Comparing formational parts of signs: a phonological approach
1. Introduction
Sign language research has long played a role in questions of human language evolution (e.g., Terrace et al., 1979) because language in a different modality provides a crucial test for understanding which aspects of language are part of a unique biological endowment and which are particular to each modality or to a particular language (or family). However, comparing the modalities will only ever be as successful as the quality of the research within each modality. In keeping with the workshop theme ‘Hands on sign language emergence: current methods’, this talk focuses on the methodological challenges and tools for studying the formational parts of signs from the perspective of descriptive phonology.
1.1. The Challenge/Problem
There are countless linguistic research questions and even whole domains of study that require comparing signs with each other on the basis of their articulatory components. This includes morphology (Fernald & Napoli, 2000; Lepic & Occhino 2018), psycholinguistics (Caselli et al., 2016), historical linguistics (Shaw & Delaporte, 2010), sociolinguistics (Fenlon et al., 2013), typology (Woodward, 1993; Parks, 2011; Omardeen 2018), etc. However, despite attempts to move toward a standard means of comparison (Hulst & Channon, 2008), individual studies typically use phonological coding that only partially maps to other coding schemas and rarely include all phonological aspects of a sign. I argue that what is needed is a comprehensive baseline or template of formational categories—units for comparison—that have empirical and theoretical validity.
Here, I describe an approach to reach such a baseline, taken in two stages. First, I show how a search for minimal pairs in one language identified a set of units for comparison, and second, I explore how these units were used to create a template in the form of a character string in order to compare the phonology of any two signs. Note that this ideally also involves a flexible database to function as both a repository for formational parts and a means of accessing signs on the basis of their parts (e.g., SignBanks; see Johnston, 2016; Crasborn et al. 2014).
2. Minimal Pairs
A somewhat novel methodology (for sign languages) was used in the analysis of Kenyan Sign Language (KSL; Morgan 2017) in order to find phonemic units within a set of 1,880 non-compound signs. First, in a FileMaker Pro database, fields were established for the basic parameters of handshape, location, movement, and orientation. These were initially based on previous phonological descriptions (e.g., Brentari 1998, Kooij 2002) and during the coding process, the specific fields and values within each field were gradually modified to fit the data in the KSL lexicon in a way that was both maximally informative and concise. In parallel to the coding, a separate dataset of possible minimal pairs was gathered one at a time through customized searches. When the phonetic coding of the lexical database was complete, an analysis of the contrastive units began. The list of potential minimal pairs was systematically evaluated and narrowed down from 926 possible pairs to 461 ‘true’ minimal pairs. These were signs that differed by only one unit of contrast, and for which video was obtained from the same signer. These pairs yielded around 30 different types of contrast (e.g., location, repetition) as well as an inventory of over 125 different KSL-specific features (e.g., [nose], [forehead], [upper-arm], [repeated], [single]).
3. Sign-to-Sign Comparison through a Character String Template
From the finite set of contrast types (‘units’) in KSL, a 40-character string was created, encompassing six phonological domains: articulator, handshape, orientation, location, core movement, and manner of movement. Except for handshape, which has two levels of coding, each unit consists of unique and non-overlapping information. Some features are conditional upon other features (e.g., [alternating] is conditional upon a sign being [2-handed]), and therefore require an algorithm to process correctly, similar to Omardeen (2018).
4. Conclusion
Solving this methodological challenge will have positive consequences for many areas of research, including language evolution. There are increasingly more tools to compare the visual parts of signs—avatars, digital 3D skeletons, machine learning. These tools are powerful and promising, yet linguists will still need a means of comparing formational units that have psychological validity for signers, and this means drawing from the work of descriptive linguistics.