Timeline for An overview of mathematical-logical approaches in formalizing natural languages
Current License: CC BY-SA 4.0
23 events
| when toggle format | what | by | license | comment | |
|---|---|---|---|---|---|
| Aug 6, 2023 at 8:56 | comment | added | fosco | Plato was quite good at pankration. I don't know how many people today could call him "pseudo-mathematician" to his face, and survive the experience | |
| Aug 6, 2023 at 0:38 | comment | added | Wlod AA | @GerryMyerson, ha-ha-ha, good for you! (Why am I not surprised? :-)) | |
| Aug 5, 2023 at 22:18 | comment | added | Gerry Myerson | I, for one, would be ecstatic to belong to a group with Plato, Aristotle, and Russell. | |
| Aug 5, 2023 at 17:37 | history | became hot network question | |||
| Aug 5, 2023 at 17:24 | answer | added | Nagase | timeline score: 8 | |
| Aug 5, 2023 at 15:26 | comment | added | Timothy Chow | Regarding LLMs, Noam Chomsky and others have been rather outspoken in arguing that LLMs don't give us significant insight into natural language. Of course, plenty of people, such as Steven Piantadosi, have written scathing criticisms of Chomsky's approach. The situation is reminiscent of the linguistics wars half a century ago. Anyway, there is certainly no consensus that LLMs are better than, or incompatible with, other analytical approaches to natural language. | |
| Aug 5, 2023 at 14:08 | answer | added | Timothy Chow | timeline score: 5 | |
| Aug 5, 2023 at 12:44 | comment | added | Joseph Van Name | Including LLMs in this post will make this question too broad and possibly outside the scope of MO. But how do we know that trained LLMs are not just performing a more formal approach? And we should be more concerned about consistency and safety than simply performance of language models. Will you get the same thing if you train two LLMs using the same training data but different initializations? If not, then we have room to improve the consistency of language models. | |
| Aug 5, 2023 at 12:23 | comment | added | fosco | @WlodAA yeah, the definition of formalization is "freezing" :D | |
| Aug 5, 2023 at 11:09 | comment | added | Wlod AA | @fosco, formalizing means freezing. (Now one can see that formalizing natural languages is an oxymoron), | |
| Aug 5, 2023 at 7:17 | comment | added | fosco | @WlodAA I suspect formalizing the interaction of people in a trade of goods, or the dialectic process of constructing of a proof, would have looked like similar oxymorons to people living one century ago, and yet here we are, with game theory, linear logic, and thinking sand. | |
| Aug 5, 2023 at 5:27 | comment | added | Heleyrine Brookvinth | Although of course, the expectation that formal methods could offer said insight (for the users of language) into language, may be too naïve, and its truth is certainly up to debate (not to mention these two approaches are not entirely mutually exclusive from what I understand), but that was generally why I excluded AI-based methods in my post. | |
| Aug 5, 2023 at 5:22 | comment | added | Heleyrine Brookvinth | Thank you, @JoelDavidHamkins for your comment. I should again preface the things I say with the fact that my understanding on both LLMs and formal approaches are shallow; but from what I understand, LLMs do not necessarily aim to provide a deeper understanding of the structure of language and its interaction with meaning. LLMs can generate text by associating and finding patterns between data points which while effective, was less interesting to me. | |
| Aug 5, 2023 at 5:05 | history | edited | Rodrigo de Azevedo | CC BY-SA 4.0 |
Prepended crossposting header.
|
| Aug 5, 2023 at 4:30 | comment | added | Wlod AA | To me, formalizing natural languages is an oxymoron. | |
| Aug 5, 2023 at 4:29 | history | edited | Heleyrine Brookvinth | CC BY-SA 4.0 |
added 3 characters in body
|
| Aug 5, 2023 at 3:19 | comment | added | Joel David Hamkins | Indeed, couldn't the spectacular success of LLMs reasonably be taken to show an essential inadequacy of many of the formal approaches to natural languages that you mention? One might find this to be fundamentally similar to what happened some time ago in AI generally, which had abandoned the formal/logical approaches to AI that were prominent in the very early days decades ago in favor of more successful essentially numerical stochastic methods. | |
| Aug 5, 2023 at 1:58 | comment | added | Joel David Hamkins | Why aren't the LLM models on your list? In light of the recent impressive advances, I would think of these as the most successful and promising mathematical approaches to natural language. | |
| Aug 5, 2023 at 1:14 | comment | added | provocateur | I'm curious - aren't category theory and lambda calculus really two sides of the same coin? How do they lead to very different pictures of semantics? | |
| Aug 4, 2023 at 18:25 | review | Close votes | |||
| Aug 6, 2023 at 16:13 | |||||
| Aug 4, 2023 at 18:18 | comment | added | Paul Taylor | Why the vote to Close? I am continually appalled by the persistent hostility n MathOverflow to anyone who sees mathematics outside the 1950s bourbachiste agenda. | |
| S Aug 4, 2023 at 17:59 | review | First questions | |||
| Aug 4, 2023 at 20:11 | |||||
| S Aug 4, 2023 at 17:59 | history | asked | Heleyrine Brookvinth | CC BY-SA 4.0 |