A workflow proposal for construction digital corpus in sign language

Abstract

The sign language corpora currently available in linguistic research and on open-access websites consist of a video recording module, since the data belongs to a language with a gesture-visual modality, and another module for “transcribing” the data through glosses that use the writing of an oral language, such as Portuguese or English. However, this method commonly used to describe and study the grammatical properties of sign languages ​​presents a problem: the lack of a module that actually performs the function of transcription, which is to associate form and meaning through a system of written annotation of the sign. This is because, when analyzing these “transcriptions” by gloss, it is clear that they actually play the role of translation and not transcription, since a writing system equivalent to phonetic writing is not used to transcribe the signs; the translations are not representations of the sign, since they do not consider its articulation in the annotation; they are merely translations into the oral language of the meanings of what was signaled in the video. This fact presents itself as a problem for linguistic studies because it can affect the analysis and description of linguistic properties, interfering with the formulation of hypotheses about the grammatical structure of sign language by the researcher. One proposal that we put forward to address this problem is the use of a writing system for sign language that represents the phonetic articulation of the sign to compose the transcription module in initiatives to build corpora for these languages. The system that best met the objective of linearly representing the phonetic articulation of mano-visual signs of signed languages ​​was the Writing System for Sign Language (SEL) (Lessa-de-Oliveira, 2023). In this context, this dissertation had as its main objective: (a) to carry out a survey of how studies in the field of Linguistics work with data from sign languages ​​(specifically Libras) and how the corpora that support these studies are constituted; (b) to evaluate the limits and possibilities of the initiatives to build corpora for sign languages ​​found; and (c) propose a workflow for building sign language corpora that meets all the initial stages of annotation, including the possibility of phonetic and/or phonological transcriptions, that follows the guidelines for building corpora for oral-auditory languages ​​with regard to annotation possibilities, and that reuses existing technologies. To this end, a research methodology was implemented in our work that is characterized as applied research, in which we followed the steps: (i) conducting a study to understand the state of the art and surveying sign language corpora, observing their scope, possibilities, and limits; (ii) surveying important requirements for annotation and control in the production of sign language corpora; (iii) defining the metadata that can guide the cataloging of data, which must be saved in a file structure as a raw corpus (Santos; Namiuti, 2019); (iv) construction of a theoretical model based on the proposed workflow for constructing sign language corpora that includes the transcription and translation module with annotation examples based on Libras data selected from the research sources; (v) performance of a comparative analysis between the model constructed based on the proposed workflow and the corpora constructed using other schemes for Libras language investigations. From this, it was possible to demonstrate that: (1) an initiative to construct corpora for sign languages ​​that follows systematized and standardized annotation patterns is necessary for research in the area of ​​Linguistics of signed languages ​​(manual-visual), in the case of Brazil Libras, to be well-founded; and (2) the Sel script is characterized as a plausible proposal to compose the transcription module in this initiative.