Salvato in:
Dettagli Bibliografici
Autori principali: Dunn, Jonathan, Eida, Mai Mohamed
Natura: Preprint
Pubblicazione: 2025
Soggetti:
Accesso online:https://arxiv.org/abs/2508.16837
Tags: Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
_version_ 1866912551243087872
author Dunn, Jonathan
Eida, Mai Mohamed
author_facet Dunn, Jonathan
Eida, Mai Mohamed
contents This paper investigates false positive constructions: grammatical structures which an LLM hallucinates as distinct constructions but which human introspection does not support. Both a behavioural probing task using contextual embeddings and a meta-linguistic probing task using prompts are included, allowing us to distinguish between implicit and explicit linguistic knowledge. Both methods reveal that models do indeed hallucinate constructions. We then simulate hypothesis testing to determine what would have happened if a linguist had falsely hypothesized that these hallucinated constructions do exist. The high accuracy obtained shows that such false hypotheses would have been overwhelmingly confirmed. This suggests that construction probing methods suffer from a confirmation bias and raises the issue of what unknown and incorrect syntactic knowledge these models also possess.
format Preprint
id arxiv_https___arxiv_org_abs_2508_16837
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle LLMs Learn Constructions That Humans Do Not Know
Dunn, Jonathan
Eida, Mai Mohamed
Computation and Language
This paper investigates false positive constructions: grammatical structures which an LLM hallucinates as distinct constructions but which human introspection does not support. Both a behavioural probing task using contextual embeddings and a meta-linguistic probing task using prompts are included, allowing us to distinguish between implicit and explicit linguistic knowledge. Both methods reveal that models do indeed hallucinate constructions. We then simulate hypothesis testing to determine what would have happened if a linguist had falsely hypothesized that these hallucinated constructions do exist. The high accuracy obtained shows that such false hypotheses would have been overwhelmingly confirmed. This suggests that construction probing methods suffer from a confirmation bias and raises the issue of what unknown and incorrect syntactic knowledge these models also possess.
title LLMs Learn Constructions That Humans Do Not Know
topic Computation and Language
url https://arxiv.org/abs/2508.16837