Learning From Failure: Data Capture in an Australian Aboriginal Community

Proceedings of the 60th Annual Meeting of the Association for Computational Linguistics Volume 1: Long Papers, pages 4988 - 4998 May 22-27, 2022 c⃝2022 Association for Computational Linguistics Learning From Failure: Data Capture in an Australian Aboriginal Community Éric Le Ferrand,1,2 Steven Bird,1 and Laurent Besacier2 1Northern Institute, Charles Darwin University, Australia 2Laboratoire Informatique de Grenoble, Université Grenoble Alpes, France Abstract Most low resource language technology devel- opment is premised on the need to collect data for training statistical models. When we follow the typical process of recording and transcrib- ing text for small Indigenous languages, we hit up against the so-called “transcription bot- tleneck.” Therefore it is worth exploring new ways of engaging with speakers which generate data while avoiding the transcription bottleneck. We have deployed a prototype app for speak- ers to use for confirming system guesses in an approach to transcription based on word spot- ting. However, in the process of testing the app we encountered many new problems for engagement with speakers. This paper presents a close-up study of the process of deploying data capture technology on the ground in an Australian Aboriginal community. We reflect on our interactions with participants and draw lessons that apply to anyone seeking to develop methods for language data collection in an In- digenous community. 1 Introduction For decades, the work of collecting data for In- digenous languages has been the province of docu- mentary and descriptive linguistics (Bouquiaux and Thomas, 1992; Vaux and Cooper, 1999; Meakins et al., 2018). This work has involved various kinds of elicitation, e.g. of word lists, phrases, etc, to sup- port description of the phonology, morphosyntax, and grammar of the language. It has also involved the collection of unrestricted text, through record- ing and transcription. In most cases, the result is audio with aligned text. Many software tools have been developed for supporting these activities (Boersma, 2001; Clark et al., 2008; Hatton, 2013; Sloetjes et al., 2013). Within the field of natural language processing, established practice is to support the linguist’s work (Michaud et al., 2018; Seifart et al., 2018; Foley et al., 2018; Cox et al., 2019). In some cases, this includes the participation of speakers in activities using apps controlled by linguists (Bird et al., 2014; Hanke, 2017; Bettinson and Bird, 2017). However, the premise is basically the same: obtain a substan- tial quantity of audio and transcribe it, or post-edit the output of an automatic transcription system. We believe that these approaches do not ade- quately address a fundamental reality of small lan- guages: they are oral. There may be an official orthography, but it has no place in the local lan- guage ecology where any written business takes place in a language of wider communication. As a result, local people are usually not confident in the orthography of the language. Furthermore, there may be low confidence in using computers and text editors, and inadequate support for the language in terms of keyboarding and spelling correction. Add to all this the …

Content truncated to 500 words