🧑‍🔬 Michael C. Frank

Michael C. Frank studies how people understand and learn language, pairing formal computational models with experiments and corpus-based simulations. Their papers here ask how a listener works out what a speaker intends when the words alone leave the meaning open, and how children come to master language from far less input than artificial systems need. The work sits where psycholinguistics & language acquisition, reasoning, and language models & pretraining meet. It moves between tightly controlled games played by human participants and neural networks trained on transcripts of speech addressed to children.

The most cited paper here is Predicting Pragmatic Reasoning in Language Games (2012), written with Noah D. Goodman. It turns informal accounts of pragmatics into testable quantitative predictions. Participants viewed a few objects differing along two dimensions and wagered on which one a single word named. In the proposed account, comprehension is probabilistic inference. It weighs how salient each object is in context, measured independently with a separate group, against the expectation that speakers choose words that single out their target, with informativeness defined through surprisal.

The newer direction carries questions about acquisition into machine learning. Is Child-Directed Speech Effective Training Data for Language Models? (2024), with Steven Y. Feng, builds GPT-2 and RoBERTa models without pretraining, using English CHILDES transcripts. It compares them with models trained on TinyDialogues, a synthetic corpus of matching size produced with GPT-4, and on subtitles, Wikipedia and the BabyLM mixture. Syntax is probed with the Zorro benchmark and meaning with word-similarity tests. The study also checks whether ordering conversations by the child's age makes a difference.

Frank is based at Stanford University, and earlier papers also carry the Massachusetts Institute of Technology.

Michael C. Frank has 2 papers in our library, in 2012–2024, most of them on Language models & pretraining, Reasoning, Psycholinguistics & language acquisition; often with Noah D. Goodman, Steven Y. Feng.