LLMs show statistical preemption for 120 verb-construction pairs, with surprisal driven by competing-form frequency rather than verb frequency, scaling as a power law with size, and causally shifted by controlled fine-tuning.
A systematic framework for generating novel experimental hypotheses from language models
3 Pith papers cite this work. Polarity classification is still indexing.
3
Pith papers citing it
abstract
Neural language models (LMs) have been shown to capture complex linguistic patterns, yet their utility in understanding human language and more broadly, human cognition, remains debated. While existing work in this area often evaluates human-machine alignment, few studies attempt to translate findings from this enterprise into novel insights about humans. To this end, we propose a systematic framework for hypothesis generation that uses LMs to simulate outcomes of experiments that do not yet exist in the literature. We instantiate this framework in the context of a specific research question in child language development: dative verb acquisition and cross-structural generalization. Through this instantiation, we derive novel, untested hypotheses: the alignment between argument ordering and discourse prominence features of exposure contexts modulates how children generalize new verbs to unobserved structures. Additionally, we also design a set of experiments that can test these hypotheses in the lab with children. This work contributes both a domain-general framework for systematic hypothesis generation via simulated learners and domain-specific, lab-testable hypotheses for child language acquisition research.
fields
cs.CL 3years
2026 3verdicts
UNVERDICTED 3representative citing papers
Collocational bootstrapping via co-occurrence regularities enables neural networks to learn subject-verb agreement robustly when input variability matches child-directed speech, indicating it as a viable acquisition strategy.
LMs develop shared yet item-sensitive filler-gap mechanisms with limited data but require substantially more data than humans to match generalizations.
citing papers explorer
-
Do Language Models Know What Not to Say? Causal Evidence for Statistical Preemption in LLMs
LLMs show statistical preemption for 120 verb-construction pairs, with surprisal driven by competing-form frequency rather than verb frequency, scaling as a power law with size, and causally shifted by controlled fine-tuning.
-
Collocational bootstrapping: A hypothesis about the learning of subject-verb agreement in humans and neural networks
Collocational bootstrapping via co-occurrence regularities enables neural networks to learn subject-verb agreement robustly when input variability matches child-directed speech, indicating it as a viable acquisition strategy.
-
Filling in the Mechanisms: How do LMs Learn Filler-Gap Dependencies under Developmental Constraints?
LMs develop shared yet item-sensitive filler-gap mechanisms with limited data but require substantially more data than humans to match generalizations.