Language Ideology Bias in Conversational Technology

The beliefs that we have about language are called language ideologies and influence how we create and use language technologies. In this paper, we explore language ideologies and their role in the process of language technology design using conversational technology as an illustrative example. We draw on two qualitative studies, both of which aim at discovering common language conceptualisations in the context of language technology design through collaborative work with study participants. In study 1, we use a survey, group discussions and co-design methods with technology developers. In study 2, we use a survey and group work with technology users. We found that standard language ideology is intertwined with a referential (language in its function to convey information) view on language data in the development process, and that a conceptualization of language as referential tool dominates the language technology landscape. However, participants in both qualitative studies are aware of other functions of language. Further we found that language ideologies are intertwined with public discourse about language technology, and upcoming policies on AI regulation will reinforce these ideologies. We argue that non-referential functions of language must be integrated into language models, and that the actual practices of both language and language technologies must be carefully considered for improved conversational AI and effective policies.