JP3369584B2 - Japanese parser - Google Patents

Description

Description: BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a phrase structure and
Japanese parser that mechanically analyzes the receiving structure
Especially in Japanese sentences containing adnominal modifier clauses.
The present invention relates to a Japanese parser capable of judging the case of Japanese. 2. Description of the Related Art Conventionally, when syntactic analysis of a Japanese sentence is performed mechanically, it is necessary to control grammatical information such as part of speech of a word and the possibility of involvement from other words. The phrase structure and the dependency structure are extracted based on the semantic constraint information and the like. Also, when the Japanese sentence includes the adnominal modification clause, that is, an embedded sentence for performing adnominal modification, the type of the adnominal modification clause is determined to determine the clause structure. [0003] By the way, the kind of the adnominal modifier clause is determined by whether or not the word modified by the adnominal modifier clause (hereinafter, this word is referred to as an antecedent) can have some case relation with the predicate in the clause.
Alternatively, the determination is made based on whether or not the antecedent has a content attribute and can receive an equal clause such as “to”. [0004] However, if the antecedent is a word such as "koto", "mono" or "no" that cannot be specified by itself, or if the polysemy has a large width and is related only to the predicate in the clause, In the case of a word whose meaning cannot be specified, the type of the adnominal modified clause cannot be determined by the above method, and the clause structure cannot be determined. The present invention provides a Japanese parsing apparatus capable of accurately determining the type of a continuation modifier in a Japanese sentence including a continuation modifier whose preceding word is an indefinite word as described above. The purpose is to: [0006] In order to solve the above-mentioned problems, in the present invention, a Japanese compound sentence including a continuity modifier subjected to morphological analysis is input, and the syntax analysis of the Japanese compound sentence is performed. In a Japanese syntax analyzer having a syntax analyzer and a part-of-speech matrix table storing information on dependency between words, the syntax analyzer converts grammatical information and semantic information of each word added by morphological analysis. Local parsing that refers to, determines the dependency relationship between words by referring to the analysis rules regarding the dependency between words stored in the part-of-speech matrix table, and creates a parse tree representing the dependency relationship between words and part analyzes the parse tree created by the local parser to determine the type of adnominal clauses contained in the Japanese complex sentences in accordance with predetermined analysis rule, the parse tree A global parser to be modified, wherein the global parser, when the word associated with the adnominal modifier is an indeterminate word, the indeterminate word and the indeterminate word are case particles. Is determined by referring to the case slot of the above-described descriptive word, and the semantic constraint information defined in the case relationship with the above-described descriptive word and stored in the above-described case slot is converted to the above-described meaning. Determining the type of the adnominal modifier clause by determining the case relation between the indefinite word and the predicate in the adnominal modifier clause based on the semantic constraint information added to the indefinite word. It is characterized by. An embodiment of the present invention will be described below with reference to the drawings. FIG. 5 is a block diagram showing a configuration of the Japanese syntax analyzer according to one embodiment of the present invention. The analytic processing apparatus 1 of this embodiment includes a morphological analysis unit 2 and a Japanese dictionary unit 4 connected to the morphological analysis unit 2.
And a syntax analysis processing device 60 for inputting a morphologically analyzed Japanese sentence and analyzing the dependency structure of the Japanese sentence, and a part-of-speech matrix table 5 storing information on dependency between words. As shown in FIG.
Is a word string that has been subjected to a morphological analysis process of dividing the Japanese sentence input by the morphological analysis unit 2 for each word and adding grammatical information and semantic information to each word with reference to the Japanese dictionary 4. Here, in the Japanese dictionary 4, grammatical information such as the part of speech, semantic feature, and case of each word and semantic information are registered. The syntactic analysis processing device 60 uses the grammatical information and semantic information of each word obtained by the morphological analysis to determine the dependency relationship between words in accordance with the analysis rules stored in the part-of-speech matrix table 5, and determines the interdependency between words. A local parsing unit 61 for creating a parse tree representing a dependency relationship, and a parse tree representing a dependency relationship created by the local parse unit 61 are analyzed, and the parse tree is corrected according to a predetermined analysis rule. And a global parsing unit 62. The global parsing unit 62 determines the type of the adnominal modifier clause when the Japanese sentence to be analyzed includes the adnominal modifier clause. In particular, when the word in the adnominal modifier is an indeterminate word, the case relation between the indeterminate word and the subsequent predicate to which the indeterminate word relates is determined, and Adding semantic constraint information defined in the case relation to the indeterminate word, and determining the case relation between the indeterminate word and the predicate in the adnominal modifier clause based on the added semantic constraint information To determine the type of the adnominal modifier clause. The part-of-speech matrix table 5 stores analysis rules corresponding to combinations of parts of speech of words related to the word, that is, rules corresponding to a combination of two words. In this embodiment, in consideration of the simplicity of the processing algorithm, etc., the parsing analysis will be described as performing dependency analysis not in terms of phrases but in the form of analyzing semantic dependency relationships between words. However, the following description can be similarly applied to syntactic analysis based on dependency analysis between phrases and phrase structure analysis. Next, the operation of this embodiment will be described with reference to FIGS. FIG. 1 shows an analysis process in a case where a sentence "bought and sold yesterday" is input to the analysis processing device 1 of the present embodiment. FIG. 1 shows a morphological analyzer 2
5 shows a process performed by the syntax analysis processing device 60 when a divided word string such as “yesterday / bought / no / a / sold” is input. The local parser 61 of the parser 60 sequentially determines the dependency relation for each of the above words. Since the predicate "buy" modifies "no" in " Detects that yesterday and "bought" form adnominal modifiers. Next, the large local sentence analyzer 62 tries to determine the type of the adnominal modifier clause, but cannot determine any case because the meaning of "no" is indefinite.
Therefore, it is assumed that this is a nominative relation clause, and the process proceeds to the next processing. Here, C1 in the figure indicates a noun modification clause. Also,
[PN] indicates a virtually placed pronoun, and (sbj) indicates that [PN] is the nominative of the predicate "buy". As a result, C1 indicates the nominative relation clause. Next, the local parsing unit 61 determines the dependency relationship between "no" and "wo", and then determines the dependency relationship between "wo" and "sold".
Here, a dependency relationship is established in both cases. At this time, the global parsing unit 62 refers to the case slot 41 of “sold” shown in FIG. 2B to determine what case “no” is related to “sold”. Here, it is recognized as a target case. Then, the semantic feature [product] defined in the target case of “sold” is added to “no”. Thus, it is determined that "no" indicates a product. Thereafter, the global parser 62
Adds the condition that "no" indicates a product, and again determines the type of case in the adnominal modifier clause C1. As shown in FIG. 2 (a), "no" indicating a product for "bought" is regarded as a target case, so that the adnominal modifier clause C1 is determined to be a target case relation clause. Corrected to case relation clause. Here, (PNJ) is the predicate "bought" in (dobj) in the figure.
Indicates that the subject is eligible. Thus, the analysis processing in the present example sentence ends. Next, with reference to FIG. 3, an explanation will be given on the analysis processing in the case where the sentence "I regret that I bought yesterday" is input to this embodiment. In this analysis example, the processing up to assuming the adnominal modification clause C2 as the nominative relation clause is the same as the analysis example in FIG. After the local parsing unit 61 determines the dependency relationship between "no" and "wo" whose meaning is indefinite, and then determines the dependency relationship between "wo" and "regret", The global parsing unit 62 refers to the case slot 41 of “regret” shown in FIG. 4 to determine what case “no” is related to “regret”. Here, it is recognized as a target case. Then, the semantic [action content] defined in the object of “regret” is added to “no”. Thus, it is determined that "no" indicates the content of the act. Thereafter, the global parser 62
Adds the condition that "no" indicates the content of the action, and determines again the type of the case of the adnominal modifier clause C2. However, FIG.
As shown in (a), "buy" has no case to be associated with the content of the act. Therefore, "no" indicating the content of the act is not related to "bought" at all. Thus, the adnominal modifier clause C2 is corrected to a content clause. Here, the content clause is a series of sentences indicating the content of the word to be modified,
It can be expressed in the form of "... (modifier)". In FIG. 3, [TH-PN] indicates a virtually placed pronoun, and the word "TH-PN" and "(") "immediately before (TH) indicate that C2 is a content clause. This is because, as described above, the sentence forming the content clause can be expressed in the form of “to (modifier)”. Note that when a phrase “to” connecting the sentence and the antecedent is explicitly expressed in the content clause, the expression [TH-P
[N] need not be added before [N]. Thus, the analysis processing in the present example sentence is completed. Here, comparing the example sentences shown in FIG. 1 and FIG. 3, when the meanings of “no” are indefinite, the verbs “sold” and “regret” are given even though the sentences have the same form. Different types of adnominal modifier clauses are different. In such a case, according to the present embodiment, the meaning of "no" can be estimated from the type of the verb by the above-described processing, and the type of the adnominal modifier can be accurately determined. As described above, the Japanese syntax analyzer of the present invention comprises a local syntax analyzer for determining the dependency relationship between words in a Japanese sentence; It has a global parser that analyzes the judgment result, judges the type of the adnominal modifier clause, and corrects the above judgment result as necessary, so that the application range of the means for judging the dependency structure of the sentence is expanded and the performance is improved. There is an effect that can be achieved. Also, the global parser determines the type of the adnominal modifier by estimating the meaning of the antecedent based on the relationship between the antecedent and the predicate behind the antecedent.
Even if the antecedent is an indefinite word such as "koto", "mono" or "no", the effect is that the type of the adnominal modifier clause relating to the antecedent can be accurately determined.

BRIEF DESCRIPTION OF THE DRAWINGS FIG. 1 is a diagram showing an example of analysis processing according to the embodiment of FIG. FIG. 2 is a diagram showing a case slot of the predicate of FIG. 1; FIG. 3 is a diagram showing another example of analysis processing according to the embodiment of FIG. 5; FIG. 4 is a diagram showing a case slot of the predicate of FIG. 3; FIG. 5 is a block diagram showing a Japanese syntax analyzer according to one embodiment of the present invention. [Description of Signs] 1: Analysis processing device 60: syntax analysis processing device 61: local syntax analysis unit 62: global syntax analysis unit C1: adnominal modifier clause C2: adnominal modifier clause

Continuation of the front page (58) Field surveyed (Int.Cl. ⁷ , DB name) G06F 17/21-17/28 JICST file (JOIS)

Claims (1)

(57) [Claims] [Claim 1] A parsing unit for inputting a Japanese compound sentence including a continuity modifier subjected to morphological analysis and performing syntax analysis of the Japanese compound sentence, and a relation between words. A Japanese part-of-speech matrix table that stores information related to reception, wherein the parsing unit refers to grammatical information and semantic information of each word added by morphological analysis, and stores the grammatical information and semantic information in the part-of-speech matrix table A local parsing unit for determining a dependency relationship between words by referring to a parsing rule regarding the dependency between words, and creating a parse tree representing the dependency relationship between words; analyzing the parse tree created by, it determines the type of adnominal clauses contained in the Japanese complex sentences in accordance with predetermined analysis rule, and a global parser to modify the parse tree, the large When the word in the adnominal modifier is an indeterminate word, the regional parser analyzes a case between the indeterminate word and the subsequent predicate in which the indeterminate word is connected via a case particle. The relationship is determined by referring to the case slot of the above-described descriptive word, and the semantic constraint information defined in the case relation with the above-described descriptive word and stored in the above-mentioned case slot is added to the above-mentioned indeterminate word, and the added above-mentioned meaning is added. A Japanese parsing apparatus, characterized in that a case relation between the indefinite word and a predicate in the adnominal modifier clause is determined based on constraint information to determine a type of the adnominal modifier clause.