Generalizations across filler-gap dependencies in neural language models
This work addresses the problem of modeling human language acquisition in AI, highlighting limitations in current neural models for linguistic generalization, though it is incremental in scope.
The study investigated whether neural language models develop shared representations for filler-gap dependencies, finding that while they can differentiate grammatical from ungrammatical cases, they rely on superficial input properties rather than structural generalizations.
Humans develop their grammars by making structural generalizations from finite input. We ask how filler-gap dependencies, which share a structural generalization despite diverse surface forms, might arise from the input. We explicitly control the input to a neural language model (NLM) to uncover whether the model posits a shared representation for filler-gap dependencies. We show that while NLMs do have success differentiating grammatical from ungrammatical filler-gap dependencies, they rely on superficial properties of the input, rather than on a shared generalization. Our work highlights the need for specific linguistic inductive biases to model language acquisition.