[22] Graham Neubig - Unsupervised Learning of Lexical Information
The Thesis Review
00:00
The Importance of Alignment in Machine Learning
With BPE you learn the dictionary in one stage and then train whatever model you want on that vocabulary. Was that the same here where you design this model, you get some dictionary and then you could train whatever machine learning model you wanted? You could do that, yes. But within the paper that I presented I also did a thing that's kind of like specific to phrase-based translation Where I didn't just memorize the phrases at the lowest level of the tree but I also memorized the phrases all the way up the tree so it could memorize longer phrases as well. And I actually have a software toolkit, PyAlign that implements this algorithm and it's definitely possible to
Transcript
Play full episode
