Модели на основе подсчета

Training neural network language mod- els over large vocabularies is computa- tionally costly compared to count-based models such as Kneser-Ney. We present a systematic comparison of neural strate- gies to …...

AbstractRecent studies of distributional semantic models have set up a competition between word embeddings obtained from predictive neural networks and word vectors obtained from abstractive count-based models. This paper is …...