Get Started
Home
Topics
Search
Library
Research questionWhich sentence statistics let neural networks learn hierarchical syntax, and how much data do they need?Local patterns can support multiple parses, while correlations across broader scales may resolve those ambiguities. It remains unclear which statistics enable hierarchical representations and how their presence changes the amount of data required.
AI
Machine Learning
Natural Language Processing
Statistical Machine Learning
Latest papersRecent research connected to this question, newest first.Deep networks learn to parse uniform-depth context-free languages from local statisticsThe evidence concerns a tunable class of probabilistic context-free grammars with controllable ambiguity and cross-scale correlations. Predictions are examined with an inference mechanism inspired by deep convolutional structure and validated on deep convolutional and transformer architectures.research paper · Sep 3, 2026
Related questions
How can noisy LLM conditionals support coherent probabilistic inference over structured variables without order-dependent generation?How do attention and recurrent state divide factual retrieval from output style in hybrid language models?How can grammar-constrained decoding preserve syntactic validity without distorting an LLM’s output probabilities?How can language models sparsely access a growing internal state without limiting it to fixed-size memory?