[44] Hady Elsahar - NLG from Structured Knowledge Bases (& Controlling LMs)
The Thesis Review
Controllability in a Low Resource Multilingual Setting
I think controllability could be a very good controllable language generation for example. So you could see that in two ways as a low resource setup where you say I could annotate 200 or 300 summarization entries in a data set and then you have an extreme low resource setup but you could tackle it from a different perspective. We have very large general purpose models and we would like them to match some certain aspects of that data set. This is a different perspective that makes the low resource setups not low resource anymore. And thinking of it this way makes you skip the problem of saying I don't need actually to have a one million example to learn this new style.
00:00
Transcript
Play full episode
Remember Everything You Learn from Podcasts
Save insights instantly, chat with episodes, and build lasting knowledge - all powered by AI.
