Because Language - a podcast about linguistics, the science of language. 73: Consequences of Language (with Nick Enfield and Morten Christiansen)
76 snips
Apr 3, 2023 
What changed when humans began using language? A look at how words let us plan, coordinate, and hold one another accountable, from everyday conversation to emergency calls. Then, researchers discuss what large language models reveal about learning grammar from exposure, and the long-running debate over whether language is innate. Plus, the surprising history of flipping the bird.
AI Snips
Chapters
Transcript
Episode notes
Language Builds Shared Common Ground
- Language created a shared pile of knowledge called common ground that participants continuously build together.
- This intersubjectivity lets people coordinate actions and intentions beyond immediate perception.
LLMs Show Grammar From Statistical Learning
- Large language models (LLMs) produce grammatical output by learning distributional regularities from massive text exposure.
- This empirical result challenges the claim that grammar must be innately specified to be acquired.
Data Volume ≠ Human Input Complexity
- LLMs require enormous data but they lack many forms of input children receive like embodied interaction and social feedback.
- Therefore LLM success doesn't straightforwardly prove children learn identically by raw text statistics.