JP2555918B2 - Context analysis method - Google Patents

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a context analysis method for a natural language processing system that analyzes input data described in a natural language such as a natural language interface system or a machine translation system and extracts the meaning of the data. .

[0002]

2. Description of the Related Art In a natural language processing system such as a natural language interface system or a machine translation system, context analysis is required to perform abbreviation complement or anaphora identification for abbreviated expressions or anaphoric expressions in an input sentence. As a conventional technology,
A context analysis device "Japanese Patent Application Laid-Open No. 2-068661" and a context processing method "Japanese Patent Application No. 2-404767" using a syntactic analysis result and target data model knowledge are known. Also, 3
Context analysis method "Japanese Patent Application No. 4-176201" in which the serial means, that is, the morphological context analysis means, the syntactic context analysis means, and the semantic context analysis means perform context analysis using the morphological / syntactic / semantic analysis results and memory knowledge, respectively. It has been known.

[0003]

Any of the above-mentioned devices
Even in the method, it is insufficient when context processing that combines both syntactic knowledge and semantic knowledge is required.

That is, when the syntactic analysis result and the target data are used, it is difficult to associate the two in the case where the syntactic structure of the synonymous expression is significantly different between the context and the input sentence. Further, in the method in which the context analysis process is divided into three serial means, the syntactic analysis and the semantic analysis are separated, so that the analysis of the expression in which the syntactic and semantic are combined is all performed in the semantic context analysis. In this case, even if the relationship based on the syntactic structure is the main and the relationship based on the meaning is subordinate, the processing is delayed until the semantic context analysis, which is inefficient.

An object of the present invention is to provide a method for efficiently performing context analysis processing, including the case where the relationship between sentences is syntactic knowledge and semantic knowledge.

[0006]

According to the present invention, a morpheme / syntactic analysis means for creating a syntactic structure by performing morphological analysis and syntactic analysis of a sentence described in natural language, and inputting the syntactic structure
Then, a syntactic context analysis means for complementing the omission by using the semantic knowledge and the context knowledge and outputting a complementary syntactic structure, and a meaning for extracting the semantic structure expressing the meaning of the sentence from the semantic knowledge and the complementary syntactic structure Analysis means, and input the semantic structure
Then , reference is made to use the semantic knowledge from the syntactic context analysis means and the semantic context analysis means for creating a complementary semantic structure by establishing and complementing the semantic structure using the semantic knowledge and the context knowledge. And a semantic knowledge storage unit that stores the semantic knowledge and is referred to by the semantic analysis unit, the semantic context analysis unit, and the knowledge prefetching unit.

Further, among the semantic knowledge , attributes or events
And partial semantic knowledge storing means for storing the knowledge acquired by the partial semantic knowledge acquiring means and making it available from the knowledge prefetching means. It is characterized by having.

[0008]

In the context analysis method of the present invention, the context analysis is separated into two phases, syntactic context analysis and semantic context analysis, and the former utilizes the common semantic knowledge, while the former is context analysis centered on the syntactic structure. In the latter, context analysis focusing on the semantic structure after degeneracy of synonymous expressions is performed. Furthermore, for syntactic context analysis, the semantic knowledge that is frequently used is extracted and used as partial knowledge to improve the processing efficiency.

[0009]

DESCRIPTION OF THE PREFERRED EMBODIMENTS The contents of the present invention will be described with reference to the drawings. FIG. 1 shows a block diagram of an embodiment for implementing the method of the present invention.
FIG. 2 is a block diagram showing the structure of the syntax context analysis means 2 in FIG. FIG. 3 shows a context analysis algorithm in the syntactic context analysis means shown in FIGS. 1 and 2. FIG. 4 shows a knowledge acquisition algorithm in the partial semantic knowledge acquisition means 8 in FIG. FIG. 5 shows a determination algorithm in the part that determines whether or not the partial semantic knowledge selection criterion in FIG. 4 is satisfied. FIG. 6 shows example dialogue sentences in the natural language interface system. FIG. 7 shows analysis intermediate data and results for the example sentence in FIG. 8 and 9, a part of the semantic knowledge in the semantic knowledge storage unit in FIG. 1 is shown as a representative example. 10 and 11 show partial semantic knowledge examples with respect to the semantic knowledge examples of FIGS. 8 and 9, respectively.

In FIG. 1, the morpheme / syntactic analysis means 1 creates a syntactic structure by subjecting an input natural language sentence to morphological analysis and syntactic analysis. Syntax context analysis means 2
Creates a complementary syntactic structure by using the syntactic structure as an input and using the context knowledge and the result of inquiry of the knowledge prefetching means 5 regarding the semantic knowledge. The semantic analysis unit 3 receives the complementary syntactic structure as an input and uses the knowledge of the semantic knowledge storage unit 6 to extract the corresponding semantic structure. The semantic context analysis means 4 receives the above-mentioned semantic structure as an input and stores the context knowledge and the semantic knowledge storage means 6
Create a complementary semantic structure using the knowledge of and. The knowledge prefetching means 5 refers to the semantic knowledge in the partial semantic knowledge storage means 7 and the semantic knowledge storage means 6 and responds to the reference result in accordance with the request from the syntactic context analysis means 2 described above. The semantic knowledge storage means 6 stores the semantic knowledge, and outputs necessary knowledge in accordance with requests from the semantic analysis means 3, the semantic context analysis means 4, the knowledge prefetching means 5 and the partial semantic knowledge acquisition means 8. The partial semantic knowledge storage means 7 stores the partial semantic knowledge created by the partial semantic knowledge acquisition means 8 and outputs the necessary knowledge according to the request from the knowledge prefetching means 5. The partial-semantic-knowledge acquiring means 8 extracts knowledge that is frequently used in context analysis from the above-mentioned semantic-knowledge storing means 6 and stores the knowledge in the partial-semantic-knowledge storing means 7.

Next, the operation of the context analysis method will be described with reference to FIGS. 1 to 5 and 6 to 11. As shown in FIG. 1, syntactic context processing is performed on the syntactic structure resulting from morpheme / syntactic analysis of an input sentence described in natural language. As shown in FIG. 2, the basic structure of the syntactic context processing is composed of a syntactic context rule which describes a context analysis engine and a context rule centering on a syntactic structure. Further, the knowledge prefetching means for utilizing the knowledge about the focus stack and meaning that accumulates and manages the focus information of the previous context in a stack form from the grammar rule is used. For the focus stack, Sidne
r, C.I. L. "Focusing in the Co
prehension of Define A
naphora "(Computational Mo
dels of Discourse, 1983). FIG. 3 shows an algorithm of syntactic context processing by this mechanism. Here, the grammatical rule corresponding to the input sentence is selected, and the previous context information is searched while tracing the focus stack accordingly. When there is a correspondence candidate between the input sentence and the previous context, the correspondence relationship and the validity of complementation are verified by the semantic knowledge. The above-mentioned knowledge prefetching means is used for utilizing this semantic knowledge. When it is determined that the complement is valid, the input sentence is complemented using the context information, and the result is used as the complementary syntax structure. If it is determined to be invalid,
Another candidate or another grammar rule is searched and the same processing is performed. If all rules fail, it is syntactically determined to be unrelated to the previous context, and the input syntax structure is directly interpreted as the complementary syntax structure. After that, after performing semantic analysis on the complementary syntactic structure, semantic context analysis is performed on the result. In this semantic context analysis, the meaning structure corresponding to the contents of the focus stack is used as the value of the focus stack, and the correspondence / complementation processing with the semantic structure of the input sentence is performed while utilizing the semantic knowledge. The basic method is similar to the syntactic context analysis means.

Next, a process of processing the pre-context and the input sentence example shown in FIG. 6 will be described. Figure 6 shows the previous context and 4 input examples,
And the interpretation result sentence. Here, the input 4 example sentences have different expressions but have the same meaning. In Examples 2 and 3, the expression equivalent to “sales ranking” in Example 1 corresponds to the phrase “ranking of sales”, and in Example 4, it is expanded into two phrases “about sales ... ranking”. Furthermore, the expressions “in Osaka” in Examples 1 and 2 are shortened to “in Osaka” in Examples 3 and 4. FIG. 7 shows a syntactic structure corresponding to the previous context, the input sentence, and the interpretation result. At the time of performing the syntactic context processing, the syntactic structure of this previous context is the value of the focus stack, and the correspondence processing between this and the syntactic structure of the input sentence is performed. When processing syntactic context, refer from knowledge prefetching means
FIG. 8 shows an example of the semantic knowledge to be used. The example 1 is correlated using "synonymous relationship" between "company" and "company", "ownership relationship" between "company" and "sales ranking", and the like, as shown in FIG. The interpretation result shown in is obtained.
On the other hand, for Examples 2 to 4, knowledge of an idiom that deals with synonymous relationships between words and phrases is required. The conventional context analysis using the knowledge of the target data model cannot handle it well because the knowledge is insufficient. In the present invention, this idiom-related knowledge is treated as a part of semantic knowledge, and an example thereof is shown in FIG. By converting the expression of the input sentence using this knowledge, the context processing of Examples 2 to 4 becomes possible. Note that, among Examples 1 to 4, Example 1-3 is handled by the syntactic context analysis means, and in Example 4, the meaning analysis means associates the topic (sales) with the target (rank) and then the meaning Process by context analysis means.

Among the examples of the syntactic context analysis described above, the input examples 1 and 2 having a large syntactic similarity can be efficiently processed by extracting and utilizing partial semantic knowledge from the semantic knowledge. FIG. 4 shows an algorithm for extracting partial semantic knowledge from the semantic knowledge. FIG. 5 shows an algorithm of partial semantic knowledge selection, which is the central part of the algorithm. From the semantic knowledge, knowledge of synonyms and broader terms related to entities, attributes and events, and the relationship between them, and idiom knowledge about attributes are selected. An example of partial semantic knowledge created according to the algorithms shown in FIGS. 4 and 5 is shown in FIGS. FIG. 10 expresses the knowledge of synonyms, superordinate words, etc. that satisfy the selection criteria among the semantic knowledge in FIG. 8 and the related knowledge between them. Therefore, enormous knowledge such as subordinate words and entities in the semantic knowledge is excluded. FIG. 11 shows an example of partial semantic knowledge regarding idioms. Of the knowledge in FIG. 9, only the knowledge about attributes is adopted, and the knowledge about entities is excluded.

An example 2 of FIG. 6 will be described as an example of using this partial semantic knowledge. Here, the attribute information of "company" called "sales ranking" is expressed by the phrase "order of sales". By utilizing the partial semantic knowledge about the idiom in FIG. 11, both can be dealt with, so that it is not necessary to refer to the semantic knowledge body, and efficient processing is possible. However, when the expression “in Osaka” corresponds to the expression “in Osaka” as in Example 3, “Osaka” is a substance, and therefore the semantic knowledge is directly referenced.

[0015]

According to the present invention, the context analysis can be performed by the syntactic context analysis means and the semantic context analysis means for an input sentence including an expression in which syntactic knowledge and semantic knowledge are combined.
In the syntactic context analysis, it is possible to analyze the context when the expression in the context and the expression of the input sentence are syntactically similar by using the semantic knowledge through the knowledge prefetching means together with the syntactic structure information. Moreover, efficient context analysis is possible by extracting a part of the semantic knowledge as partial semantic knowledge and utilizing it through the knowledge prefetching means. Furthermore, when the syntactic structures are quite different, it is possible to complement the appropriate expression by performing the semantic context analysis based on the result of the semantic analysis.

[Brief description of drawings]

FIG. 1 is a functional block diagram showing the method of the present invention.

FIG. 2 is a block diagram showing the configuration of the syntactic context analysis means in FIG.

FIG. 3 is a diagram showing a context analysis algorithm in the syntactic context analysis means in FIGS. 1 and 2.

FIG. 4 is a diagram showing a knowledge acquisition algorithm in the partial semantic knowledge acquisition means in FIG.

5 is a diagram showing a determination algorithm in a portion for determining whether or not the partial semantic knowledge selection criterion in FIG. 4 is satisfied.

FIG. 6 is a diagram showing an example sentence of dialogue in a natural language interface system.

7 is a diagram showing analysis intermediate data and analysis results for the example sentence in FIG.

FIG. 8 is a diagram showing a typical example of a part of the semantic knowledge in the semantic knowledge storage means in FIG.

FIG. 9 is a diagram showing a typical example of a part of the semantic knowledge in the semantic knowledge storage unit in FIG.

10 is a diagram showing an example of partial semantic knowledge with respect to the example of semantic knowledge in FIG.

11 is a diagram showing an example of partial semantic knowledge with respect to the example of semantic knowledge in FIG.

[Explanation of symbols]

 1 morpheme / syntactic analysis means 2 syntactic context analysis means 3 semantic analysis means 4 semantic context analysis means 5 knowledge look-ahead means 6 semantic knowledge storage means 7 partial semantic knowledge storage means 8 partial semantic knowledge acquisition means 11 context analysis engine 12 syntactic context grammar rules 13 Focus stack

Claims (2)

(57) [Claims] 1. A morpheme / syntactic analysis unit for creating a syntactic structure by performing morphological analysis and syntactic analysis of a sentence described in natural language; and omitting the syntactic structure as an input using semantic knowledge and context knowledge. and syntax contextual analysis means for outputting a complement perform complementary syntax structure, and semantic analysis means for extracting a semantic structure representing the meaning of the sentence from the complementary syntactic structure and the semantic knowledge, and input the semantic structure, the semantic knowledge And a context knowledge, the meaning context analysis means for creating a complementary meaning structure by establishing and complementing the meaning structure, and the knowledge look-ahead means referenced by the syntactic context analysis means for using the meaning knowledge And a semantic knowledge storing means that stores the semantic knowledge and is referred to by the semantic analyzing means, the semantic context analyzing means, and the knowledge prefetching means. . 2. Of the semantic knowledge, such as attributes or events
And a partial semantic knowledge storage unit for storing the knowledge acquired by the partial semantic knowledge acquisition unit and making it available to the knowledge prefetching unit. The context analysis method according to claim 1, wherein: