2. General method

2.1. Item selection

We first identified a list of 104 concrete noun–verb pairs in ASL. Of these pairs, 72 followed the movement reduplication + size rule, where verbs are articulated with a single movement in a large signing space, and nouns are articulated with a reduplicated movement in a reduced signing space (Fig. 1). The remaining 32 followed the movement size rule, where both verbs and nouns are produced with reduplicated movement, but nouns are articulated in a smaller signing space relative to verbs. We coded whether signs were pantomimic or not (Fig. 1). We operationalized pantomimic signs as being handling signs or conventional pantomimes that involve manipulating objects (e.g., holding a Y handshape for telephone/to_call). Fifty-eight of the pairs included pantomimic iconicity and 46 did not.

We then identified separate black and white line drawings of images that could depict either the object or the action associated with the object for each of the noun–verb pairs (see Fig. 2 for an example of images depicting a chair and sitting on a chair). Using 3 batches of pictures we asked Amazon Mechanical Turk (mTurk) master workers (Batch 1: n = 12, Batch 2: n = 15, Batch 3: n = 10) to rate individual line drawings on how object-like or action-like the image was, using a scale of 1 (object) to 7 (action). Workers were paid in accordance with US national minimum wage ($7.25/hour) and did not participate in any subsequent experiments reported in this paper. We selected pictures that were rated more object-like (average rating of 2 or lower) to pair with nouns and pictures that were rated more action-like (average rating of 6 or higher) to pair with verbs. We were able to identify a subset of 61 noun–verb pairs in ASL for which we had good line drawings to represent both the object and actions. We used items from this subset of noun–verb pairs for all experiments (see Appendix). The picture and video stimuli are available on OSF (https://osf.io/q3hjg/?view_only=55c01f66aace43dc89555297de1359fb).

Fig. 2. Images of the two methods of assessing sensitivity to the movements that distinguish nouns from verbs (A) match-to-sample (Experiment 1a) and (B) forced-matching (Experiments 1b-2c).The arrows in (B) denote the path from the image to the correct location below the video.

3. Experiment 1: nonpantomimic signs

3.1. Experiment 1a: testing the sensitivity of the movement reduplication + size rule with a match-to-sample design

3.1.1. Method

Participants

Thirty-seven of the 40 recruited MTurk workers completed all trials and passed all checks. The final sample included 22 males and 15 females (27 White, 4 Black or African American, 3 Asian, 1 Hispanic or Latino, 1 mixed-race, and 1 no answer). We also recruited 14 ASL signers (4 M and 10 F; 9 native signers and 5 early signers) through personal networks and snowball sampling. The deaf ASL signers included 12 White, 1 Asian, and 1 American Indian. The deaf participants were entered into a lottery to win one of two $50 gift cards.

Design and procedure

We selected 20 noun–verb pairs that had no pantomimic iconicity and that followed the movement reduplication + size rule for inclusion in the study (see Appendix). In a within-subjects, match-to-sample paradigm presented in Qualtrics, participants saw a randomized presentation of either the noun sign (n = 10) or the verb sign (n = 10) from the noun–verb minimal pair, alongside the line-drawn images of both the related object and action referent (Fig. 2a). Placement of the line drawings, was randomized across trials. Participants were instructed to select the picture that best illustrated the sign. We generated two counterbalanced orders of administration such that if the noun sign of the minimal pair was presented in one order, then its verb counterpart was presented in the other, and vice versa. Participants were randomly assigned to complete one of the two orders of administration (20 nonsigners and 7 signers completed one order, and 17 nonsigners and 7 signers completed the second order; the three nonsigners who did not complete all trials or pass the attention checks happened to be randomly assigned to the second order). It took participants on average 7 minutes to complete this study.

3.1.2. Results and discussion

We ran a mixed-effects logistic regression to examine participants’ success in matching object pictures with noun signs and action pictures with verb signs. We included item and participants as random intercepts, and lexical class, signing status, and the interaction between lexical class and signing status as fixed effects. All fixed factors were sum coded (lexical class: noun = −1, verb = +1; signing status: nonsigner = −1, signer = +1). We observed significant main effects of lexical class and signing status, and a significant interaction between signing status and lexical class (Table 1 and Fig. 3). The pattern of the data illustrates that nonsigners correctly matched verbs to action pictures, but not nouns with object pictures. Thus they did not use differences in movement to distinguish actions from objects. Instead, they treated most signs, regardless of movement reduplication, as actions – a bias that the match-to-sample design could not overcome. Signers, on the other hand, applied their knowledge of the ways that ASL distinguishes between nouns and verbs and correctly matched the signs to their corresponding meaning.

Table 1. Fixed and random effects predicting accuracy with the movement reduplication + size in the match to sample paradigm in Experiment 1a

Note. Table generated using the tab_model function from the SJPlot package version 2.6.2 for the R programming environment. Fixed factors were sum coded: (lexical class: noun = −1, verb = +1; signing status: nonsigner = −1, signer = +1) glmer(correct ~ Lexical_Class × Hearing_status+ (1 | ResponseID) + (1 | ITEM), data = df, family = binomial, control = glmerControl(optimizer = “bobyqa”)).

Fig. 3. Proportion of correct matches made by non-signers and signers with the movement reduplication + size rule tested using the match-to-sample paradigm. Dots represent an individual participant’s mean. Whiskers represent the maximum and minimum scores except for extreme values. Dots outside of the whiskers are outliers. Dashed line indicates chance performance. Figure created using the ggplot2 package (Wickham, Reference Wickham2016) in the R programing environment.

The results from the nonsigners could be explained in two ways. First, in the absence of a larger context, humans may be biased to treat any movement in a gesture as indicating an action. Previous work has documented this bias among nonsigners (Ortega et al., Reference Ortega, Schiefner and Özyürek2019). Second, the match-to-sample paradigm does not provide enough information for nonsigners to detect the relevant, albeit subtle contrasts in sign movement. Here, nonsigners saw either the noun or the verb variant, and had no opportunity to compare the movement information across the two signs. As such, for nonsigners the variations in movement were not necessarily contrastive. For signers fluent in ASL, this posed no challenge as they could draw on their knowledge of the systemic morphophonological contrasts in the language to answer correctly.

3.2. Experiment 1b: testing the sensitivity to the movement reduplication + size and movement size only rule in a between-subjects design with a forced-matching paradigm

To try to overcome the nonsigners’ bias to interpret all hand movements as actions and to provide participants with enough information to detect the relevant movement contrasts, we created a forced matching paradigm where participants saw both the noun and verb signs in each trial. Participants were asked to match each sign video with either a line drawing of an object or a line drawing representing the action associated with that object (Fig. 2b). Participants could not proceed to the next trial without matching both pictures to both signs.

3.2.1. Method

Participants

All 80 recruited MTurk workers completed all trials and passed all checks. The final sample included 52 males, 27 females, and 1 undisclosed (61 White, 7 Asian, 4 Hispanic or Latino, 1 Middle Eastern, 3 mixed race, and 4 undisclosed). Half of the participants viewed signs that differed according to the movement reduplication + size rule, and half of the participants viewed signs that differed according to the movement size only rule. We collected data for each condition separately, running the movement reduplication + size rule first, then recruiting new participants for the movement size only rule condition.

To confirm that deaf signers were sensitive to the movement size only rule, we tested deaf ASL signers in this condition alone. Nine of the deaf ASL signers that participated in Experiment 1a also completed Experiment 2 (males = 3, females = 6; 8 White, 1 Asian).

Design and procedure

Using Qualtrics, we presented both the noun and verb signs on the screen. The location of the sign videos on the screen was randomized across trials. They could play each video as often as they wanted. Participants also saw line-drawings of the object and action referents related to the signs. Location of the images was randomized across trials. Participants were asked to match the videos with their corresponding images by clicking on one drawing, then dragging it to the location under the video to which it best corresponded. They had to match one picture with one video and the other picture with the second video in order to advance to the next trial. It took participants an average of 7 minutes to complete this study.

3.2.2. Results and discussion

After data collection, we identified that one pantomimic sign (bath) had been erroneously included among the exclusively nonpantomimic items. We removed the item from our analysis, although the pattern of findings remained qualitatively the same. We ran mixed-effects logistic regressions predicting accuracy in the forced-matching paradigm using the outlined model (see Section 2.3). We found that while nonsigners were significantly better at making correct mappings in the movement size only condition than in the movement reduplication + size condition, the intercept of the model indicates that their overall performance was significantly below chance. Thus, nonsigners did not systematically interpret these noun–verb contrasts in a sign-like way (see Fig. 4 and Table 2). Visual inspection of Fig. 4 illustrates two patterns. First, nonsigners systematically make the incorrect mapping, overwhelmingly preferring to map reduplicated movements to actions and nonreduplicated movements to objects. Second, there seems to be a trend for nonsigners to be more ASL-like in the movement size only condition.

Fig. 4. Proportion of correct matches made by non-signers across the forced matching paradigms that assessed sensitivity to each rule, both between- and within-subjects, as a function of whether the signs included non-pantomimic or pantomimic elements across all experiments. Dots represent an individual participant’s mean. Whiskers represent the maximum and minimum scores except for extreme values. Dots outside of the whiskers are outliers. Dashed line indicates chance performance. Figure created using the ggplot2 package (Wickham, Reference Wickham2016) in the R programing environment.

Table 2. Fixed and random effects predicting accuracy Experiments 1b–2b

Note. Table generated using the tab_model function from the SJPlot package version 2.6.2 for the R programming environment. Fixed factor of Rule Type was sum coded (movement reduplication + size = −1, movement size only = +1). Bolded p-values are statistically significant.

^a glmer(correct ~ rule_type + (1 | ResponseID) + (1 | ITEM), data = df, family = binomial, control = glmerControl(optimizer = “bobyqa”)).

^b glmer(correct ~ rule_type + (1 + rule_type|ResponseID) + (1 | ITEM), data = df, family = binomial, control = glmerControl(optimizer = “bobyqa”).

Because the deaf ASL signers only completed one condition, we compared their performance to chance, finding that they correctly mapped ASL verbs to images of actions and ASL nouns to images of objects, and did so at very high rates (M = .89, SD = .17, t(8) = 6.84, p < 0.001, 95%CI[0.76–1.02]).

4. Experiment 2: pantomimic signs

As noted above, signs with pantomimic iconicity where the hand represents a hand are preferentially associated with actions (Ortega & Özyürek, Reference Ortega and Özyürek2019; Padden et al., Reference Padden, Hwang, Lepic and Seegers2015; van Nispen et al., Reference van Nispen, van de Sandt-Koenderman and Krahmer2017) and are seen as more iconic by nonsigners (Sehyr & Emmorey, Reference Sehyr and Emmorey2019). As such, pantomimic iconicity might override any existing biases to interpret some types of sign movements as associated with actions and others with objects. For example, the bias to treat pantomimic signs as actions may lead nonsigners to ignore relevant movement differences that contrast nouns and verbs. Parallel to Experiment 1, we explored nonsigners’ sensitivity to the movement reduplication + size and the movement size only rules in a between-subjects design (Experiments 2a) and in a within-subjects design (Experiment 2b), when signs had pantomimic elements.

5. Item analyses

To better understand the effect of item, we combined the data from across all of the experiments and plotted accuracy for each item in Fig. 5. In addition, for each item, we compared performance to chance using an exact binomal test applying a Bonferroni correction for multiple comparisons, setting the p-value as 0.0008. Performance on all of the items was at or significantly below chance in their mapping of the movement patterns to nouns and verbs (see Supplementary Table S1). What is clear from Fig. 5 is that for items where reduplication is contrastive, nonsigners were less likely to make an ASL-like mapping, preferring to treat reduplicated movement as “action-like” instead of “object-like.”

Fig. 5. Proportion correct for each item collapsed across all experiments. The dotted line represents chance performance. The error bars represent standard error (SE). Figure created using the ggplot2 package (Wickham, Reference Wickham2016) in the R programing environment.

6. Investigating the effect of pantomimic iconicity

Our plan from the outset was to investigate nonpantomimic and pantomimic iconicity separately. Visual inspection of the data presented in Fig. 4 does not indicate that pantomimic iconicity improved participants’ ability to make correct matches. But to explicitly test this possibility we combined the data from all of the experiments and ran a mixed effects logistic regression (see Section 2.3). We continued to observe the main effect of rule type (see Table 3) and no effect of iconicity type. There was no interaction between rule type and iconicity type – the effect of rule was the same for both types of iconicity. Thus, participants’ choices were systematically less ASL-like when the relevant contrast involved movement reduplication + size than with movement size alone and pantomimic iconicity did nothing to shift this pattern of performance.

Table 3. Fixed and random effects predicting accuracy combining data from all experiments for nonsigners

Note. Table generated using the tab_model function from the SJPlot package version 2.6.2 for the R programming environment. glmer(correct ~ rule_type × Iconicity_type_sc + (1 + rule_sc| ResponseID) + (1 | item), data = df, hearing, family = binomial, control = glmerControl(optimizer = “bobyqa”)).

7. General discussion

Across a series of experiments, we investigated whether morphophonological distinctions between nouns and verbs in ASL are iconically driven by the way humans construe objects and actions. We found some evidence that nonsigners were more accurate in mapping small movements to objects and large movements to actions than in mapping small reduplicated movements to objects and large nonreduplicated movements to actions. Nevertheless, their overall mapping accuracy was either at or below chance indicating that contrary to our original prediction, they did not automatically make the same mappings observed in ASL.

Nonsigners’ performance exhibited a set of biases that deviated from the patterns observed in sign languages. First, in the match-to-sample paradigm (Experiment 1a) where participants were presented with a single sign, nonsigners treated the majority of signs as verbs, reflecting a bias to treat any human movement as verb-like. People typically construe human movements as depicting events, and they attend to small changes in movement properties such as acceleration and velocity to reliably identify boundaries between events (e.g., Zacks et al., Reference Zacks, Kumar, Abrams and Mehta2009). Ortega et al. (Reference Ortega, Schiefner and Özyürek2019) also found that nonsigners showed a strong bias to assume that signs from Sign Language of the Netherlands (NGT) refer to actions and not objects. A similar bias has been seen in nonsigners’ creation of novel gestures about objects; when nonsigners are asked to silently gesture about objects, they prefer to depict the action affordances of an object (e.g., how a toothbrush is handled) instead of its physical features (e.g., the size and shape of a toothbrush; Ortega & Özyürek, Reference Ortega and Özyürek2019; van Nispen et al., Reference van Nispen, van de Sandt-Koenderman and Krahmer2017).

Second, when we made the morphophonological contrasts more salient by presenting both nouns and verbs simultaneously, nonsigners systematically interpreted movement reduplication as more action-like and nonreduplicated movement as more object-like (Fig. 4), displaying a sensitivity to the movement reduplication and size contrast between nouns and verbs. Their interpretation of this contrast, however, is the opposite of how reduplication is used to derive nouns from verbs across many sign languages. Thus, the bias that nonsigners have in interpreting reduplicated movement is likely not the source of the morphophonological distinction used to distinguish nouns from verbs found in mature sign languages.

The analysis consistently revealed more ASL-like mappings for items that followed the movement size only rule than for the movement reduplication + size rule. However, this improvement in performance rarely rose to above chance levels. We suggest that the significant difference between conditions is a result of nonsigners’ systematic below chance performance in the movement reduplication + size condition where they regularly interpreted reduplication as action-like, and not as a result of treating small movements as more object-like (e.g., Wilcox, Reference Wilcox2004). Nevertheless, at least one other study has shown that nonsigners slightly prefer to treat gestures with long, continuous movement as more action-like and those with small constrained movements as more object-like (Verhoef & Lepic, Reference Verhoef, Lepic, Ravignani, Barbieri, Martins, Flaherty, Jadoul, Lattenkamp, Little, Mudd and Verhoef2020). Thus, whether nonsigners treat movement size as a meaningful cue to lexical category merits further exploration.

We predicted that when the two rules were presented simultaneously, as in Experiments 1c and 2b, any sensitivity to the mappings that nonsigners showed when the rules were presented on their own might disappear because it would be difficult to extract the relevant contrasts. Visual inspection of the results does not yield a straightforward account. With nonpantomimic signs, nonsigners continued to interpret reduplicated movements as actions and nonreduplicated movements as objects, but their performance in the movement size only condition was now less accurate. With pantomimic signs, nonsigners no longer systematically interpreted reduplication as verb-like, and their performance with both rule types was at chance. Regardless, nonsigners’ choices did not parallel the linguistic mappings associated with either rule in sign languages.

Given the robust evidence that nonsigners treat pantomimic iconicity as more action-like, we speculated that we might see improved rule mapping with pantomimic signs. Pantomimic iconicity, however, did not affect the biases that we observed with nonpantomimic signs. One difference between our study and other studies that have previously demonstrated a strong association between pantomimic iconicity and action concepts is that our nouns and verbs differed as a function of movement and not handshape. The preferential association of pantomimic iconicity with actions may be limited to cases where handshape is a contrastive feature distinguishing nouns and verbs (Ortega & Özyürek, Reference Ortega and Özyürek2019; Padden et al., Reference Padden, Hwang, Lepic and Seegers2015; Verhoef et al., Reference Verhoef, Leipic and Padden2018).

Investigation of sign languages at their earliest stage of emergence can help us understand whether sign languages’ morphophonological distinctions used to differentiate nouns and verbs might arise from a widespread iconic bias among humans to see large articulated movements as actions and small reduplicated movements as deemphasizing actions and inviting an object interpretation of signs (Lepic & Padden, Reference Lepic and Padden2017; Wilcox, Reference Wilcox2004). Homesigners, deaf children who create their own family sign communication, seem to show some evidence of the movement size bias. They tend to produce larger sign movements using more proximal joints (e.g., shoulders and elbows) to communicate about actions, and smaller sign movements when they communicate about objects (Abner et al., Reference Abner, Flaherty, Stangl, Coppola, Brentari and Goldin-Meadow2019; Goldin-Meadow et al., Reference Goldin-Meadow, Butcher, Mylander and Dodge1994). Emerging sign languages, such as the village sign language Al-Sayyid Bedouin Sign Language (ABSL) and the community sign language Nicaraguan Sign Language (NSL), also show an early emergence of large sign movements to express actions, although NSL seems to exhibit this pattern more reliably than ABSL (Abner et al., Reference Abner, Flaherty, Stangl, Coppola, Brentari and Goldin-Meadow2019; Tkachman & Sandler, Reference Tkachman and Sandler2013). Our observation that in at least one case (Experiment 1b), nonsigners more systematically interpreted large movements as actions for nonpantomimic signs lends some support for an iconic bias for the movement size only morphophonological distinction early in language emergence; however, this bias is by no means robustly present at the inception of a sign language.

The use of the movement reduplication + size distinction is far less prevalent than the movement size only distinction in homesign and emerging sign languages. In a study of a single homesigner in the United States, Goldin-Meadow et al. (Reference Goldin-Meadow, Butcher, Mylander and Dodge1994) found that gestures for objects tended to be “abbreviated” relative to those used for actions, and the nature of this abbreviation was, in part, the elimination of reduplicated or “bidirectional” movement (pp. 283–284). That is, the homesigner exhibited the same bias nonsigners exhibited, associating reduplicated movement with actions. In a recent study of four homesigners in Nicaragua, three homesigners used reduplication in a contrastive way to distinguish objects from actions, but they did so only for iterable actions (Abner et al., Reference Abner, Flaherty, Stangl, Coppola, Brentari and Goldin-Meadow2019). Only with later cohorts of NSL signers do we see a more systematic use of reduplication to mark nouns across both iterable and noniterable events (Abner et al., Reference Abner, Flaherty, Stangl, Coppola, Brentari and Goldin-Meadow2019). Thus, an initial bias to treat reduplication as action-like may make it more difficult for a sign language to immediately adopt reduplication as a strategy for noun derivation.

There are several possible reasons for the more gradual adoption of these morphophonological distinctions in sign languages. First, these distinctions are not readily available to nonsigners who have a bias to interpret reduplication as action-like instead of object-like. Second, handshape may be a more salient cue than movement to distinguish objects from actions for nonsigners and homesigners during language creation (Hunsicker & Goldin-Meadow, Reference Hunsicker and Goldin-Meadow2013; Padden et al., Reference Padden, Hwang, Lepic and Seegers2015; Verhoef et al., Reference Verhoef, Leipic and Padden2018). Finally, as a language emerges, other linguistic pressures may shift the language away from any initial bias. For example, many different sign languages leverage space and movement for verbal morphology, marking agreement, number, and aspect (see Sandler & Lillo-Martin, Reference Sandler and Lillo-Martin2006 for a review). Relying on movement reduplication to mark actions could constrain the ways a signer can use space and directional movement to mark other grammatical functions. Interestingly, the use of space for grammatical features emerges over time (Kocab et al., Reference Kocab, Pyers and Senghas2015; Pyers et al., Reference Pyers, Shusterman, Senghas, Spelke and Emmorey2010; Pyers & Senghas, Reference Pyers and Senghas2007; Senghas, Reference Senghas2003; Senghas & Coppola, Reference Senghas and Coppola2001), and there may be an inverse relationship between the emergence of spatial grammar and the use of reduplication to represent actions, although we need additional research to explore this possibility. Relatedly, the use of movement reduplication to mark objects may arise alongside a phonological shift restricting the size of the signing movements associated with nouns; some preliminary evidence shows that nouns in ASL are articulated in a smaller signing space compared to verbs (Sehyr et al., Reference Sehyr, Edinger and Emmorey2019), and reduplication may increase the perceptual salience of signs articulated with smaller movements.

The robustness of various morphophonological distinctions between noun–verb pairs in sign languages is certainly debatable. While elicited production (Abner et al., Reference Abner, Flaherty, Stangl, Coppola, Brentari and Goldin-Meadow2019) and assessed comprehension (Experiments 1a and 1b) of these noun–verb pairs in ASL elicits reliable distinctions from fluent signers, the occurrence of these distinctions may be less robust in connected discourse or naturalistic conversation (Bouchard et al., Reference Bouchard, Dubuisson and Parisot2005; Johnson, Reference Johnston2001; Voghel, Reference Voghel2005). Signers instead may favor other features like word order, spatial grammar, and context over morphophonology to communicate and interpret the meaning of a potential noun or verb form (Voghel, Reference Voghel2005). One intriguing possibility is that linguistic competition with human cognitive construals may lead to less consistent use of those linguistic features in naturalistic conversation, although we need more research to explore this possibility.