overgeneralization

This paper disentangles statistical preemption from entrenchment in language models' avoidance of overgeneralizations through controlled experiments, finding that LMs exhibit abstract preemption rather than verb-specific preemption, with implications for human language learning.