120 - Evaluation of Text Generation, with Asli Celikyilmaz
NLP Highlights
00:00
The Challenges of Pushing Neural NLG Systems in Production
There's also other issues like, like I mentioned, like these neural decoders are not easily controllable. So it is still a big challenge when we consider these ethical or bias issues that might originate from either training data or the biases that we introduce into these models. One example is that a chatbot, for instance, can just answer a yes. Yes could be very, you know, okay, but it could also be an offensive given the context. So how do we act?
Transcript
Play full episode
