Botchat

Abstract

This paper investigates the ability of three chatbots to detect and correct errors in language correctness and appropriateness. The study specifically examines how these AI models respond to various text styles and contextual
cues when tasked with editing prompts. Through an open letter addressed to Orsolya Nádor, the author presents
findings derived from experiments conducted with these AI models. The primary research hypothesis posits that
chatbots are sensitive to contextual and stylistic factors, leading them to prioritize corrections that favour linguistic
appropriateness over strict adherence to language correctness.
The results indicate that chatbots are indeed sensitive to context and text style, demonstrating a preference for
linguistic appropriateness over rigid conformity to the rules of language correctness. However, in decontextualized
sentences and addressing strongly stigmatized language forms, the chatbots exhibited a greater tendency to prioritize language correctness. Additionally, the chatbots consistently applied corrections based on spelling rules,
regardless of context.
These findings confirm the hypothesis that chatbots are capable of discerning nuanced linguistic patterns and
making editing decisions accordingly, reflecting their increasing sophistication in natural language processing.