UMass CS685 (Advanced NLP) F20: Neural language models

preview_player
Показать описание
word embeddings, composition functions, softmax layers, fixed-window neural language models, recurrent neural language models

Рекомендации по теме
Комментарии
Автор

Hi professor, thanks for sharing these great videos, i have a question, the backoff part, if the denominator is zero(doesn't occur in the training set), we use backoff, from looking ahead 3 words to looking ahead 2 words, even only one word, so in the example of the video, backoff means we count(opened their) or count(their)

zhonghuahe