Comments on: Beyond Chomsky normal form: Extending Strong learning algorithms for PCFGs https://icgi2020.lis-lab.fr August 23-27, 2021 Tue, 31 Aug 2021 11:10:14 +0000 hourly 1 https://wordpress.org/?v=7.1 By: François Coste https://icgi2020.lis-lab.fr/beyond-chomsky-normal-form-extending-strong-learning-algorithms-for-pcfgs/#comment-18 Mon, 23 Aug 2021 09:47:02 +0000 https://icgi2020.lis-lab.fr/?page_id=466#comment-18 I have the same kind of question as Rémi that I would like to complete from the formal language point of view: do you have any intuitions about the class of languages that can be represented, for instance in relation to substitutable languages (forgetting weights)?

]]> By: Alex https://icgi2020.lis-lab.fr/beyond-chomsky-normal-form-extending-strong-learning-algorithms-for-pcfgs/#comment-16 Mon, 23 Aug 2021 07:48:13 +0000 https://icgi2020.lis-lab.fr/?page_id=466#comment-16 In reply to Rémi.

Yes I think anchoring is too strong in general. I think one problem is that some categories done have any yields of length 1. Local unambiguity seems ok though ( I think)

]]> By: Rémi https://icgi2020.lis-lab.fr/beyond-chomsky-normal-form-extending-strong-learning-algorithms-for-pcfgs/#comment-5 Thu, 19 Aug 2021 09:08:40 +0000 https://icgi2020.lis-lab.fr/?page_id=466#comment-5 My question is likely a bit naive due to my lack of knowledge in linguistic: How compatible is the anchoring restriction with the current understanding of natural languages? Your examples are quite convincing, but intuitively we may imagine that some abstract categories exist that do not admit a corresponding terminal.
On the same line of idea, is the local non ambiguity compatible with commonly share understanding of natural languages?

]]>