Edward Gibson: Human Language, Psycholinguistics, Syntax, Grammar & LLMs | Lex Fridman Podcast #426
Topics covered
Popular Clips
Questions from this episode
- Asked by 8 people
- Asked by 4 people
- Asked by 3 people
- Asked by 2 people
- Asked by 1 person
Episode Highlights
Inner Voice
Edward Gibson and discuss the concept of the inner voice and its prevalence among individuals. Gibson reveals that while most people report having an inner voice, he does not, which he finds perplexing. This leads to a broader conversation about the relationship between language and thought, highlighting the stability of language networks in the brain.
Most people have this strong percept of hearing sound in their heads when they're just thinking.
---
Fridman and Gibson also touch on the challenges of studying these phenomena, especially in young children, due to the complexities of brain scanning techniques 1 2.
Counting
The discussion shifts to how language influences counting abilities. Gibson explains that some cultures, like the Pirahã, lack specific counting words, which limits their ability to perform exact counting tasks. This absence of counting words leads to approximate counting beyond the number four.
Numbers are really another fascinating source of information here where you might naively, I certainly thought that all humans would have words for exact counting and the piraha don't.
---
Fridman and Gibson explore the implications of this limitation, suggesting that the development of counting systems may be tied to agricultural needs 3 4.
Optimization
Gibson delves into the optimization of human language for communication amid noise. He explains that word order and other linguistic structures may have evolved to enhance comprehension and production in noisy environments. This idea is rooted in communication theory, particularly the work of Claude Shannon.
Language, if it's a communication system, we are trying to optimize, in some sense, the passing of the message from one side to the other.
---
The conversation also touches on the evolution of language structures and the role of dependency lengths in making language more efficient 5 6.
Form vs. Meaning
The final segment focuses on the distinction between the form and meaning of language, particularly in the context of large language models (LLMs). Gibson argues that while LLMs excel at mimicking the form of language, they struggle with understanding its meaning. This gap poses significant challenges for developing truly intelligent language systems.
What they're good at is form. And that's, that's why they're good, because they can do form; meaning's hard.
---
Fridman and Gibson discuss the limitations of LLMs, including their inability to grasp deeper meanings and their susceptibility to errors in understanding context 7 8.