Determining the distribution of implicit causality (IC) biases in American Sign Language (ASL) is necessary for future psycholinguistic studies. In addition, understanding implicit causality in a sign language is important given claims that the foundation for this phenomenon is similar across languages and cultures (Hartshorne et al., 2013). This is particularly so in the light of findings from past work in sign language linguistics, namely the suggestion that the visual-manual modality imposes certain restrictions on the mapping between verb semantics and syntax that do not exist in spoken languages (Edge & Herrmann, 1977; Kegl, 1990; Meir et al., 2007). Such restrictions might exert a strong influence on thematic roles and IC biases in ASL.
To discover how IC biases are distributed in ASL, we conducted a norming study with a large set of verbs. Experiment 1 provides a valence norming study, with the goal of defining a set of transitive ASL verbs (videos of the verb signs and transitivity ratings are freely available at the Open Science Framework: https://osf.io/bwjq2/). This set of verbs was used to build the stimuli for Experiment 2, which aimed to discover the direction of the verbs’ IC biases (resulting bias scores can also be found on the Open Science Framework: https://osf.io/bwjq2/). Finally, we analyzed whether thematic role predicts IC biases in a subset of the verbs from Experiment 2 and how this is affected by transitivity preferences and by the modality-specific phenomenon of body-anchoring.
Experiment 1
Participants, materials and procedure
Five Deaf ASL signers rated the acceptability of 292 ASL signs in transitive contexts. With the exception of one participant who learned ASL as young adult, all were native signers.Footnote 10 We constructed the list of 292 signs by using the 305 verbs from the stimuli used by Ferstl et al. (2011). We also added a number of verbs from other sources (Baker-Shenk & Cokely, 1980; Padden, 1988), as well as verbs that we knew to occur in ASL but that do not have (lexical) English translation equivalents. We then evaluated these verbs with the help of a Deaf native ASL consultant. The evaluation procedure was as follows: (1) we discarded verbs that did not have sign equivalents, including certain verbs of sound and hearing (e.g., ‘echoed’) that have little relevance in a visual language (Anderson & Reilly, 2002: 86); (2) we kept only one sign where multiple English verbs were glossed with the same ASL sign; and (3) for English verbs whose sign equivalents shared a manual form but differed in mouthing, we kept both signs. In order to avoid excluding potential verbs from norming, we kept all remaining signs. This included glosses that work as more than one part of speech in English (e.g., ‘wow’, ‘value’), even when we assumed that they would only function as one category in ASL. The verb sign videos are available at the Open Science Framework (https://osf.io/bwjq2/).
The rating procedure was as follows. We introduced pictures of two characters, named MAYA and LISA, and we asked the signers to imagine the two names as two referents in a (transitive) frame (i.e., MAYA VERB LISA). Leaving the pictures of the characters visible on the screen, we then showed video clips of the verbs from our list one at a time, creating transitive contexts with different verbs. The signers judged whether the verb was acceptable in the presented transitive frame and noted their answer in a booklet by ticking either a ‘yes’ or a ‘no’ box. The signers were informed that they could have more time for specific verb signs if they needed it. Some signers occasionally made use of this possibility. The task took 45–55 minutes.
Results
We calculated an acceptance score for each verb by summing the number of signers who found the verb to be acceptable in the provided context. A score of 5 indicates that all signers found the verb to be acceptable in a transitive context, while a score of 0 indicates that no signer found the verb acceptable in this context. The results (Table 1) show that 5.48% of the verbs were rejected by all signers, and 53.42% were accepted by all signers. Thus, the signers deemed the majority of the verbs fully acceptable.
Table 1 Number and proportion of verbs by acceptance score Verbs that were accepted by a majority of signers (that is, by at least three out of five signers) comprised 81.85% of the total. Of the verbs, 18.15% (n = 53) were deemed unacceptable by three or more signers. Table 2 shows which verbs were rejected by all five signers, by four signers, and by three signers.Footnote 11 The full verb distribution can be found at the Open Science Framework (https://osf.io/bwjq2/).
Table 2 Verbs rejected by all five, by four, and by three signers Experiment 2
We used the verbs from Experiment 1 that were deemed acceptable by at least three out of five signers as the basis for the stimuli in Experiment 2. For each of these verbs, we created a sentence fragment to be used in a sentence completion task intended to reveal the verbs’ biases.
Participants and procedure
Eight Deaf native signers (mean age: 31.63, range: 22–50) participated in the experiment. After giving informed consent and filling out a background questionnaire, the signers were shown a prerecorded experiment introduction video, which provided instructions in ASL. We gave each signer three practice trials and offered suggestions or clarification where necessary. When the signer felt comfortable with the demands of the task, we turned on the video camera to record their sentences, and they proceeded to the experiment proper, which lasted around one hour.
Materials
We created 239 sentence fragments from the verbs that were found to be acceptable transitives in Experiment 1. The sentence fragments were of the type ‘NP V NP WHY? …’Footnote 12, which is the ASL equivalent of the English ‘NP V NP because … ‘, e.g., ‘#THOM LOVE #LEAH WHY? …’ (Fig. 3).Footnote 13
The English ‘Name Verb Name because …’ sentence frame has been used extensively in research of spoken language verb biases (Garvey & Caramazza, 1974; Koornneef & van Berkum, 2006; Stevenson et al., 1994, among others). In Experiment 2, each NP consisted of a name drawn from a pool of 49 names. Each name consisted of four letters fingerspelled using the hand alphabet. The sentence fragments were presented in randomized order. The participants’ task was to use the sentence fragments as context for free sentence continuations. The instructions were to watch the fragments and complete the sentence with the first continuation that came to mind.
Sentence completion tasks have traditionally been conducted in writing. This has the advantage of giving participants the option of revisiting the context sentence while composing their answer. This ensures that the sentence context remains accessible to participants. Because ASL has no widely accepted written form, the possibility of conducting the experiment in writing was not available to us. We opted instead to ensure participants’ familiarity with the context sentence by asking them to repeat it before providing their own continuation.
Coding and evaluation
After collection, a Deaf, native signer coded the participants’ responses for next-mention, that is, which referent they mentioned as the subject in their free continuation. Next-mention was coded as NP1, NP2, Both, Unclear, or Other (see Examples (3)–(7) below. The next-mention is shown in bold).
-
(7)
NP1:
#LOLA-R APPROACH #GALE WHY? IX-R NOTICE MANY THINGS CONFLICT APPROACH DISCUSS RESOLVE
‘Lolai approached Galej, because shei noticed a lot of conflicts, so shei approached (herj) to discuss and resolve (them)’
-
(8)
NP2:
#KATE ASSIST IX-R #ANDY WHY? IX-R BROKE #LEG CRIPPLE
‘Kate assisted Andy, because he broke his leg and was unable to walk’
-
(9)
Both:
#SAUL BEGUILE #ROSS IX-L WHY? BOTH WANT BRING WATCH #CONCERT MUSIC PERFORMANCE
‘Saul beguiled Ross, because they both wanted to go watch a concert, a music performance.
-
(10)
Unclear:
#TYRA ATTRACT LENA WHY? Ø REALLY LIKE IX-L
‘Tyra attracts/is attracted to Lena, because she really likes her’
-
(11)
Other:
#NOAH SCARE #EARL WHY? #ITS HALLOWEEN COSTUME SCARE
‘Noah scares Earl, because it’s Halloween, he puts on a costume and scares him
Trials that were accidentally skipped and those where the participant did not provide a meaningful answer (e.g., responded with ‘I don’t know’) were discarded (n = 38). We also discarded trials where the participant had understood the verb incorrectly (n = 52).Footnote 14 These two categories together amounted to 5% of the total trials.
Independently, another Deaf, native signer coded 50% of the data (four full subjects). The inter-coder agreement was found to be kappa = 0.916. Differences were resolved through discussion where possible; otherwise, the first coder’s choice was retained. The coders were instructed to be conservative and code potential ambiguities as ‘unclear’.
Results
We first provide an overview of how participants’ responses were distributed (Table 3). As mentioned above, 5% of trials were discarded from further analysis because they did not provide a meaningful response or contained an incorrect verb meaning. In addition, 40 responses were coded as ‘Unclear’, 135 responses as ‘Both’, and 80 responses as ‘Other’. Responses with these three labels were also excluded from further analysis, amounting to an additional 13% of the total data set. As shown in Table 3, the remaining responses were categorized as NP1-biased (38%) and NP2-biased (44%). A statistical analysis showed a trend towards more NP2 continuations than NP1 continuations (t(238) = −1.85, p = 0.065).
Table 3 Coding of responses We next provide an overview of the biases for individual verbs by calculating a bias score (100*(no.NP1 − no.NP2)/( no.NP1+ no.NP2)) for each verb, collapsing across subject variation.Footnote 15 Across verbs, the calculated bias scores covered the full range of values (M = −7.35, SD = 61.27, range = −100 to +100). Similar to the pattern in the continuations overall, the negative mean bias score suggests a preference towards NP2-biased verbs over NP1-biased verbs.
Table 4 bins the bias scores by strength and shows the number of verbs in each bin, excluding verbs for which we had fewer than five responses (n = 26). Overall, fewer verbs were biased towards NP1 (41%, n = 88) than towards NP2 (54%, n = 116). Moreover, fewer verbs were strongly biased towards NP1 (n = 34) than were strongly biased towards NP2 (n = 55), based on bias scores of above 50 and below −50. Taken together, these results show converging tendencies pointing towards a stronger presence of NP2-biases in the present data set.
Table 4 Bias score distribution How verb semantics influences implicit causality in ASL
While Experiment 2 suggested that there may be a relatively low proportion of NP1-biased continuations in ASL, it is not clear why this is. It is possible that the observed distribution of IC biases in ASL indicates that thematic roles do not predict IC biases in the manual modality as they do in the spoken modality, possibly because of body-anchoring, and possibly because the relationship between semantic structure and IC bias is still developing in a language as relatively young as ASL. However, given previous research, another likely explanation for the low proportion of NP1-biased verbs is a dispreference in ASL for lexicalizing verbs with stimulus-experiencer structure.
To assess these possibilities, we first asked whether verbs that may have stimulus-experiencer structure in other languages are lexicalized as such in American Sign Language (ASL). We focused on the verbs in our stimuli for which the glosses are classified as stimulus-experiencer verbs in Ferstl et al. (2011).Footnote 16 We first examined whether these verbs are more likely to be intransitive, rather than transitive, as compared to verbs from other categories. We did this by analyzing the verbs that were rejected by the majority of signers in Experiment 1. This analysis shows that 55% (n = 53) of the rejected verbs had glosses that were classified as stimulus-experiencer by Ferstl and colleagues. Nevertheless, 58% of all verbs with stimulus-experiencer glosses (n = 69) were judged to be acceptable by three or more signers. As these verbs were included in Experiment 2, we can examine the distribution of their bias scores. This examination helps us assess whether these verbs were consistently biased in one direction over the other. The analysis revealed bias scores distributed on a continuum from strongly NP2-biased (scores of −100 to −66: CHEER, COMFORT, ANNOY, AMAZE, DISGUST, INTEREST, AMUSE, FASCINATE, ENTICE), to not clearly biased (scores of −25 to 25: e.g., ENCOURAGE, TEMPT, SCARE, CONFUSE), to strongly NP1-biased (scores of 100 to 66: ATTRACT, HARASS, INSULT, CHARM, FLATTER, EMBARRASS, PAIN).
There is an even distribution of NP2-biased (n = 15) and NP1-biased (n = 16) verbs (Fig. 4). The fact that the majority of these verbs show clear biases suggests overall agreement among signers about the bias of individual verbs rather than a pattern based on ideolectal differences. On the other hand, the distribution encompasses both types of biases, suggesting that this group of verbs is not homogenous in ASL.
We next used the sentence contexts produced by the signers to determine the thematic roles most frequently occurring with each of the potential stimulus-experiencer verbs (Table 5). We used the transcription of each response (excluding those labeled as N/A or wrong verb) to categorize the thematic roles occurring with the verb as AgP, StimExp, ExpStim, Unclear, or AgP/StimExp (ambiguous between Agent-Patient and Stimulus-Experiencer). Aggregating over signers, we then assessed which thematic role constellation was most frequent for each verb. If there was no one constellation used by a majority of signers, we coded the predominant thematic role as Unclear. This analysis revealed that the split in the bias direction within this category of verbs is largely accompanied by a split in thematic structure. Specifically, as a group the NP1-biased verbs in the current data set are more frequently lexicalized with stimulus-experiencer structure compared to other structures. Eight out of 16 verbs occurred more often with stimulus-experiencer structure than with any other structure, including ‘INSULT’, which was ambiguous between stimulus-experiencer and agent-patient structure.
Table 5 Biases and thematic structures in the stimulus-experiencer category The reverse was true for the NP2-biased verbs, which were most frequently lexicalized with experiencer-stimulus structure in the current data. In this group, eight out of 15 verbs occurred with experiencer-stimulus structure more frequently than they occurred with any other structure. Despite their glosses suggesting that they should be stimulus-experiencer verbs, some verbs, such as ‘ANNOY’ and ‘AMAZE’, in fact had experiencer-stimulus structure. This suggests that when they are used in transitive contexts (‘MAYA ANNOY LISA’), they are understood along the lines of ‘Maya is annoyed with/by Lisa’ rather than ‘Maya annoys Lisa’ (Example (12), Fig. 5). In sum, in this group of potential stimulus-experiencer verbs, those that are biased towards NP2 tend to also have experiencer-stimulus structure. This is in line with previous research findings, namely that experiencer-stimulus verbs tend to be NP2-biased. Consequently, although the direction of bias is unexpected compared with the English verb matching the gloss, the direction of the bias in these ASL verbs is nevertheless predictable based on thematic structure.
Within the NP2-biased verbs as well as the NP1-biased verbs, we also find examples with neither stimulus-experiencer nor experiencer-stimulus structure. In the NP2-biased group, some signers used the first NP in a more agentive role instead, such that the verbs are best interpreted as agent-patient verbs (Example (13)).
-
(12)
#DANA ANNOY #JADE WHY? #JADE ALWAYS COMPLAIN ANNOY
‘Dana is annoyed with Jade, because Jade always complains. [Dana is] annoyed’
-
(13)
#TROY CHEER #NORA WHY? IX-R SCORE
‘Troy cheered Nora, because she scored’
Although stimulus-experiencer occurred as the predominant role more often than did any other thematic structure, we also found a number of other structures for NP1-biased verbs. Importantly, many involved the agent-patient thematic roles. Some structures were ambiguous between agent-patient and stimulus-experiencer structures. In other instances, a verb that is lexicalized as a stimulus-experiencer verb in other languages was clearly used as an agent-patient verb in this sample of ASL. This is similar to what Kegl (1990) reported for ‘SCARE’. In the examples below, the subject is an agent intentionally causing something to happen to the patient in (14), and the subject is a potentially unwitting stimulus for an emotion in the experiencer in (15).
-
(14)
#MARA HURT #GALE WHY #MARA MAD HIT HIT #GALE
‘Mara hurt Gale, because Mara got angry and hit Gale’
-
(15)
#MARA HURT #GALE WHY IX-L DECIDE CUT DONT-WANT
IX-R IN POSS-L LIFE ANY MORE
‘Mara hurt Gale, because she decided to cut her off, she didn’t want her in her life anymore’
All told, there are 16 actual stimulus-experiencer verbs within the group of verbs with stimulus-experiencer glosses (including six with ambiguous stimulus-experiencer/agent-patient structure). This corresponds to 7.5% of the 213 verbs in Experiment 2 for which we calculated bias scores. This suggests a low proportion of stimulus-experiencer verbs in ASL overall and begs the question whether this is a result of body-anchoring. In the final analysis, we therefore investigated whether certain thematic roles were more or less likely to occur in body-anchored verbs in the current data set.
Table 6 shows how body-anchoring interacts with bias and actual thematic role in the potential stimulus-experiencer verbs in the current study. These data do not suggest that ASL verb bias is a function of body-anchoring. All experiencer-stimulus verbs are body-anchored, and the majority of them are NP2-biased (8/11). Most stimulus-experiencer verbs are body-anchored as well, and most of the body-anchored verbs are also NP1-biased (5/7). Thus, there is no evidence in the present data set that the experiencer is the preferred referent in continuations of sentences containing body-anchored verbs.
Table 6 Body-anchoring and thematic roles within the assumed stimulus-experiencer category The distribution of verbs in Table 6 also does not suggest a strong role for body-anchoring in determining whether verbs are lexicalized with stimulus-experiencer or experiencer-stimulus structure. If body-anchoring required the experiencer to appear as the first argument, we would not expect to find body-anchored verbs with stimulus-experiencer structure.
However, we do find such verbs (e.g., ‘DISAPPOINT’, ‘EMBARRASS’, and ‘SHOCK’, Figs. 6, 7 and 8). As Table 6 shows, eight verbs were body-anchored out of the 16 verbs that had stimulus-experiencer structure or were ambiguous between stimulus-experiencer and agent-patient structure. While all of the verbs that have experiencer-stimulus structure are in fact body-anchored (11/11), nearly all of the actual stimulus-experiencer verbs are body-anchored as well (7/10). Therefore, the results of the present study do not support the idea that the phonological feature of body-anchoring in some signs is the primary predictor in ASL for whether a verb will be lexicalized as a stimulus-experiencer verb or not.
On the other hand, the verbs which are ambiguous between an agent-patient and a stimulus-experiencer reading are not body-anchored. The pattern of stimulus-experiencer and experiencer-stimulus verbs being mostly body-anchored suggests a correlation between potential psychological verbs being articulated on the body and being interpreted as involving experiences or emotions rather than actions.Footnote 17