Reproduce “Understanding Epistemic Language with a Language-augmented Bayesian Theory of Mind” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Lance Ying, Tan Zhi-Xuan, Lionel Wong, Vikash Mansinghka, Joshua B. Tenenbaum
Researchers at MIT and Harvard developed LaBToM, a computational cognitive model that integrates formal linguistic representations with a Bayesian Theory of Mind to interpret epistemic language. The model accurately predicts human plausibility judgments of statements about others' beliefs, achieving a Pearson's r of 0.76 with human ratings and outperforming multimodal Large Language Models.
Key insights
- Explicitly structured cognitive models, such as Bayesian Theory of Mind, are crucial for robust and accurate interpretation of grounded epistemic language, outperforming black-box Large Language Models.
- The "Language of Thought" hypothesis is supported by ELoT's effectiveness as a compositional, probabilistic representation for translating natural epistemic language.
- A neuro-symbolic approach, where Large Language Models act as grammar-constrained semantic parsers feeding into a structured cognitive model, provides a powerful framework for grounded language understanding.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.