the agent), as mentioned earlier. However, prominence can be attributed to other elements, attracting attention, in certain linguistic contexts. An example of this are passive constructions, where the agent is demoted to the object position while the (initial) subject position is occupied by the patient. Here, a clear competition between arguments is evident, with the agent, being inherently prominent, losing syntactic prominence to the patient, which in turn occupies the subject position, and in most cases, the first position in the sentence. It is worth noting that the notion of subject and object is not clear on independent grounds for the most part in the syntax of sign languages. An investigation of grammatical categories based on consistent syntactic behavior has been carried out for only a few sign languages (e.g., Jantunen 2008 for Finnish Sign Language [FinSL]; Johnston 2019 for Australian Sign Language [Auslan]). In both of these studies, there was little evidence of strict grammatical structures or morphosyntactic markers that would allow the attribution of argument status as subject and object, independent of semantic roles and constituent order. Based on the association between agent and subject as reflecting prominence in semantic and syntactic terms, we assume that arguments in first position are more prominent than arguments in non-first position and that agents are typically subjects.

2.2 Verb modification

Although the analyses vary, from agreement to cliticization to hybrid gestural accounts (Pfau et al. 2018), there is consensus that verbs can be spatially modified in different ways, through changes in palm orientation and/or direction of movement of the hand(s), to reflect their arguments. There are different accounts of how arguments may be realized or identified, including association of the location of the hands with locations associated with referents (Lillo-Martin and Meier 2011); the use of eye gaze directed at referents, in particular, the object referent (Neidle 2000); and the use of the body to represent referents, in particular, the subject referent (Meir et al. 2007). Verbs that have been called agreement or indicating verbs in the literature (Pfau et al. 2018) undergo spatial modification, i.e., movement in space, in relation to both the subject and the object referents. For DGS, the modification of verbs that move through space for argument marking has been analyzed as being obligatory (Oomen 2021, Steinbach 2022). Note, however, that this is based on corpus analysis findings by Oomen (2021) that roughly 93% of spatially modifiable verbs are “at least congruent” (p. 227) with their subject and object arguments, where congruent means that locations in space coincide with the citation form production of the verb forms. For British Sign Language (BSL), in contrast, verb modification has not been found to be obligatory and has been described as depending on semantic factors, such as animacy (Fenlon et al. 2018). With respect to the spatial modification of verbs to indicate the patient/object argument, in particular, Fenlon et al. (2018) found that verbs are modified more frequently when the patient is human than when the patient is inanimate. The BSL data also contain proportions of verbs that are clearly modified or congruent with citation form production, but as Oomen (2021) notes, the proportion of clearly unmodified tokens is higher in the BSL data (roughly 30%) reported by Fenlon et al. (2018). Interestingly, Pfau et al. (2018) note a lack of investigation of the nature of argument marking for DGS, but assume similar factors motivating analyses of optional marking (as for BSL; also for Australian Sign Language by De Beuzeville et al. 2009) to hold for DGS, as well.

From the perspective of prominence, the modification of the verb to indicate the patient can be seen as providing additional information that generates active attention to the patient (Himmelmann and Primus 2015). On an assumption of optional marking, as in BSL, modification is additional morphosyntactic marking that happens only in certain contexts. For BSL, these contexts are conditioned by features traditionally relevant to prominence, namely, the animacy of referents. The behavior of the verb is thus influenced by the animacy of the object, reflecting prominence-lending features of the object. Based on these findings, we assume that the spatial modification of verbs to indicate arguments is a prominence-marking device. In particular, the spatial modification of verbs to indicate the object can be interpreted as attributing prominence to the object.

2.3 Omission of arguments

Many sign languages have been described as allowing argument omission, akin to pro-drop languages (Lillo-Martin 1991, Glück and Pfau 1998, McKee et al. 2011, Jantunen 2008, Johnston 2019, Oomen 2021). This phenomenon is related to a large extent to verb types. On the one hand, agreement (or indicating) verbs are said to license argument omission (e.g., Lillo-Martin 1991). Similarly, due to an association of the body with the subject (Meir et al. 2007), body-anchored plain verbs have been described as licensing omission of subjects (Oomen 2017, Oomen and Kimmelman 2019). Specifically, Oomen (2017) shows that body-anchored psych verbs in Sign Language of the Netherlands (NGT) allow null subjects only for first-person subjects, but not for non-first-person arguments, due to the iconically motivated association between first person and the body of signer. The same pattern is found by Oomen and Kimmelman (2019) for all types of body-anchored verbs in DGS and Russian Sign Language, with the same iconicity-based (body as first person) explanation. Other plain verbs, on the other hand (i.e., neutral plain verbs, which do not move through space, but which can agree with (or indicate) an argument through placement in space), allow all types of null subjects. On this account, we can infer that a non-overt first person as subject is easy to recover, but that the same may not be true for (non-overt) objects. Proske (2022) shows that body-anchored transitive verbs rarely omit the object. She refers to Oomen (2021) to explain that in body-anchored verbs, the body-as-subject phenomenon is particularly strong (Meir et al. 2007). With the subject mapped onto the signer’s body, the object tends to be realized overtly to saturate the sentence. The possibility of subject omission arises from the understanding that the signer’s body can be interpreted as the realized subject (Meir et al. 2007). In the same vein, subjects that are not first person tend to be marked morphologically, to preclude the interpretation that the signer is first person.

In terms of prominence, it is possible to consider how the overtness of an argument can give it more or less prominence. From a discourse perspective, given and definite referents are more prominent than new and indefinite referents. Argument drop is thus also linked to how a referent can be recovered in discourse. The higher the discursive prominence of a referent, the greater the possibility of its omission (e.g., Ariel 1990). On a prominence account, prominent elements license linguistic operations (Himmelmann and Primus 2015). A prominent referent based on discourse cues may be zero-marked (and non-overt arguments may be higher in prominence than overt arguments), while a prominent element based on semantic cues (e.g., animacy) may receive special morphosyntactic marking (and a referent with marking may be more prominent than a referent without marking, as with differential object marking). There is no one-to-one correspondence between the prominence of referents and the linguistic expression or operation that marks the prominence.

2.4 Analysis of pam in DGS

The original analysis of the sign glossed as pam in DGS as an agreement auxiliary is reflected in its name: pam stands for Person Agreement Marker (Rathmann 2000) and denotes a sign that functions as a subject–object agreement auxiliary (Steinbach and Pfau 2007), or simply agreement auxiliary, in contexts where the verb cannot move through space to indicate its arguments. In a new analysis, based on acceptability judgments and sentence translations from German, Bross (2020) suggests that pam is not an agreement auxiliary, but rather a preposition-like element that is used as a differential object marker and related to animacy, definiteness, and affectedness. He finds that pam is used obligatorily with emotionally/mentally affected animate objects and that pam forces a definite reading of the object when used with verbs of (effective) action and perception (verbs high on Tsunoda’s (1985) affectedness hierarchy). Syntactically, Bross (2020) assumes that two usage patterns associated with pam exist, a clause-internal pattern (i.e., either S-pam-O-V or S-V-pam-O) and a clause-final pattern (i.e., S-O-V-pam). Furthermore, he notes a preference for clause-internal patterns by the signers consulted for the study. Bross (2020) leaves open the possibility that pam may function as an auxiliary in some dialects of DGS, stressing the applicability of his analysis to what he calls Southern DGS. Although not focusing specifically on the use of pam, Proske (2022) and Oomen (2021) find support for Bross’ (2020) differential object marking account for the use of pam on the grounds of animacy restrictions (see also Murmann 2012, Macht 2016, de Souza Santos et al. 2025). In addition, Proske (2022) finds that the use of pam in preverbal position (a clause-internal pattern) shows spatial modification only for the object argument, which is consistent with a differential object-marking analysis.

Based on a review of previous literature on the behavior of pam, Steinbach (2022) concludes that two versions of pam exist. The first is an agreement marker, glossed by Steinbach (2022) as _x pam _y, and the second is a differential object marker, glossed as pam _x. Steinbach (2022) distinguishes their use based on different syntactic behavior (and structure): the agreement marker occurs postverbally and agrees with (or indicates) both the subject and object arguments (through spatial modification), while the differential object marker occurs preverbally and marks only the object argument. Steinbach (2022) notes that an animacy constraint applies to the use of both forms, given that both forms are assumed to be grammaticalized from the nominal sign person. On Steinbach’s (2022) derivation, the postverbal position of _x pam _y allows agreement with (modification for) subject and object arguments in the same way as agreement (indicating) verbs. In contrast, preverbal pam _x only marks the object and is DP-internal or PP-internal (thus remaining neutral on the word class of pam _x). Steinbach (2022) notes that the different structures associated with _x pam _y and pam _x are compatible with the co-occurrence of agreement (indicating) verbs inflected for subject and object arguments. On the original account of pam as an agreement auxiliary, this behavior is unexpected, as the auxiliary was assumed to be used to mark arguments when the verb could not. However, a ‘doubling up’ of argument marking – using both spatially modifiable verbs and pam – has been previously noted in the literature (Rathmann 2003, Steinbach and Pfau 2007, Bross 2020, Proske 2022). In fact, Proske (2022) found pam used more frequently with indicating verbs than with plain verbs (that do not move through space). However, Proske (2022) found pam to be used primarily preverbally, which would suggest pam _x (the differential object marker) and not _x pam _y (the agreement marker) in Steinbach’s (2022) terms.

Overall, the verbs that occurred with pam in Proske (2022) – using a sentence repetition task, in which spontaneous instances of pam insertion were observed – are consistent with findings by Bross (2020) and de Souza Santos et al. (2025) supporting a differential object-marking analysis. Both Bross (2020) and de Souza Santos et al. (2025) discuss evidence for the use of pam being related to object affectedness, albeit in substantially different ways. Bross (2020), as mentioned previously, relates the use of pam to affectedness based on Tsunoda’s (1985) verb hierarchy. Specifically, he claims that the verb classes of Pursuit, Knowledge, and Feeling (alternatively: verbs with mentally/emotionally affected objects) occur obligatorily with pam, while verbs higher in object affectedness result in a definite reading of the object when occurring with pam – i.e., pam marks highly affected definite objects. De Souza Santos et al. (2025) also observe the strong connection between animacy and definiteness traits: pam is predominantly associated with human and definite, i.e., highly individuated (Aissen 2003), objects, but with little evidence of affectedness playing a systematic role. Instead, they argue that the degree to which the object argument exhibits semantic properties prototypically attributed to the agent is important. In particular, their corpus analysis of pam reveals three main contexts in which pam is used: (i) verbs with stimulus objects (Malchukov 2005) causing a change in the mental/emotional state of an experiencer subject (e.g., love); (ii) interactional content verbs where pam marks the recipient (indirect object) of information (e.g., recommend); and (iii) verbs with non-prototypical, animate, agent-like patients (e.g., help). De Souza Santos et al. (2025) suggest that the use of pam is related to the linguistic prominence of the object. Under this view, the use of pam as a differential object marker is interpreted as an element that assigns syntactic prominence to the verb’s object, particularly in the case of highly individuated objects (animate and definite). pam is thus a marker associated with arguments with more prominent characteristics, such as individuation and initiation, which, according to Himmelmann and Primus (2015), are characteristics of the agent, which is inherently prominent, drawing passive attention. Therefore, when the patient, in certain cases, has some of the agent’s characteristics, it becomes more prominent than the agent, attracting active attention.

4.1 Data: Public DGS corpus

Our analysis is based on naturalistic data available in the Public DGS Corpus (Konrad et al. 2020). The public corpus consists of 50 hours of video recordings, comprising data collected from pairs of signers engaged in free-form dialogues on various topics from across Germany (Berlin, Brandenburg, Bremen, Hamburg, Hesse, Lower Saxony, Mecklenburg-Vorpommern, North Rhine-Westphalia, Rhineland-Palatinate, Saarland, Saxony, Saxony-Anhalt, and Schleswig-Holstein) and balanced across four age groups (18–30 years; 31–45 years; 46–60 years; and 61 and above) (Schulder et al. 2024). All data were collected in a recording studio with a blue background, using three cameras positioned to the side of and in front of the interlocutors. The corpus is composed of a range of textual genres, including narratives and free conversations about topics such as experiences of being deaf or discussions about well-known individuals. All videos from the public corpus are available for download, accompanied by ELAN files containing data annotations. These annotations include glosses of the signs in English and German, separately for the right hand and left hand for each of the signers in the dyad, as well as translations (roughly on an utterance level) into English and German.

4.2 Data selection for the analysis

Data analyses were carried out through comprehensive searches in ELAN (version 6.6 2023) and the corpus website. For the initial analysis, all the ELAN files in the corpus were downloaded, and a structured multiple search was performed within the software to identify instances of pam (which is glossed in the corpus as on-person1), yielding 669 occurrences in the corpus. All occurrences were checked by members of the research team, including two native DGS signers. This resulted in the identification of erroneous glossing of on-person1 in the corpus, based primarily on native signer intuition. These signs were reclassified as being (a version of the sign) person, pronominal pointing signs (index), or other phonologically similar, but unrelated signs (e.g., the verb say) (n = 233; see de Souza Santos et al. 2025). The erroneous glossing can be attributed to similar formational properties between these different sign types, with coarticulation effects obscuring phonetic distinctions, and a need for additional research, as in the case of person.^[1] Going the other way, and for similar reasons of phonological similarity, we identified (mainly by chance) additional occurrences of pam that were annotated as other signs (e.g., index) in the corpus (n = 14). A total of 450 occurrences of pam were included in the final analysis.

4.3 Data coding

4.3.2 Verb type and modification

We first coded all verbs occurring in clauses with pam based on their ability to move through space to indicate arguments. Verbs that can move to indicate both subject and object arguments were coded as BOTH (e.g., the DGS signs ask, give; Figure 2a). Verbs that have a fixed initial location on the body and can be spatially modified only to indicate (non-first) object arguments were coded as AWAY (e.g., the DGS signs thank, say; Figure 2b). Verbs that are executed in neutral space, but whose location can be modified to agree with one argument, either the subject or the object, were coded as NEUTRAL-PLAIN (e.g., the DGS signs pay, break; Figure 2c). Body-anchored verbs that cannot be moved through or localized in space were coded as BODY-PLAIN (e.g., the DGS sign know, love; Figure 2d). Clauses with non-lexical, constructed action or predicates from the productive lexicon, including classifier predicates (Brentari and Padden 2001), were coded as PRODUCTIVE (Figure 3a), while single-sign predicates^[4] that represent quoted content were coded as QUOTATION (Figure 3b). Finally, we coded as NO-VERB clauses that did not have an explicit verb sign associated with pam (Figure 3c). This includes utterances with larger quotations (or constructed dialogue), coded as A3 (Section 4.3.1), when no verb was mentioned (Figure 4).

Figure 2

Example of DGS verbs coded as (a) BOTH: verbs that can move to indicate both subject and object arguments, as with ask (dgscorpus_nue_08 | 18-30f); (b) AWAY: verbs that move from a fixed initial location on the body to indicate (non-first) object arguments, as with thank (dgscorpus_koe_01 | 18-30f); (c) NEUTRAL-PLAIN: verbs produced in neutral space that can be modified to agree with one argument, as with break (dgscorpus_koe_02 | 46-60f); and (d) BODY-PLAIN: verbs that cannot be moved through or localized in space, as with love (dgscorpus_koe_16 | 46-60f).

Figure 3

Example of predicates coded as: (a) PRODUCTIVE: verbs that are not in the established lexicon (dgscorpus_mue_13 | 46-60m); (b) QUOTATION: single-sign predicates that represent quoted content (dgscorpus_goe_07 | 18-30f); (c) NO-VERB: for clauses with no explicit verb sign (dgscorpus_koe_13 || 61+m). In the example in (c), pam appears before the manual sign not and is accompanied by the mouthed verb verlassen (to leave someone) in a context meaning ‘I do not leave her (alone)’.

Figure 4

Example of quotation (constructed dialogue) as a complement argument of a verb of saying/telling, but without an explicit verb sign (dgscorpus_mst_10 | 46-60f).

For the verb types that could be modified to indicate arguments (i.e., BOTH, AWAY, NEUTRAL-PLAIN), we then coded whether they were in fact modified, and if so, how. We followed the coding scheme proposed in Fenlon et al. (2018). Verbs that were spatially modified for the object/patient or for both subject/agent and object/patient arguments were coded as ‘modified’. This was based on the use of locations in space that were associated with subject and/or object referents. Verbs that were not spatially modified in accordance with referent–location associations were coded as ‘unmodified’. The ‘congruent’ label was used to mark cases where the realization of the verb looked the same as the citation form of the sign, such that it was not possible to interpret the verb form as being clearly modified or not.

5.1 pam and constituent order

With respect to the use of pam and constituent order, we are particularly interested in the relationship between pam and the verb. Accordingly, we determined whether pam occurred preverbally or postverbally. We were also interested in understanding the relationship between preverbal and postverbal occurrences and clause-internal and clause-final patterns (as described by Bross 2020), as these categories do not overlap (footnote 3). In Figure 5, we see that the majority of occurrences of pam are postverbal (67% in total; n = 295), comprised of all clause-final orders (56%; n = 243) and about one-quarter (12%; n = 52) of clause-internal orders. The remaining clause-internal orders (33%; n = 142) make up the proportion of preverbal pam occurrences.

Figure 5

Proportion of clauses with pam occurring preverbally or postverbally.

Because Bross (2020) notes that the dominant pattern in his data elicitation with signers from the south of Germany is clause-internal, interpreted by Steinbach (2022) as corresponding to a dominant preverbal pattern, it is interesting to compare the distribution of preverbal vs postverbal as well as clause-internal and clause-final occurrences of pam across the different regions of data collection in the DGS corpus project (Figure 6). The difference between preverbal and postverbal pam does not seem to differ for regions in the south of Germany compared to other parts of Germany; certainly, the proportion of preverbal placement of pam is not higher in the southern areas of DGS corpus data collection. When we group cities into the major regions of North (SH, HB, HH), East (MVP, BER, LEI), West (GOE, MST, KOE, FRA), and South (NUE, STU, MUE) (following Macht 2016), we see a relatively even distribution of preverbal and postverbal occurrences of pam, as shown in Table 1. The use of preverbal pam, however, seems somewhat lower in the eastern regions of Berlin (BER) and Leipzig (LEI), but also in Bremen (HB), and comparatively high in Hamburg (HH). Note that this differs from findings by Macht (2016), whose analysis of DGS corpus data revealed a higher proportion of use of preverbal pam in the south of Germany (see also the summary in Steinbach 2022). When we look at the distribution of clause-final and clause-internal pam, we see that both structures are equally favored in the southern regions of STU (Stuttgart) and MUE (Munich), which correspond most closely to signer origin in the data analyzed by Bross (2020). This does not differ substantially from most other regions in Germany; however, the clause-final pattern seems more strongly favored in some regions, e.g., BER (Berlin) and LEI (Leipzig).

Figure 6

Distribution of preverbal and postverbal (in a) and clause-internal and clause-final (in c) occurrence of pam in 13 different regions of data collection across Germany, showing the acronym and the total number of occurrences of pam for each city/region and (in b) the location of each city/region within Germany. Key to the city/region acronyms: SH = Schleswig-Holstein; MVP = Mecklenburg-Vorpommern (city: Rostock); BER = Berlin; LEI = Leipzig; HB = (Hanseatic) Bremen; HH = (Hanseatic) Hamburg; GOE = Göttingen; MST = Münster; KOE = Köln (Cologne); FRA = Frankfurt; NUE = Nürnberg (Nuremberg); STU = Stuttgart; MUE = München (Munich).

We also looked at whether pam occurs adjacent to the verb or not. In both preverbal and postverbal positions, pam placement is predominantly immediately adjacent to the verb, i.e., right before or right after the verb, with only a small proportion occurring with intervening material (6% preverbal, n = 25; 3% postverbal, n = 14). Looking at pam and A2, taking into consideration only clauses with overt realization of A2 (n = 112), we also predominantly find adjacency between pam and A2 (78%, n = 87). In more than half of these cases (62%; n = 54), pam is also adjacent to V. This is nearly always a V-pam-A2 order, i.e., the postverbal, clause-internal (not clause-final) pattern, which makes up one-third of occurrences, as noted earlier. Finally, we looked at the position of A1 and whether A1 appeared in first or non-first position in each clause. The first thing to note is that close to half of all clauses did not have an overt A1 (43%; n = 199). When it was overt, A1 occurred in first position in the vast majority of cases (92%; n = 243). In only very few cases was A2 in first position, by means of either an overt nominal or a pronominal index form (3%; n = 13). There were, however, a considerable number of cases in which A2 was associated with first position by virtue of object marking with pam, i.e., where pam itself appeared in first position (16%; n = 75). We discuss argument omission again in Section 5.3.

5.2 pam and verb modification

In this subsection, we first present the distribution of different verb types that were coded (Figure 7). We see that body-anchored plain verbs (BODY-PLAIN), which cannot move through space to indicate arguments, represent the largest proportion of verbs (42%; n = 192). A similar proportion is represented by verbs that can move through or be located in space to indicate (or agree with) at least one of their arguments (i.e., BOTH, AWAY, and NEUTRAL-PLAIN verbs; 43%, n = 195). Of these, 19% represent verbs that can be spatially modified for both subject and object arguments (BOTH), while 9% represent verbs that allow movement only away from the body and whose movement can thus indicate only the object argument (AWAY). Verbs with no movement, but whose execution at a location in space can indicate (or agree with) an argument – either subject or object, but more typically the object – represent 15% of occurrences (NEUTRAL-PLAIN). The remaining verb types comprise the final 15% of clauses with pam (QUOTATION: 4%, n = 20; PRODUCTIVE: 6%, n = 30; NO-VERB: 5%, n = 25).

Figure 7

Distribution of verb types across clauses with pam.

When we look at the use of pam together with verbs that can be modified for arguments, we see that these verbs do also get modified in contexts of occurrence with pam. For BOTH verbs, which can be modified for both subject and object arguments, we find modification for both A1 and A2 arguments in about one-third of cases. In another one-third of cases, the verb form was coded as congruent with its citation form, such that it is difficult to say whether there is modification or not. For verbs that agree only with the object (AWAY verbs), there was a similar occurrence of clear modification for A2 in about one-third of cases.

We now turn to the spatial modification behavior of pam itself. Of particular interest here is any differential modification behavior of pam in preverbal or postverbal placement. In Figure 8, we see the proportion of modification of preverbal and postverbal pam for A1 and A2 arguments. Unsurprisingly, we see modification for A2 in the majority of cases for both preverbal and postverbal pam. For cases labeled as congruent, there is no independent spatial evidence for the spatial modification of the sign to indicate its arguments. If we collapse congruent with modified, however as also done by (Fenlon et al. 2018), there are only very few cases in which pam is not spatially inflected (i.e., unmodified). For the A1 argument, the picture is slightly different. Here, we see comparatively more cases of unmodified pam in postverbal position. For A1, occurrences of pam coded as congruent are more frequent; clear modification for A1 is more difficult to determine, especially for first-to-non-first arguments. If we consider the predominance of pam modification for A2, it is interesting to look specifically at the modification behavior for A1 in these cases and the relationship to preverbal and postverbal occurrence of pam, as shown in Table 2. When pam occurs preverbally (and is modified for A2), A1 is more likely to be modified (64%; n = 79).

Figure 8

Modification behavior of pam for A1 and A2 arguments by preverbal and postverbal occurrence.

When we look at the distribution of arguments in terms of person for A1 and A2 across clauses with pam, the majority of clauses were either first-to-non-first person (39%; n = 181) or non-first-to-first person (41%; n = 189). The remaining clauses had non-first-to-non-first-person arguments (20%; n = 92). In terms of the modification of pam for person, pam was most likely to be modified in cases of non-first-to-first person. Here, pam was modified for first-person A2 in all cases and for non-first-person A1 in 81% of cases (n = 153). For first-to-non-first clauses, pam was clearly modified for non-first A2 in over half of cases (59%; n = 106) and only clearly unmodified in 3% (n = 6) cases; otherwise, the coding was congruent for A2 and always congruent for first-person A1. For non-first-to-non-first clauses, spatial modification of pam for A2 was clearly apparent in a majority of cases (78%; n = 72) and only unmodified in very few cases (4%; n = 4); otherwise, it was congruent. For the non-first A1s here, however, pam was substantially more likely to be unmodified (52%; n = 48), with only 27% clear modification (n = 25). An overview of these results is given in Table 3.

5.3 pam and argument omission

In this subsection, we look at the omission of arguments in clauses in which pam occurred. Due in part to the prevalence of argument omission, there were a total of 29 different sign orders in the data. A very large proportion of utterances with pam occur without overt encoding of A2 with a nominal and/or pronominal index (77%; N = 346). Depending on the analysis of pam, it may itself count as an overt realization of A2. We return to this point in the discussion. The most frequently represented clause type was A1 V pam, with overt encoding of A1 and A2 marking only with pam (N = 112; 25%). Equally frequent was a clause consisting only of V pam, with omission of both A1 and A2 (N = 110; 24%).