Скачать или смотреть Found in Translation (FOTRAN): Joerg Tiedemann LADAL Webinar 2022

This talk was recorded on Dec 5, 2022 as part of the LADAL Webinar Series 2022.

LADAL Website: https://ladal.edu.au
LADAL Webinar Series 2022: https://ladal.edu.au/webinars2022.html
LADAL on Twitter: @slcladal
Facebook: https://www.facebook.com/profile.php?...
Contact: [email protected]

Abstract
The goal of language technology is to create computational models that can understand and generate language in a way humans can do. One of the strategies is to learn such communication abilities from real-world data and in that way somewhat resemble humans and their capability of picking up language skills through practical experience. However, the crucial question is what kind of experience is needed and what kind of tasks have to be practiced to build an understanding of human language signals. We are currently running a project that studies the use of translations that form natural semantic mirrors of original texts in other languages as a means of providing information about the underlying latent meaning that corresponds to the observable language string. The big question is what kind of abstractions can be learned from this cross-lingual signal and how much does that reflect our knowledge about linguistic properties on various levels. Part of this question is how much language diversity can be used to push abstraction levels even further. In this talk I will present some of our results and try to connect this kind of neural “black-box” NLP with questions in general linguistics and cognition.