Language and AI practitioners: a sociolinguistic assessment of practices

Opening remarks by Massimiliano Bampi, Director of the Department of Humanities
Discussants: Andrea Binelli and Antonio Bibbò

AI networks are now widely used to automatise and enhance a wide range of activities relating to language. Large Language Models (LLMs) chatbots are commonly used ‘to quickly look up’ a wide range of private and professional information and LLMs are integrated into various technologies that aim to automatise time-consuming processes such as summarising, translation of texts and various customer service scenarios. Some of these applications generate outputs that appear to closely resemble human language usage. However, issues remain. Language technologies continue to hallucinate, provide structurally and particularly interactionally problematic language output especially with regard to non-referential language use. One effect of this is the homogenisation of language varieties such as Irish English practices. I provide evidence of how LLMs ‘deal with’ Irish varieties of English and the assess the reasons for this by assessing the workings of current language technologies from a sociolinguistic perspective. Contrasting how AI or NLP practitioners on the one hand and sociolinguists on the other conceptualise and ‘deal’ with language, I argue that many of the issues encountered with AI technologies are not accidental but due to the persistent belief in modernist understandings of language within the industry that are invested in standardisation, homogenisation and dataism. The aim is to raise awareness about these issues and to suggest how a critical sociolinguistics perspective can help to improve the current state of play.

Organized by Center of Irish Studies in Trento.