Gated word-character recurrent language model

Yasumasa Miyamoto, Kyunghyun Cho

Research output: Chapter in Book/Report/Conference proceedingConference contribution

Abstract

We introduce a recurrent neural network language model (RNN-LM) with long short-term memory (LSTM) units that utilizes both character-level and word-level inputs. Our model has a gate that adaptively finds the optimal mixture of the character-level and word-level inputs. The gate creates the final vector representation of a word by combining two distinct representations of the word. The character-level inputs are converted into vector representations of words using a bidirectional LSTM. The word-level inputs are projected into another high-dimensional space by a word lookup table. The final vector representations of words are used in the LSTM language model which predicts the next word given all the preceding words. Our model with the gating mechanism effectively utilizes the character-level inputs for rare and out-of-vocabulary words and outperforms word-level language models on several English corpora.

Original languageEnglish (US)
Title of host publicationEMNLP 2016 - Conference on Empirical Methods in Natural Language Processing, Proceedings
PublisherAssociation for Computational Linguistics (ACL)
Pages1992-1997
Number of pages6
ISBN (Electronic)9781945626258
StatePublished - Jan 1 2016
Event2016 Conference on Empirical Methods in Natural Language Processing, EMNLP 2016 - Austin, United States
Duration: Nov 1 2016Nov 5 2016

Publication series

NameEMNLP 2016 - Conference on Empirical Methods in Natural Language Processing, Proceedings

Conference

Conference2016 Conference on Empirical Methods in Natural Language Processing, EMNLP 2016
CountryUnited States
CityAustin
Period11/1/1611/5/16

Fingerprint

Table lookup
Recurrent neural networks
Long short-term memory

ASJC Scopus subject areas

  • Computer Science Applications
  • Information Systems
  • Computational Theory and Mathematics

Cite this

Miyamoto, Y., & Cho, K. (2016). Gated word-character recurrent language model. In EMNLP 2016 - Conference on Empirical Methods in Natural Language Processing, Proceedings (pp. 1992-1997). (EMNLP 2016 - Conference on Empirical Methods in Natural Language Processing, Proceedings). Association for Computational Linguistics (ACL).

Gated word-character recurrent language model. / Miyamoto, Yasumasa; Cho, Kyunghyun.

EMNLP 2016 - Conference on Empirical Methods in Natural Language Processing, Proceedings. Association for Computational Linguistics (ACL), 2016. p. 1992-1997 (EMNLP 2016 - Conference on Empirical Methods in Natural Language Processing, Proceedings).

Research output: Chapter in Book/Report/Conference proceedingConference contribution

Miyamoto, Y & Cho, K 2016, Gated word-character recurrent language model. in EMNLP 2016 - Conference on Empirical Methods in Natural Language Processing, Proceedings. EMNLP 2016 - Conference on Empirical Methods in Natural Language Processing, Proceedings, Association for Computational Linguistics (ACL), pp. 1992-1997, 2016 Conference on Empirical Methods in Natural Language Processing, EMNLP 2016, Austin, United States, 11/1/16.
Miyamoto Y, Cho K. Gated word-character recurrent language model. In EMNLP 2016 - Conference on Empirical Methods in Natural Language Processing, Proceedings. Association for Computational Linguistics (ACL). 2016. p. 1992-1997. (EMNLP 2016 - Conference on Empirical Methods in Natural Language Processing, Proceedings).
Miyamoto, Yasumasa ; Cho, Kyunghyun. / Gated word-character recurrent language model. EMNLP 2016 - Conference on Empirical Methods in Natural Language Processing, Proceedings. Association for Computational Linguistics (ACL), 2016. pp. 1992-1997 (EMNLP 2016 - Conference on Empirical Methods in Natural Language Processing, Proceedings).
@inproceedings{e1614b678b3c4977be41e67e98b47eb4,
title = "Gated word-character recurrent language model",
abstract = "We introduce a recurrent neural network language model (RNN-LM) with long short-term memory (LSTM) units that utilizes both character-level and word-level inputs. Our model has a gate that adaptively finds the optimal mixture of the character-level and word-level inputs. The gate creates the final vector representation of a word by combining two distinct representations of the word. The character-level inputs are converted into vector representations of words using a bidirectional LSTM. The word-level inputs are projected into another high-dimensional space by a word lookup table. The final vector representations of words are used in the LSTM language model which predicts the next word given all the preceding words. Our model with the gating mechanism effectively utilizes the character-level inputs for rare and out-of-vocabulary words and outperforms word-level language models on several English corpora.",
author = "Yasumasa Miyamoto and Kyunghyun Cho",
year = "2016",
month = "1",
day = "1",
language = "English (US)",
series = "EMNLP 2016 - Conference on Empirical Methods in Natural Language Processing, Proceedings",
publisher = "Association for Computational Linguistics (ACL)",
pages = "1992--1997",
booktitle = "EMNLP 2016 - Conference on Empirical Methods in Natural Language Processing, Proceedings",

}

TY - GEN

T1 - Gated word-character recurrent language model

AU - Miyamoto, Yasumasa

AU - Cho, Kyunghyun

PY - 2016/1/1

Y1 - 2016/1/1

N2 - We introduce a recurrent neural network language model (RNN-LM) with long short-term memory (LSTM) units that utilizes both character-level and word-level inputs. Our model has a gate that adaptively finds the optimal mixture of the character-level and word-level inputs. The gate creates the final vector representation of a word by combining two distinct representations of the word. The character-level inputs are converted into vector representations of words using a bidirectional LSTM. The word-level inputs are projected into another high-dimensional space by a word lookup table. The final vector representations of words are used in the LSTM language model which predicts the next word given all the preceding words. Our model with the gating mechanism effectively utilizes the character-level inputs for rare and out-of-vocabulary words and outperforms word-level language models on several English corpora.

AB - We introduce a recurrent neural network language model (RNN-LM) with long short-term memory (LSTM) units that utilizes both character-level and word-level inputs. Our model has a gate that adaptively finds the optimal mixture of the character-level and word-level inputs. The gate creates the final vector representation of a word by combining two distinct representations of the word. The character-level inputs are converted into vector representations of words using a bidirectional LSTM. The word-level inputs are projected into another high-dimensional space by a word lookup table. The final vector representations of words are used in the LSTM language model which predicts the next word given all the preceding words. Our model with the gating mechanism effectively utilizes the character-level inputs for rare and out-of-vocabulary words and outperforms word-level language models on several English corpora.

UR - http://www.scopus.com/inward/record.url?scp=85072841020&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=85072841020&partnerID=8YFLogxK

M3 - Conference contribution

AN - SCOPUS:85072841020

T3 - EMNLP 2016 - Conference on Empirical Methods in Natural Language Processing, Proceedings

SP - 1992

EP - 1997

BT - EMNLP 2016 - Conference on Empirical Methods in Natural Language Processing, Proceedings

PB - Association for Computational Linguistics (ACL)

ER -