I never agreed with his views on syntaxis, but the (his?) idea that large parts of our language capabilities are innate is almost beyond doubt. Are people still arguing against it?
I think it's about the details. Chomsky argued a lot of grammar must be innate but the ability of LLMs to do grammar quite well with only a basic artificial neural network argues against that.
Are you familiar with the 'poverty of stimuli' argument? The amount of language we get to process, all aural, is the tiniest of fractions of the amount of data an LLM gets to train on. And in much less processing time, too. So no, LLMs do not argue against that.
universal grammar is probably partially correct but Chomsky's position is too wide-sweeping. Grammar just doesn't demand the kind of complexity and precision that he implies.
IMO the problem is that his theories are elaborate logical justifications to sugarcoat some cringe supremacy beliefs about languages and politics. The sugar has always been useful but the core is pure poison.
Chomsky doesn't have any supremacist ideas about language, AFAIK. And I doubt his political views can be classified as such either. What poison do you speak of?
Saying that Chomskyan linguistics 'only works on a few languages' is such a ridiculous claim that only is stated by people who haven't engaged with generative linguistics since the 1960s. There's enormous work on typologically diverse languages such as Japanese, Salish languages, Greenlandic, Basque, Gungbe or Kwa. I can provide references if you'd like.
I never agreed with his views on syntaxis, but the (his?) idea that large parts of our language capabilities are innate is almost beyond doubt. Are people still arguing against it?
I think it's about the details. Chomsky argued a lot of grammar must be innate but the ability of LLMs to do grammar quite well with only a basic artificial neural network argues against that.
Are you familiar with the 'poverty of stimuli' argument? The amount of language we get to process, all aural, is the tiniest of fractions of the amount of data an LLM gets to train on. And in much less processing time, too. So no, LLMs do not argue against that.
4 replies →
universal grammar is probably partially correct but Chomsky's position is too wide-sweeping. Grammar just doesn't demand the kind of complexity and precision that he implies.
IMO the problem is that his theories are elaborate logical justifications to sugarcoat some cringe supremacy beliefs about languages and politics. The sugar has always been useful but the core is pure poison.
Chomsky doesn't have any supremacist ideas about language, AFAIK. And I doubt his political views can be classified as such either. What poison do you speak of?
I wouldn't say that "very upset" is a correct or fair characterization for disputes in linguistics.
Chomsky's universal grammar work was based on too few languages, too little data, and doesn't hold up when you look at all human languages and usage.
See also Jenny Saffran's empirical work on infant statistical language learning.
The broad idea that some things are innate doesn't vindicate Chomsky's specific theories.
Saying that Chomskyan linguistics 'only works on a few languages' is such a ridiculous claim that only is stated by people who haven't engaged with generative linguistics since the 1960s. There's enormous work on typologically diverse languages such as Japanese, Salish languages, Greenlandic, Basque, Gungbe or Kwa. I can provide references if you'd like.
[dead]
He should have stopped there in his career.