Dissociating language and thought in large language models

arxiv:2301.06627

Original source · Paper JSON

Reasoning and the "physics" of language models · 2023-01-16

Cites (2) · Cited by (3)

Complete cleaned Markdown · Source Markdown including bibliography

Read in parts

1. Part 1
2. Part 2
3. Part 3
4. Part 4
5. Part 5

Contents

• Dissociating language and thought in large language models
• Abstract
• The language-thought conflation
• Formal vs. functional linguistic competence
• What does linguistic competence entail?
• Formal linguistic competence
• Functional linguistic competence
• Motivation for the distinction between formal vs. functional linguistic competence
• The language network supports language processing in the human brain
• The language network does not support non-linguistic cognition
• LLMs have largely mastered formal linguistic competence in English
• Statistical language models: some fundamentals
• Large language models learn core aspects of human language processing
• LLMs perform well on benchmarks of diverse linguistic phenomena
• LLMs learn hierarchical structure
• LLMs learn linguistic abstractions
• LLMs learn constructions
• LLMs are predictive of activity in the human language network
• Using LLMs as models of formal linguistic competence in humans
• Excessive reliance on statistical regularities
• Unrealistic amounts of training data
• Insufficient tests on languages other than English
• Non-augmented LLMs fall short on functional linguistic competence
• A. Fine-tuning on the task and the challenge of closed models
• B. Generalizable, robust performance
• Formal Reasoning
• World models 1: factual and commonsense knowledge
• World models 2: situation tracking
• Social reasoning
• Language input can bootstrap functional competence capabilities
• Toward models that use language like humans
• Concluding remarks
• Glossary
• Highlights
• Outstanding Questions
• Acknowledgements
• Funding Sources
• Conflicts of Interest