Ethical Considerations for Machine Translation of Indigenous Languages: Giving a Voice to the Speakers

Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics Volume 1: Long Papers, pages 4871–4897 July 9-14, 2023 ©2023 Association for Computational Linguistics Ethical Considerations for Machine Translation of Indigenous Languages: Giving a Voice to the Speakers Manuel Mager♥∗ Elisabeth Mager♯ Katharina Kann♠ Ngoc Thang Vu♦ ♥AWS AI Labs ♯Universidad Nacional Autónoma de México ♠University of Colorado Boulder ♦University of Stuttgart Abstract In recent years machine translation has be- come very successful for high-resource lan- guage pairs. This has also sparked new interest in research on the automatic translation of low- resource languages, including Indigenous lan- guages. However, the latter are deeply related to the ethnic and cultural groups that speak (or used to speak) them. The data collection, mod- eling and deploying machine translation sys- tems thus result in new ethical questions that must be addressed. Motivated by this, we first survey the existing literature on ethical con- siderations for the documentation, translation, and general natural language processing for In- digenous languages. Afterward, we conduct and analyze an interview study to shed light on the positions of community leaders, teachers, and language activists regarding ethical con- cerns for the automatic translation of their lan- guages. Our results show that the inclusion, at different degrees, of native speakers and com- munity members is vital to performing better and more ethical research on Indigenous lan- guages. 1 Introduction With the advancement of data-driven machine trans- lation (MT) systems, it has become possible to, with varying degrees of quality, to translate be- tween any pair of languages. The only precondition is the availability of enough monolingual (Lample et al., 2018; Artetxe et al., 2018) or parallel data (Vaswani et al., 2017; Bahdanau et al., 2015). There are many advantages to having high-performing MT systems. For example, they increase access to information for speakers of indigenous languages (Mager et al., 2018) and can assist revitalization efforts for these languages (Zhang et al., 2022). Research on machine translation as well as nat- ural language processing (NLP) more generally is ∗Work done while at the University of Stuttgart. moving towards low-resourced setups and multilin- gual models. Thus, the NLP community needs to open the discussion of repercussions and best prac- tices for research on indigenous languages (that in most cases are also low-resourced) since non- artificial languages cannot exist without a commu- nity of people that use (or have traditionally used) them to communicate. Indigenous languages further differ from more widely used ones in a crucial way: they are com- monly spoken by small communities, and many communities use their language (besides other fea- tures) as a delimiter to define their own identity (Palacios, 2008; Enríquez, 2019), and have in many cases also a certain degree of endangerment. Fur- thermore, in some cases, highly sensitive infor- mation – such as secret aspects of their religion – has been encoded with the help of their language (Barron-Romero et al., 2016). This is why, in re- cent years, discussions around ethical approaches to studying …

Content truncated to 500 words