68 - Neural models of factuality, with Rachel Rudinger
NLP Highlights
00:00
Using a Rule Based System to Determine Factuality
We're working witha by directional noral networks. We use the smooth l on loss in it to train our models. And so end weated the factuality prediction on this minus any of these token states that we want to make, and pass that hidden state through a two layer network. Am, but they actually work fairly well for relatively simple architecture. So i should also note that we used glovedings for the imbedded clauses as well; those are going to be things like claws imbedding verbs for outside context, or nouns for inside context, like hallucination. The data sets do both go from language to a set of events using some process and then e and
Transcript
Play full episode
