Edward Gibson: Human Language, Psycholinguistics, Syntax, Grammar & LLMs | Lex Fridman Podcast #426

MIT psycholinguist Edward Gibson argues LLMs are good at linguistic form but not meaning, and that language is invented rather than innate — directly challenging Chomsky’s universal grammar.

• The Pirahã people of the Amazon have no words for exact numbers — not even ‘one’; their quantifiers mean ‘few,’ ‘some,’ and ‘many.’
• ~95% of the world’s 1,000 documented languages follow harmonic word-order patterns (verb-initial + prepositions or verb-final + postpositions), explained by dependency-length minimization.
• Gibson rejects Chomsky’s ‘movement’ theory for how questions are formed, favoring lexical copying — individual words having declarative and interrogative forms.
• Chomsky’s Universal Grammar argument (language is innate because it’s unlearnable otherwise) is undermined if movement rules are dropped from grammar theory.
• Claude Shannon wanted to work on language at MIT in the 1940s but was crowded out by Chomsky’s rising influence and moved to Bell Labs instead.
• Language death is driven by economics: communities switch to Spanish, English, or Mandarin because those languages access larger economies, not because of grammar quality.
• Gibson is skeptical that human language is uniquely compositional — he argues we simply cannot yet decode whale, crow, or primate communication systems well enough to compare.

Guests: Edward Gibson (Ted), psycholinguistics professor at MIT, head of MIT Language Lab · 2024-04-17 · Watch on YouTube