Diachronic changes the Nicaraguan sign language classifier system: Semantic and phonological factors

2.1 LSN

LSN exemplifies a rare phenomenon in the study of language emergence and development. Its origins can be traced back to the late 1970s and early 1980s, when the government in Nicaragua established special education programs for deaf children. Prior to these programs, deaf individuals primarily utilized individual homesign systems – personal gesture systems typically developed within individual families. The congregating of these children in educational settings allowed for the interaction of these individual homesign systems, leading to horizontal transmission of signs and the spontaneous creation of a new, shared sign language (Coppola 2021, Polich 2005).

The initial phase of LSN’s development was characterized by the contributions of the first cohort of signers (Cohort One, referred to as LSN1), consisting of older children and adolescents. These individuals brought their personal homesign systems into a communal environment, thereby engaging in horizontal transmission. This nascent stage of LSN featured basic grammatical structures and simplified syntax, as documented in numerous studies (Senghas et al. 2004, Senghas and Coppola 2001).

As younger children, forming the second cohort (Cohort Two, referred to as LSN2), entered the educational programs, they were afforded the opportunity to learn LSN from first-cohort signers. This vertical transmission allowed these LSN2 signers to introduce further linguistic innovations, resulting in more complex grammatical features and a richer language structure. Research has shown that the second cohort contributed significantly to the linguistic complexity of LSN (Senghas 2003, Senghas and Coppola 2001). The progression and refinement of LSN has been well-documented, affording linguists a look at the dynamic nature of language development and the role of human interaction in shaping linguistic systems (Coppola 2021).

The creation of a new sign language in Nicaragua over the past five decades offers us the opportunity to observe the earliest stages of conventionalization of classifier predicates as they emerge. Before the 1970s, deaf Nicaraguan individuals had little contact with one another (Kegl and Iwata 1989, Polich 2005, Senghas et al. 2004, Coppola and Brentari 2014, Coppola and Newport 2006, 2021, Brentari et al. 2012, Goldin-Meadow et al. 2015). Since the earliest days of LSN research, the participation of these three groups – individual homesigners, LSN1 and LSN2 – has been critical to our understanding of the emergence of the grammar of this language.

Homesigner: An individual homesigner is a deaf individual who has not been exposed to a conventional sign language and instead develops a unique system of gestures and signs to communicate with those around them, typically their immediate family. These individuals come from diverse social situations, and as such, it is impossible to present an all-encompassing description of their backgrounds. In the current study, all homesign participants would be most accurately characterized as individual homesigners – users of a linguistic system created by a single deaf individual – rather than users of family or communal homesign system (Goico and Horton 2023). In individual homesign contexts, these systems are created independently, within each household, and are not shared across a broader community. The majority of deaf individuals in Nicaragua are not part of a signing community and do not know LSN. Due to a variety of social, geographic, and financial obstacles, they do not attend a deaf school or regularly interact with other deaf people. Individual homesigners represent the initial stage of language development, providing the raw material necessary for gesture systems to coalesce into a more standardized form of communication when these individuals are brought together (Coppola and Newport 2005, Goldin-Meadow 2003, 223–5).

Cohort One: Cohort One (LSN1) refers to the first group of children who were brought together in the newly established special education programs for the deaf in Nicaragua in the late-1970s and early-1980s. These individuals, who had developed their own individual homesign systems, began to interact and form a communal sign language. Their combined gestural inputs and communicative needs led to the creation of the earliest form of LSN. This cohort laid the foundational grammatical and lexical structures of the emerging sign language (Senghas 1995, Kegl 1999).

Cohort Two: Cohort Two (LSN2) is made up of the children who joined the Nicaraguan special education programs in the mid-1980s to the early-1990s. These children learned LSN from the first cohort and, in doing so, introduced further innovations and complexities into the language.

Their contributions included the refinement of grammatical rules and the expansion of the linguistic repertoire. The linguistic input from LSN1, combined with the natural language acquisition processes for the younger LSN2 signers, resulted in a more structured and sophisticated sign language, showcasing a clear developmental progression among stages of language development (Senghas and Coppola 2001, Senghas 2003). Subsequent cohorts have been developing LSN utilizing the same mechanisms as LSN2, by interacting with members of their own cohorts as well as those of previous cohorts (Coppola 2021).

2.2 Concepts under investigation

Whole-entity classifiers in manual modality languages are a type of classifier representing an entire object or entity rather than its parts or specific attributes (Engberg-Pedersen 1993). This type of classifier uses a specific handshape that semantically corresponds to the general category of the referent, such as a human, vehicle, or small animal. The primary phonological elements of the classifier handshape at issue include the selected fingers, joint configurations, and the possible hand-internal movement of those joints. Both of the handshapes that are the target of these investigations are whole entity classifiers:

Cl-1 classifiers: Cl-1 () classifiers are characterized by the extended index finger pointing upwards. Morphologically, this handshape can refer to a variety of elongated items, but here only tokens in which the handshape represents the upright human torso and is used to denote anthropomorphic referents, including humans and fictional human-like animals, are considered. It iconically depicts the general form and posture of an upright human figure (Supalla 1986). Morphologically, these forms will be referred to as Cl-1, and phonologically, these forms will be referred to as the ‘1-handshape,’ which refers specifically to the selected finger group using the index finger.

Cl-2 classifiers: Cl-2 () classifiers are characterized by the extended index and middle fingers pointing downwards. Morphologically, this handshape represents the legs of an anthropomorphic figure and is used to denote similar referents to Cl-1, including humans and fictional human-like animals. The two fingers iconically depict the general form and movement of the legs (Brentari 2019, Emmorey 2003). Morphologically, these forms will be referred to as Cl-2, and phonologically, these forms will be referred to as the ‘2-handshape,’ which refers specifically to the selected finger group using the spread index and middle fingers.

2.3 Variations upon the two primary handshapes

Horizontal (‘h’; e.g., ): Horizontal handshapes are those in which the active part of the hand (i.e., the palm side of the finger) is oriented parallel with the plane of movement. A horizontal classifier need not be produced horizontally. Within our dataset, horizontal classifiers were regularly used to indicate Sylvester’s climb up a drainpipe. Because the plane of movement (i.e., supporting surface, the drainpipe) runs vertically, the motion must track in parallel with that of the supporting surface (i.e., vertically) in order to be considered a horizontal production of the classifier.

Bent (‘b’; e.g., ): A bent handshape is one which includes any discernible bend in the proximal interphalangeal joints (PIP, the middle finger joint), or distal interphalangeal joints (DIP, the joint closest to the fingertip).^[2] This categorization excludes the slight bending that occurs passively when the hand is inactive.

Stacked (‘k’: e.g., ): In a stacked handshape, the fingers are not in the same plane. In Mandel’s (1981) dissertation, he refers to the fingers as ‘stair-stepped’ or ‘differentially extended;’ by definition, this variation is only relevant to the Cl-2.

Movement (‘m’: e.g., ): Hand-internal movement is defined as any increase or decrease in the angle of the metacarpophalangeal joints (MCP, or knuckle joint), PIP, or DIP, over the course of the production of the sign.

Based on previous work on the phonological structure of signs alone, we might predict that because the handshape of Cl-1 has a generally simpler structure (see Brentari 1998, Engberg-Pedersen et al. 2025), it would (i) be more frequent overall (Battison 1978, Eccarius 2008, Hara 2003, Rozelle 2003), (ii) have more subtypes in terms of joints and orientation (Brentari 1998), and (iii) occur more frequently in 2-handed signs with different handshapes, since it can also appear on the non-dominant hand (Battison 1978, Eccarius and Brentari 2007).

From the perspective of semantics, the notion of volitionality will be important for the analyses that follow, as it is the characteristic we will be tracking among cohorts and the notion relevant to the development of LSN over apparent historical time (Labov 1994). For our purposes, volitionality refers to the characteristic of a referent in a predicate indicating whether the referent intended to perform the action or, in the case of stative classifier predicates, intended to be in a particular state or location (Dowty 1991, Jackendoff 1992, Van Valin 2005).

In the case of predicates demonstrating motion or action, the notion of volitionality can be quite intuitive; Sylvester climbing the inside of a drainpipe would be considered volitional, as this was his intended action; however, Sylvester falling down that same drainpipe would be considered non-volitional. Stative classifiers, on the other hand, can be much less intuitive. Our annotation criteria defined volitionality in a stative classifier as being characterized by the actions that brought a referent to their location. Sylvester standing on a windowsill would be annotated as volitional, as the actions bringing him to that location were his own and performed intentionally. Sylvester sitting in a pile of garbage, after having been thrown out of the building, would be annotated as non-volitional, as the actions bringing him to that location were not his own and were performed without intention on his part.

Based on an analysis of ASL and Danish Sign Language (DTS) (Engberg-Pedersen et al. 2025), we hypothesize that there will be an association between (i) the volitionality of the referent and the choice of the classifier, and (ii) the use of hand-internal movement.

3.1 Populations

Nicaraguan Participants included four individual homesigners (three male, one female; ages 20, 24, 28, and 29), seven LSN1 signers (four male, three female, ages 33–43; year of entry 1974–1982); and five LSN2 signers (three male, two female, ages 21–26; year of entry 1989–1994). All participants were deaf from birth or early childhood, and the LSN2 signers all began signing by the age of 5, typically when they entered school.

ASL Signers: Fifteen ASL signers also participated (nine male, six female; age, 24–67). The narratives were collected in three different data collection periods: 2003, 2006, and 2013. The data from 2003 and 2013 were collected in the Brentari lab, and the 2006 data were collected in the signers’ homes. All of the signers were native signers with two deaf parents, or early learners, who acquired ASL before age 5.

3.2 Procedures

Participants were invited to watch the entire Canary Row cartoon. The cartoon was then shown a second time, episode by episode, with intervals for retellings of each episode after each episode was shown. The retellings were carefully recorded for subsequent analysis, ensuring that the natural signing behaviors of the participants were preserved. For individual homesigners, the task involved explaining the plot of each episode to a family member or friend who was familiar with the individual homesigner’s system, reflecting the naturalistic context in which individual homesign systems typically operate. Participants from LSN1, LSN2, and ASL interacted with either another member of their cohort or a researcher. All procedures were held constant, and only interlocutor type – family member, researcher, cohort member – was changed between groups. Importantly, retellings were always presented to an interlocutor rather than directly to the camera, maintaining the interpersonal dynamics of natural signed communication. This methodological choice aimed to capture authentic language use, providing a more accurate reflection of how classifiers are employed in real-world interactions.

3.3 Annotation

To systematically analyze the use of Cl-1 and Cl-2 classifiers, we employed ELAN (Crasborn and Sloetjes 2008) to annotate the productions of the participants. This tool enabled us to document each token of a Cl-1 or Cl-2 classifier, including variations upon those two primary handshapes, in our datasets. In addition to creating a count of Cl-1 and Cl-2 use for each of our sub-corpora, a collection of factors, ranging from the phonological to the discourse level, was annotated for each occurrence of the two target classifiers: Cl-1 and Cl-2. For the annotation tiers used in this crosslinguistic project, refer to Kimmelman (2025).

We now turn to the phonological and semantic analyses of the variation in the use of Cl-1 and Cl-2. We ask (i) Does the complexity of the overall resulting phonological structure constrain the choice of classifier (or properties of classifiers) – in general, or across LSN cohorts? (ii) Do specific formal meanings, such as volitionality, or combinations of meanings, constrain the choice of classifiers (or properties of classifiers) – in general, or across LSN cohorts?

4.1 Phonological analysis of Cl-1 and Cl-2

The purpose of the phonological analysis is to determine the phonological distribution of the two targeted handshapes for Cl-1 and Cl-2; they will be called ‘1-handshape’ and ‘2-handshape’ to emphasize their status as selected finger groups, not as morphemes. We will address their distribution on the dominant hand alone and in combination with the non-dominant hand as whole signs.

In terms of the analysis of the handshapes themselves, we consider the feature structures needed to capture the 1-handshape and 2-handshape (Figure 1). The phonological structure of the 1-handshape is simpler than that of the 2-handshape.

Figure 1

Phonological feature trees in the Prosodic Model (Brentari 1998): (a) the 1-handshape and (b) the 2-handshape.

The two main classes of handshape features (Selected Finger and Joint Configuration) can be divided into low, medium, and high complexity forms (Brentari et al. 2012, 2017). The 1-handshape (Figure 1a) is a low-complexity selected finger group from a structural point of view because it has a non-branching selected finger structure and employs the default ‘extended’ joint configuration (not shown).^[3] In addition to their simpler structural configurations, low-complexity handshapes are the most frequent handshapes cross-linguistically (Hara 2003, Eccarius and Brentari 2007) and are the earliest handshapes acquired by native signers (Boyes Braem 1981). The 2-handshape (Figure 1b) is a medium complexity handshape because it includes an extra finger (and an extra feature), and because the fingers are spread, adding a non-default feature for joint configuration.

For the analysis of 1-handshape’s and 2-handshape’s participation in two-handed signs, we consider their distribution in Battison’s four categories of signs, based on the complexity of their structure (Battison 1978). The categories are as follows: Type 0 signs are 1-handed signs (e.g., think). Type 1 signs are those in which both hands are active and perform identical motor acts (e.g., meet). Type 2 signs are those in which one hand is active (the dominant hand, called ‘H1’) and one hand is passive (the non-dominant hand, called ‘H2’), but both hands have the same handshape (e.g., sit). Type 3 signs are those in which H1 moves (active) and H2 is static (passive), and the two hands have different handshapes (e.g., touch). The handshapes that typically appear on H2 are restricted to a few variants of the whole hand or the index finger.^[4] These handshapes are called ‘unmarked’, and all other handshapes ‘marked.’ Examples of each type of sign are provided in Figure 2 ^[5] (Hulst 1996).

Figure 2

Examples of Battison’s types of signs as represented in the data: (a) a Type 0 sign produced by an LSN1 signer (one-handed), (b) a Type 1 sign produced by an LSN1 signer (two-handed, same handshape on H1 and H2, both moving), (c) a Type 2 sign produced by a homesigner, (two-handed, same handshape on H1 and H2, one hand moving) and (d) a Type 3 sign produced by an LSN2 signer (two-handed, different handshapes, one hand moving).

Simpler phonological structures are more frequent across the languages of the world, both signed and spoken (Ladefoged and Maddieson 1996, Brentari 1998, Hara 2003, Rozelle 2003). Based on previous work on the phonological structure of signs alone, we might formulate the following hypotheses:

(1) Hypotheses concerning phonological complexity:

Hypothesis 1: 1-Handshape will be more frequent overall than 2-handshape (Battison 1978, Hara 2003, Rozelle 2003).

Hypothesis 2: 1-Handshape will have more subtypes in terms of joints and orientation than 2-handshape (Brentari 1998).

Hypothesis 3: 1-Handshape will occur more frequently in Type 3 signs than 2-handshape (Eccarius and Brentari 2007).

We know that classifier handshapes interface with the morphology, syntax, and discourse components of the grammar as well, such as the interaction of classifiers with argument structure (Benedicto and Brentari 2004, Gökgöz 2024), causativity (Tang et al. 2018), and viewpoint (Perniss and Özyürek 2008); therefore, other factors besides phonology might influence the choice of handshape. By analyzing data from the Nicaraguan groups and ASL, we can determine how phonological factors are or are not prioritized in classifier choice in a young sign language, and also take a snapshot of the phonological systems of each of the groups – individual homesigners, LSN1 signers, and LSN2 signers. Note that all statistical results are based on the Mann–Whitney U-test of ranked comparisons.

4.1.1 Handshape distribution

First, we examine the 1-handshape and the 2-handshape overall, without consideration of the type of sign in which they occur. We hypothesized that 1-handshape be more frequent overall; however, that is not what we found. Pairwise comparisons across groups revealed no significant differences between pairs of groups for either 1-handshape or 2-handshape. Within-group comparisons showed that LSN1 is the only group with a significant preference, and that is for using 2-handshape (Mann–Whitney U: U = 0, z = −3.0666, ***p = 0.0002) (Figure 3).

Figure 3

Proportion of 1-handshape (1-hs, green bars) and 2-handshape (2-hs, black bars) across groups (The proportions are calculated by dividing proportion of 1-hs or 2-hs forms by the total of 1hs + 2hs.).

Next, we hypothesized that Cl-1 would have more subtypes than Cl-2, so we counted the number of bent, stacked, and moving variants for both 1-handshape and 2-handshape. This hypothesis was not confirmed, as shown in Table 1. In all groups, there are more subtypes of 2-handshape than 1-handshape.

Table 1

Number of subtypes for 1-handshape and 2-handshape across groups

	1-hs sub-types	2-hs sub-types
Homesign	4	7
LSN1	5	9
LSN2	5	9
ASL	6	9

We now turn to the distribution of 1-handshape and 2-handshape in Battison’s four types of signs: Type 0, Type 1, Type 2, and Type 3. We start by collapsing across 1- and 2-handshape how often Battison’s types appear in the dataset for all of the target groups. The results in Table 2 show that Type 3 signs are the most prevalent, and Type 0 signs are the second most prevalent across all groups in this dataset.

Table 2

Proportions and data points for each of the four types of Battison’s sign categories

	Type 0	Type 1	Type 2	Type 3
Homesign	0.14 (N = 11)	0.03 (N = 2)	0.20 (N = 14)	0.63 (N = 45)
LSN1	0.31 (N = 62)	0.13 (N = 26)	0.10 (N = 19)	0.46 (N = 90)
LSN2	0.24 (N = 55)	0.06 (N = 15)	0.14 (N = 24)	0.56 (N = 133)
ASL	0.06 (N = 18)	0.05 (N = 15)	0.06 (N = 19)	0.83 (N = 279)

Finally, we turn to the proportion of 1-handshapes and 2-handshapes that are produced in the two most frequent categories in the dataset (Type 0 and Type 3), which are also the categories that are the most distinct – one-handed signs vs two-handed signs that have different handshapes. The results are shown in Figure 4.

Figure 4

The proportion of 1-handshape (blue bars) and 2-handshape (yellow bars) across Nicaraguan groups and ASL in Type 0, Type 3, and overall.

The dataset analyzed here can provide insight into the way that Battison’s signs are treated in the overall morphophonological system, even though it addresses only 2 classifier handshapes. First, since there are differences across populations, we can infer that the degree to which Battison’s sign types are active in a relatively new language, such as LSN, can vary; we count the number of handshapes that are marked (i.e., they are not from the set of ‘unmarked’ handshapes typically found on H2) BASCO15. The homesigners use ‘marked’ handshapes more often than the other populations (homesigners: 17, LSN1: 5, LSN2: 18, and ASL: 1). This indicates that the restrictions on Type 3 two-handed signs are not as strong as they are in LSN groups or in ASL.

The phonological profile of each of these groups in the current analysis offers insights into the use of morphophonology, and we see that phonology alone will not explain the distribution of the 1-handshape and 2-handshape as classifiers. Returning to the hypotheses in (1), we find that even though previous work has shown that the 1-handshape is more frequent than the 2-handshape across the whole lexicon (Battison 1978, Hara 2003, Rozelle 2003), this is not the case in this classifier dataset. We must therefore be cautious not to overgeneralize from this small classifier dataset to the whole classifier system or to the whole lexicon. These analyses show that when these handshapes are used as classifiers their distribution can be quite different than in the lexicon as a whole.

4.2 Semantic analysis of volitionality

The semantic analysis aimed to uncover patterns of form-meaning pairings in these two classifiers (CL-1 and CL-2) by analyzing the characteristics of the lexical entries in the collected dataset. The characteristics annotated include volitionality, handshape configurations, and hand-internal movement. As stated in Section 1, volitionality refers to the characteristic of a referent in a predicate indicating whether the referent intended to perform the action or, in the case of stative classifier predicates, intended to be in a particular state or location (Dowty 1991, Jackendoff 1992, Van Valin 2005). Based on an analysis of ASL and DTS (Engberg-Pedersen et al. 2025), our hypothesis is that we would find an association between the volitionality of the referent and (i) the choice of the classifier, or (ii) the use of hand-internal movement.

The annotated data were first exported from ELAN as a tab-delimited file (TDF). This TDF was subsequently converted into a comma-separated values (CSV) format using Visual Studio Code (Microsoft Corporation 2024) leveraging Python (Python Core Team 2019) for the conversion process. The resulting CSV file was then imported to R Studio (Posit Team 2024) for statistical processing and analysis using the R statistical programming language. This workflow ensured that the data were accurately formatted and prepared for comprehensive statistical analysis, facilitating the exploration of patterns and correlations within the dataset.

The R packages dplyr (Wickham et al. 2023), tidyverse (Wickham et al. 2019), esquisse (Meyer and Perrier 2024), and report (Makowski et al. 2023) were utilized for statistical processing and analysis. The Large Language Model ChatGPT (OpenAI 2024) was used to debug and simplify code throughout.

To quantify the statistical correlations between volitionality and handshape (i.e., selected finger group), volitionality and hand-internal movement, as well as between hand-internal movement and handshape, Fisher’s exact tests were conducted. Fisher’s exact test was chosen over the more common chi-squared test to accommodate the idiosyncratic nature of the smaller datasets available in an emerging language context such as LSN. These tests allowed us to examine the association between categorical variables and determine the strength and significance of these relationships.

By applying Fisher’s exact tests, we were able to statistically validate whether the observed patterns in hand-internal movement and handshape configuration during classifier predicate production were influenced by volitionality. A classifier predicate with a volitional referent is one in which the referent has chosen their state, location, or the action they are performing. A classifier predicate with a non-volitional referent is one in which the referent has not chosen their state, location, or the action they are performing. The statistical processing of the data provided the following insights into the factors influencing classifier selection and sub-morphemic linguistic structures within and across Nicaraguan groups and ASL.

4.2.1 Increase in non-volitional classifier across Nicaraguan groups

As shown in Figure 5 (as compared with Figure 3), we see that while individual homesigners use both Cl-1 (1-handshape) and Cl-2 (2-handshape), they do so almost exclusively in volitional contexts. They did not incorporate non-volitional classifier predicates into their individual homesign systems at the frequency of later cohorts. While all groups used classifiers in non-volitional contexts, and all groups demonstrated hand-internal movement, non-volitional classifier predicates represented only 4.1% of the classifier predicates elicited by Canary Row retellings among individual homesigners. LSN1 more than doubled the individual homesigners’ prevalence of non-volitional classifier predicates, with non-volitional classifier predicates representing 10.8% of classifier predicate productions. LSN2 nearly doubled LSN1’s use of classifiers in non-volitional contexts yet again, with 18.4% of classifier predicates used non-volitionally. The ASL participants align much more closely with LSN2, with 15.8% of classifiers being produced in non-volitional contexts.

Figure 5

Proportion of tokens from each corpus which are non-volitional on a 0% through 25% scale. 95% confidence interval in teal; total tokens used to calculate percentage in white.

A pairwise z-test for proportion was conducted between the four groups. The statistical analysis reveals interesting patterns in the proportion of classifier predicates produced non-volitionally across groups. There is no statistically significant difference in the proportion of classifier predicates produced non-volitionally between individual homesigners and LSN1 (p = 0.285) despite a doubling in the absolute proportion. This indicates that while there is a noticeable increase in non-volitional classifier predicate use in LSN1 compared to homesigners, the difference is not substantial enough to reach statistical significance. Similarly, there is no statistically significant difference between LSN1 and LSN2 (p = 0.157), indicating a gradual, non-dramatic increase in the proportion of non-volitional classifier predicates as the language evolves.

However, the comparison between individual homesigners and LSN2 shows a statistically significant difference (*p = 0.028). This finding suggests that after only one generation of vertical transmission, the proportion of classifier predicates used non-volitionally becomes significantly different from the original input of the homesign systems. With this slow and steady increase in non-volitional use between generations, the accumulated difference by the second cohort of LSN signers marks a clear departure from the patterns seen in homesign systems, pointing to the influence of transmission and community on language creation.

4.2.2 LSN1 contextual preferences

LSN1 signers produced many more non-volitional classifier predicates than the individual homesigners (Figure 5). However, this increase in production is limited in its context. First, LSN1 non-volitional classifier predicates are almost entirely produced with the Cl-2 handshape, as seen in Figure 6a. Fisher’s exact test bears out the statistical significance of the correlation at **p = 0.002508.

Figure 6

(a) Percent of Cl-1 and Cl-2 signs observed to be volitional or non-volitional (by fill color) in the LSN1 dataset. Total annotated tokens in white. (b) Percent of volitional and non-volitional classifier predicates observed to contain hand-internal movement (by fill color) in the LSN1 dataset. Total annotated tokens in white.

Second, looking exclusively at LSN1 we observed no hand-internal movement in non-volitional classifier predicates, as seen in Figure 6b. A Fisher’s exact test demonstrates the significance of this correlation at ***p = 0.0008418.

4.2.3 LSN2 expands the contexts for using Cl-1 and Cl-2

The increase in production of non-volitional Cl-1 and Cl-2 classifiers was also observed in LSN2; however, the specificity of the context required for a classifier predicate to be used non-volitionally was not.

As shown in Figure 5, LSN2 demonstrates an increased use of non-volitional classifier predicates over both individual homesigners and LSN1. Moreover, LSN2 produces both Cl-1 and Cl-2 in non-volitional contexts, not Cl-2 exclusively as in LSN1 (Figure 7a). In addition, hand-internal movement is used in both volitional and non-volitional contexts (Figure 7b). Fisher’s exact tests show no statistically significant correlation between volitionality and handshape or volitionality and hand-internal movement in LSN2, at p = 0.3637 and p = 0.1196, respectively.

Figure 7

(a) Proportion of volitional and non-volitional classifier predicates (by fill) observed in Cl-1 and Cl-2 in LSN2. (b) Hand-internal movement (by fill color) produced in non-volitional and volitional contexts in LSN2. Total annotated tokens in white.

4.2.4 Movement as the domain of Cl-2

Despite these changes in usage patterns among LSN2 signers: (i) non-volitional classifier predicates being produced using Cl-1, and (ii) hand-internal movement being used in non-volitional contexts, no significant increase in hand-internal movement, as a proportion of Cl-1 classifier predicate productions, was observed (Figure 8a and b).

Figure 8

(a) Proportion of classifier predicates which contained hand-internal movement (by fill color) in LSN1 by handshape. (b) Proportion of classifier predicates which contained hand-internal movement (by fill color) in LSN2 by handshape. Total annotated tokens in white.

The lack of an increase between LSN1 and LSN2 in hand-internal movement in the Cl-1 handshape, despite the increasing use of hand-internal movement non-volitionally and increasing non-volitional uses of the Cl-1, raises an important question: Does LSN2 still make a volitional/non-volitional distinction?

4.2.5 Hand-internal movement in non-volitional contexts

The infiltration of hand-internal movement in non-volitional classifier predicates that occurs in LSN2 occurs exclusively in the Cl-2 handshape. To account for this phenomenon, we explored two possible explanations.

The first explanation might be an increase in the variety of Cl-2 forms acceptable in non-volitional contexts. Variants of the Cl-2 handshape, novel to LSN, being added to the LSN2 classifier inventory could account for this increase in hand-internal movement in the Cl-2 handshape as a whole by providing additional opportunities for non-volitional use of hand-internal movement. However, the only addition Cl-2 variant observed in LSN2 was a form of the stacked Cl-2 handshape produced with hand-internal movement, which appeared only once in our dataset (for ‘falling’). The rarity of this variant makes it a poor explanation for the increasing flexibility in Cl-2 usage observed in LSN2.

The second possible explanation considered was an increase in hand-internal movement in the already available phonological forms. Specifically, it is possible that LSN2 signers were expanding the use of hand-internal movement in the currently available variants of the Cl-2 handshape. The least marked version of the Cl-2 handshape demonstrating hand-internal movement is the inverted form with hand-internal movement at the MCP joints, iconically representing the legs of an upright human or anthropomorphic referent walking (i.e., the 2m (Cl-2 moving) handshape) (Eccarius and Brentari 2007, Hara 2003). In order to expand the use of hand-internal movement, LSN2 signers would have to increase hand-internal movement in variants outside of this 2m form.

To investigate this, a two-proportion Z-test was conducted to determine whether there was a statistically significant difference in the proportions of Cl-2 handshape classifiers performed in their bent and stacked forms between LSN1 and LSN2.

Results showed a statistically significant increase in the proportion of moving Cl-2 classifiers outside of the 2m variant *p = 0.0430672. LSN1 signers produced 8.96% their Cl-2 classifiers using non-2m variants, while LSN2 nearly doubled that percentage, using 15.53% non-2m variants of Cl-2 classifiers (Figure 9). These proportions represent a near doubling in absolute count as well (Table 3).

Figure 9

Proportion of non-2m hand-internal movement by cohort (by fill). 95% confidence interval in orange.

Table 3

Proportions of non-2m classifiers showing hand-internal movement

Cohort	Total tokens	Non-2m	Proportion	Percentage
LSN1	223	20	0.089	8.97
LSN2	251	39	0.1553785	15.53

5.1 For future investigation

This study has raised several questions that warrant further investigation to deepen our understanding of classifier predicate usage in LSN and related developing signing systems.

First, an exploration of which strategies homesigners are employing in contexts in which LSN1 or LSN2 signers would typically produce a classifier predicate would be helpful in outlining the development of classifier systems as a whole.

Second, further research is needed to examine how LSN2 signers are currently using the Cl-1 handshape in non-volitional contexts. With hand-internal movement restricted to the Cl-2 handshape, a more detailed analysis of which other contexts and phonological factors might condition the use of the Cl-1 handshape non-volitionally is called for.

Third, future studies should extend the scope of investigation to LSN3 to determine whether the trends observed in LSN2 continue in the next generational iteration of the language. Analyzing LSN3 could help clarify whether the observed changes are part of a broader trajectory of language evolution or specific to LSN2.