How linguistics learned to stop worrying and love the language models

2025/01/28 by Richard Futrell, Kyle Mahowald, Futrell, Richard +1 · 6 voices · 21 citations
Arts and Humanities · #Linguistic Education and Pedagogy

paper · pdf · doi:10.1017/s0140525x2510112x

Abstract

Abstract Language models (LMs) can produce fluent, grammatical text. Nonetheless, some maintain that language models don’t really learn language and also, even if they did, that would not be informative for the study of human learning and processing. On the other side, there have been claims that the success of LMs obviates the need for studying linguistic theory and structure. We argue that both extremes are wrong. LMs can contribute to fundamental questions about linguistic structure, language processing, and learning. They force us to rethink arguments and ways of thinking that have been foundational in linguistics. While they do not replace linguistic structure and theory, they serve as model systems and working proofs of concept for gradient, usage-based approaches to language. We offer an optimistic take on the relationship between language models and linguistics.

Citations

• What enables human language? A biocultural framework
• Both Direct and Indirect Evidence Contribute to Dative Alternation Preferences in Language Models
• Deep Learning is Not So Mysterious or Different
• What’s Surprising About Surprisal
• LLMs as a synthesis between symbolic and distributed approaches to language
• The metaphors of artificial intelligence
• Reclaiming AI as a Theoretical Tool for Cognitive Science
• Human-like Linguistic Biases in Neural Speech Models: Phonetic Categorization and Phonotactic Constraints in Wav2Vec2.0
• Decomposing modal thought.
• AI models collapse when trained on recursively generated data
• Filtered Corpus Training (FiCT) Shows that Language Models Can Generalize from Indirect Evidence
• The Platonic Representation Hypothesis
• Bridging the Empirical-Theoretical Gap in Neural Network Formal Language Learning Using Minimum Description Length
• Power Hungry Processing: Watts Driving the Cost of AI Deployment?
• Over-reliance on English hinders cognitive science
• On Logical Inference over Brains, Behaviour, and Artificial Neural Networks
• Understanding deep learning (still) requires rethinking generalization
• Why Neurons Have Thousands of Synapses, a Theory of Sequence Memory in Neocortex
• Expectation-based syntactic comprehension
• Critical period effects in second language learning: The influence of maturational state on the acquisition of English as a second language
• The TRACE model of speech perception

Cited by

• Can Transformers Really Do It All? On the Compatibility of Inductive Biases Across Tasks
• Linear representations of grammaticality in neural language models
• Language Models Fail to Introspect About Their Knowledge of Language
• Constructions are Revealed in Word Distributions
• Do Language Models Know Who Did What to Whom?
• Both Direct and Indirect Evidence Contribute to Dative Alternation Preferences in Language Models
• LLMs as a synthesis between symbolic and distributed approaches to language
• From Zipf's Law to Neural Scaling through Heaps' Law and Hilberg's Hypothesis
• Tracing the ongoing emergence of human-like reasoning in Large Language Models
• HSKBenchmark: Modeling and Benchmarking Chinese Second Language Acquisition in Large Language Models through Curriculum Tuning
• Cognitive Convergence: Deep Similarities Between Large Language Models and Human Cognition
• Contouring Emotion Dynamics Through Metaphor: A Related-Event Approach to Emotion Nouns in Japanese and English
• France or Spain or Germany or France: A Neural Account of Non-Redundant Redundant Disjunctions
• WUGNECTIVES: Novel Entity Inferences of Language Models from Discourse Connectives
• Biasless Language Models Learn Unnaturally: How LLMs Fail to Distinguish the Possible from the Impossible
• Which course? Discourse! Teaching Discourse and Generation in the Era of LLMs
• The Philosophy of Language Models
• Language Models Identify Ambiguities and Exploit Loopholes
• UNVEILING: What Makes Linguistics Olympiad Puzzles Tricky for LLMs?
• Human-like fleeting memory improves language learning but impairs reading time prediction in transformer language models
• Decomposing Representation Space into Interpretable Subspaces with Unsupervised Learning

Discussions

• LMs need linguistics! New paper, with @futrell.bsky.social, on LMs and linguistics that conveys our excitement about what the present moment means for linguistics and what linguistics can do for LMs. [bsky, 110 points, 3 comments]
• Linguistics Learned to Stop Worrying and Love the Language Models [hn, 3 points, 1 comments]
• There's going to be an upcoming issue of behavioral and brain sciences on a paper by @futrell.bsky.social @kmahowald.bsky.social "How Linguistics Learned to Stop Worrying and Love the Language Models" [bsky, 2 points, 1 comments]
• This is in response to the recent target article by @futrell.bsky.social & @kmahowald.bsky.social, a wonderful read! Thanks to the authors for writing such a clear and inspiring piece to start this di [bsky, 1 points, 0 comments]
• One finding that speaks to this is whether LLMs learn “impossible” languages as easily as humans. Unfortunately, there has been some confusion about the findings, and the recent BBS paper by Futrell a [bsky, 1 points, 1 comments]
• Now reading: "How Linguistics Learned to Stop Worrying and Love the Language Models" to vibe with the spectre that is Chomskian syntax. It did help to get an "optimistic take on the relationship betw [bsky, 0 points, 0 comments]

Related

• Language Models as Models of Language
• Modelling Language
• Language Models Model Language
• Do Language Models' Words Refer?
• Large Language Models