**1. Introduction**

Classifiers present a ubiquitous, rich, and productive morphological category of structures within sign languages. Despite the myriad ways in which they exploit iconic properties of the referent they depict, similar to sublexical units of a language, their features may also be constrained by phonology. As phonological reorganisation takes place over time (Frishberg 1975; Brentari et al. 2012; Senghas et al. 2004), our study compares results from the same production task between signers of Cena, an emerging sign language of north-eastern Brazil in its third generation, and Libras, the national sign language of Brazil, to determine how handshape complexity and variation fare in two languages of different ages and sociolinguistic profiles. We aim to investigate the question of whether in a language of relative youth, we find more complex and varied classifier handshapes given that classifiers are likely unconventionalised, thus putting a greater burden on recoverability through strategies such as iconic depiction. We also explore how signers choose to encode manner and path in motion events. Considering existing findings from signers of Nicaraguan Sign Language (NSL) illustrating that later-cohort signers show a

**Citation:** Stoianov, Diane, Diná Souza da Silva, Jó Carlos Neves Freitas, Anderson Almeida-Silva, and Andrew Nevins. 2022. Comparing Iconicity Trade-Offs in Cena and Libras during a Sign Language Production Task. *Languages* 7: 98. https://doi.org/10.3390/languages 7020098

Academic Editors: Wendy Sandler, Mark Aronoff and Carol Padden

Received: 15 October 2021 Accepted: 18 March 2022 Published: 15 April 2022

**Publisher's Note:** MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations.

**Copyright:** © 2022 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (https:// creativecommons.org/licenses/by/ 4.0/).

greater preference for encoding manner and path sequentially relative to earlier-cohort signers, we are interested in whether this departure from iconic depiction represents a developmental stage of emerging sign languages generally, or whether it may be specific to the conditions under which NSL emerged.

In Section 1, we provide background on classifiers in sign languages, providing a brief summary of common categorisations of classifier types. The section also details how the phonological features of classifier complexity may be reorganised in predictable ways over time and provides a model of quantifying complexity based on the prosodic model (Brentari 1998). Finally, Section 1 contextualises both languages in the study and the cultural contexts in which they are used. Section 2 details the Haifa Clips communicative production task, developed for Sandler et al. (2005), in which signers describe basic motion events to another native signer who must correctly identify the events, and our subsequent analysis methods. Section 3 provides results, finding that, overall, Cena signers do not appear to make as much use of more complex handshapes as we may have expected given young sign languages' propensity for iconic depiction (Sandler et al. 2011; Hou 2016), and although we do find more handshape variants in one type of classifier in Cena, these may be accounted for by assimilation. Our results also show how in their encoding of movement, both Libras and Cena signers pattern more like gesturers, early-cohort NSL signers, and signers of urban sign languages. Section 4 presents discussion of the findings. We consider and further break down results in terms of complexity and variation in this section, before discussing our conclusions in Section 5.

#### *1.1. Classifiers in Sign Languages*

#### 1.1.1. An Overview

Classifiers are handshapes that denote a broad class of referents such as vehicles, people, or round objects within a given sign language. In this way, they function similarly to those of spoken languages, although in the interests of space we will not include a detailed comparison here1. Handshape is usually described as one of the sub-lexical building blocks of signs that combines with location and movement to form meaningful signs, much in the same way meaningless sounds combine to form words in spoken language. However, handshapes as they exist in classifiers are morphemic; simply put, they are meaningful. Classifiers may be combined with particular movement or location specifications to depict verb events, forming what we will call *classifier constructions*, although labels in the literature vary (c.f. Engberg-Pedersen 2010 for *classifier signs*; Cormier et al. 2012 for *depicting constructions*) depending on which model of proposed classifier representation and structure one may subscribe to. Examples of classifier constructions from British Sign Language (BSL) (Sutton-Spence and Woll 1999) and Hong Kong Sign Language (HKSL) (Tang and Yang 2007) are shown in Figures 1 and 2. In Figure 1, the signer first produces the lexical sign CAT in the image on the left, and on the right depicts an action performed by the referent with a BSL animal classifier and a motion verb. The construction in Figure 2 combines a vehicle classifier with a verb of motion producing *a vehicle arrives*. Both classifiers pick out some visual characteristic of the referent—the vehicle classifier depicts the overall shape, whereas the animal classifier highlights the salient property of its legs. This exemplifies what is known as *iconicity,* which features heavily in classifiers—some motivated relationship between form and meaning.

**Figure 1.** An animal classifier used in a classifier construction 'a cat sits' in BSL. Reprinted with permission from Sutton-Spence and Woll. Cambridge University Press 1999.

**Figure 2.** A vehicle classifier used in a classifier construction in HKSL. Reprinted with permission from Tang and Yang. 2007 Elsevier.

Classifiers vary in the properties of their referents they pick out and how they do so. In his early analysis of American Sign Language (ASL) classifiers, Supalla (1982, 1986) proposed five types: (i) size and shape specifiers (hereby SaSSes), which denote a referent by depicting its size or physical form; the hands may statically show the outer edges of the object to show its height or width (such as the thickness of a book), or they may trace its shape (such as the outline of a Christmas tree); (ii) semantic classifiers, depicting a general semantic class of objects such as an animal in Figure 1 or vehicle in Figure 2, but may be manipulated to show additional or specific details; Supalla (1986) gives the example of a tree classifier—a signer is free to modify this broad semantic category classifier to show the type of tree, be it a palm or weeping willow; (iii) body classifiers, which use the body of the signer to denote the whole body of an animate referent; (iv) body part classifiers in which parts of the body represent themselves; and (v) instrument classifiers, indirectly denoting a referent through depicting its handling or manipulation. In the manipulation of the referent object, the hands can represent themselves, or a tool being manipulated (one may think of a flat hand moved in a sawing motion to represent a knife).

Since Supalla's work, many other categories have been proposed for ASL and other languages, and terminology varies (see Tang et al. 2021 for a recent overview). Studies also vary in the number of proposed sub-types of classifiers, but the same two main types of classifiers persist throughout much of the literature even when additional categories are also proposed:


Classifiers are often discussed as a freer class of items in a sign language relative to lexical signs, in that some (not all) featural specifications of the same classifier may vary across usages based on the semantic properties of the specific referent. The movement features for an entity classifier within a classifier construction depicting a person walking may depend on their style or speed of walking, just as the type of tree depicted may determine specifics of the chosen handshape. However, whilst a signer has relatively more articulatory choices at their behest when using classifiers, it is not completely unconstrained. They must become conventionalised within a given language. Classifiers for the same group of referents vary crosslinguistically and are not always transparent despite their tendency to take advantage of iconic relations—compare the BSL vehicle classifier handshape]<sup>2</sup> to thatofASL

This crosslinguistic variation might in part be due to the iconic capacity of handshape being more varied than that of movement and location. There are often many aspects of an entity available for iconic depiction, and choices are influenced by various factors including conceptual salience (Tkachman et al. 2020), i.e., which visual aspect of the referent may be most salient such as the beak of a chicken as opposed to its feet. The choices for movement and location are not so broad, if one wants to depict iconically. In reality, referents are only 'specified' (to use a phonological metaphor) for one location at any given time, relative to other potential referents. The same may be said about orientation, and to a lesser extent, movement3. It cannot be said that an entity is specified only for one shape, however. A human being or a car could be deconstructed into many shapes, including but not limited to their overall shape. The overall shape of a referent is merely one representational choice available out of many in constructing an entity classifier. This variety of choice can be observed in experimental contexts. Schembri et al. (2005) found that when depicting the same referents, the handshapes of classifiers used by signers of unrelated sign languages varied much more than their movement or location specifications. In short, handshapes for the same referent vary widely between languages. We turn next to what may influence system-wide choices in handshape in languages, particularly those of different ages and stages of conventionalisation.

#### 1.1.2. The Phonology of Classifiers

 .

Such variety of possibility in classifier handshape may serve as one motivation to stray further away from iconicity. As users of any communicative system mediated through anatomy, signers and speakers are subject to biological principles of energy conservation and a temptation towards the path of least resistance. It is known that signs become less iconic over time (Frishberg 1975), and that such phonological reorganisation can be motivated by ease of articulation. Eccarius (2008) notes the older form of the airplane classifier handshape in HKSL is the highly marked . Over time, young signers are replacing this handshape with , wherein the index finger is extended rather than the middle. What this newer variant may lose in iconicity, it gains in ease. Only the thumb, index, and pinky fingers are controlled by muscles that allow them to extend independently with no adjacent digit to support them (Ann 2006, p. 94). This pull towards articulatorily ease appears

strong enough to transcend the lone domain of handshape. The movement features of VIDEOTAPE-RECORDER in ASL have shifted from being asymmetrical—depicting the way in which the reels really move—to being symmetrical (Figure 3). The motivation from articulatory ease is clear. Human physiology is marked by bilateral symmetry, and as such movements that are symmetrical from the midline of the body can be specified for only one path of movement, rather than the two that asymmetrical movements require. Empirically, studies on symmetry in gesture (in hearing non-sign language-learning infants) and sign (in deaf sign language-learning infants) in young children support the idea that two-handed symmetrical movement is articulatorily easier than two-handed asymmetrical movement (Fagard 1994; Cheek et al. 2001; Pettenati et al. 2010). We take this as strong evidence of their relative ease, analogously to how factors such as infant substitutions and error shape conclusions about the relative difficulty of sounds in spoken language.

**Figure 3.** The evolution of ASL VIDEOTAPE-RECORDER. Adapted from Klima and Bellugi (1979). Reprinted with permission from Klima and Bellugi. 1979 Harvard University Press.

In short, the balance of the trade-off between faithfulness to iconicity and pressures of phonology may shift over time. Such an idea is also supported by more recent work. Brentari et al. (2012) present evidence that the phonological features of classifiers show a predictable distribution in terms of complexity. Naturally, a quantifiable measure of complexity is needed for such a claim. The authors compare two types of complexity— finger complexity and joint complexity—and define them as follows. Finger complexity is concerned with which fingers are selected for a given handshape, as articulatory difficulty varies in part because of how different muscle groups support the extension of the digits of the hand. For example, it is less strenuous for the middle, ring, and pinky fingers to all share the specification of flexion or extension in a handshape (see Ann 2006 for an anatomical explanation of why). Brentari and colleagues describe the different criteria one can use to arrive at a notion of low finger complexity in handshapes: early acquisition, crosslinguistic frequency, and representational simplicity—in this case under the prosodic model (Brentari 1998)4. These criteria all overlap to capture three selected finger groups of low complexity (all, index, and thumb) shown in Table 1. Handshapes with medium complexity are those that have a single non-radial finger extended, i.e., the middle, ring, or pinky finger. Additionally, the medium complexity category captures handshapes with two selected fingers. Representational complexity determines this criterion; medium finger complexity handshapes differ from low complexity handshape by one additional feature specification. The prosodic model utilises one of the central ideas of dependency phonology5: features dominate over other features to yield possible contrasts. This is analogous to representations of vowel systems wherein the place feature [high] alone might be realised as [i], but if [high] dominates over [low] this results in [I]. For a handshape where the index and middle fingers are extended, for selected fingers [one] dominates [all]. Similarly, different place features combine in dominance relations to yield handshapes such as the ring finger alone being extended. As such feature interactions require two features, this additional feature forms the criteria for medium complexity. High complexity captures

all other possible selected finger groups that differ in the type and number of additional feature configurations they need.


**Table 1.** Handshapes demonstrating finger complexity scores according to Brentari et al. (2012).

Whilst joint complexity was not the authors' focus, they define it as follows. Fully open and fully closed handshapes are given the lowest score of 1, in which the selected fingers are fully extended, or all fingers are fully closed respectively. Flat and spread handshapes receive a higher score of 2. Flat handshapes are formed by bending the finger(s) at the base joint while the other finger joints remain extended, and spread handshapes are any in which the extended fingers are spread apart from one another. Curved and bent handshapes receive a higher score still (3), wherein all the selected finger joints are flexed to a greater (bent) or lesser (curved) degree. The category of highest joint complexity, 4, is reserved for stacked (in which each selected finger is increasingly flexed) and crossed (when selected fingers are crossed) handshapes. Accompanying examples of all groups can be found in Table 2. These complexity scores are again based on a notion of representational complexity under the Prosodic Model (Brentari 1998), but the resulting stratification accurately predicts patterns we would expect given such categorisation (Brentari et al. 2016). In acquisition, the handshapes that are the earliest acquired by ASL- (Boyes Braem 1990) and BSL-learning children (Morgan et al. 2007) are of low joint complexity. Those that are of high joint complexity, conversely, are among the latest acquired and among the most infrequent crosslinguistically (Rozelle 2003). However, whilst all handshapes of low complexity may be those that are crosslinguistically frequent and earliest acquired, the reverse does not necessarily hold. That is, handshapes that may be frequent or easily acquired may not always receive scores of low complexity. One such exception is the curved handshape6, which is considered unmarked (Battison 1978) and is frequent crosslinguistically in classifiers (Zwitserlood 2012). Its relative frequency in classifiers across languages is likely at least partially grounded in its ubiquity as a manual configuration for handling objects. As the model of Brentari et al. is motivated primarily by representational complexity, such a model will overlook influences of this type.

Moving away from notions of complexity defined by linguistic criteria, Brentari et al.'s model generally overlaps with a model of articulatory ease based on the anatomy and physiology of the hand proposed by Ann (2006), with some minor differences7. Whilst keeping in mind that representational complexity is not automatically the same as articulatory difficulty, it is pertinent to determine to what extent a phonological measure of complexity overlaps with purely motoric articulatory difficulty. In this case, this model of representational complexity does largely overlap with conceptions of articulatory difficulty based on acquisition, crosslinguistic distribution, and anatomy.


**Table 2.** Handshapes demonstrating joint complexity scores according to Brentari et al. (2012).

Considering the findings of Eccarius (2008) and Brentari and Eccarius (2010), that handshapes in entity classifiers (or *object classifiers* as they call them) have greater finger complexity and handshapes in handling classifiers have greater joint complexity across unrelated sign languages, Brentari et al. (2012) compare hearing gesturers, homesigners, and ASL and Italian Sign Language signers to shed light on whether this pattern is one imposed by some aspect of the linguistic system, or a general tendency in codification shared by signers and non-signers alike, based on iconic properties available to all. The authors found the former: gesturers demonstrated the inverse of the results from the crosslinguistic studies (i.e., greater joint complexity in object handshapes and greater finger complexity in handling handshapes). The homesigners' results mirrored those of signers but with less polarised differences, and the signers in the study replicated findings from previous research. In other words, there is a unidirectional pattern of change in phonological complexity through gesturers, homesigners, and signers respectively. Finger complexity in entity classifier handshapes increases, as does joint complexity in handling classifier handshapes. Differences in finger complexity distribution across signers, homesigners, and gesturers found by Brentari et al. (2012) can be seen in Figure 4, in which the asterisk denotes a statistically significant difference. Taken together with examples from Frishberg (1975) and Eccarius (2008), this seems to sugges<sup>t</sup> that even in a realm as iconic as classifiers, we still observe that something like handshape complexity is not distributed equally across classifier types and is subject to, at least in part, predictable phonological organisation. We take this as our starting point for the current study.

**Figure 4.** Finger complexity in object<sup>8</sup> and handling classifier handshapes across groups. Adapted from Brentari et al. (2012). Reprinted with permission from Brentari et al. 2012 Springer Nature.

#### 1.1.3. Manner and Path in Motion Events

Entity classifiers allow signers to set up a referent in space and have it undergo or perform actions. As such, they are often used by signers to depict motion events. If motion occurred, there is always a start and end point between which a referent moved; this is what is referred to as *path* movement. There is also a *manner* of movement—how a referent go<sup>t</sup> from A to B (such as rolling or walking). Perceptually, both these aspects of movement are perceived simultaneously and thus one might imagine that any linguistic expression of motion events would reflect this simultaneity. However, comparisons made among spoken languages demonstrate that they tend to segmen<sup>t</sup> a motion event into two linguistic structures, one encoding the path and another the manner (Talmy 1985). More recent research on signers and gesturers has aimed to answer the question on whether this tendency to segmen<sup>t</sup> and linearise is a general property of language, or an effect of modality. Visual information such as path and manner of movement can be easily 'stacked' in sign languages—encoded simultaneously mirroring the way it is visually experienced. Sequential encoding is of course equally possible (see Figure 5 for both methods). Nicaraguan Sign Language (NSL) offers a unique vantage point into the dynamic early stages of a developing sign language and how motion information is encoded (Senghas et al. 2004); much like hearing Spanish-speaking, non-signing gesturers, first-cohort child NSL signers tended to encode events holistically, representing the simultaneous exhibition of path and movement features as they occur in the real observed event. In second- and third-cohort child signers, significantly less simultaneous encoding was observed, and was replaced by sequential encoding at a comparable rate of frequency (Figure 6).

However, one should not rush to conclude that there is simply a unidirectional change in a language from simultaneous encoding towards sequential encoding as a language develops over time. When tested with the same materials as the NSL signers, signers of Spanish Sign Language encoded manner and path primarily simultaneously (Senghas and Littman 2004), demonstrating the tendency of urban sign languages to depict manner and path simultaneously rather than sequentially. There are mentions, however brief, in existing literature on other conditions in which linear segmentation of movement may occur. Supalla (1990) describes two types of constraints on simultaneity of manner and path in ASL classifiers: motoric constraints, where manner and path physically cannot be depicted simultaneously, and grammatical constraints, wherein existing constraints on movement prohibit a particular combination of a verb of motion with a selected classifier9. Newport and Meier (1985), De Beuzeville (2004), and Tang et al. (2007) all report instances of sign language-learning children breaking down motion events into linear constructions, the individual parts of which sequentially encode aspects of its path or manner. Newport and Meier (1985) argue that this is motivated by articulatory ease, that children may struggle to articulate classifier handshapes (which are often marked) and the path or manner of the motion verb simultaneously. Concerning the choice between linear segmentation and simultaneous encoding of manner and path, the latter is the more iconic choice in terms of temporality. That is, manner and path are simultaneous in a motion event itself, so to encode them as such is faithful to reality. Among children acquiring sign languages, in the trade-off between iconicity and articulatory ease, ease may prevail when certain aspects of manner or path are difficult to produce simultaneously for the young signer.

What the cases of later-cohort NSL signers and children in the acquisition stage seem to sugges<sup>t</sup> is that segmentation is the exception rather than the rule. The language model available to later-cohort NSL signers (who exhibit linear segmentation of motion events) was one that was undergoing rapid and multifaceted restructuring as various homesigners came together. Children in the acquisition stage are also in an atypical linguistic situation relative to other language users; they are in a transitory and temporary period when they do not ye<sup>t</sup> have a full grasp of their native language. What both these cases have in common is some unique environment in terms of language input and/or acquisition stage, and both cases involve child signers. None of this applies to signers in our study. On the other hand, further investigation may sugges<sup>t</sup> that these are not conditioning factors in accounting for

sequential encoding of manner and path, and the preference may be also found in second and third cohorts of adult signers of emerging sign languages. The current study takes one step towards answering this question.

**Figure 5.** (**A**) A gesturer encoding manner and path simultaneously; (**B**) a NSL signer encoding manner and path sequentially. Adapted from Senghas et al. (2004). Reprinted with permission from (Senghas et al. 2004) The American Association for the Advancement of Science.

**Figure 6.** Proportion of simultaneous (**A**) and sequential (**B**) movement encoding across groups in Senghas et al. (2004). Reprinted with permission from (Senghas et al. 2004) The American Association for the Advancement of Science.

#### *1.2. The Current Study*

So far, we have presented evidence that classifiers are indeed an area in which the phonological domains of handshape and movement are subject to (re)organisation over time. Subsequently, we present an exploratory analysis of classifier handshape and movement in two sign languages differing in age and sociolinguistic profile. Existing work by Sandler et al. (2011) on Al-Sayyid Bedouin Sign Language (a village sign language of Israel, henceforth ABSL) in accounting for phonetic variation details how signers of an emergen<sup>t</sup> sign language aim for highly iconic holistic depictions of referents, in lieu of contrastive primitives or phoneme-like units in a phonological system. This appears to be true of other typologically similar sign languages (e.g., Hou 2016). If it is the case that classifiers are subject to pressures from articulatory ease over time, and that signers of emerging sign

languages tend to aim for holistic iconic depictions, we may expect greater complexity and variation in Cena considering its youth relative to Libras, as Cena signers aim for specific and unconventionalised iconic depictions.

To test this, we compare responses to a video depicting a bottle falling from Cena and Libras signers. In the stimulus, the bottle falls without intervention from a visible human agen<sup>t</sup> so it elicited mostly entity classifiers and SaSSes (our criteria for assigning a classifier each of these labels is explained in Section 2.3). We present the analysis of these types separately, to tease apart the distribution of handshapes across different types of classifiers considering there may be aspects of each type of depiction (whole entity vs. size and shape) that may influence the selection of handshapes, as we have seen happens between entity and handling handshapes. The variants will be coded for complexity following Brentari et al. (2012) and Brentari (1998) and compared across languages. We will also assess variation by way of the number of handshape variants per language and their distribution. Last, we present an analysis of how movement manner and path are encoded in descriptions of motion events considering the relevant variable of language time-depth (cf. Senghas et al. 2004). The domain of classifiers is well-suited to our aims. The high degree of iconicity often found in classifiers provides an opportunity to observe other factors that may pull sign form away from faithfulness to semantics or an iconic representation, such as articulatory ease, or the emergence of sequential encoding.
