Is Chat-GPT a grammatically competent informant?
Hubert Haider
May 2023 

This is a report on Chat-GPT's qualities as a ‘native speaker’ based on a series of grammar competence tasks on English and partly on German. The chatbot passes the tests in style, be it restrictions on wh-movement such as That-trace-effect, wh-in-situ restrictions, movement of IOs, or the Subject condition, as well as other grammatical constraints like the LLC or exceptional subcategorization restrictions (ECM infinitives). The bot's ability to resolve ambiguities is good but not perfect.

Format:	[ pdf ]
Reference:	lingbuzz/007285 (please use that when you cite this article)
Published in:
keywords:	universal grammar, no negative evidence, large language model, computational modeling, statistical learning, wh-movement, superiority, that-trace effect, subject condition, ecm, llc, ambiguity resolution, syntax, universal grammar, machine learning, semantics, morphology, cognitive science
previous versions:	v2 [May 2023] v1 [May 2023]
Downloaded:	505 times

 

[ edit this article | back to article list ]