Lexical Iconicity is differentially favored under transmission in a new sign language: The effect of type of iconicity

Abstract

Observations that iconicity diminishes over time in sign languages pose a puzzle--why should something so evidently useful and functional decrease? Using an archival dataset of signs elicited over 15 years from 4 first-cohort and 4 third-cohort signers of an emerging sign language (Nicaraguan Sign Language), we investigated changes in pantomimic (body-to-body) and perceptual (body-to-object) iconicity. We make three key observations: (1) there is greater variability in the signs produced by the first cohort compared to the third; (2) while both types of iconicity are evident, pantomimic iconicity is more prevalent than perceptual iconicity for both groups; and (3) across cohorts, pantomimic elements are dropped to a greater proportion than perceptual elements. The higher rate of pantomimic iconicity in the first-cohort lexicon reflects the usefulness of body-as-body mapping in language creation. Yet, its greater vulnerability to change over transmission suggests that it is less favored by children’s language acquisition processes.

1.0. Introduction

Iconic symbolism entails an analogical, part-to-part mapping between the form of a symbol and its referent, such as an array of squares on a piece of paper representing the arrangement of seats in a theater. Iconicity is a common characteristic of signed and spoken languages, found at many levels, including their lexicons and grammars (Meir, Padden, Aronoff & Sandler 2013; Perlman, Little, Thompson, B & Thompson, R 2018; Perniss, Thompson & Vigliocco 2010; Taub 2001). Because words in sign languages are essentially configurations of visible parts of the body -- including the hands, arms, upper torso, and face -- signing provides multiple opportunities for iconic mappings between the body and many aspects of meaning (Meir, Padden, Arnoff & Sandler 2007). These include using the whole body to represent another body and its movement, or using a part of the body, such as the hand, to represent a characteristic of a referent, such as its flatness. The prevalence of iconicity in signed languages reflects this rich opportunity; all sign languages documented to date exhibit extensive iconicity in their lexicons (Cuxac & Sallandre 2007; Perlman et al. 2018; Pizzuto & Voltera 2000). Iconicity seems to help language acquisition and processing in many ways: Iconic signs are acquired at a younger age (Caselli & Pyers 2017; Caselli & Pyers 2019; Thompson, Vinson & Vigliocco 2012) and retrieved from memory more quickly (Thompson, Vinson & Vigliocco 2009). In novel word learning paradigms, children leverage iconicity to identify, learn, and remember novel signs and words (Kantartzis, Imai, Evans & Kita 2019; Magid & Pyers 2017; Namy 2008; Tolar, Lederberg, Gokhale & Tomasello 2007). Even non-signing, hearing adults asked to create gestured expressions will generate iconic forms (Goldin-Meadow, McNeill & Singleton 1996; Padden, Hwang, Lepic & Seegers 2015; Ortega & Özyürek 2019; Van Nispen, Van de Sandt-Koenderman & Krahmer 2017). Clearly, iconicity is a useful, highly available tool for creating and learning signs. Yet, the historical study of sign languages suggests that iconicity diminishes over time (Battison 1978; Frishberg 1975). Apparently, many signs that originated as iconic forms have changed over generations of learning and use in ways that have reduced the transparency and salience of the iconic mappings between lexical forms and their meanings. This presents a puzzle: why should something so evidently functional and frequent decrease? The present study addresses this question by capturing the use and change of iconicity in the earliest stages of the emergence of a new sign language in Nicaragua, Nicaraguan Sign Language (NSL). Is iconicity leveraged in the creation of new signs, and does it indeed decrease over time as signs are taken up by new generations of learners? To address this question, we analyzed lexical signs at two moments in the early stages of NSL. Signs used today by older signers, who first created NSL, reveal how iconicity is leveraged to generate new lexical items; signs used by younger signers, who learned the language two decades later, reveal how lexical iconicity is affected by acquisition processes.

Current research on iconicity in language recognizes the importance of the qualitative nature of the mapping between form and referent, rather than mere transparency or degree of iconicity (Dingemanse et al. 2015; Caselli & Pyers 2019). The type of iconic mapping likely plays different roles in the creation as opposed to the learning of signs. We consider two types of iconic mapping between the body and its referent. The first entails mapping the body of the signer to another body, as in when the signer personifies or simulates interaction with the referent. In this case, the link from form to meaning is direct, body-to-body. We refer to this type as pantomimic¹ iconicity (Caselli & Pyers 2019; Magid & Pyers 2017; Ortega, Sümer & Özyürek 2017; Tolar et al. 2007). The Nicaraguan signs LION and CORN (Figure 1a & 1b) are both pantomimic. In LION, the signer personifies the referent, such that the arm and hand map to the front leg and paw of a lion. In CORN, the signer simulates handling the referent, such that the hand and mouth map to the hand and mouth of someone holding and eating corn on the cob. The second type of iconic mapping entails using a part of the body to represent observable, perceptual features of the referent. Thus the link from form to meaning is not body-to-body, but body-to-object. We refer to this type as perceptual² iconicity (Caselli & Pyers 2019; Magid & Pyers 2017; Ortega et al. 2017; Tolar et al. 2007). The Nicaraguan sign HAMMOCK (Figure 1c) uses perceptual iconicity, in that the signer’s hands represent the flat and gently curved fabric of the hammock arcing forward and back. A sign can include both pantomimic and perceptual³ iconicity, either as two separate sign-units within a compound sign, or as two phonological elements of a single sign, as in the Nicaraguan sign PLANTAIN (Figure 1d).

Figure 1.

Examples of types of iconicity. Pantomimic signs entail a body-to-body correspondence, and any movement of the articulators corresponds to (A) action features of the referent, as in pantomiming the baring of teeth and swiping movement of the hand for LION, or (B) someone enacting interacting with the referent, as in cupping the hands near the mouth and making biting movements for CORN. (C) Perceptual signs entail a correspondence between the articulators and observable features of the referent, such as its size and shape, and the sign movement corresponds to the extent or movement of the referent, such as flat hands swinging outward for HAMMOCK. A sign was coded as (D) both if it simultaneously included pantomimic and perceptual iconicity, such as a flat hand representing the shape of a knife (perceptual) with a simultaneous motion that demonstrates a cutting movement (pantomimic) for PLANTAIN.

Both pantomimic and perceptual iconicity appear in abundance in mature sign languages around the world. A question is whether they are equally employed as a sign language is created and matures over generations. There is reason to expect that pantomimic iconicity will be favored as signers build a new sign language. When asked to create gestural representations, non-signers overwhelmingly make use of body-to-body mappings in the signs they create (Goodglass & Kaplan 1963; Ortega & Özyürek 2019; Motamedi, Schouwstra, Smith, Culbertson & Kirby 2019). However, we don’t know what kind of changes in iconicity to expect over time as a language is used and passed down over generations. Frishberg (1975) noted that American Sign Language (ASL) declined in iconicity; however, the language she studied had already been in use for a century, and she did not explore the nature of the iconic mappings involved. Are pantomimic signs more apt to lose their iconicity? They may have more to lose; pantomimic signs are perceived to be more highly iconic than perceptual signs (Caselli & Pyers 2019; Sehyr & Emmorey 2019). Furthermore, pantomimic signs may engage more of the body, even the entire body, and therefore be more subject to reduction under motoric and perceptual pressures relative to perceptual signs, which may start out more compact. Because we are interested in changes in the type of iconic mapping used, more than in the degree, in this study we attended to which components of representations were retained or dropped. Do signs change in iconicity even in the earliest stages of emergence, and if so, is such change influenced by the type of iconic mapping between body and meaning? We turn to a newly emerging sign language in Nicaragua to answer this question. To our knowledge this is the first examination of the function of iconicity in the emergence of a natural sign language lexicon.

NSL was created by deaf children and adolescents, starting with an initial cohort of 50 individuals who arrived in a new special education school in Managua in the 1970s (Kegl et al. 1999; Polich 2005; Senghas, R. 2003). The pedagogical approach at the school focused on teaching written and spoken Spanish, but as the children socialized in their free time in the schoolyard and on the buses, they made use of shared gestures and the various individual home sign systems that they had developed with their families (Coppola & Newport 2005; Goldin-Meadow 2005). The natural processes that were engaged in their daily social interaction led to the creation of a new sign language. Each year, new children arrived at the school, learned the system used by the older children, and continued to develop the language further (Coppola & Senghas 2010; Kocab, Pyers & Senghas 2015; Kocab, Senghas & Snedecker 2016; Senghas 2010; Senghas & Coppola 2001). From the early 1980s, the school has consistently enrolled about 200 deaf children in grades ranging from pre-kindergarten to sixth grade. Today over 1500 deaf signers in and around Managua use NSL as their primary, everyday language. We can group these signers by the decade in which they arrived, calling those who joined the community in its first, second, and third decades the first cohort, second cohort, and third cohort, respectively. By comparing the variants of NSL across age cohorts, we can track the development of the language over time, as it was transmitted to new generations of learners.

In the present study, we examined changes in the lexicon of NSL with respect to the role that iconicity plays in motivating the forms of signs. We directly compared citation forms of common, everyday signs elicited from first-cohort signers in 2002 and 2003 to those elicited from third-cohort signers fifteen years later, in 2017. Signers were all adults at the time of elicitation; everyone had been signing since entering preschool and had at least 20 years of experience using NSL as their primary language. In this way, we ensured that the signing we captured reflected the outcome of each cohort’s early acquisition and lifelong use. Importantly, because of the approximately 20-year age gap between the first- and third-cohort signers, they did not attend the school at the same time, and did not interact directly. The primary language models for the third-cohort signers were second-cohort signers, who learned the earliest form of NSL from the first-cohort signers. We were particularly interested to see how pantomimic and perceptual iconicity evident in the signs of the first cohort changed as the language was transmitted to the cohorts that followed. We anticipated that change would be rapid in these early years of emergence of the NSL lexicon, as the community quickly converged on conventional sign forms for each meaning. If so, such changes would be evident as differences between the cohorts. Accordingly, the nature of the first-cohort signs would reveal the iconic resources that were recruited as signs were created, and differences between first- and third-cohort signs would reveal the impact of acquisition by two successive cohorts of learners. Specifically, we considered how acquisition might make use of, or alter, iconic patterns in NSL.

There are several characteristics of child language acquisition that might play a role in the conventionalization of an emerging lexicon. First, children don’t match their input exactly; when presented with variable input, they produce something more regular and systematic. Adult learners under the same conditions don’t reorganize their language in this way (Hudson Kam & Newport 2009). Second, children are particularly adept at detecting word-internal patterns, and can do so based on a very small amount of evidence (Saffran, Aslin & Newport 1996). Finally, child language learners are able to leverage sublexical features of words; this effect has been shown in children learning ASL (Caselli & Pyers 2017), while older learners of ASL struggle to detect and integrate phonological regularities (Mayberry & Fischer 1989). We sought to capture whether these and other aspects of child acquisition might play a role in conventionalization.

Our first line of inquiry addressed the degree of conventionalization of the NSL lexicon over the period spanned by the study. We examined the number of distinct signs produced for each concept. A shift from a high number of distinct signs for a concept among the first-cohort signers to a smaller number of distinct signs for that concept among the third-cohort signers would indicate lexical convergence over that period. We then examined the qualitative changes in iconicity from the first to the third cohort by documenting the prevalence of different types of iconic signs within each group. Finally, wherever an element was dropped (or added) from a sign, we noted the type(s) of iconicity evident in that element. In this way, we could determine whether pantomimic and perceptual iconicity were equally subject to change.

2.0. Methods

2.1. Participants

The participants in the present study included eight deaf signers of NSL: four from the first cohort (2 female, 2 male) and four from the third cohort (2 female, 2 male). All participants had joined the signing community before the age of six, upon entering school (range 2;1–5;8, M=3;11 (years;months)). First-cohort signers had joined within the years 1974–1983, and third-cohort signers had joined between 1993–1998. Elicitations were carried out with first-cohort participants in 2002 and 2003 (when the language was entering its third decade), and with third-cohort participants in 2017 (when the language was in its fourth decade). All participants were adults at the time of elicitation (age range 23–35y, M= 27y).

2.2. Stimuli and Procedure

Participants were videotaped seated, producing signs one at a time, over multiple sessions. For each item, they saw an image representing the targeted concept, accompanied by a Spanish word or phrase. The 26 signs that were analyzed for this study identified predominantly concrete concepts, many of which were animals and foods, but also included two more abstract concepts, hungry and jealous (see Table 1). These were a subset of signs taken from a larger ongoing study that included 300 signs and expressions, including both concrete and abstract concepts, as well as proper names and places. The signs in the current study had been successfully elicited from at least seven of the eight signers, and met the following additional criteria: (1) they labeled imageable concepts, (2) they did not label manually manipulated tools, which are being analyzed for a separate study, (3) they were not apparently borrowed from another sign language⁴, and (4) they did not apparently originate as initialized signs (that is, signs whose handshape corresponded to the first letter of a written word).

Table 1.

List of concepts elicited

banana	hammock	milk
bed	heart	pizza
chicken	horse	plantain
coffee	hungry	rice
corn	ice	soda-pop
cow	jealous	sugar
dog	lion	turtle
egg	mango	watermelon
frog	meat

2.3. Coding

2.3.1. Variability

We determined the diversity of responses produced within each cohort by first summing the different signed articulations (minus one) produced by signers across the group for a single concept, then dividing that sum by the number of signers in the group (minus one) who produced a sign for that item.

$$\mathit{\text{Variability}}\text{=}\underset{\left(n-1\right)}{\underset{\_}{\left(\mathit{\text{unique}}\mathit{\text{signs}}\text{-}\text{1}\right)}}$$

Thus, variability scores ranged from 0 to 1, such that all signers within a cohort producing the same sign would yield a score of 0 (no variability) and each signer producing a different, unique sign would yield a score of 1 (maximum variability).

2.3.2. Prevalence of iconicity type

We then examined each signer’s response to each item with regard to the type(s) of iconicity evident in the sign, considering both manual and non-manual articulators. A sign was coded as pantomimic if its iconic mapping entailed a body-to-body correspondence and if any movement of the articulators corresponded to action features of the referent or someone interacting with the referent. A sign was coded as perceptual if the iconic mapping entailed a correspondence between the articulators and observable features of the referent, such as its size and shape, and the movement in the sign corresponded to the extent or position of the referent in space or the movement or path of the referent. A sign was coded as both if it simultaneously included both types of iconicity (see Figure 1). A second coder coded 30% of the data. Interrater reliability was high, with 92.9% agreement (Cohen’s Kappa=.912, p<.001).

For compound signs, with multiple independent sign-units strung together in a sequence, we coded each sign-unit separately for type of iconicity using the same criteria. In these cases, a sign could include multiple types of iconicity non-simultaneously. For example, one two-sign variant of the compound WATERMELON includes one pantomimic sign-unit in which the signer imitated holding and eating from a slice of watermelon, followed by a both sign-unit in which the signer pantomimed spitting out a seed, while simultaneously tracing the path of the seed perceptually with the index finger (see Figure 2). This multi-sign unit was accordingly coded as pantomimic + both. We coded each response with regard to whether it included signs with pantomimic, perceptual, or both types of iconicity. In this way, we determined the prevalence of each type of iconicity in the signing of each cohort.

Figure 2.

An example of a compound sign. Here, the two-sign compound WATERMELON includes one pantomimic sign-unit in which the signer imitated holding and eating from a slice of watermelon, followed by a both sign-unit in which the signer pantomimed spitting out a seed with his mouth, while simultaneously tracing the path of the seed perceptually with the index finger. This compound sign was accordingly coded as pantomimic + both.

2.3.3. Changes in iconic elements within multi-sign compounds

We then identified changes in the iconic elements included in the signs for each concept. For each sign, we examined the variants produced by the third cohort relative to the first, in order to determine whether either type of iconicity was more susceptible to change over time. For the majority of concepts at least one variant was a compound sign, that is, a sequence of multiple sign-units. We considered all cases in which a sign-unit was present as part of a compound sign produced by a member of the first cohort, but not the third (dropping), and all cases in which a sign-unit was part of a third-cohort compound sign, but not the first (adding). In each case, we noted whether the affected sign element had expressed pantomimic iconicity, perceptual iconicity, or both. In this way, we could determine whether one of these types of iconic signs was more subject to change as the lexicon was passed down.

3.0. Results

3.1. Variability

In order to capture the extent of conventionalization of each sign, we measured the variability of signs used for each item within each cohort. The first-cohort signers produced a more diverse set of signs for each concept, with a mean variability score of .73, relative to the third-cohort signers (M =.31, t(49.97)=5.13, p<.001, Figure 3). Eleven of the 26 items received the maximum variability score of 1 among the first-cohort signers, meaning that each signer produced a different, unique sign for that concept. Among the third-cohort signers, in contrast, such high variability was still present for only two of those 11 items (HEART, SUGAR). The majority (21 out of 26) of the third-cohort signers’ responses included only one or two variants for a concept.

Figure 3.

Variability in the signs produced for each concept by 4 first-cohort and 4 third-cohort signers of NSL. A score of 0 indicates that all signers produced the same sign; a score of 1 indicates that each signer produced a different sign. There is significantly more variability among the first-cohort signers than the third-cohort signers.

Notably, for the majority of concepts (22/26, or 88.5%), the signs produced by the third cohort were selected from among the signs already present in the signing of the first cohort; that is, they were a subset of the first-cohort’s signs. The four exceptions were all cases in which a new element was added to a pre-existing first-cohort sign to form a multi-sign compound. These included the signs DOG and CHICKEN, which both added a borrowed sign to a pre-existing iconic sign, the sign RICE, which added an initialized color sign (WHITE) to a pre-existing iconic sign, and the sign ICE, for which one third-cohort signer added a new pantomimic sign (holding) to a pre-existing pantomimic sign (crushing).

3.2. Prevalence of iconcity type

In our second analysis, we computed the proportion of signs produced by each cohort that exhibited pantomimic iconicity, perceptual iconicity, and/or both (Figure 5). Note that multi-sign compounds counted toward multiple types; e.g., the sign WATERMELON (Figure 3) which included a pantomimic sign-unit and a both sign-unit counted toward the proportion of pantomimic and to the proportion of both. Across both cohorts, pantomimic signs were the most prevalent. Over half the responses (M=.54) from all signers, regardless of cohort, included a sign with pantomimic iconicity only; the remaining signs were relatively equally distributed between perceptual iconicity (M=.27) and the combination of both types of iconicity (M= .30).

Figure 5.

Average proportion of participant responses exhibiting sign-units with each type of iconicity, by cohort. Multi-sign compounds counted toward multiple types e.g., a multi-sign compound that included a pantomimic sign-unit and a both sign-unit counted toward the proportion of pantomimic and to the proportion of both. Pantomimic iconicity was the most prevalent type across cohorts, and was the type that decreased the most as the language was passed from the first to the third cohort.

When comparing the two cohorts, we saw a decrease in the prevalence of iconicity in the transition from the first to the third cohort, particularly in pantomimic iconicity, where it was included in 63% of the first cohort’s signs, but only in 45% of the third-cohort’s responses (t(6)=3.95, p=.008). The decreases in the other types of iconicity were non-significant (perceptual: t(6)=0.96, p=.393; both: t(4.13)=1.23, p=.287).

3.3. Changes in iconic elements within multi-sign compounds

Our third analysis compared first-cohort signs to third-cohort signs to determine changes in the iconic elements that were present in multi-sign compounds. Specifically, we examined the type of iconicity of sign elements that were dropped from, or added to, the signs of first-cohort signers. Logically, multi-sign compounds could reflect any combination of different types of iconicity, e.g., two pantomimic signs, a pantomimic and a perceptual sign, a pantomimic and both sign, etc. For each sign element dropped or added, we noted whether it expressed pantomimic iconicity, perceptual iconicity, or both.

Many more iconic elements were dropped than added (Figure 6). Of the 26 items, 21 elicited a multi-sign compound with a pantomimic sign-unit among the first-cohort signers, and of those, 81% (n=17) dropped a pantomimic sign-unit in the productions of the third-cohort signers. Eleven of the items elicited a compound with a perceptual sign-unit, and of those 27% (n=3) were dropped. Seven elicited a compound with a sign exhibiting both types of iconicity, and of those, 57% (n=4) were dropped. In only 4 of the 26 items (15%) did third-cohort signers add an element to a first-cohort sign, and, as mentioned above, only one of those additions was an iconic element: the addition of a pantomimic holding sign (4%) to the sign ICE. While this sign-unit was not observed in any of the first-cohort variants of the sign ICE, it was observed in other signs for foods. The remaining 3 additions were borrowings⁵ (n=2) and initialization (n=1). As a result of these changes in multi-sign compounds, particularly the extensive dropping of iconic elements, many fewer items elicited compound signs (n=8) from any one of the third-cohort signers, compared to the first-cohort signers (n=22).

Figure 6.

Proportion of signs that changed from the first cohort to the third cohort, as a result of dropping or adding sign-units exhibiting each type of iconicity. Negative values represent dropping a sign from a multi-sign compound; positive values represent adding a sign to form a multi-sign compound.

An example of a sign dropping a pantomimic sign-unit across cohorts is presented in Figure 7. First-cohort signs for JEALOUS (Figure 7A & 7B) always include the pantomimic gesture of biting a knuckle, as if to suppress emotion. A third-cohort version of JEALOUS (Figure 7C) lacks the biting element, and is produced by touching the knuckle to the chin rather than between the teeth. Interestingly, the variant used by the third-cohort signer appears in the productions of some first-cohort signers, but never without the more pantomimic biting element (e.g., Figure 7B). It is interesting to note that the variant for JEALOUS preferred by the third-cohort signers is the less transparently iconic one that no longer appears as part of a multi-sign compound.

Figure 7.

An example of a multi-sign compound dropping an iconic element in the third cohort relative to the first. (A) A first-cohort signer produces a single sign JEALOUS that includes the pantomimic element of biting a knuckle of her index finger. (B) A first-cohort signer produces a multi-sign compound JEALOUS that includes biting a knuckle followed by touching the knuckle to the chin. (C) A third-cohort signer produces a single sign JEALOUS that has dropped the pantomimic biting element, retaining only the movement of the knuckle to the chin seen in (B).

4. Discussion

By spanning fifteen years early in the development of NSL, our study captured evidence of rapid change and conventionalization in an emerging lexicon. The first-cohort signers, members of the initial group that created NSL, produced several sign variants for each item they labeled, reflecting exuberant production of sign forms that, in their iconic representation, mapped to multiple aspects of a referent. The productions of the third-cohort signers were considerably less variable, reflecting a conventionalization process that pruned away sign forms, selecting just one or two forms for each concept. This pattern of many variants reducing to a smaller set of signs parallels findings observed in another emergent language, Al-Sayyid Bedouin Sign Language (ABSL) (Meir et al. 2010, Sandler et al. 2011).

We examined the role of iconic representation in this emerging lexicon, and found both pantomimic and perceptual iconicity in abundance. Indeed, all of the signs elicited in our study included at least one of these types of iconicity, and many used both. We noticed that pantomimic iconicity, with its body-to-body symbolic mappings, was used especially where signers could easily personify animate referents, such as in LION, or enact the manipulation of a referent, as in CORN; perceptual iconicity was used especially where signers could easily represent the visible characteristics of a referent with part of the body, such as the hands representing the curved woven fabric in HAMMOCK. Our analyses revealed that pantomimic iconicity was the most prevalent; however, it was also the most susceptible to decline over time. A frequent pattern of change was the dropping of iconic signs from multi-sign compounds. In this process, pantomimic signs were more likely to be dropped than perceptual ones.

These changes reveal the impact of both creation and conventionalization processes on a natural human language as it emerges. The signs we elicited in our study were a sample of the outcome of a decades-long, creative language-building process that took place over the course of regular communication within a community of hundreds of peers. Thus, we are not examining the creation of novel labels by individuals on the fly, or in the laboratory; rather, we observe creation as a protracted, interactive process. The first-cohort signers in this study were among the original group that carried out this creative language-building. Because we sampled everyday terms, we expect that the signs we captured were established fairly early in that process, and highly familiar to all of our participants, having been produced countless times. The productions of third-cohort signers include the outcome of this same creative process, with the added effect of transmission of the language over two decades. The vehicle for this transmission was children’s natural language-learning abilities. In other words, the lexicon that we sampled from the third cohort reflects the cumulative effect of hundreds of instances of language acquisition operating on the first cohort’s creation. The differences between our two groups suggest that the processes of creation and acquisition do not take advantage of iconicity in the same way.

While the first cohort evidently exploited iconicity extensively to create the everyday signs we elicited, by the third cohort, signers were not continuing to creatively add to the iconicity that was already there. Indeed, their signs had dropped many of their original iconic elements as they were learned and relearned. Thus, symbol creation makes a greater use of iconicity than symbol acquisition. Furthermore, the form of early signs indicates that the sign-creating process favors pantomimic iconicity, which directly maps the signer’s body to another body (human or otherwise), in a part-for-part fashion. This preference may be due to the directness of this mapping (Gentner 1983; Emmorey 2014), which may make it psychologically more available than perceptual iconicity. Moreover, embodiment is a powerful resource for symbol generation (Arbib 2008), that could lead to a bias toward pantomimic iconicity when producing signs. Similarly, both the directness of mapping and the availability of embodied motor representations may make pantomimic iconicity more transparent for comprehension as well, and therefore highly effective in co-creating and communicating meaning.

Given these benefits for sign creation, it is surprising that acquisition and transmission processes do not similarly favor preserving pantomimic iconicity when selecting which sign elements to retain. After all, experimental studies investigating the effect of type of iconicity in development have shown that preschoolers recognize, learn, and produce gestures or signs with pantomimic iconicity earlier than those with perceptual iconicity (Magid & Pyers 2017; Namy 2008; Ortega et al. 2017; Tolar et al. 2007), suggesting that the iconic mapping of body-to-body is most readily accessible to children. Some recent work has shown that gestures and signs that combine both pantomimic and perceptual iconicity into a single sign may be better learned by children than signs with just pantomimic iconicity, likely because they encode more semantic information into the phonological form (Caselli & Pyers 2019; Chen, Magid & Pyers 2016). Nevertheless, in our analyses we find that the proportion of iconic signs, particularly pantomimic ones, dropped during the process of transmission from the first to the third cohort. If pantomimic iconicity is so useful in acquisition, why do we observe fewer signs with pantomimic iconicity following transmission, which operates by means of acquisition?

Of course, iconic representation is not the only force influencing the form of a sign. We expect that iconicity competes with other forces that operate on signs over time, including motoric influences (e.g., Battison 1978; Meier, Mauk, Cheek & Moreland 2008), visual processing effects (Siple 1978), and pressures of communicative efficiency (Kanwal, Smith, Culberston & Kirby 2017). However, the pressures from these forces should yield similar effects over protracted creation, acquisition, and repeated use. Furthermore, the existence of these forces doesn’t explain why pantomimic and perceptual signs would be differentially affected. Perhaps perceptual signs, for independent reasons, are more likely to have features that make them less susceptible to erosion. For example, they might be more likely to use fewer parts of the body, more distal joints, and a more central location -- characteristics that make them already more compact or efficient than pantomimic signs, which tend to engage more of the body. We need to examine more closely the nature of the changes within individual signers, and within each cohort over time in order to better capture the differential impact of these forces on language creation, use, and acquisition.

One open question is whether the type of semantic category affects the pattern of iconic mappings that appear in an emerging lexicon. When nonsigners create silent gestures, they vary the type of iconic gestures produced as a function of semantic class; pantomimic iconicity appears more frequently among gestures for manipulable objects and perceptual iconicity is favored for non-manipulable objects (Ortega & Özyürek 2019). Such systematicity has also been observed in mature sign languages. For example, one study found that across sign languages, signs for tools tend to capture how a tool is handled, signs for animals tend to involve personification, and signs for fruits and vegetables tend to show object characteristics (Hwang et al. 2017). The data that we have available do not allow us to systematically contrast these semantic categories in NSL; however, we did elicit signs for six animals and five fruits and vegetables. A quick look at the productions of signers of the third cohort, who for the most part produced single signs, show that the signs for three of the six animals had some personification (two with pantomimic-only iconicity and one with both types of iconicity); the signs for the remaining three animals depicted only their perceptual features. Of the five fruit and vegetable signs, three included pantomimic iconicity that showed how a person consumes the food, and two showed both perceptual iconicity (e.g., the shape of the referent (BANANA), or the tool used to cut it (PLANTAIN) plus pantomimic iconicity (how a banana is peeled, and how a plantain is held and cut). The data we have do not reveal overwhelming preferences for one type of iconicity over the other within these semantic categories. A more comprehensive and systematic study using a larger set of signs would be a next step to understanding whether and how iconicity and semantic class interact in the earliest stages of the emergence of a lexicon.

While we did not design our study to examine compounding, it became clear that the use of multi-sign compounds interacted with the use of iconicity as the NSL lexicon was created and transmitted. Compounding was used more liberally by the first-cohort signers. Of the 26 signs elicited, first-cohort signers produced multi-sign compounds for 22 (85%), while third-cohort signers produced multi-sign compounds for only 8 (31%).

Concatenation is a basic combinatorial process, so it should not be surprising that it appears early as a word-building strategy. A signer can optimize understanding by representing multiple characteristics of a referent over multiple signs, leveraging different types of iconic mappings. In a study of spontaneous gestures produced in the laboratory (Ortega & Özyürek 2019) the majority of representations for objects (as opposed to actions), were multi-gesture combinations. Furthermore, people were better at guessing the meanings of gestures for objects that were multi-gesture combinations that leveraged different types of iconicity, such as pantomimic + perceptual (‘acting + drawing’). It is easy to see how such production and comprehension effects could influence the coining of signs, making multi-sign compounds more likely in the initial stages of language emergence. The question of why NSL compounds would reduce to single signs over transmission is similar to the question of why iconicity would ever decrease, despite being useful. Because the first cohort continues to produce the multi-sign compound variants even after decades of use, it appears to be the influence of repeated language acquisition that led to the reduction in the third cohort. Again, the effects of visual processing and motoric effort should affect everyone’s signing similarly over time, and not just the more recent cohorts’ signing. Perhaps the relevant factor is sensitivity to emerging phonological and morphological systems. In a study of ABSL, Meir et al. (2010) documented the reduction of multi-sign phrases and compounds with the emergence of compound-internal structure. Moreover, Sandler et al. (2011) proposed that conventionalization of a sign form across a group of signers may signal the emergence of phonological structure. In the case of NSL, as successive children acquired the lexicon, they were actively integrating each sign they learned into emerging phonological and morphological systems. Because part of the meaning of a word is encoded in the grammatical structure that accompanies it (including word-internal structure), signers could optimize for efficiency and systematicity in communicating meaning while producing fewer sign elements. A next step in pursuing this account would be to identify the emerging phonological and morphosyntactic structures in NSL that take some of the communicative load off of the strategy of extensively concatenating descriptive (iconic) signs.

In sum, it is clear that the human capacity for embodiment is a powerful tool for building meaning, readily applied to the creation of new symbols. The body-to-body mapping that underlies pantomimic iconicity takes advantage of an analogical representation that is most robustly available; if a concept affords a body-to-body mapping, we are highly likely to use it when generating a symbol for that concept. The robustness of body-to-body mapping does not preclude the body-to-object mapping captured by perceptual iconicity. While not as widespread as pantomimic iconicity, it too is a highly available tool for language creation. Furthermore, it is evidently less vulnerable to erosion than pantomimic iconicity during transmission. While children may be aided by iconicity as they learn new lexical items, they are also clearly skilled at acquiring forms with little or no recognizable iconic mapping, and they may also be influenced by other factors that make signs easy to learn and remember, such as having fewer parts, or systematically distinguishable forms. The differences between pantomimic and perceptual iconicity that are evident as signs are created, and emerge as they are passed down, suggest that the pressures that compete with iconic motivations for form affect language creation and acquisition differently.

To understand how a system evolves, we need to understand the details of how it replicates itself. For natural languages, the means of replication is child acquisition. The nature of children’s cognitive abilities, including their working memory and analogical reasoning, will determine which characteristics of a language are replicated, which are altered, and which are left behind. The effects of these processes will whittle away at the wealth of content in the input, sculpting it into a tightly organized system. Ultimately, the nature of a language is determined by capacities for both language creation and language acquisition. What we have captured here in the earliest stages of the emergence of Nicaraguan Sign Language reveals the combined power of these two remarkable human abilities.

Figure 4.

The different signs elicited for the concept horse. This concept elicited highly variable signed expressions from the four first-cohort signers, who produced four different sign variants, and less variability among the third-cohort signers, with three of the four signers producing the same form. The fourth signer produced a slightly different version, in which the two hands assumed a similar handshape.