Speaker
Description
The poster demonstrates how – within a very short time – LLMs underpinning text-generation and machine translation have become powerful agents for new types of language standardisation. As regards Standard English, LLMs help to entrench North American standards world-wide, although it needs to be borne in mind that the norms engineered into the algorithms do not fully correspond to traditional prescriptive notions of educated usage. As several studies have shown (e. g. Bender et al. 2021, Blaschke et al. 2024, Liu et al. 2024), LLMs tend to discriminate against small and technologically less well-resourced languages and against nonstandard varieties of the larger and well-resourced languages. To this list of targets of potential discrimination the present poster adds Standard British English, currently still one of the two global reference standards for English usage in the offline world. Other current and emerging standard varieties of the pluricentric global language tend to have more limited geographical reach and/or less international prestige and are therefore likely to fare even worse. On the other hand, LLMs can be shown to be a very friendly environment for at least some nonstandard varieties of English and multilingual practices involving English, especially when – as is the case for Jamaican Creole or ‘Spanglish’ – they are associated with the global media and entertainment industries. The poster demonstrates the resulting standardisation paradox for English. It is very likely, though, that similar developments are affecting other pluricentric languages, as well.
References:
- Bender, E. M., Gebru, T., McMillan-Major, A., & Shmitchell, S. 2021. On the dangers of stochastic parrots: Can language models be too big? In Proceedings of the 2021 ACM Conference on Fairness, Accountability, and Transparency (pp. 610–623). New York: Association for Computing Machinery.
- Blaschke, V., Purschke, Ch., Schütze, H. &. Plank, B. 2024. What do dialect speakers want? A survey of attitudes towards language technology for German dialects. ArXiv preprint, abs/2402.11968.
- Liu, Ch., Gurevych, I. & Korhonen, A. 2024. Culturally aware and adapted NLP: A taxonomy and a survey of the state of the art. ArXiv:2406.03930v1 [cs.CL] 6 June.
