Login| Sign Up| Help| Contact|

Patent Searching and Data


Title:
SYMBOL STRING GENERATOR, SENTENCE COMPACTION DEVICE, SYMBOL STRING GENERATION METHOD, AND PROGRAM
Document Type and Number:
WIPO Patent Application WO/2019/167642
Kind Code:
A1
Abstract:
A symbol string generator that, when a first symbol string x representing a sentence is inputted, generates, via a pre-learned neural network, a second symbol string y that accords with a prescribed purpose that corresponds to the sentence, wherein the symbol string generator is characterized in that: the neural network has an encoding unit that converts elements xi of the inputted first symbol string x to a first hidden state, an attention mechanism unit that applies a weighting to the first hidden state and outputs the weighted first hidden state as a second hidden state, a decoding unit that outputs a third hidden state on the basis of the tth element xt of the first symbol string x, the t-1th element yt-1} of the second symbol string y, and the second hidden state, and an output unit that generates the tth element yt of the second symbol string y on the basis of the second hidden state and the third hidden state and outputs the generated element yt; the attention mechanism unit calculates each of first probabilities Pparent (xj}|xt,x) that the parent of an element xt included in the first symbol string x is an element xj other than the element xt in a dependency structure tree that corresponds to the sentence, calculates, using the calculated first probabilities Pparent (xj}|xt,x), each of second probabilities αd,t,j that the dth order parent of the element xt is a term xj other than the element xt, and outputs γd,t derived by weighting the first hidden state using the calculated second probabilities αd,t j, as the second hidden state.

Inventors:
KAMIGAITO, Hidetaka (9-11 Midori-cho 3-chome, Musashino-sh, Tokyo 85, 〒1808585, JP)
NAGATA, Masaaki (9-11 Midori-cho 3-chome, Musashino-sh, Tokyo 85, 〒1808585, JP)
HIRAO, Tsutomu (9-11 Midori-cho 3-chome, Musashino-sh, Tokyo 85, 〒1808585, JP)
Application Number:
JP2019/005283
Publication Date:
September 06, 2019
Filing Date:
February 14, 2019
Export Citation:
Click for automatic bibliography generation   Help
Assignee:
NIPPON TELEGRAPH AND TELEPHONE CORPORATION (5-1 Otemachi 1-chome, Chiyoda-ku Tokyo, 16, 〒1008116, JP)
International Classes:
G06F17/27; G06F16/00; G06F16/30; G06N3/04
Other References:
FILIPPOVA, KATJA ET AL.: "Sentence Compression by Deletion with LSTMs", PROCEEDINGS OF THE 2015 CONFERENCE ON EMPIRICAL METHODS IN NATURAL LANGUAGE PROCESSING, September 2015 (2015-09-01), pages 360 - 368, XP055423653, DOI: 10.18653/v1/D15-1042
NALLAPATI, RAMESH ET AL.: "Abstractive Text Summarization using Sequence-to-sequence RNNs and Beyond", PROCEEDINGS OF THE 20TH SIGNLL CONFERENCE ON COMPUTATIONAL NATURAL LANGUAGE LEARNING, August 2016 (2016-08-01), pages 280 - 290, XP055441008, DOI: 10.18653/v1/K16-1028
RUSH, ALEXANDER M. ET AL.: "A Neural Attention Model for Abstractive Sentence Summarization", PROCEEDINGS OF THE 2015 CONFERENCE ON EMPIRICAL METHODS IN NATURAL LANGUAGE PROCESSING, September 2015 (2015-09-01), pages 379 - 389, XP055546567, DOI: 10.18653/v1/D15-1044
KAMIGAITO, HIDETAKA ET AL.: "Neural sentence compression considering chaining of the dependency structure", PROCEEDINGS OF THE 24TH ANNUAL MEETING OF THE ASSOCIATION FOR NATURAL LANGUAGE PROCESSING, 13 March 2018 (2018-03-13), pages 1096 - 1099, XP055634119
Attorney, Agent or Firm:
ITOH, Tadashige et al. (16th Floor, Marunouchi MY PLAZA 1-1, Marunouchi 2-chome, Chiyoda-k, Tokyo 05, 〒1000005, JP)
Download PDF: