Many studies on the state-of-the-art AI technologies, such as LLMs, have found that they pose interesting challenges to a democratic state, such as the problem of unequal access to democratic participation and deliberation, unjustifiable control unfairly imposed on linguistic groups, and unequal cultural competence. As the analysis shows, LLMs offer clear benefits, such as enabling interlinguistic translation for linguistically diverse democratic states, which may otherwise lack practical support to enable functioning cross-linguistic communication.
This paper deepens the analysis by looking at it with an intralinguistic lens, where the focus is drawn not between different languages but within a linguistic group. Namely, we look at differences between dialects or varieties. AI technologies are found to exhibit serious bias to a specific type of language within a supposedly homogenous linguistic group. This points to concerns of intralinguistic justice: generative AI risks reinforcing dominant norms found within a linguistic group by privileging already hegemonic forms of language while marginalizing others.
This chapter maps the terrain of recent worries in the field, namely, how the issue of access to state-of-the-art AI technologies should be mitigated, as they pose a site where problematic linguistic power dynamics emerge. We deepen the analysis of the previous chapter by looking at intralinguistic discrepancies that arise due to the employment of LLMs, that is, discrepancies that arise within a supposedly homogeneous linguistic group (e.g., the German-speaking group within the Austrian state, or the English-speaking group within the United States).
Speaker: Seunghyun Song