A Data-free Universal Prior over Syntactic Structures
It is shown that a universal prior over syntactic structures emerges from a model of human language production, in which words are progressively integrated into syntactic structure, providing a data-independent structural bias for probabilistic models, including large language models.