Toward More Meaningful Resources for Lower-resourced Languages

Findings of the Association for Computational Linguistics: ACL 2022, pages 523 - 532 May 22-27, 2022 c⃝2022 Association for Computational Linguistics Toward More Meaningful Resources for Lower-resourced Languages Constantine Lignos and Nolan Holley∗ and Chester Palen-Michel∗ and Jonne Sälevä∗ Michtom School of Computer Science Brandeis University {lignos,cpalenmichel,jonnesaleva}@brandeis.edu nrh2@williams.edu Abstract In this position paper, we describe our perspec- tive on how meaningful resources for lower- resourced languages should be developed in connection with the speakers of those lan- guages. Before advancing that position, we first examine two massively multilingual re- sources used in language technology devel- opment, identifying shortcomings that limit their usefulness. We explore the contents of the names stored in Wikidata for a few lower- resourced languages and find that many of them are not in fact in the languages they claim to be, requiring non-trivial effort to correct. We dis- cuss quality issues present in WikiAnn and eval- uate whether it is a useful supplement to hand- annotated data. We then discuss the importance of creating annotations for lower-resourced lan- guages in a thoughtful and ethical way that includes the language speakers as part of the development process. We conclude with recom- mended guidelines for resource development. 1 Introduction Recent years have seen increased interest from the natural language processing community in devel- oping both models and datasets for what may be termed “lower-resourced” languages. Advances in transfer learning and the increased availability of data and benchmarks in these languages have made it straightforward to create what appear to be high-performing models for these languages with little or no annotated data. Yet despite the popularity and apparent effec- tiveness of these systems, the unique challenges and best practices for developing datasets and mod- els for lower-resourced languages are rarely dis- cussed alongside the system themselves. In this position paper, through commentary on our experi- ences working with existing datasets and a discus- sion of current trends, we explore what resources *Denotes equal contribution. we believe will be most useful for the develop- ment of meaningful language technology for lower- resourced languages, advancing our perspective that open data and models and a participatory ap- proach to research are the keys to progress. Before we can elaborate further, we must address a terminological issue which creates a stumbling block when attempting to discuss resources and lan- guage technology for the languages of interest to us. Throughout this paper, we will use the term lower- resourced language to refer to languages that have received fewer resources—as measured in any num- bers of dimensions such as models, datasets, papers, funding, etc.—than the most popularly-studied lan- guages in the field of natural language processing. We explicitly use the comparative lower rather than low to emphasize the continuum that exists across languages regarding the resources available for de- veloping language technology. Whether a language is lower-resourced in a spe- cific context may depend on what is available for the task at hand. Due to singular efforts, a lan- guage that may have otherwise been …

Content truncated to 500 words