INTENTION ESTIMATING DEVICE AND INTENTION ESTIMATING METHOD

• US 20150255064A1
• Filed: 10/09/2013
• Published: 09/10/2015
• Est. Priority Date: 11/30/2012
• Status: Active Grant

• Current Assignee: Mitsubishi Electric Corporation
• Original Assignee: Mitsubishi Electric Corporation
• Inventors: Fujii, Yoichi, Ishll, Jun
• Granted Patent: US 9,530,405 B2
• Time in Patent Office: Days
• US Class Current: 1/1
• CPC Class Codes: G06F 40/268 Morphological analysis
• CPC Class Codes: G06F 40/30 Semantic analysis
• CPC Class Codes: G10L 15/18 using natural language modelling
• CPC Class Codes: G10L 15/1822 Parsing for meaning understanding

Abstract

An intention estimation unit extractor that extracts one or more intention estimation units each of which is a unit on which an estimation of the intention is to be performed from an inputted language, an intention estimator that estimates a partial intention of a part which constructs each of the extracted intention estimation units, an intention co-occurrence weight calculator that calculates an intention co-occurrence weight based on a relationship between partial intentions, and an intention sequence estimator that generates an intention sequence corresponding to the inputted language by using one or more partial intentions, and generates an intention estimation result corresponding to the inputted language by using both a score showing a likelihood of the generated intention sequence and an intention co-occurrence weight which the intention co-occurrence weight calculator calculates for the partial intentions which construct the generated intention sequence are disposed.

Claims

1. An intention estimating device that estimates a user'"'"'s intention from the user'"'"'s language input, said intention estimating device comprising:

an intention estimation unit extractor that extracts one or more intention estimation units each of which is a unit on which an estimation of said intention is to be performed from said inputted language;

an intention estimator that estimates a partial intention which is an intention of a part which constructs each of the intention estimation units extracted by said intention estimation unit extractor from said each of the intention estimation units; and

an intention co-occurrence weight calculator that calculates an intention co-occurrence weight based on a relationship between said partial intentions estimated by said intention estimator, wherein

said intention estimating device includes an intention sequence estimator that generates an intention sequence corresponding to said inputted language by using one or more said partial intentions estimated by said intention estimator, and generates an intention estimation result corresponding to said inputted language by using both a score showing a likelihood of said generated intention sequence and an intention co-occurrence weight which said intention co-occurrence weight calculator calculates for said partial intentions which construct said generated intention sequence.

2. The intention estimating device according to claim 1, wherein said intention co-occurrence weight calculator refers to intention hierarchical layer graph data in which said intentions are hierarchized and defined according to the relationship between said intentions to calculate said intention co-occurrence weight by using a weight which is defined for each of said intentions which constructs each hierarchical layer.

3. The intention estimating device according to claim 2, wherein said intention hierarchical layer graph data have a relationship in which an intention in a lower hierarchical layer gives a concrete form to an intention in a higher hierarchical layer, and said intention co-occurrence weight calculator refers to said intention hierarchical layer graph data to, when a common intention in a higher hierarchical layer exists in two continuous partial intentions which construct said intention sequence generated by said intention sequence estimator, calculate said intention co-occurrence weight by using a weight defined for said intention in a higher hierarchical layer.

4. The intention estimating device according to claim 2, wherein said intention hierarchical layer graph data have a relationship in which an intention in a lower hierarchical layer gives a concrete form to an intention in a higher hierarchical layer, and said intention co-occurrence weight calculator refers to said intention hierarchical layer graph data to, when a hierarchical relationship exists between two continuous partial intentions which construct said intention sequence generated by said intention sequence estimator, set a preset value as said intention co-occurrence weight.

5. The intention estimating device according to claim 1, wherein said intention estimating device includes an intention sequence conversion table that holds an intention sequence conversion rule for converting said partial intentions according to a relationship between a contiguous sequence of said partial intentions which construct said intention sequence, and an intention sequence convertor that converts said partial intention having a relationship matching the intention sequence conversion rule described in said intention sequence conversion table, among the contiguous sequence of said partial intentions which construct said intention sequence generated by said intention sequence estimator, according to said intention sequence conversion rule.

6. An intention estimating method of estimating a user'"'"'s intention from the user'"'"'s language input, said device comprising:

a step of an intention estimation unit extractor extracting one or more intention estimation units each of which is a unit on which an estimation of said intention is to be performed from said inputted language;

a step of an intention estimator estimating a partial intention which is an intention of a part which constructs each of said extracted intention estimation units from said each of the intention estimation units;

a step of an intention co-occurrence weight calculator calculating an intention co-occurrence weight based on a relationship between said estimated partial intentions; and

a step of an intention sequence estimator generating an intention sequence corresponding to said inputted language by using said estimated one or more said partial intentions, and generating an intention estimation result corresponding to said inputted language by using both a score showing a likelihood of said generated intention sequence and an intention co-occurrence weight which said intention co-occurrence weight calculator calculates for said partial intentions which construct said generated intention sequence.