Language understanding apparatus, language understanding method, and computer program

• US 8,117,023 B2
• Filed: 12/04/2007
• Issued: 02/14/2012
• Est. Priority Date: 12/06/2006
• Status: Active Grant

• Current Assignee: Honda Motor Co., Ltd. (Honda Motor Company)
• Original Assignee: Honda Motor Co., Ltd. (Honda Motor Company)
• Inventors: Funakoshi, Kotaro, Nakano, Mikio
• Primary Examiner(s): Neway, Samuel G
• Application Number:
  US11/999,236
• Publication Number:
  US 20080140389A1
• Time in Patent Office: 1,533 Days
• Field of Search: 704/9
• US Class Current: 704/9
• CPC Class Codes: G06F 40/30 Semantic analysis

Abstract

This language understanding apparatus includes a storage section, a phrase interpreter, and a bidding section. The bidding section reads information on a slot of a semantic frame corresponding to a concept indicated by an information read by the phrase interpreter and information on a concept corresponding to the slot by referring to the storage section, specifies, for each semantic frame read, from the concept representation detected by the phrase interpreter, a concept representation corresponding to a concept corresponding to each slot of the semantic frame or a subordinate concept of that concept, and a concept representation whose slot of another semantic frame is not bidden, and repeats bidding the slot for the specified concept representation, thereby generating slot allocation table data indicating a bidding result.

Claims

1. A language understanding apparatus comprising:

a storage section that stores;

concept structure data indicating a correlation between a tree structure of a concept to be used in a domain and a concept representation specifying the concept to be used in the domain,

frame definition data corresponding to the concept to be used in the domain and indicating a correlation between one or more semantic frames for representing the concept to be used in the domain by another concept,

one or more variables to be used in the one or more semantic frames, and

variable definition data indicating a concept to be used as a variable;

a phrase interpreter that receives input utterance data detects a concept representation included in an utterance content indicated by the input utterance data by referring to the storage section, and reads information of a concept corresponding to the detected concept representation from the storage section; and

a bidding section that;

reads information on a variable of a semantic frame corresponding to the concept indicated by the information read by the phrase interpreter and information on a concept corresponding to the variable by referring to the storage section, and

allocates by writing to a variable allocation table from the concept representation detected by the phrase interpreter, for each semantic frame read;

a concept representation associated with a concept corresponding to each variable of the semantic frame or a subordinate concept of that concept, and

a concept representation whose variable of another semantic frame is not written, and

generates variable allocation table data indicating a bidding result.

2. The language understanding apparatus according to claim 1, further comprising

an access table generator that receives the input utterance data, performs parsing of the utterance content indicated by the received input utterance data to acquire a dependency relationship among morpheme sequences constituting the utterance content, and generates access table data indicating an accessibility among the morpheme sequences based on the acquired dependency relationship, wherein

when making a bid for each semantic frame, the bidding section specifies, from the concept representation detected by the phrase interpreter, a concept representation corresponding to a concept corresponding to each variable of the semantic frame or a subordinate concept of that concept, and a concept representation whose variable of another semantic frame is not bidden, and bids the variable for the specified concept representation when the access table data generated by the access table generator indicates that the concept representation corresponding to the semantic frame is accessible to the specified concept representation.

3. The language understanding apparatus according to claim 2, wherein

the access table generator separates or combines the morpheme sequences in such a way that one of the morpheme sequences undergone the parsing includes one parsed concept representation detected by the phrase interpreter, and generates the access table data using a resultant morpheme sequence.

4. The language understanding apparatus according to claim 1, further comprising:

a frame combination generator that, by referred to the storage section, specifies a semantic frame corresponding to the concept read by the phrase interpreter, and, when there are a plurality of semantic frames specified in correspondence to one concept, generates available semantic frame combinations, wherein the bidding section generates the variable allocation table data for each of the semantic frame combinations generated by the frame combination generator; and

a scoring section that gives a score based on a state of bidding a variable for a concept representation by referring to the variable allocation table data corresponding to each of the semantic frame combinations generated by the frame combination generator, and selects a semantic frame combination used in an utterance content based on the given score.

5. The language understanding apparatus according to claim 1, wherein:

the variable definition data further includes information on a concept representation specifying a variable;

referring to the storage section, the phrase interpreter detects a concept representation included in the utterance content indicated by the received input utterance data and reads a concept or a variable corresponding to the detected concept representation; and

the language understanding apparatus further comprises a variable collision solving section that, when a plurality of variable of a same semantic frame are bidden for one concept representation by the bidding section, detects if another concept representation specifying a variable accompanies the concept representation in the utterance content, and, upon detecting the accompanying another concept representation, selects a variable specified by the accompanying another concept representation as a variable to be bidden for the concept representation and rewrites the variable allocation table data based on a selection result.

6. The language understanding apparatus according to claim 1, wherein:

the variable definition data further includes information on a marker specifying a variable; and

the language understanding apparatus fiber comprises a variable collision solving section that, when a plurality of variable of a same semantic frame are bidden for one concept representation by the bidding section, detects if a marker accompanies the concept representation in the utterance content by referring to the storage section, and, upon detecting the accompanying marker, selects a variable specified by the marker as a variable to be bidden for the concept representation and rewrites the variable allocation table data based on a selection result.

7. The language understanding apparatus according to claim 1, further comprising

a variable collision solving section that, when a plurality of variables of a same semantic frame are bidden for each of a plurality of concept representations corresponding to a same concept by the bidding section, selects a variable to be bidden for each of the plurality of concept representations in such a way that an order of appearance of the plurality of concept representations in the utterance content corresponds to an order of variables indicated by information on a variable appearance order stored in the storage section, and rewrites the variable allocation table data based on a selection result.

8. The language understanding apparatus according to claim 1, further comprising

a syntax error parser that, when there is a blank variable which has not been bidden for any concept representation by the bidding section, bids the blank variable for a concept representation which matches with a concept corresponding to the blank variable and for which another variable is not bidden, and rewrites the variable allocation table data based on a bidding result.

9. The language understanding apparatus according to claim 1, further comprising

an intrasentential omission solving section that, for a concept representation which corresponds to a semantic frame and for which a variable of the semantic frame is not bidden by the bidding section, specifies another concept representation appearing before the concept representation in the utterance content and corresponding to a same concept as the concept representation, detects that the specified another concept representation and the concept representation form a parallel structure, detects that a semantic frame corresponding to the specified another concept representation has a same variable as the unbidden variable of the semantic frame corresponding to the concept representation, and, when detecting that the detected variable of the semantic frame corresponding to the specified another concept representation is bidden for a further concept representation by the bidding section, bids the unbidden variable of the semantic frame corresponding to the concept representation for the detected further concept representation and rewrites the variable allocation table data based on a bidding result.

10. The language understanding apparatus according to claim 1, further comprising

a semantic tree generator that, referring to the variable allocation table data, generates a semantic tree from a concept representation having a semantic frame and a concept representation for which a variable of the semantic frame is bidden.

11. The language understanding apparatus according to claim 10, wherein

the semantic tree generator generates a semantic tree by arranging an enumeration of concept representations for which individual variables of a semantic frame is to be bidden at a lower node of a concept representation corresponding to the semantic frame, and, when the semantic frame corresponds to an arranged concept representation, repeating arranging the concept representations for which individual variables of the semantic frame of the arranged concept representation is to be bidden at a lower node of the arranged concept representation.

12. The language understanding apparatus according to claim 11, wherein

the semantic tree generator arranges a single concept representation for a bidding destination when there is the single concept representation for the bidding destination or an enumeration of group nodes indicating an enumeration when there are a plurality of variables to be bidden, at a lower node of a concept representation corresponding to a semantic frame, arranges an enumeration of a plurality of concept representations for which variables corresponding to the group nodes are to be bidden, under command of the group nodes indicating the enumeration, and, when the semantic frame corresponds to an arranged concept representation, repeats a similar process for the semantic frame of the arranged concept representation to generate a semantic tree, and converts the generated semantic tree to a semantic tree having only top nodes as group nodes when there is a group node in the generated semantic tree.

13. The language understanding apparatus according to claim 11, wherein

when detecting a repetitive pattern comprised of a plurality of variables in a same semantic frame, the semantic tree generator groups the repetitive patter and generates the semantic tree for each group.

14. The language understanding apparatus according to claim 11, further comprising

a task representation generator that generates a task execution instruction from the semantic tree generated by the semantic tree generator.

15. A language understanding method for use in a language understanding apparatus which converts an utterance content of a natural language into a semantic representation to be used in computer processing and has a storage section that stores concept structure data indicating a correlation between a tree stricture of a concept to be used in a domain and a concept representation specifying the concept to be used in the domain, frame definition data corresponding to the concept to be used in the domain and indicating a correlation between one or more semantic frames for representing the concept to be used in the domain by another concept, and one or more variables to be used in the one or more semantic frames, and variable definition data indicating a concept to be used as a variable, the method comprising:

a phrase interpreting step of receiving input utterance data, detecting a concept representation included in an utterance content indicated by the input utterance data by referring to the storage section, and red information of a concept corresponding to the detected concept representation from the storage section; and

a bidding step of;

reading information on a variable of a semantic frame corresponding to the concept indicated by the information read at the phrase interpreting step and information on a concept corresponding to the variable by referring to the storage section,

allocating by writing to a variable allocation table from the concept representation detected at the phrase interpreting step, for each semantic frame read;

a concept representation associated with a concept corresponding to each variable of the semantic frame or a subordinate concept of that concept, and

a concept representation whose variable of another semantic frame is not written, and generating variable allocation table data indicating a bidding result.

16. A non-transitory medium holding computer-executable instructions that when executed cause at least one computing device to:

store in a storage section;

concept structure data indicating a correlation between a tree structure of a concept to be used in a domain and a concept representation specifying the concept to be used in the domain,

frame definition data corresponding to the concept to be used in the domain and indicating a correlation between one or more semantic frames for representing the concept to be used in the domain by another concept, and

one or more variables to be used in the one or more semantic frames, and variable definition data indicating a concept to be used as a variable;

receive with a phrase interpreter input utterance data, detects a concept representation included in an utterance content indicated by the input utterance data by referring to the storage section, and reads information of a concept corresponding to the detected concept representation from the storage section; and

with a bidding section;

read information on a variable of a semantic frame corresponding to the concept indicated by the information read by the phrase interpreter and information on a concept corresponding to the variable by referring to the storage section,

allocate by writing to a variable allocation table from the concept representation detected by the phrase interpreter for each semantic frame read;

a concept representation associated with a concept corresponding to each variable of the semantic frame or a subordinate concept of that concept, and

a concept representation whose variable of another semantic frame is not written, and

generate variable allocation table data indicating a bidding result.