Dissociating language and thought in large language models
2023/01/16 by Kyle Mahowald, Mahowald, Kyle, Anna A. Ivanova +9 · 12 voices
#cs.CL #cs.AI
paper · pdf · doi:10.48550/arxiv.2301.06627
Abstract
Large Language Models (LLMs) have come closest among all models to date to mastering human language, yet opinions about their linguistic and cognitive capabilities remain split. Here, we evaluate LLMs using a distinction between formal linguistic competence -- knowledge of linguistic rules and patterns -- and functional linguistic competence -- understanding and using language in the world. We ground this distinction in human neuroscience, which has shown that formal and functional competence rely on different neural mechanisms. Although LLMs are surprisingly good at formal competence, their performance on functional competence tasks remains spotty and often requires specialized fine-tuning and/or coupling with external modules. We posit that models that use language in human-like ways would need to master both of these competence types, which, in turn, could require the emergence of mechanisms specialized for formal linguistic competence, distinct from functional competence.
Discussions
- Dissociating language and thought in large language models [hn, 42 points, 4 comments]
- The must-read paper on LLMs, language, and thought that I reference here: Dissociating language and thought in large language models arxiv.org/abs/2301.06627 by @kmahowald.bsky.social @neuranna.bsky.s [bsky, 16 points, 0 comments]
- I think this may be one of the best contemporary arguments in support of the idea that there's a meaningful distinction to be made between next-word prediction and other cognitive (even linguistic) ta [bsky, 10 points, 1 comments]
- Formal vs functional linguistic competence? arxiv.org/pdf/2301.06627 [bsky, 3 points, 1 comments]
- Dissociating language and thought in LLMs: a cognitive perspective [hn, 3 points, 0 comments]
- Dissociating language and thought in large language models [hn, 2 points, 0 comments]
- Dissociating language, thought in large language models: a cognitive perspective [hn, 2 points, 0 comments]
- Here’s one example of a fairly even-handed discussion. While the authors are more optimistic than I am regarding LLM linguistic capabilities beyond next-word prediction, they also acknowledge that the [bsky, 1 points, 1 comments]
- さっきの「ボチョムキン理解」の根拠として、こちらの論文： Dissociating language and thought in large language models arxiv.org/abs/2301.066... 言語の能力を、 「言語規則やパターンの知識を指す「形式言語能力」と、現実の世界で言語を理解し使用する能力を指す「機能言語能力」」 に分けたうえで、 「LLMsは形式的言語能 [bsky, 1 points, 0 comments]
- Dissociating language and thought in large language models [hn, 1 points, 1 comments]
- Dissociating language and thought in large language models arxiv.org/abs/2301.06627 [bsky, 0 points, 0 comments]
- Dissociating language and thought in large language models https://arxiv.org/abs/2301.06627 [bsky, 0 points, 0 comments]
Related