Notes on ‘Modern language models refute Chomsky’s approach to language (Steven T. Piantadosi)’

LLM and Linguistics
LLM and Linguistics
Author
Affiliation
Published

October 10, 2023

Nature Machine Intelligence 在最近的主編的話 Language models and linguistic theories beyond words 再度把 LLM 和語言學的(沒有)關係拿到檯面上。雖然浮光掠影的談過,但是對於語言學知識較為友善的計算語言學家心裡的疑惑,是時候該拿出來討論了。

簡單說,語言學在當前的AI語言模型發展進程中,基本上處在邊陲的角色。為什麼?是語言學(或主流的形式語言學範式)有什麼問題?我們又需要什麼樣的語言學?還是 LLM 只是曇花一現的隨機鸚鵡?

姑且不論背後的語言與心智假說是否合理,

這些都是大問題,我想嘗試開始用寫作筆記的方式幫自己整理頭緒。

首先是 UC Berkeley 心理與腦神經科學教授 Steven t. Piantadosi 今年三月開始在LingBuzz上的一篇標題即開戰的文章:Modern language models refute Chomsky’s approach to language.

摘要幾個重點

Citation

BibTeX citation:
@online{hsieh2023,
  author = {Hsieh, Shu-Kai},
  title = {Notes on “{Modern} Language Models Refute {Chomsky’s}
    Approach to Language {(Steven} {T.} {Piantadosi)}”},
  date = {2023-10-10},
  url = {https://loperntu.github.io/posts/2023-10-09-quarto-blogs/},
  langid = {en}
}
For attribution, please cite this work as:
Hsieh, Shu-Kai. 2023. “Notes on ‘Modern Language Models Refute Chomsky’s Approach to Language (Steven T. Piantadosi)’.” October 10, 2023. https://loperntu.github.io/posts/2023-10-09-quarto-blogs/.