Argument Marking in ASL

ASL uses both word order and morphological marking of arguments to indicate subject and object. Morphological marking of arguments is limited to one class of verbs, known as agreeing or indicating verbs (Padden, Reference Padden1988). In morphological marking on agreement verbs, the movement direction or palm orientation of the verb indicates the subject- and object-hood of the nouns; in this case, word order can deviate from the canonical SVO order used in ASL. ASL additionally marks arguments through the use of word order. For the class of plain verbs (Padden, Reference Padden1988), which cannot be modified with morphological marking described above, arguments are primarily indicated via word order. Sign languages, like the majority of spoken languages, are classified as having basic word order of either SVO or SOV (Napoli & Sutton-Spence, Reference Napoli and Sutton-Spence2014). ASL uses the basic word order of SVO (Coons, Reference Coons2022; Fischer, Reference Fischer and Li1975; Liddell, Reference Liddell2003). In example (1), taken from Fischer (Reference Fischer and Li1975), arguments are disambiguated by their order of occurrence in the sentence. The man is the subject because this sign occurs in the syntactic location of the subject while the child is in the syntactic position of object.

In ASL, the basic word order of SVO is the only word order allowed in sentences with reversible arguments in the absence of explicit grammatical marking (Fischer, Reference Fischer and Li1975). In addition to the morphological marking of arguments on agreeing verbs described above, alternative word orders of O,SV and OV,S (comma indicates a prosodic break) are formed through topicalization (Liddell, Reference Liddell1980). Grammatical nonmanual marking of word order variations include prosodic breaks, grammatical facial expressions, head tilts or shakes, and shoulder or body movements. Topicalization allows for fronting of the object or verb phrase with nonmanual marking of the raised arguments.

Thus, the use of SVO to indicate which argument is the Subject and which is the Object in dual argument reversible structures provides evidence of abstract rule learning, the gateway to learning complex sentence structure.

Childhood Acquisition of Word Order

Young children typically acquire the language of their environment, ultimately matching adult-like use of grammatical structures including argument marking. Preferential looking studies have shown that children are sensitive to basic word order of their native language in comprehension as young as 17 months (Naigles et al., Reference Naigles, Gleitman and Gleitman1993). Their production of basic word order occurs as soon as they utter full sentences. This is true for hearing children learning spoken language (Braine, Reference Braine1963; Bloom & Capatides, Reference Bloom and Capatides1993; Tomasello, Reference Tomasello1992), and importantly for the present study, for deaf children learning sign language (Coerts, Reference Coerts, Chamberlain, Morford and Mayberry2000; Pichler, Reference Pichler2001; Pizzio, Reference Pizzio2013).

As described above, basic word order is a key building block in the later acquisition of more complex syntactic forms (Diessel, Reference Diessel2004; Slobin & Bever, Reference Slobin and Bever1982). It has been argued that children’s acquisition of word order is facilitated by pragmatic and semantic factors, such as animacy and real-world knowledge. Theories impute different underlying mechanisms by which children transition from their initial understanding of individual word meanings associated with concrete objects to the use of abstract word order and morphosyntactic marking. But most theories posit that linguistic input is mapped onto the basic categories of syntactic structure through language experience (Bates & MacWhinney, Reference Bates, MacWhinney, Gleitman and Wanner1982; Pinker, Reference Pinker1996; Tomasello, Reference Tomasello2003). The competition model (Bates & MacWhinney, Reference Bates and MacWhinney1989) highlights the fact that, in most cases, the word order and morphosyntactic marking of arguments is redundant with semantic and pragmatic cues. For example, in sentences such as the ‘the boy kicks the ball,’ knowledge about the real world and whether animate or inanimate things tend to be kickers can result in comprehension without reliance on word order or morphosyntactic markers. These cues remain available to adults. In fact, when word order and morphological marking of arguments is artificially stripped from sentences in an experimental setting, adults rely on such cues and can correctly interpret the meaning of most event types (Mahowald et al., Reference Mahowald, Diachek, Gibson, Fedorenko and Futrell2022). The fact that in many instances word order and morphosyntactic marking are redundant with semantic and pragmatic cues can facilitate the mapping of these cues onto the word order and morphosyntax rules of a language.

But semantic and pragmatic cues can only take the language user so far because they are unavailable in reversible SVO sentences such as ‘the boy hits the girl.’ The only way to know which argument, the boy or the girl, did the act, the hitting, is to know the word order rule. In this way, the comprehension and production of sentences with two reversible arguments provides stronger evidence for abstract rule learning than sentences with irreversible arguments.

Some studies use scenarios where semantic or pragmatic cues conflict with word order and morphosyntactic marking to determine which cues the children rely more on when dealing with two argument structures. Young children have been shown to rely more on event knowledge (Chan et al., Reference Chan, Lieven and Tomasello2009; Strohner & Nelson, Reference Strohner and Nelson1974), discourse context (Hargrove & Panagos, Reference Hargrove and Panagos1982), and animacy (Chapman & Kohn, Reference Chapman and Kohn1978; Dodson & Tomasello, Reference Dodson and Tomasello1998) than on the morphosyntactic or word order markings of their language in these comprehension paradigms. However, by age six or seven, children who have experienced language from birth no longer depend on these cues and instead produce sentences with grammatical word order (Braine, Reference Braine1963; Bloom & Capatides, Reference Bloom and Capatides1993; Tomasello, Reference Tomasello1992; Coerts, Reference Coerts, Chamberlain, Morford and Mayberry2000; Pichler, Reference Pichler2001; Pizzio, Reference Pizzio2013) and use morphosyntactic and word order cues in comprehension (Chan et al., Reference Chan, Lieven and Tomasello2009; Chapman & Kohn, Reference Chapman and Kohn1978; Dodson & Tomasello, Reference Dodson and Tomasello1998; Hargrove & Panagos, Reference Hargrove and Panagos1982; Strohner & Nelson, Reference Strohner and Nelson1974).

Current Study

Previous studies of early word combinations produced by late first language learners provide evidence that these individuals are developing sentence structure (Berk & Lillo-Martin, Reference Berk and Lillo-Martin2012; Ferjan Ramírez et al., Reference Ferjan Ramírez, Lieberman and Mayberry2013; Morford, Reference Morford2003). However, recent experiments suggest that the acquisition of two argument structures in ASL may be limited when language acquisition begins after early childhood (Mayberry et al., Reference Mayberry, Hatrak, Ilkbasaran, Cheng, Huang and Hall2023). To test this hypothesis, we examined the sign language productions of eleven LL1 signers of ASL in response to pictured line drawings of three types of events: intransitive, irreversible transitive, and reversible transitive events. We predict that LL1 signers would master single argument clauses prior to dual argument clauses. For dual argument clauses, then, if the LL1 signers have mastered the syntactic word order rule, they should use it for both irreversible and reversible events. And if LL1 signers have acquired basic word order in ASL, then their argument ordering should pattern with that of the native signers. Alternatively, if the late L1 learners have not learned basic ASL clausal structure, then they should not pattern with the native signers. This would suggest that there is a sensitive or critical period for the learning of basic clause structure ending after early childhood.

Testing Procedure

Participants were tested on a computer by either a deaf native or a hearing near-native ASL signer. As described above, the elicited production task was interleaved with an ASL elicited imitation task in an effort to prime sentence structure – that is, to provide multiple exposures to the target structure to encourage the participant to produce it. A total of 56 elicited production and 56 elicited imitation sentences were collected for each participant. During the instructions for each block, participants viewed a picture which was a line drawing paired with a video clip of a model sentence using the targeted structure. After instructions, participants saw a picture on a computer screen and described it – elicited production. Second, they saw a video of an ASL sentence using the target structure and repeated it – elicited imitation. Third, participants signed their description of the next picture designed to elicit the target structure and, fourth, repeated the stimulus ASL sentence describing the picture, again using the same target sentence but with different lexical items. This sequence of elicited production and imitation was repeated four times within each block. By the fourth elicited production trial in a block the participant had a total of four previous exposures to the syntactic structure in ASL (one model sentence in the instructions and three previous elicited imitation trials) along with an associated picture designed to depict the structure (Figure 1). Data for the current study were taken from the elicited production task. The order of the pictures within each block was randomized across participants. The order of blocks was also randomized.

Figure 1. Sample of elicitation procedure for SVO syntactic block.

Coding Procedures

Stimulus picture selection. In order to select stimulus pictures depicting simple events with one or two arguments from the elicited production stimulus set, all of the pictures were coded according to the semantic relationships of the elements in each picture. For example, a picture showing a boy pulling a woman who was sitting in a wagon was coded for having an actor (transitive), patient, and instrument. A picture of a girl sleeping in bed was coded as having an actor (intransitive) and location. To test the hypothesis under investigation, we selected pictures consisting of three event types: intransitive, irreversible transitive, and reversible transitive. Hence, stimulus pictures depicting more than one event were excluded from the analysis, as were pictures illustrating reciprocal action. The signed productions were elicited by 20 pictured events: 8 intransitive, 12 transitive (9 reversible events and 3 irreversible). The uneven number of items per event type was due to the fact that larger study was designed to elicit a variety of syntactic structures in ASL so that event type was not controlled.

Coding of participants’ signed productions. The participants’ signed responses were coded by a native or near-native signer using ELAN. Each sign that the participant produced was glossed into English. In addition, nonmanual and prosodic marking was indicated in the gloss. After this initial coding, each sign in the response was additionally coded for whether it referred to the actor, patient, recipient, theme, or act of the pictured event. The sequence of signs was analyzed to discover ordering patterns as a function of event condition.

The participants’ response to each stimulus picture was then coded using the semantic relationships from the pictures, following Coppola (Reference Coppola2002), Coppola and Newport (Reference Coppola and Newport2005), and Neveu (Reference Neveu2016). The act from the event was also coded. The coding categories included:

Some participants included more than one act in their responses. Therefore, acts were additionally coded for who was doing the action. For example, act(a) is an act completed by the actor from the picture while act(p) is an act completed by the patient from the picture. Because the events were shown in the stimuli, all responses were coded according to the roles of the arguments in the picture. We did not attempt to abstract the semantic roles from the signing of the participants but used the semantic roles from the event in the picture to analyze each participant’s signed response. In order to differentiate repetitions of the same act from additional acts, if a second or additional act was signed that was not a repetition, it was coded with a number such as act2 or act3. This coding can be seen in example (2) which was a response to a drawing of a woman kissing a man on the cheek and example (3) which was a response to a drawing of a man pulling a woman in a wagon. Note that in example (2), the second act is also marked as act(p) because it is completed by the patient from the picture. This will be discussed in the results section in more detail because it breaks the event into two clauses, one describing the action of the actor and one describing the action of the patient.

Similarly, when a property was included in the response, it was coded as describing the actor or patient with the same notation ‘(a)’ or ‘(p)’ as acts. In example (4), the LL1 signer produced signs coded as properties for both the actor and the patient. The property referring to the woman who was the patient is marked with (p) and that referring to the actor is marked with (a).

In order to determine the order of arguments, each production was coded for the sequential order of the signs using the coded categories described above. Therefore, for intransitive sentences the coding was act-actor, actor-act, or unable to be determined (in cases where the act or actor was repeated such as act-actor-act). With transitive events, the order was coded over all three elements present in the picture (act, actor and patient or theme). When each element was signed one time in a sentence, this would lead to six possible orders. In addition, each production was coded for any omissions of arguments that were present in the picture and for repetitions of any arguments. The order of arguments was then determined for the elements explicitly signed. In example (5), which is a description of the picture of a girl eating a cookie, the signer omits the sign for cookie. Therefore, the order of arguments, for example (5), is actor-act (omit theme).

Intransitive Events

The groups produced mostly actor-act order in response to drawings of intransitive events. Nonetheless, there was a statistically significant association between the signer group and the distribution of word orders, X ²(1) = 29.72, p < 0.001. Effect size calculated using Cramer’s V, was 0.43, which is a medium to large effect. Specifically, all the native signers produced only the argument order of actor-act. The LL1 signers also tended to produce this order (68.75% of all productions) but they also produced other orders, as shown in Table 2. These other orders included few (2.5%) act-actor. The remainder of the productions either repeated or omitted the actor or act so that they could not be reliably judged as either actor-act or act-actor.

Table 2. Argument order for intransitive events, percent (count)

Transitive Irreversible Events

For transitive events which were not reversible, there was again a significant association between the signer group and the distribution of word orders, X ²(3) = 12.63, p = 0.006. The effect size calculated using Cramer’s V was 0.5 and is a large effect, indicating a strong association between group and ordering patterns. These events all included an animate actor and an inanimate theme. The most commonly produced pattern for both groups was actor-act (omit theme), as in example (5), repeated from above, given by a LL1 participant as a response to a drawing of a girl eating a cookie.

The other order produced by native signers was actor-act- theme. While this order was produced by the LL1 group, it was not the second most common order, unlike for the native signers.

The second most common order of arguments for transitive irreversible events among the LL1 group was actor-theme-act, which accounted for 20.7% of all responses, but was never produced by native signers. Other orders produced by LL1 signers, but not by native signers, included theme-act-actor; theme-act (omit actor); multiple clauses, and indeterminate. The rate of production of argument orders is shown in Table 3.

Table 3. Argument orders for transitive irreversible events, percent (count)

Thus, when native signers produced overt signs for all of the arguments in irreversible transitive events, they only produced actor-act-theme order. However, when LL1 signers produced overt signs for all arguments in the event, they produced a variety of orders of arguments. This was not only true for group level analysis: individual participants were not consistent in their argument ordering. Examples (6) and (7) were both produced by the same LL1 participant as responses to drawings of a girl eating a cookie and a boy reading a book.

Thus, when only one argument was produced by the native signers and LL1 signers, they produced the order actor-act, as in intransitive events. However, when all arguments of the event were produced, the native signers all produced the same order, actor-act-theme, while the LL1 group used a variety of argument orders.

Transitive Reversible Events

For reversible transitive events, there was again a significant association between group and argument ordering distribution X ²(3) = 28.07, p < 0.001. Effect size calculated using Cramer’s V was 0.34, which is a medium effect, indicating a moderate association between the variables. These pictured events all included two animate arguments (actor and patient). The most common ordering pattern for both groups was actor-act-patient as seen in example (8), which was given as a response to a picture of a man kissing a woman on the cheek.

This order of arguments aligns with the basic SVO word order of ASL and with the order used by native signers for irreversible events when both arguments were produced (actor-act- theme). Interestingly, neither native signers nor LL1 signers tended to drop the patient in these events although both groups tended to drop the theme in irreversible events. The full list of argument orders for transitive reversible events is found in Table 4.

Table 4. Argument order for transitive reversible events, percent (count)

The second most common production order for native signers was patient-actor-act, as seen in example (9). Topicalization of object in ASL produces O,SV word order with obligatory nonmanual markers including the prosodic break symbolized here by a comma (Liddell, Reference Liddell2003). When these patient-actor-act sentences were produced by LL1, they did not include the obligatory nonmanual or prosodic markers for topicalization.

The second most common order of arguments for the LL1 group was assigning an action to each animate argument. This pattern was seen in both groups, though more often in the LL1 signers. Since some participants did not produce consistent prosodic cues that could be used to determine clausal boundaries, responses were coded as having multiple clauses when an act was produced for both the actor and patient for the primary event pictured. As seen in example (10), the participant produced more than one clause with an act for both the actor and patient of the original event, which was a woman pulling a boy in a wagon.

LL1 signers also produced other argument orders that were produced once or not at all by any native signer, including: actor-patient-act; patient-act (omit actor); patient-act-actor; actor-act (omit patient); and Indeterminate order.

To summarize, across all event types, there was a significant association between the order of arguments produced and the signer group. Although the LL1 signers did produce orders that were produced by the native signers, they did so at different rates. In addition, they produced argument orders that were not present in the signing of the native group. Both groups tended to produce actor-act order when only one argument was produced; however, the groups differed in their argument ordering when two arguments were produced. Native signers overwhelmingly produced actor-act-theme and actor-act-patient orders both of which correspond to ASL basic word order of SVO. By contrast, LL1 signers showed more variety in their argument ordering: the most common order for sentences containing both overt actor and overt theme was actor-theme-act order while the most common order for reversible events was actor-act-patient.

Influence of Syntactic Context

While some similarity in argument ordering was observed in the native and LL1 signers, there were significant differences between the groups for the argument ordering as a function of event type. We next examined the factors that can potentially influence the variation in argument ordering. As previously explained, the data for the current analysis were taken from a larger study which was organized into syntactic blocks. For each block, participants saw a picture and described it. Next, they did an elicited imitation task, repeating a sentence about the picture they were seeing signed by a native signer. Recall that the pictures used here were selected based on analysis of the pictured elements to include only one event and one actor; therefore, the pictures appeared in a variety of syntactic blocks shown in Table 5. Thus, we initially looked at whether the participant’s production during the description task matched the syntactic block of the larger experiment, which was determined by the imitation task. For example, when the syntactic block was SVO order, the question is whether the participants also produced SVO order. Similarly, when the syntactic block was a topic-comment construction, did they produce O,SV order? To answer the question, we calculated Fisher’s exact test to determine if there was a significant association between group (native or LL1) and the number of signed productions which conformed or not to the structure of the syntactic block. As expected, the native signer group was significantly associated (Fisher’s exact test: p<0.001) with the production of utterances matching the syntactic block, producing matching sentence structures in 80.5% of all responses. By contrast, the LL1 group matched the syntactic structure in 31.8% of their responses, despite being exposed to the target structure four successive times.

Table 5. Number of pictures from various syntactic blocks

Individual Patterns

Due to the variation in argument orders produced by the participants for transitive events, we used Fisher’s exact test to determine if the variation in orders was equally distributed among the participants or if each participant tended to use some orders more than others. For each group and event type, matrices were created with each participant as a row and each argument ordering as a column. Cells were filled with the number of times the participant produced that argument ordering. For native signers, the results show independence between individual signers and argument ordering for irreversible (p=0.916) and reversible (p=0.08) transitive events. This suggests that the native signers were using grammatical structures similar to one another in response to the task because the variation in orders of arguments is equally distributed across participants. For LL1 signers, there was also evidence for equal distribution of argument orders for irreversible events (p=0.50). However, for reversible events, certain patterns were significantly more likely for some signers (p< 0.001). For example, half of the responses using actor-act-patient order came from just 3 of the 11 LL1 signers, while some LL1 signers never produced this order of arguments. Similarly, over half (52.4%) of productions of multiple clauses for reversible events were produced by 2 of the 11 signers, and this order was never used by 3 of the signers. These results indicate that the argument ordering of LL1 signers for single and dual argument events converge neither with one another nor with the ordering of arguments in basic clauses in the language they are learning, ASL.