Abstract: Presupposition projection remains a critical area of linguistic research, particularly in understanding how meaning is inferred beyond explicit assertion. This study explores the processing of presuppositions in conditional sentences by large language models (LLMs) in both English and Mandarin, evaluating their alignment with established linguistic theories such as Satisfaction Theory (ST) and Discourse Representation Theory (DRT). Through controlled experiments inspired by Romoli’s (2011) human subject study, we reveal considerable variation across models, both within and across languages, challenging the assumption that LLMs uniformly approximate human-like pragmatic competence. While some models exhibited patterns aligning with ST, others diverged significantly, suggesting that LLMs can produce contextually appropriate text without a structured, human-like understanding of presupposition.
Paper Type: Long
Research Area: Discourse and Pragmatics
Research Area Keywords: linguistic theories, probing, anaphora resolution, computational psycholinguistics
Contribution Types: Reproduction study
Languages Studied: English, Mandarin
Submission Number: 3402
Loading