Title:
CONTEXT ANALYSIS DEVICE AND COMPUTER PROGRAM THEREFOR
Document Type and Number:
WIPO Patent Application WO/2018/043598
Kind Code:
A1
Abstract:
[Problem]
To provide a context analysis device that is able to utilize contextual features comprehensively and efficiently to perform context analysis with high precision.
[Solution]
A context analysis device 160 that includes: an analysis control unit 230 that detects a predicate from which a subject or the like has been omitted, and a supplement candidate therefor; and an anaphor/omission analysis unit 216 that determines a word that should be supplemented. The anaphor/omission analysis unit 216 includes: word vector generation units 206, 208, 210, and 212, which, for supplement candidates, generate multiple types of word vectors from a text 204; a learned weighting neural network 214 (or LSTM) which, for each supplement candidate, inputs a word vector and calculates a score indicating a probability that the word vector is the omitted word; and a list storage unit 234 and a supplement processing unit 236, which determine the supplement candidate having the best score. The word vectors at least include multiple word vectors each of which is extracted using character strings in the entire text other than the analysis subject and the candidates. Other words, such as deictic words, can be processed in the same manner.
Inventors:
IIDA RYU (JP)
TORISAWA KENTARO (JP)
KRUENGKRAI CANASAI (JP)
OH JONGHOON (JP)
KLOETZER JULIEN (JP)
TORISAWA KENTARO (JP)
KRUENGKRAI CANASAI (JP)
OH JONGHOON (JP)
KLOETZER JULIEN (JP)
Application Number:
PCT/JP2017/031250
Publication Date:
March 08, 2018
Filing Date:
August 30, 2017
Export Citation:
Assignee:
NAT INST INF & COMM TECH (JP)
International Classes:
G06F17/27; G06F17/30
Foreign References:
| US20090326919A1 | 2009-12-31 |
Other References:
EHARA, TERUMASA ET AL., ZERO-SUBJECT RESOLUTION BY PROBABILISTIC MODEL, vol. 1, no. 44, 15 February 1997 (1997-02-15), pages 12 - 28
OONO, MASAYUKI ET AL.: "Contextualization of selective preference model based on distributed representation", RESEARCH REPORT OF THE INFORMATION PROCESSING SOCIETY OF JAPAN OF NATURAL LANGUAGE PROCESSING, 15 January 2016 (2016-01-15), pages 1 - 9
OONO, MASAYUKI ET AL.: "Contextualization of selective preference model based on distributed representation", RESEARCH REPORT OF THE INFORMATION PROCESSING SOCIETY OF JAPAN OF NATURAL LANGUAGE PROCESSING, 15 January 2016 (2016-01-15), pages 1 - 9
Attorney, Agent or Firm:
SHIMIZU, Satoshi (JP)
Download PDF: