‘Nous fêterons’ or ‘On va fêter’?

Mimicking Age-Sensitive Variation with ChatGPT

Abstract

This study explores ChatGPT’s capability to mimic age-sensitive linguistic variation in contemporary French, particularly focusing on older adult speech. Our investigation aimed to assess whether ChatGPT could (1) align its naive responses with age-related language use, (2) demonstrate explicit knowledge of age-related linguistic variation, and (3) modify responses based on such knowledge. Using contexts from the LangAge corpus, ChatGPT was prompted to answer questions from the perspective of speakers of different ages (30– 90) in different interview years (1980–2020), with a specific focus on the use of first-person plural subject clitics (nous/on) and future tenses (futur simple/proche). The results revealed that ChatGPT’s responses predominantly favored formal linguistic variants across all ages. While expert-knowledge injection significantly increased the usage of formal variants, there was no systematic influence of age, birth year, or interview year on variant selection. A partial exception is represented by speakers aged 70 for whom ChatGPT displayed heightened linguistic uncertainty in the naive answer. By contrast, the variant distribution in (3) is mainly motivated by ChatGPT’s expert knowledge generated in (2). These findings highlight the potential and limitations of current LLMs in capturing age-specific variation while encouraging further integration of sociolinguistic methods into LLM research.