NLP Highlights cover image

116 - Grounded Language Understanding, with Yonatan Bisk

NLP Highlights

00:00

The Simplest Way to Get the Shortest Path Trajectory in a Second Bot?

Tut: I've studied a lot recently on generalization in semantic parsing. So we are mapping language to the sequences of actions. They're not unlike a simulated space, but it is theorys related. And what we find is that the closer your action space is to like, predicate argument structure in language, the better you generalize. For instance, if turning left requires a sequen, a long sequence of incremental like manipulation steps, then my modelwl will tend to memorize that long sequence and generalize poorer. But still, from a modelling perspective, it probably would be useful to have something higher level in between that you then plan from to those lower level things. It's

Transcript
Play full episode

The AI-powered Podcast Player

Save insights by tapping your headphones, chat with episodes, discover the best highlights - and more!
App store bannerPlay store banner
Get the app