Meaning generation method, meaning generation apparatus, and storage medium

Patent metadata

Field	Value
Publication number	US-2017365252-A1
Application number	US-201715618383-A
Country	US
Kind code	A1
Filing date	Jun 9, 2017
Priority date	Jun 17, 2016
Publication date	Dec 21, 2017
Grant date	—

Abstract

A meaning generation method, in a meaning generation apparatus, includes acquiring meaning training data including text data of a sentence that can be an utterance sentence and meaning information indicating a meaning of the sentence and associated with the text data of the sentence, acquiring restatement training data including the text data of the sentence and text data of a restatement sentence of the sentence, and learning association between the utterance sentence and the meaning information and the restatement sentence. The learning includes learning of a degree of importance of a word included in the utterance sentence, and the learning is performed by applying the meaning training data and the restatement training data to a common model, and storing a result of the learning as learning result information.

First claim

What is claimed is:

1 . A meaning generation method in a meaning generation apparatus, the meaning generation method comprising:
acquiring first meaning training data including text data of a first sentence that can be an utterance sentence, the first meaning training data also including meaning information associated with the text data of the first sentence and indicating a meaning of the first sentence; acquiring first restatement training data including the text data of the first sentence and text data of a second sentence associated with the text data of the first sentence, the second sentence being a restatement sentence of the first sentence; learning association between the utterance sentence and the meaning information and between the utterance sentence and the restatement sentence, the learning including learning of a degree of importance of a word included in the utterance sentence, the learning performed by applying the first meaning training data and the first restatement training data to a common model; and storing a result of the learning as learning result information.

2 . The meaning generation method according to claim 1 , further comprising:
acquiring text data of a third sentence uttered by a user; and generating meaning information corresponding to the third sentence based on the learning result information.

3 . The meaning generation method according to claim 1 , wherein the first meaning training data is acquired from a first corpus in which two or more pieces of meaning training data are accumulated.

4 . The meaning generation method according to claim 1 , wherein the first restatement training data is acquired from a second corpus in which two or more pieces of restatement training data are accumulated.

5 . The meaning generation method according to claim 1 , wherein the model is a neural network model.

6 . The meaning generation method according to claim 5 , wherein the learning is performed by performing error backpropagation learning between (i) the meaning information and the second sentence associated with the first sentence and (ii) the posterior probability, calculated using the model, for the meaning information and the second sentence corresponding to the first sentence.

7 . The meaning generation method according to claim 1 , wherein the learning of association between the utterance sentence and the restatement sentence is performed using internal information obtained in the learning of the association between the utterance sentence and the meaning information.

8 . The meaning generation method according to claim 7 , wherein the model is a neural network model, and
the internal information is a weight between layers in the neural network model.

9 . The meaning generation method according to claim 1 , wherein at least one of the acquiring of the first meaning training data, the acquiring of the first restatement training data, and the learning of the association, and the storing of the learning result information is performed by a processor.

10 . A meaning generation apparatus comprising:
a meaning training data acquirer that acquires first meaning training data including text data of a first sentence that can be an utterance sentence, the first meaning training data also including meaning information associated with the text data of the first sentence and indicating a meaning of the first sentence; a restatement training data acquirer that acquires first restatement training data including the text data of the first sentence and text data of a second sentence associated with the text data of the first sentence, the second sentence being a restatement sentence of the first sentence; a learner that performs learning of association between the utterance sentence and the meaning information and between the utterance sentence and the restatement sentence, the learning including learning of a degree of importance of a word included in the utterance sentence, the learning performed by applying the first meaning training data and the first restatement training data to a common model; and a storage that stores a result of the learning as learning result information.

11 . A non-transitory computer-readable storage medium storing a program causing a computer to execute a process comprising:
acquiring first meaning training data including text data of a first sentence that can be an utterance sentence, the first meaning training data also including meaning information associated with the text data of the first sentence and indicating a meaning of the first sentence; acquiring first restatement training data including the text data of the first sentence and text data of a second sentence associated with the text data of the first sentence, the second sentence being a restatement sentence of the first sentence; learning association between the utterance sentence and the meaning information and between the utterance sentence and the restatement sentence, the learning including learning of a degree of importance of a word included in the utterance sentence, the learning performed by applying the first meaning training data and the first restatement training data to a common model; and storing a result of the learning as learning result information.

Assignees

• Panasonic Ip Man Co Ltd

Inventors

• Ushio Takashi
• Yamagami Katsuyoshi

Classifications

• G06N3/044

  Recurrent networks, e.g. Hopfield networks · CPC title

• G10L15/1815

  Semantic context, e.g. disambiguation of the recognition hypotheses based on word meaning · CPC title

• G10L15/063 Primary

  Training · CPC title

• G10L15/16

  using artificial neural networks · CPC title

• G06N3/08

  Learning methods · CPC title