JPH04220767A - Context processing system - Google Patents

Abstract

PURPOSE:To efficiently realize a context processing by acknowledging the object of indication indicated by a demonstrative pronoun which appears in an input sentence and complement an omitted noun phrase by means of information retrieval with a natural language. CONSTITUTION:By utilizing a meaning indicator dictionary 2 in which the meaning indicator of respective words is registered, an input sentence 1 is expressed in a meaning expression by a syntax analizing means 3. The pattern checking is conducted for the meaning expression by means of the meaning expression in a preceding context obtained by the preceding sentence context storage means 6 and a pattern collating means 5, the noun phrase in the input sentence closely related to the noun phrase in the preceding context is detected and the meaning expression for which the pattern check has already been executed is outputted to a result output means 8. At this time, a data model 4 for retrieval information and a function expression table 7 are used as reference for the pattern check. Since the meaning indicator dictionary 2, the data model 4 for retrieval information and the function expression table 7 are the data taking into account only the close relationship between the words, they can be easily prepared.

Description

[Detailed description of the invention]

0001

[Industrial Application Field] The present invention relates to natural language context processing that identifies the referent of a demonstrative pronoun appearing in a natural language input sentence and complements omitted noun phrases in a natural language information retrieval system. Regarding the method.

0002

2. Description of the Related Art In natural language information retrieval, a user's natural language search is input and converted into an information retrieval command language. By using information retrieval in natural language, even users who do not know the command language for information retrieval can search a database in, for example, Japanese. In a natural language information retrieval method, it is necessary to identify the referent of a demonstrative pronoun in a natural language input sentence and to complement omitted noun phrases.

Conventionally, context processing has been realized by referring to the semantic relationships between words in a semantic network. Figure 2 shows an example of a semantic network. For example, in the context of ``Taro loves Hanako,''``she'' in the sentence ``She is beautiful'' refers to ``Hanako.'' It is recognized by being subordinate to "woman".

For example, as described in Japanese Patent Application Laid-Open No. 62-206670, context processing is realized by referring to a semantic network that has semantic relationships such as superior-subordinate relationships, attribute relationships, deep case relationships, and synonym relationships. Ta. FIG. 3 shows a configuration diagram of the context processing method described in Japanese Patent Application Laid-Open No. 62-206670.

[0005]

[Problems to be Solved by the Invention] In the conventional context processing method that refers to the semantic relationships between words in a semantic network, it is necessary to prepare a huge semantic network, which makes it difficult to efficiently search data in the network. It cannot be realized. Furthermore, as the number of words in natural languages increases, it becomes virtually impossible to prepare a semantic network.

[0006]

[Means for Solving the Problems] The first invention is a method for analyzing a natural language input into a computer for information retrieval and converting it into a command language for information retrieval, in which a meaning marker for each word is registered. a semantic marker dictionary, a parsing means that receives an input sentence, refers to the semantic marker dictionary, and outputs a first semantic expression; and a third semantic expression that performs pattern matching between the first semantic expression and the second semantic expression. The present invention is characterized by comprising: a pattern matching means for outputting a semantic expression; and a preceding context storage means for storing a third semantic expression outputted from the pattern matching means and outputting it as a second semantic expression.

[0007] The second invention uses a pattern matching means that performs pattern matching between a first semantic expression and a second semantic expression and outputs a third semantic expression by referring to a data model of search information. It is characterized by

[0008] The third invention uses a pattern matching means that performs pattern matching between the first semantic expression and the second semantic expression and outputs a third semantic expression by referring to a table of functional expressions. It is characterized by

[0009] The fourth invention is a pattern that performs pattern matching between a first semantic expression and a second semantic expression by referring to a data model of search information and a table of functional expressions, and outputs a third semantic expression. It is characterized by using a verification means.

0010

In the context processing method of the present invention, the pattern matching means performs pattern matching between the semantic expression and the preceding context by referring to the semantic indicator of the word, the data model of the search information, and the functional expression table. The semantic marker dictionary, the data model of the search information, and the functional expression table are data that focuses only on sibling relationships between words, and can be easily prepared.

[0011] The semantic marker dictionary requires only a description amount on the order of the number of words. A database schema is used as a data model for search information. A database schema is determined at the time the database is defined, and the amount of description thereof is limited. The number of functional expressions described in the functional expression table is also limited. Overall, context processing can be achieved with less computer memory.

Furthermore, in the pattern matching means for pattern matching between the semantic expression and the preceding context, the semantic indicator dictionary of the word to be referred to, the data model of the search information, and the functional expression table are described in a distributed manner. The probability of successful pattern matching is high and the processing is efficient.

[0013]

[Examples] Examples of the present invention will be shown below. FIG. 1 is a diagram showing the configuration of the context processing method of the present invention. In Figure 1,
1 is the input sentence, 2 is the meaning indicator dictionary, 3 is the syntactic analysis means, 4
is the data model of the search information, 5 is the pattern matching means, 6
7 is a preceding context storage means, 7 is a function expression table, and 8 is a result output means.

a. Meaning Marker Dictionary In the present invention, the meaning mark of a word is defined by the generic concept of the word. For example, since the superordinate concept of the noun phrase "calculator" is "thing", the semantic marker of the noun phrase "calculator" is defined as "thing". All possible meaning markers are prepared in advance, and the meaning markers of each word are registered in a meaning marker dictionary. Figure 4 shows a list of possible semantic markers for a noun phrase. Figure 4
In , semantic markers are expressed using abbreviations. For example, the abbreviation for the meaning marker "thing" is "mo". FIG. 5 shows an example of word meaning markers in the meaning marker dictionary.

b. Data Model of Search Information The data model of search information is realized by a database schema when the search information is stored in a database. A database schema is defined by data items of the database in which information to be searched is stored. For example, the schema of the database shown in FIG. 6 is as shown in FIG. Schemas of databases that have items in common with other databases are associated with schemas of other databases by the common items and defined together. For example, as shown in Figure 8, "Name"
The schemas of two databases that have a common item ``name'' are related and defined by a common item ``name''.

c. Function Expression Table In the natural language input sentence, there is a function expression that expresses only a search function that does not depend on the content of the information to be searched. for example,
The noun phrase ``average value'' in ``Show me the average value of prices'' and the noun phrase ``average value'' in ``Show me the average value of height'' are both used in the search function. Expressing the average value. In this way, a functional expression table of words serving as functional expressions is prepared in advance. FIG. 9 shows an example of a table of functional expressions of noun phrases.

d. Syntax analysis means The syntax analysis means receives an input sentence and outputs a semantic expression. The semantic expression expresses the search content in natural language. FIG. 10 shows an example of a meaning expression.

e. Preceding Context Storage Means The preceding context storage means stores the preceding context retrieved in the user's previous information search. In the present invention, the means for expressing the preceding context is the same as the semantic expression. For example, when the previous information search is "Show me the sales of company A", the preceding context storage means stores the meaning expression of the sentence "Show me the sales of company A".

f. Pattern Matching Means The pattern matching means inputs the semantic representation of the preceding context obtained from the preceding context storage means and the semantic representation of the input sentence obtained from the syntactic analysis means, and performs pattern matching to identify noun phrases and siblings in the preceding context. A related noun phrase in an input sentence is detected, and a pattern-matched meaning expression is output to a result output means. Further, if there is no noun phrase in the input sentence that has a sibling relationship with a noun phrase in the preceding context, the pattern matching means outputs the semantic representation of the input sentence from the syntactic analysis means to the result output means. Noun phrases that have a sibling relationship are: a. Noun phrases with the same semantic marker. Noun phrases in the same schema or in related schemas. A noun phrase registered in the functional expression table.

For example, the noun phrase "Taro" and the noun phrase "Hanako"
Since they both have the semantic marker ``hi'', they are in a sibling relationship with each other. Figure 5 shows the meaning markers of the noun phrase ``Taro'' and the noun phrase ``Hanako.'' Further, since the noun phrase "price" and the noun phrase "age" are both in the associated database schema, they have a sibling relationship with each other. FIG. 8 shows the relationship between the noun phrase "price" and the noun phrase "age." Further, since the noun phrase "average value" and the noun phrase "maximum value" are both in the functional expression table, they have a sibling relationship with each other. FIG. 7 shows an example of a table of functional expressions containing the noun phrases "average value" and "maximum value."

[0021] In the following, when a noun phrase in the input sentence and a noun phrase in the preceding context have a sibling relationship, they will be referred to as noun phrase 1 and noun phrase 2, respectively. Pattern matching means is noun phrase 1
Pattern matching is performed between the noun phrase in the input sentence that modifies noun phrase 2 and the noun phrase in the preceding context that modifies noun phrase 2, and at the same time the noun phrase in the input sentence that modifies noun phrase 1 and the preceding context that modifies noun phrase 2. Perform pattern matching with noun phrases in .

When performing pattern matching between a noun phrase that modifies noun phrase 1 and a noun phrase that modifies noun phrase 2, the pattern matching means compares the noun phrase that modifies noun phrase 1 and the noun phrase that modifies noun phrase 2. Detect whether there is a sibling relationship. If the noun phrase that modifies noun phrase 1 is not in a sibling relationship with the noun phrase that modifies noun phrase 2, the pattern matching means uses noun phrase 2.
A noun phrase that modifies noun phrase 1 is added to the newly input semantic expression as one that modifies noun phrase 1. For example, if the semantic expression ``Please tell me Taro's age'' is in the preceding context, and ``The address is'' is input next, ``Address'' is ``Taro's age''.
Because it is in the same database schema as ``Age'',
They have a sibling relationship. Since "Taro", which modifies "age", has no sibling relationship with other parts that modify "address", the pattern matching means newly adds "Taro" to "age" as a modifier, and adds " The meaning expression "Taro's address is" is output to the result output means.

[0023] Furthermore, if "the maximum price is" is in the preceding context and "its average value is" is input next, then "maximum value" and "average value" are both in the functional expression table. Therefore, they have a sibling relationship. As a result of pattern matching between the input semantic expression and the preceding context, the pattern matching means adds a new "price" that modifies the "maximum value" to the "average value",
Outputs the semantic expression "average price".

[0024] Furthermore, if there is a preceding context of ``Please tell me Taro's age,'' and then ``What is Hanako's address?'' is input, ``address'' and ``age'' are in a sibling relationship. However, since "Taro" which modifies "age" and "Hanako" which modifies "address" are siblings, the pattern matching means uses the semantic expression of the input sentence "Hanako's address is" as the result output means. Output.

[0025] If the noun phrase modified by noun phrase 1 is not in a sibling relationship with the noun phrase modified by noun phrase 2, the pattern matching means modifies the noun phrase modified by noun phrase 2 into noun phrase 1. is added to the newly input semantic expression. For example, if the input sentence "Please tell me Taro's age" is in the preceding context, and "Hanako wa" is input next, "Taro" and "Hanako" can be distinguished by matching the semantic expression and the preceding context. A sibling relationship is detected. Since "age" is modified by "Taro" and there is nothing modified by "Hanako", the pattern matching means uses the newly input meaning of "age" as modified by "Hanako". In addition to the expression, the meaning expression "Hanako's age is" is output to the result output means.

FIG. 11 shows an example of pattern matching performed by the pattern matching means. In Figure 11, (1) is the semantic representation of the preceding context, (2) is the semantic representation of the input sentence input from the syntactic analysis means, and (3) is the semantic representation shown in (1) and (2) in the pattern. It is a semantic expression obtained by matching.

g. Result Output Means The result output means inputs the semantic expression from the pattern matching means and outputs it to the preceding context storage means as a semantic expression of the preceding context.

[0028]

According to the present invention, the referent of a demonstrative pronoun can be recognized and the completion of a noun phrase can be realized in a natural language information retrieval system. By referring to the semantic marker dictionary and the data model and function expression table of the search information, it is possible to describe the data necessary for context processing with less computer description than in the conventional method, and it is possible to improve processing efficiency. can.

[Brief explanation of the drawing]

FIG. 1 is a diagram showing the configuration of a context processing method according to the present invention.

FIG. 2 is a diagram showing an example of a semantic network.

FIG. 3 is a diagram illustrating an example of a conventional context processing method.

FIG. 4 is a diagram showing a list of meaning indicators of noun phrases.

FIG. 5 is a diagram showing an example of word meaning markers in a meaning marker dictionary.

FIG. 6 is a diagram showing an example of a database.

FIG. 7 is a diagram showing an example of a database schema.

FIG. 8 is a diagram illustrating an example of a schema of associated databases.

FIG. 9 is a diagram showing an example of a function expression table.

FIG. 10 is a diagram showing an example of a meaning expression.

FIG. 11 is a diagram showing an example of pattern matching performed by pattern matching means.

[Explanation of symbols]

1 Input sentence 2 Semantic marker dictionary 3 Syntax analysis means 4 Data model of search information 5 Pattern matching means 6 Preceding context storage means 7 Function expression table 8 Result output means

Claims (4)

[Claims] Claim 1: A method for analyzing natural language input into a computer for information retrieval and converting it into a command language for information retrieval, comprising: a semantic marker dictionary that registers the meaning markers of each word; and a semantic marker dictionary that receives input sentences. a syntactic analysis means for outputting a first meaning expression by referring to the meaning marker dictionary; and a pattern matching means for performing pattern matching between the first meaning expression and the second meaning expression and outputting a third meaning expression. and a preceding context storage means for storing the third semantic expression outputted from the pattern matching means and outputting it as a second semantic expression. 2. Using a pattern matching means that performs pattern matching between the first semantic expression and the second semantic expression and outputs the third semantic expression by referring to a data model of search information. The context processing method according to claim 1, characterized in that: 3. Using a pattern matching means that performs pattern matching between the first semantic expression and the second semantic expression and outputs the third semantic expression by referring to a table of functional expressions. The context processing method according to claim 1, characterized in that: 4. A pattern that performs pattern matching between the first semantic expression and the second semantic expression and outputs the third semantic expression by referring to a data model and a functional expression table of search information. The context processing method according to claim 1, characterized in that a matching means is used.