[11] Jacob Andreas - Learning from Language
The Thesis Review
00:00
The Limits of Formal Semantics in Language Learning
In the initial stage of the project I think basically like it felt too hard to simultaneously learn how to predict these layouts. And so we went with a fairly rigid model in which basically these like very, very coarse module layouts were derived from an off-the-shelf dependency parsers. But since then there's been more work from a couple of other groups that in my mind have just like at least in the case of question answering solve that problem much more satisfactorily using a very different class of structure predictors.
Transcript
Play full episode
