JP2600081B2 - Natural language dialogue device - Google Patents

Description

Description: BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a natural language dialogue apparatus for performing question answering and searching.

2. Description of the Related Art In a conventional natural language dialogue apparatus for performing question answering / searching, etc., it is assumed that a user of the apparatus is fixed at that level, and a device for various users having various levels is provided. Was not considered as.

Now, a database search apparatus using a natural language which is generally performed will be described.

Database search in Japanese is usually performed by inputting the logical sum and logical product of keywords, but this keyword is extracted from natural language input sentences,
It has been experimentally performed to perform a database search based on that and display or generate an answer. For example, from the phrase “Licensor is company A * Technology introduction contract”, the logical formula of the keyword Licensor (Company A) * Technology introduction contract is f (“Noun 1? Is noun 2?”) → Noun 1? (Noun 2?) F (modification sentence? @Noun?) → f (modification sentence?) * F (noun?) However,? A word with a symbol indicates a character string that can be classified into its syntactic category, and a symbol indicates concatenation of character strings. It is generated using the syntax / logical expression interaction rule f defined in such a manner. Such a conversion is effective only when a word corresponding to an attribute already defined in the domain of the database is directly used.

On the other hand, for example, from the phrase “contract where company A sold a patent”, the logical expression of the keyword “licensor (company A) * technology introduction contract” is g (“noun 3 sells noun 4?”) → 'Sell (Subject: Noun 3 ?, Target: Noun 4?)' G (Dependent sentence ?？ Noun?) → 'Noun?' ＠ g (Dependent sentence?) # H ('Contract' ＠ 'Sell (Subject: Noun) 3? ＠Corporate #, Target:
Noun 4? ＠Patent #) ') → Licensor (Noun 3?) * Technology introduction contract However,? A word with a symbol indicates a character string that can be classified into its syntactic category, a portion enclosed in single quotation marks (') indicates a semantic expression, and a type of a character string with a ＠ indicates the following # Is restricted by the semantic expression with.

There is also a technique of generating using a syntax / semantic correspondence rule g and a semantic / logical expression correspondence rule h expressed in such a form. Such a conversion is also effective when input using undefined words in the domain of the database,
In this case, it is implicitly assumed that the user always uses words that are not directly used in the database.

As shown in the above two examples, the natural language dialogue device according to the related art exerts its effect only on a certain level of user assumed for each device.

Problems to be Solved by the Invention In a conventional natural language dialogue apparatus, question answering / searching and the like are performed assuming that the level of the user is fixed, and it is possible to respond to various users having various levels. However, it lacks versatility for the user level.

Also, in order for the system to solve the problem given by the dialogue with the user, information is obtained from the user, but the information indicating how much information is sufficient from the user is sufficient. It is necessary to consider two aspects of sufficiency and the expressiveness of information, which is what kind of knowledge expression should be applied to the state where the user is placed.

In view of the above, in addition to the sufficiency of the information required for the system, the present invention realizes a flexible dialogue according to the level of the user in consideration of the expressibility of the information, and realizes various levels of the user. An object of the present invention is to provide a natural language dialogue device that can respond.

Means for Solving the Problems In order to achieve the above object, a natural language dialogue device of the present invention uses knowledge obtained by analyzing a user's natural language input sentence by a syntax / semantic analysis device (2). An information knowledge extraction device (3) for extracting information and knowledge possessed by a user from a semantic expression that is an expression in the form of a semantic expression (31) similar to the above, Classify as a set of rules having a condition described in the form of the semantic expression necessary for performing a database search and a premise and consequent part described in the form of the semantic expression, and a set of the rule and the condition. A plurality of sets of rules and conditions stored in the first storage means, and a subset of the rules and conditions stored in the first storage means. And stored for each set The second storage means (5) compares the semantic expression (31) with a plurality of sets of rules and conditions stored in the first storage means (4). Search for a set of rules and conditions that completely or partially match the rules, identify the one with the highest degree of conformity with the premise or condition of the rule, and apply the specified rules and conditions to the user. First judging means for recognizing rules and conditions for classifying the target problem and outputting a set of the rules and conditions and a result of applying the contents of the semantic expression (31) to the rules and conditions (6) and a set of rules and conditions output by the first determination means (6) and a result of applying the contents of the semantic expression (31) to the rules and conditions as target problem classification information knowledge. A target problem classification information knowledge storage device (7) to be stored; The second storage means (corresponding to the set of rules and conditions stored in the first storage means (4) whose taste expression (31) is specified by the first determination means (6). Compared with a plurality of rule and condition sets stored in 5), if there is more than one that has the highest degree of conformity with the rule or condition, or if there is more than one that has the same degree of conformity, the number of rules and conditions A second set of rules and conditions for classifying the level of the user, and outputting the set of rules and conditions;
A level classification information knowledge storage device (9) for storing the output of the determination means (8) as level classification information knowledge;
Of the rules and conditions stored in the storage means (4), the target problem classification information knowledge specified by the first determination means (6) and stored in the target problem classification information knowledge storage device (7) The condition to which it belongs belongs to the semantic expression (3
Directly by 1) or the semantic expression (31)
Is not indirectly satisfied by applying to the rule belonging to the target problem classification information knowledge, the semantic expression (31), the target stored in the target problem classification information knowledge storage device (7). From the problem classification information knowledge and the level classification information knowledge stored in the level classification information knowledge storage device (9), the target problem classification information knowledge stored in the target problem classification information knowledge storage device (7) is satisfied. And a semantic expression included in the level classification information knowledge stored in the level classification information knowledge storage device (9) and related to the content included in the semantic expression (31). An utterance intention / utterance content generation device (12) for identifying a certain semantic expression and generating an utterance intention / utterance content containing the specified semantic expression as an utterance necessary to satisfy the condition; It is those equipped.

In the above configuration, a natural language sentence input by the user is analyzed, and information and knowledge possessed by the user are extracted and compared with the inference rules and conditions stored in the first and second storage means. By determining the target problem and the level of the user in question, and generating a sentence corresponding to the intent / content of the next question to the user according to the target problem / level, In addition to generating dialogue expressions according to the level, it is also possible to efficiently obtain information necessary for problem solving and to answer and search questions using a natural language dialogue device that can respond appropriately to various users at various levels. Can be performed.

Embodiment An embodiment of the present invention will be described with reference to FIG. 1 and FIG.

In FIG. 1, 1 is an input device such as a keyword for performing Japanese input, and 2 is a syntax / semantic analysis of the input Japanese to obtain its semantic expression as a knowledge expression using terms. 3. A syntactic and semantic analysis device for extracting information / knowledge from a semantic expression of an input sentence, which extracts information / knowledge held by a user regarding a contract document to be searched in the same semantic expression form as described above. Device.

Reference numeral 4 denotes a condition described in the semantic expression format necessary for performing a database search for classifying a target problem in which the user is placed, and a premise described in the semantic expression format. Column of the target problem classification inference rule / condition storage device storing a plurality of pairs of inference rules having
Is column 4 of the target problem classification inference rule / condition storage device.
For each target problem classification inference rule and condition stored in each device, associate a subset of the inference rules and conditions as the level of user knowledge, and for each target problem classification inference rule and condition storage device Is a row of the level classification inference rule / condition storage device stored in the table.

Numeral 6 is for placing the user by comparing the information / knowledge obtained by the information / knowledge extraction device 3 with the inference rules / conditions stored in each column of the target problem classification inference rule / condition storage device 4. The inference rules / conditions stored in the determined target problem classification inference rules / condition storage device and the information obtained by the information / knowledge extraction device 3 for the inference rules / conditions are determined. A target problem determination / target problem classification information knowledge generation / correction device 7 for outputting the result of applying the content of knowledge, and the inference rule 7 output by the target problem determination / target problem classification information knowledge generation / correction device 6
A target problem classification information knowledge storage device for storing, as target problem classification information knowledge, a condition and a result obtained by applying the information / knowledge content obtained by the information / knowledge extraction device 3 to the inference rule / condition; Corresponds to information / knowledge obtained by the knowledge extraction device 3 and one of the target problem classification / inference rules determined by the target problem classification information knowledge generation / correction device 6 / column 4 of the condition storage device. The user's level is determined by comparing the inference rules and conditions stored in the column of the attached level classification inference rules and condition storage devices, and the determined level classification inference rules and condition storage devices are further determined. The level judgment / level classification information knowledge generation / correction device 9 which outputs the inference rules / conditions stored in A level classification information knowledge storage device for storing power and inference rules and conditions as the level classification information knowledge.

Reference numeral 10 denotes a database relating to contract documents. Reference numeral 11 denotes a database for searching a database by referring to the target problem classification information knowledge stored in the target problem classification information knowledge storage device 7 according to an instruction of the target problem determination / target problem classification information knowledge generation and correction device 6. This is a database search device for generating an instruction, searching the database 10, and obtaining a semantic expression of a search result.

Reference numeral 12 denotes the information / knowledge obtained by the information / knowledge extraction device 3, the target problem classification information knowledge stored in the target problem classification information knowledge storage device 7, and the level classification information knowledge storage device 9.
From the level classification information knowledge stored in the database, a semantic expression corresponding to the utterance intention and utterance content for the user is generated, and the utterance intention and utterance content for the user are obtained from the semantic expression of the search result output from the database search device 11. A speech intention / speech content generation device for generating a semantic expression corresponding to
13 is a sentence generation device used to generate an utterance sentence for the user using the semantic expression obtained from the utterance intention / utterance content generation device 12, and 14 is for displaying the Japanese language obtained by the sentence generation device 13. It is a display device 14 such as a VDT.

Hereinafter, the portion enclosed in single quotes (') in the description indicates a semantic expression description which is a knowledge expression using a term.

In column 4 of the target problem classification inference rule / condition storage device and column 5 of the level classification inference rule / condition storage device of the device according to the present embodiment, they are described in advance in the form of a semantic expression necessary for performing a database search. A plurality of sets each of the set conditions and inference rules having a premise part and a consequent part described in the form of a semantic expression in the form of “premise part → consequent part” are stored. The inference rules and conditions stored in column 4 of the target problem classification inference rules and conditions storage device are necessary inference rules regardless of whether or not the user knows the target problem in which the user is located.・ It is a condition. On the other hand, the inference rules and conditions stored in column 5 of the level classification inference rules and conditions storage device are those of the inference rules and conditions necessary for the target problem in which the user is located. Are the inference rules and conditions actually held.

Below, in the rules, a question mark (?) Is added after the uppercase letter
The ones marked with represent variables. Now, column 4 of the target problem classification inference rule / condition storage device as shown in FIG.
It is assumed that the following inference rules and conditions are stored in one of the target problem classification inference rules and conditions storage device 41.

Inference rules: Use patent (subject: X ?, patent: Y ?, product: Z ?, time: W1?), Own (subject: V ?, target: Y ?, time: W2?) → licensor ( Body: V?), Licensee (Body: X?), Contracted patent (Body: Y?), Contracted product (Body: Z?), Condition: Use patent (Subject: X ?, Patent: Y ?, Product: Z ?, Time: W1?) Owned (Subject: V ?, Target: Y ?, Time: W2?) Duplicated (Time 1: W1 ?, Time 2: W2?) Licensor (Body : V?), Licensee (main body: X?), Contracted patent (main body: Y?), Contracted product (main body: Z?), And target problem classification inference rules and conditions as shown in Fig. 2 The target problem classification and inference rules stored in the device 41
It is assumed that one level classification inference rule / condition storage device 51 in column 5 of the level classification inference rule / condition storage device associated with the condition stores the following inference rules / conditions.

Inference rules: Use patent (subject: X ?, patent: Y ?, product: Z ?, time: W1?), Own (subject: V ?, target: Y ?, time: W2?) → licensor ( Body: V?), Licensee (Body: X?), Contracted patent (Body: Y?), Contracted product (Body: Z?), Condition: Use patent (Subject: X ?, Patent: Y ?, Product: Z ?, Time: W1?) Owned (Subject: V ?, Target: Y ?, Time: W2?) Duplicated (Time 1: W1 ?, Time 2: W2?) Licensor (Body : V?), Licensee (main body: X?), Contracted patent (main body: Y?), Contracted product (main body: Z?), And also the target problem classification inference rules and conditions as shown in Fig. 2. The level classification inference rule corresponding to the target problem classification inference rule / condition stored in the storage device 41
It is assumed that the following inference rules and conditions are stored in the level classification inference rule and condition storage device 52 in the column 5 of the condition storage device.

Inference rule: none Conditions Use patent (subject: X ?, patent: Y ?, product: Z ?, time: W1?), Own (subject: V ?, target: Y ?, time: W2?) (Time 1: W1 ?, Time 2: W2?) Now, when the user inputs a sentence "I want to know the contract in which Company A uses a patent" by using the input device 1, it becomes a character string 11. Is output. When the syntactic / semantic analysis device 2 receives the character string 11, it outputs a semantic expression 21 indicating the intention of the user of the sentence as shown in (21) below.

The information / knowledge extraction device 3 outputs a semantic expression (31) as shown in (31) as information / knowledge possessed by the user in response to the semantic expression 21.

(31) 'Use a patent (Subject: Company A)' Target problem determination / target problem classification information knowledge generation and correction device 6
Compares the semantic expression 31 with the inference rules and conditions stored in each device in column 4 of the target problem classification inference rules and conditions storage device, and completely matches the premise or condition of the inference rules.
Alternatively, search for a set of inference rules and conditions that partially match,
By specifying the one that has the highest degree of conformity with the premise or condition of the inference rule, the target problem classification inference rule that stores the inference rules and conditions appropriate to indicate the target problem for the user One storage device in the column 4 of the condition storage device is determined, the target problem classification inference rule / condition stored in the determined storage device, and the meaning obtained as a result of applying the inference rule to the semantic expression 31. The expression and the target problem classification information knowledge storage device 7 as the target problem classification information knowledge 61
And the semantic expression 31 is passed to the level determination / level classification information knowledge generation / correction device 8.

In this example, the target problem classification inference rule shown in FIG.
The target problem classification inference rules and conditions stored in the condition storage device 41 are selected, and the target problem classification information knowledge and the semantic expressions obtained as a result of applying the inference rules to the semantic expressions 31 are selected.
61 is stored in the target problem classification information knowledge storage device 7.

Of the inference rules and conditions included in the target problem classification information knowledge stored in the target problem classification information knowledge storage device 7, the conditions are directly depending on the semantic expression 31, and by applying the inference rules. Since it has not been satisfied indirectly yet, the level judgment / level classification information knowledge generation / correction device 8 next selects the semantic expression 31 and the inference rules / conditions stored in the level classification inference rule / condition storage device 5 described above. Of the target problem classification inference rules and conditions stored in the inference rules and conditions storage device 41 and a plurality of sets of inference rules and conditions associated with each other as a subset of the inference rules and conditions. If there is more than one with the highest degree or with the same degree of relevance, it is appropriate to indicate the level of the user by specifying one with a small number of inference rules and conditions. The storage device storing the inference rules and conditions is specified, and the level classification inference rules and conditions stored in the specified storage device are stored in the level classification information knowledge storage device 9 as the level classification information knowledge 81, and the semantic expression 31 is stored. Is passed to the speech intention / speech content generation device 12.

In this example, the level classification inference rules and conditions stored in the level classification inference rules and condition storage devices 51 and 52 shown in FIG. 2 are compared with the semantic expressions 31, respectively. The level classification inference rules and conditions stored in the storage device 52 are selected and stored in the level classification information knowledge storage device 9 as the level classification information knowledge 81.

Next, the utterance intention / utterance content generation device 12 stores the semantic expression 31, the target problem classification information knowledge stored in the target problem classification information knowledge storage device 7, and the level classification information knowledge storage device 9.
Is necessary to satisfy the database search conditions described in the target problem classification information knowledge using the level classification information knowledge stored in the The utterance intention / speech content that uses the expression and is related to the semantic expression 31 for the purpose of collecting information from the user is generated and passed to the sentence generation device 13.

In this case, first, the conditions described in the target problem classification information knowledge stored in the target problem classification information knowledge storage device 7 are stored in the semantic expression 31 and in the semantic expression 31 in the same device 7. By applying the inference rules, only the variable X? Portion of the conditions stored in the original target problem classification inference rule / condition storage device 41 is filled in as shown below. I have.

Use a patent (Subject: Company A, Patent: Y ?, Product: Z ?, Time: W1?) Own (Subject: V ?, Target: Y ?, Time: W2?) Duplicate (Time 1: W1:?, Time 2: W2?) Licensor (main unit: V?) Licensee (main unit: Company A) Contracted patent (main unit: Y?) Contracted product (main unit: Z?) Thus, target problem classification information In order to satisfy the conditions stored in the knowledge storage device 7, it is necessary to obtain information on variables Y ?, Z ?, and V? From the user.

Next, the inference rules and conditions described in the level classification information knowledge stored in the level classification information knowledge storage device 9 are as follows: Inference rules: None Condition: Use a patent (subject: X ?, patent: Y ?, product : Z?, Time: W1?) Own (Subject: V?, Target: Y?, Time: W2?) These variables are duplicated (Time1: W1?, Time2: W2?) To obtain information about Y ?, Z ?, and V ?, use a patent (patent: Y?) Use a patent (product: Z?) Own (Subject: V?) Own (target: Y?) ) Can be a candidate. Further, the semantic expression 31 for the immediately preceding input sentence is (31) 'Use a patent (subject: Company A)', and it is assumed that the content included in this semantic expression 31 is relevant from the above four candidates. Select Use a patent (patent: Y?). This is because the same expression as the knowledge expression included in the semantic expression 31 is used. Therefore, for the purpose of listening to Y ?, in this embodiment, as shown in (121), a semantic expression 121 indicating utterance intention and utterance content is generated for the purpose of obtaining detailed information on the patent.

(121) 'I want to know (target: patent)' The sentence generation device 13 receives the above semantic expression 121 and
31) shows a state in which a Japanese utterance sentence 131 as shown in 31) is passed to the display device 14, and the display device 14 presents the utterance sentence 131 to the user.

(131) "Tell me about patents" The user makes further input in response to this sentence, and as a result, the same processing as above is performed.

In this example, the level classification inference rule / condition storage device 52 is selected as the storage device for storing the level classification inference rule / condition. However, the user responds to the above utterance sentence 131 by saying that the contracted patent is Japanese. If an input sentence such as "Company A wants to know the contract using a patent as a licensee" is obtained instead of the first input sentence 11, The judgment / level classification information knowledge generation / correction device 8 selects the level classification inference rule / condition storage device 51.

As described above, the level determination / level classification information knowledge generation / correction device 8 can change the level classification inference rule / condition storage device that stores the inference rules / conditions to be used according to the input sentence of the user during the dialogue. . When it is changed, the level classification information knowledge stored in the level classification information knowledge storage device 9 is also corrected, and the utterance intention / utterance content generation device 12 is used by the user according to the corrected level classification information knowledge storage device 9. The utterance intention and utterance contents for are determined.

Target problem determination / target problem classification information knowledge generation / correction device 6
In the comparison process, if all the conditions described in the target problem classification information knowledge stored in the target problem classification information knowledge storage device 7 are satisfied, the target problem determination / target problem classification information knowledge generation correction The device 6 generates a command 62 for starting a search and passes it to the data search device 11. The database search device 11 generates a database search command 111 using the conditions described in the target problem classification information knowledge stored in the target problem classification information knowledge storage device 7 at that time, and searches the database 10. ,
The semantic expression 112 as the search result is passed to the utterance intention / utterance content generation device 12. The utterance intention / utterance content generation device 12 receives the semantic expression 112, determines the utterance intention / utterance content corresponding to the semantic expression 122, and determines the sentence generation device 13
Sentence generator 13 sends the Japanese sentence 1 based on the semantic expression 122.
32, and the display device 14 presents the Japanese sentence 132 to the user.

As described above, the present invention uses the information and knowledge of the user's problem target area to determine both the user's target problem and the user's level for the problem. , By controlling the operation related to the target area according to both the target problem and the level of the user, and by selecting the content related to the immediately preceding user's utterance, A natural language dialogue with a natural flow corresponding to the user at the target problem and level can be performed.

Further, the utterance intention / utterance content (12) identifies the condition that the level classification inference condition should satisfy the target problem inference condition, and the content prompting the user to input to satisfy the condition is the next question content. Therefore, the utterance contents suitable for obtaining the information required for the system can be generated.

That is, even if the problem to be solved by the system is the same,
By generating the next question with the natural language dialogue device according to the present invention, what information is necessary for solving the problem can be efficiently obtained from the user.

[Brief description of the drawings]

FIG. 1 is a block diagram of a database search system according to an embodiment of the present invention. FIG. 2 is a column 4 of a target problem classification inference rule and condition storage device and a level classification inference rule and condition storage device shown in FIG. 5 is a conceptual diagram showing a configuration of a column 5 of FIG. 1 input device 2 syntactic / semantic analysis device 3 information / knowledge extraction device 4 target subject class inference rule / condition storage device column 5 level classification inference rule / condition storage device ..., Target problem determination / target problem classification information knowledge generation / correction device, 7: target problem classification information knowledge storage device, 8
... Level judgment / level classification information knowledge generation / correction apparatus, 9
...... Level classification information knowledge storage device, 10 ... Database, 11 ... Database search device, 12 ... Speech intention / speech content generation device, 13 ... Sentence generation device, 14 ... Display device,
41,42 …… Target problem classification inference rule / condition storage device, 51,5
2,53 ... Level class inference rule / condition storage device.

Claims (1)

(57) [Claims] 1. A method for extracting a condition necessary for performing a database search based on an input sentence in a natural language, performing a search in accordance with the condition, and generating a response natural language corresponding to the search result. In the language dialogue device, the information and knowledge possessed by the user are obtained from the semantic expression, which is a knowledge expression using a term obtained by analyzing the user's natural language input sentence by the syntax / semantic analysis device (2). Information knowledge extraction device (3) that extracts in the form of a similar semantic expression (31)
A condition described in the form of the semantic expression necessary for performing a database search, and a premise part and a consequent part described in the form of the semantic expression necessary for performing a database search. A first storage unit (4) storing a plurality of sets of rules and conditions, classified as a set of rules; and a plurality of sets of rules and conditions stored in the first storage unit, A subset of the rules and conditions
Second storage means (5) for storing each set in association with a user's knowledge level, and rules and conditions stored in the first storage means (4) for the semantic expression (31) Search for a set of rules and conditions that completely or partially match the prerequisites or conditions of the rule, and find the one that has the highest degree of conformity with the prerequisites or conditions of the rule. The specified rules and conditions are identified as rules and conditions for classifying the target problem in which the user is located, and the set of rules and conditions and the semantic expression for the rules and conditions (31) First determination means (6) for outputting a result obtained by applying the content of (1), a set of rules and conditions output by the first determination means (6), and the semantic expression (31) for the rules and conditions The result of applying the content of the subject as the target problem classification information knowledge A target problem classification information knowledge storage device (7) to be stored; rules and conditions stored in the first storage means (4) in which the semantic expression (31) is specified by the first determination means (6) Is compared with a plurality of sets of rules and conditions stored in the second storage means (5), which correspond to the set of rules. If there are a plurality of the same items, a second determination means for specifying a rule and condition set having a small number of rules and conditions as a set of rules and conditions for classifying the user level and outputting the set of rules and conditions (8), a level classification information knowledge storage device (9) that stores the output of the second determination means (8) as level classification information knowledge, and a rule stored in the first storage means (4). And the condition are determined by the first determination means (6). Then, the condition belonging to the target problem classification information knowledge stored in the target problem classification information knowledge storage device (7) is either directly by the semantic expression (31) or subject to the semantic expression (31). When not satisfied indirectly by applying to the rules belonging to the problem classification information knowledge,
From the semantic expression (31), the target problem classification information knowledge stored in the target problem classification information knowledge storage device (7), and the level classification information knowledge stored in the level classification information knowledge storage device (9), The level classification information knowledge necessary to satisfy the target problem classification information knowledge stored in the target problem classification information knowledge storage device (7) and stored in the level classification information knowledge storage device (9) And the semantic expression related to the content included in the semantic expression (31), and the intention and the utterance content having the specified semantic content satisfy the condition. Utterance intentions generated as utterances necessary for
An utterance content generation device (12) is provided, and utters to the user based on the content generated by the utterance intention / utterance content generation device (12), and prompts the user for input. Natural language dialogue device with improved interactivity with the user for the purpose.