Search papers, labs, and topics across Lattice.
This study investigates how language models (LMs) avoid overgeneralizations, specifically examining the roles of preemption and entrenchment through controlled rearing experiments on LMs trained with child-caregiver conversations. The findings reveal that while LMs successfully avoid overgeneralizations, they do not exhibit verb-specific preemption; instead, they demonstrate weak evidence of abstract preemption and treat competing structures as indirect positive evidence. This challenges existing theories about how LMs process language and suggests the need for further human experiments to explore the concept of abstract preemption in learning.
Language models avoid overgeneralizations not through specific preemptive evidence, but by treating competing structures as indirect positive cues, reshaping our understanding of language acquisition mechanisms.
How do learners avoid overgeneralizations such as Tom laughed me without explicit negative evidence? Constructionists have posited two proposals that describe indirect negative evidence against overgeneralizations: preemption (which privileges exposure to near-synonymous construction---e.g., she made him laugh) vs. entrenchment (all exposures to a verb's grammatical usages, including cases like He laughed). We disentangle these hypotheses by running controlled rearing experiments on LMs trained on child-caregiver conversations, where we systematically remove preemptive vs. non-preemptive evidence. We find that while LMs avoid overgeneralizations, they do not show preemption at a verb-specific level, instead showing weak but non-zero evidence of abstract preemption. Combined with results from analyzing the LMs'training dynamics, we find that LMs treat competing structures as indirect positive---as opposed to negative---evidence in the verb-specific condition. Insofar as preemption is the more plausible route to avoiding overgeneralizations in humans, our results point the need for there to be sensitivities to indirect negative evidence in neural network learners, and suggest new human experiments to test abstract preemption.
