*2.4. Statistics*

#### 2.4.1. Dependent Variables

*Frequency of type of non-manual use*. For each participant we computed the number of each of the coded non-manuals produced with each elicited question. This computation allows us to see the change in frequency of non-manual use across cohorts.

*Duration of non-manual production.* For every non-manual that co-occurred with an interrogative, we measured the time from non-manual onset to offset.

*Coarticulation of non-manuals with question words.* For the subset of questions produced with a wh-question word and with a non-manual, we categorized whether the non-manual overlapped with an overtly articulated (spoken or signed) wh-question word.

#### 2.4.2. Model Specification

For all analyses we ran mixed effect regression models using R version 4.0.3 (R Core Team 2021) and the lme4 package (Bates et al. 2015). For all models we included participants and items as random intercepts. We operationalized the change in the language by cohort. Cohort was dummy coded as hearing, first, second, and third, with hearing as the reference level. Non-manual type was dummy coded with chin lift as the reference level, because we had no theoretical motivation to preset any specific non-manual as the reference level, and that was the default reference level selected by the program. For each central question, we specify the models used.

To test the role of frequency in non-manual adoption, we ran a mixed effects linear regression predicting the total number of non-manuals produced alongside every question produced (*N* = 676) with participants and items as random intercepts, and cohort and non-manual type, and the interaction between cohort and non-manual type, as fixed factors. In addition, we ran six separate mixed effects linear regressions predicting the total number of each of the six non-manuals and cohort as a fixed factor. These analyses should allow us (a) to identify if the facial gestures of Nicaraguan Spanish speakers are the seed for the non-manual markers of wh-questions in LSN, and (b) to isolate the candidate grammatical markers of wh-questions in LSN–those that appear most frequently.

To test the role of *duration* in non-manual adoption, we ran a mixed effects linear regression predicting the duration of every non-manual produced (*N* = 1050), with participants and items as random intercepts and cohort and non-manual type, as well as the interaction between cohort and non-manual type, as fixed factors.

Finally, to capture any tendency to coordinate the timing of the non-manual with the production of a question word, we ran a mixed effects logistic regression using the glmr() function, predicting non-manual *coarticulation* with the question word, with participants and items as random intercepts and cohort and non-manual type, as well as the interaction between cohort and non-manual type, as fixed factors. For this analysis we considered only the subset of non-manuals that were produced during interrogatives where a wh-question word was also produced (*N* = 711).

When any models failed to converge, we removed participants as a random effect and changed the optimizer.

## **3. Results**

Table 3 presents the total number of questions and non-manuals elicited from each cohort.


**Table 3.** Sum of questions, wh-question words, and non-manuals elicited by cohort.

#### *3.1. Frequency of Non-Manual Occurrence*

To answer the question of how the available facial gestures in the hearing community have been taken up and repurposed as wh-question markers by Nicaraguan signers, we ran a mixed effects linear regression predicting total non-manual use with all questions, and then predicting the use of each of the six non-manuals (Table 4; see Statistics for model specification).

**Table 4.** Table of fixed and random effects predicting total non-manual production. Table generated using the tab\_model function from the SJPlot package version 2.6.2 for the R programming environment (Lüdecke 2021).


Note: lmer(total\_nonmanual ~ cohort + (1|item) + (1|ID), data = df).

While first-cohort signers were statistically similar to the hearing non-signers in their total non-manual production, second- and third-cohort signers produced more nonmanuals than the non-signer participants (Table 4). Turning to the patterns of the individual non-manuals, we see a tendency for them to increase or stabilize across the four groups, with the nose wrinkle and the brow furrow increasing to a significant degree (Table 5). Though it ultimately stabilizes at a relatively low level, the initial increase in the nose wrinkle is intriguing, given that it was virtually nonexistent among the hearing non-signers. The most dramatic change was in the use of the brow furrow, which was rare among the hearing non-signers, but came to dominate in its use by second- and third-cohort signers (Figure 2).

**Figure 2.** Average frequency each non-manual per question across cohorts. The black line indicates the median value for each non-manual type. Whiskers indicate the maximum and minimum values, excluding extreme outliers. Non-manuals are ordered according to mean frequency over all groups combined, with the more frequent to the right. Figure created using the ggplot2 package (Wickham 2016) in the R programming environment.

Visualization of the data allows us to better understand which facial gestures from the hearing community are taken up by the sign language, and how the distribution of non-manual use changes over time (Figure 2). Intriguingly, the brow furrow is the most dominant non-manual across all cohorts of signers, with a dramatic increase in its use from the first to second cohort. However, this non-manual was one of the least prevalent among the hearing participants. The non-manual most highly preferred by the hearing non-signers is the head tilt, followed by the chin lift and the brow raise. Among the signers, the head tilt emerges as the second most frequently used non-manual, but the chin lift and the brow raise remain relatively constant across groups. Thus, the primary markers of wh-questions for the later users of LSN are the brow furrow and the head tilt, even though only the head tilt, and not the brow furrow, is heavily represented among the facial gestures produced by non-signers.


#### *3.2. Duration of Non-Manual*

We asked whether non-manuals with the longest duration among the hearing nonsigners might be especially salient, or better able to serve grammatical functions, and therefore be favored for selection. We did not observe any simple effects of group or non-manual type on duration (See Table S1 in Supplementary Materials), indicating that duration is highly unlikely to be driving the effect of cohort on frequency observed in the analyses above. What is crucial is whether we see interactive effects between cohort and non-manual type, specifically for the non-manuals that we saw most frequently among the later cohorts of signers, the brow furrow and the head tilt. Indeed, we did see an interactive effect for cohort and brow furrow, such that first and second cohort signers produced the brow furrow for a longer duration than hearing signers and relative to the chin lift (first cohort x furrowed brow: *β =* 0.63, 95% CI [0.12–1.15], *p* = 0.016; second cohort x furrowed brow: *β =* 0.65, 95%CI [0.24–1.07], *p* = 0.002). We also observed a similar interactive effect for the second-cohort signers and their production of the head tilt; they held the head tilt for longer, relative to the chin lift, than hearing signers (second cohort x head tilt: *β =* 0.39, 95% CI [0.05–0.73], *p =* 0.023).

Visual inspection of the data (Figure 3) shows that among later cohorts of signers, the head tilt and brow furrow are indeed held the longest, with the brow furrow showing the greatest variability in duration. Thus, for the two non-manuals that emerge as likely candidates for grammatical markers, we observe that they are produced for a longer duration by LSN signers. The upper bounds of the whiskers in Figure 3 indicate that the brow furrow, in particular, has the potential to be held for much longer among the LSN signers, compared to the other non-manuals, and compared to the non-signers. The later emergence of a longer duration for the brow furrow suggests that duration is not driving the selection of the form; rather, it may be an indicator of increasing systematization that follows along with its increase in frequency.

**Figure 3.** Average duration of each non-manual produced across cohorts. The black line shows median duration for each non-manual. Whiskers represent maximum and minimum scores, excluding extreme outliers. Non-manual categories are ordered according to frequency, as represented in Figure 2, with the more frequent to the right. Figure created using the ggplot2 package (Wickham 2016) in the R programming environment.

#### *3.3. Coarticulation of Non-Manual with Wh-Question Word*

Our third test examined whether the coarticulation of a non-manual with the whquestion word made it more likely to be taken up as a marker of wh-questions. The position of the wh-word in the elicited questions varied within and across cohorts, with the wh-word appearing in sentence-initial, sentence-medial, and sentence-final positions.

We ran a mixed-effects logistic regression (see Statistics section for model specification) on the subset of the data that included the questions that had both a manually articulated question word (Figure 1) and a non-manual. The model showed that the three cohorts did not significantly differ from non-signers in their coordination of the non-manual with the wh-question word (see Table S2 in the Supplementary Materials). Regardless of cohort, head tilt was significantly more likely than the chin lift to be coarticulated with the whquestion word (*OR* = 0.39, 95%CI [0.18–0.83], *p =* 0.015). The most frequently produced non-manual, the brow furrow, was not significantly more likely than the chin-lift to be produced at the same time as the wh-question word.

Visual inspection of the data (Figure 4) reveals that, while on average the number of questions where the brow furrow is coarticulated with the wh-question word is relatively constant across the groups, with the later cohorts we observe more cases above the median, suggesting that within these later cohorts, more signers are articulating the question words and brow furrow simultaneously, although this effect is not significant.

**Figure 4.** Number of each non-manuals type coarticulated with the wh-question word. The black line shows the median number of questions for which the non-manual was coarticulated with the wh-question word. Whiskers represent maximum and minimum scores, excluding extreme outliers. Non-manual categories are ordered according to mean frequency as represented in Figure 2, with the more frequent to the right. (The unusually skewed bar for the shoulder raise in the third cohort results from a single participant who repeated questions multiple times per item, raising the shoulders multiple times per question.) Figure created using the ggplot2 package (Wickham 2016) in the R programming environment.

## **4. Discussion**

We began our inquiry into the emergence of the grammar of LSN by seeking the origins of the forms used to mark wh-questions. We looked for the seed of non-manual markers in the facial gestures of members of the hearing community as they asked questions in Spanish

and followed the use of these non-manual expressions in the signing of three age cohorts of LSN, representing the first three decades of the language's birth and growth. Our crosssectional comparison of the frequency and duration of several non-manuals captured the progressive adoption of two of these forms: the brow furrow and the head tilt. Intriguingly, these two non-manual forms were not the ones most frequently produced by the Spanish speakers; nor were they initially sustained for the longest duration; nor were they the forms most apt to overlap in timing with the spoken wh-word. Thus, while the seed of non-manual markers can be found in the diversity of candidate facial gestures produced by the surrounding Nicaraguan hearing community, the adoption of the dominant forms into LSN is not driven by their frequency, duration, or overlap among non-signers. Rather these metrics appear to be indicators, not drivers, of language evolution.

The changes that we documented over the early years of LSN sugges<sup>t</sup> a path of grammaticalization of selected non-manual forms. However, none of our three measures of the initial seed effectively determined which non-manuals would ultimately be selected for grammaticalization. We consider three possible alternative mechanisms that merit further exploration.

First, there may be other visual-perceptual or articulatory affordances, not measured in our current study, that favored these particular non-manuals for selection. For example, coarticulation with other facial gestures could work in multiple ways. The appearance of a facial gesture in isolation may make it more salient to a learner; alternatively, selection may favor facial gestures that can be easily co-articulated with other facial gestures. In addition, there are other characteristics of facial gestures, such as the intensity of the articulated form (e.g., Domaneschi et al. 2017), that may increase its salience. The affordances of particular non-manual expressions, such as the brow furrow, may allow them to be maintained over stretches of discourse. These characteristics may favor certain non-manuals in the service of prosodic functions, such as marking the boundaries of clauses and sentences, and pragmatic functions such as turn taking and back-channeling (Brentari et al. 2018; Mesch 2016; Sandler 2010; Wilbur 2013). Exploration of these other aspects of articulation would be especially informative.

Second, the selection of the brow furrow and the head tilt may have more to do with communicative salience. The brow furrow and, to a lesser extent, the head tilt, may be capitalizing on human universals in non-verbal communication. Eyebrow movements, including the brow furrow, have been observed to signal that a speaker feels uncertain or perplexed (Campbell et al. 1999; Domaneschi et al. 2017; Ekman 1979; Swerts and Krahmer 2005), two mental states that underpin requests for information. The articulatory affordance and/or communicative function of the brow furrow might explain its consistent association with wh-questions across a variety of sign languages around the world (Zeshan 2004). However, our data did not reveal a preference for hearing non-signers to produce the brow furrow with questions. Why we don't empirically find an overwhelming preference to produce this nonmanual is unclear. The previous work showing an association between uncertainty and the brow furrow among non-signers did not measure the frequency of the production of the brow furrow with questions but rather the intensity of facial expression production (Domaneschi et al. 2017) or the interlocutor's interpretation of these facial expressions (Campbell et al. 1999; Swerts and Krahmer 2005). The one study that quantified the number of brow furrows accompanying wh-questions produced by hearing non-signers in the U.S. also showed very low rates of production (~5% of the time; Pyers and Emmorey 2008). Thus, it may be the case that while hearing non-signers produce the brow furrow, they generally do so less often than previously estimated. Instead of the brow furrow, nonsigners in the present study preferred the lateral head tilt, which also carries communicative significance. While the use of the lateral head tilt in wh-questions is less documented in sign languages, research on spoken communication has proposed that a lateral head tilt functions to make a speaker appear more friendly by offsetting direct eye contact with the listener (Costa and Bitti 2000), perhaps making a question more likely to be answered. Thus, communicative salience may account for the facial expressions that are produced by

hearing non-signers when asking questions. Once a facial expression is taken up into a language, additional considerations, such as articulatory affordance, may then come into play.

Third, the gestures that hearing people produce when they speak to each other are just one of several possible seeds of LSN. It is unclear the degree to which deaf children would be able to access the ambient spoken conversations in their environment. There are likely other gestural behaviors in the communication between hearing family members and deaf children, such as enactments, pointing, and other gestures unaccompanied by speech, in the communication between hearing family members and deaf children that are taken up in the creation of the homesign systems that arise in families with a deaf child (Coppola 2020; Goldin-Meadow 2005). Indeed, parents of Nicaraguan homesigners overwhelmingly prefer to communicate with their children using manual gestures without any accompanying speech (Coppola et al. 2006). The nature of facial gestures produced without speech is an open question. However, such voice-off homesign systems were probably a more direct precursor to first-cohort signing.

Given that the brow furrow is observed in many other sign languages around the world, one might wonder about direct influence from other sign languages on LSN. Over the years, there have been several lexical signs adopted from other sign languages, some from direct contact and some via dictionaries, which we have been documenting in other studies. However, the history of language contact and the pattern of emergence of nonmanuals sugges<sup>t</sup> that international contact is a less likely source than local gestures. In the 1970s and 1980s in Nicaragua, and even into the 1990s, international contact was limited to a few individual signers, all members of the first cohort, for limited periods. Yet the dominance of the brow furrow is not evident in their signing. Its dominance begins with the second cohort, whose primary exposure was to that first cohort, and their own peer community. This signature suggests that it arose locally as the language was taken up by younger signers.

Whatever their pattern of initial use, the non-manuals were deployed differently by signers from different cohorts. The brow furrow appears to have shifted early from disfavor into favor. The first-cohort signers used it more frequently than non-signers, and occasionally sustained it longer (though this is not a statistically detected difference). However, the first cohort dropped any tendency to coarticulate the brow furrow with the wh-question sign. Indeed, they showed little coarticulation of any non-manual with a wh-question sign. As they took up the language, the second and then third cohorts began coarticulating the brow furrow, in particular, with wh-question signs. At the same time, they increasingly differentiated their two favored forms, the brow furrow and the head tilt, from the others in frequency. These changes over cohorts in the coarticulation of non-manual and manual signs are likely to have interacted with other changes in the use of lexical signs for questions. Consider the data in Table 3 that reveal a striking shift from first-cohort signers who rarely produced manual wh-question signs, to second- and third-cohort signers who often produced more than one manual wh-question sign in a single question. It would be a useful pursuit in future inquiry to carry out an analysis of the changing syntactic structure of wh-questions in LSN, in coordination with the changing use of non-manuals.

What does this pattern across cohorts today reveal about the nature and history of the emergence of LSN? Recall that a synchronous, cross-sectional "apparent time" analysis can be taken to reflect a diachronic history of the language, with older age cohorts reflecting language of longer ago, and younger age cohorts reflecting a more recent variant. While this pattern indicates the changes that took place, it does not fully explain why new developments in the language do not spread to all members of the community, and why an earlier variant of the language persists among older members. The three cohorts in this study roughly and categorically represent a constant, continuous influx of new members into the LSN signing community across three decades. While new arrivals learn from older members, all members continue to interact within and across age groups, and any

communication between younger and older individuals would logically have an equal potential of influence in either direction. We would expect an emerging language to undergo reorganization and signal compression, in response to articulatory and perceptual pressures that favor communicative and processing efficiency (Kirby et al. 2008; Kanwal et al. 2017). However, everyone, regardless of age, is subject to these pressures. Accommodations that result during peer interaction and transmission, that are generated by or learnable by adults, should be evident in longitudinal, real-time snapshots of the language over the years, but not show up in "apparent time" cross-sectional analyses today. In contrast, differences between age cohorts that are still evident decades later point to the effect of the changing nature of learning across the life span. In such cases, their different natures will lead children and adults to arrive at different accommodations to these pressures. The measurable differences between age cohorts that we have documented, in both the selection of brow furrow by the first cohort, and the coarticulation of non-manuals by the second and third cohorts, thus reveal the nature of children's learning in particular. Early childhood appears to be a time when key language-learning mechanisms are available, and each age cohort, in sequence, took their turn in this childhood stage, passing on a changed system.

In following these changes, we appear to have captured two tipping points in the emergence of LSN. The first, with the first cohort, entailed the selection of a small number of potential non-manual markers from the variety of facial gestures that accompanied spoken questions in the local non-signing community. As the nonmanuals were adopted into LSN, their relative frequency and use did not correspond to their use as facial gestures by the non-signers. They were less differentiated, not reflecting the variability of their source. At the same time, the non-manuals became more separate, rather than co-articulated with the wh-word in the sentence.

The second tipping point occurred when the second cohort took up LSN, and a subset of the non-manuals started to dominate. This change is particularly interesting in light of differences between adult and child learners when presented with certain kinds of variability in their input. Research using artificially created languages in the laboratory has found that when the input includes several alternative forms whose use is undifferentiated, child learners will acquire a smaller set of forms, and apply them more systematically (Hudson Kam and Newport 2005). Adult learners are not as quick to reorganize a language under the same conditions (Hudson Kam and Newport 2009). This particular solution of narrowing and systematizing appears to be an imprint of typical child learning on LSN.

As a few forms started to dominate, LSN signers of the second and third cohort increased the co-articulation of the non-manual wh-markers with the signed wh-word. The timing of this change is notable given the ecology of the language at that time. The systematization of the grammar of wh-questions in LSN happens concurrently with the establishment of a lexicon, the organization of argumen<sup>t</sup> structure, and the coordination of discourse, among other linguistic features. Previous work has documented that the transition from the first to the second cohort was characterized by the appearance of grammatical features that depend on simultaneous, rather than sequential, production (Senghas et al. 1997; Senghas and Coppola 2001; Senghas 2003; Kocab et al. 2015, 2016). The grammar of LSN non-manuals apparently emerged in coordination with other simultaneous aspects of the grammar.

A fuller consideration might explore these and other aspects of the changing ecology of LSN, and how the brow furrow and other non-manuals fit in as the full repertoire of signs and their syntax is being created. For example, at the transition from the first to the second cohort of LSN signers, the system for marking syntactic objects was changing (Senghas et al. 1997; Senghas 2003). Research on ASL has shown that wh-question words can be absent from the surface structure of a signed question, leaving the non-manual with a greater role in indicating the syntax of the sentence (Petronio and Lillo-Martin 1997). It would be useful to examine non-manuals' temporal overlap with other arguments in question sentences, aside from wh-question words. We informally observed that non-manuals were sometimes co-articulated with the sign for the queried item or object. There are many such

possible ways, as LSN was transmitted from one generation to the next, that the syntax of non-manual question markers may have been changing dynamically with the rest of the language.

By closely examining subtle movements of the face during a simple communication task, we have captured key aspects of the earliest stages of the evolution of markers for wh-questions in LSN. Though it represents only a sliver of the grammar, this small piece can reveal mechanisms responsible for the birth and growth of a grammar. The seeds of language are borne of countless acts of communication, in which individuals leverage the expressive power of the body to represent and transmit information. Once initial forms are available, they evolve through transmission from one person to the next. This is how a language is created. Even an element as small as the furrowing of the eyebrows, once produced, is available to be shaped into language. The changes that we have captured here reveal the progressive effects of language acquisition processes, applied by hundreds of children over several decades, with the language of the younges<sup>t</sup> LSN signers today reflecting their combined, cumulative impact.

**Supplementary Materials:** The following are available online at https://www.mdpi.com/article/10 .3390/languages7020137/s1, Table S1: Fixed and Random effects predicting non-manual duration. Table S2: Fixed and random effects predicting the coordination of the non-manual with the whquestion word.

**Author Contributions:** Conceptualization, A.K., J.P. and A.S.; Methodology, A.K., J.P. and A.S.; Investigation, A.K. and J.P.; Data Curation, J.P.; Writing—Original Draft Preparation, A.K., J.P. and A.S.; Writing—Review & Editing, A.K., J.P. and A.S.; Visualization, J.P.; Supervision, A.K. and J.P.; Funding Acquisition, A.K., J.P. and A.S. All authors have read and agreed to the published version of the manuscript.

**Funding:** This work was supported by grants from NSF GRFP (A.K.), NIH/NIDCD gran<sup>t</sup> R01 DC005407 (A.S.), the James S. McDonnell Foundation (J.P.) and the Wellesley College Provost Office (J.P.).

**Institutional Review Board Statement:** The study was conducted according to the guidelines of the Declaration of Helsinki, and approved by the Institutional Review Board of Barnard College on 29 April 2008 (and annually henceforth).

**Informed Consent Statement:** Informed consent was obtained from all subjects involved in the study.

**Data Availability Statement:** Deidentified data presented in this study are available on request from the corresponding author.

**Acknowledgments:** We are grateful for the Nicaraguan Deaf community for their ongoing support. We thank Kristie Casanova-Artl, Sofia Fioretti, Sara Maria Hasbun, Shannon Howard, Catherine Jaramillo, Audrey Lin, Kathryn Marton, Kristianny Ruelas-Vargas, Andrea Takahesu-Tabori, Rafael Treviño, and Caroline Zola for research assistance on this project.

**Conflicts of Interest:** The authors declare no conflict of interest.
