Attentive Language Models

Giancarlo Salton, Robert Ross, John Kelleher


Abstract
In this paper, we extend Recurrent Neural Network Language Models (RNN-LMs) with an attention mechanism. We show that an “attentive” RNN-LM (with 11M parameters) achieves a better perplexity than larger RNN-LMs (with 66M parameters) and achieves performance comparable to an ensemble of 10 similar sized RNN-LMs. We also show that an “attentive” RNN-LM needs less contextual information to achieve similar results to the state-of-the-art on the wikitext2 dataset.
Anthology ID:
I17-1045
Volume:
Proceedings of the Eighth International Joint Conference on Natural Language Processing (Volume 1: Long Papers)
Month:
November
Year:
2017
Address:
Taipei, Taiwan
Venue:
IJCNLP
SIG:
Publisher:
Asian Federation of Natural Language Processing
Note:
Pages:
441–450
Language:
URL:
https://www.aclweb.org/anthology/I17-1045
DOI:
Bib Export formats:
BibTeX MODS XML EndNote
PDF:
http://aclanthology.lst.uni-saarland.de/I17-1045.pdf