JP2005284723A - Natural language processing system, natural language processing method, and computer program - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To obtain an accurate syntax semantic analysis result on the whole original sentence by accurately connecting the syntax analysis result of each element after dividing a long sentence to make a syntax semantic analysis. <P>SOLUTION: Divided positions in an input sentence are determined based on morphemic information of one morpheme or two or more continuous morphemes included in the input sentence, and a syntax semantic analysis is made on each element of the input sentence divided in the divided positions, to acquire the modification relation between words and the kind of the relation for every element. Based on morphemic information in the divided positions of the input sentence, the connected positions and modification relation in the syntax semantic analysis result of each divided element are determined to connect the syntax semantic analysis results of the respective elements to one another. <P>COPYRIGHT: (C)2006,JPO&NCIPI

Description

  The present invention relates to a natural language processing system and a natural language processing method for mathematically handling a natural language used by human beings for daily communication, and a computer program, and in particular, a natural language sentence is divided into words and affixes. A natural language processing system, natural language processing method, and computer language which perform syntactic and semantic analysis processing after obtaining morphological analysis such as recognition and part of speech recognition, and acquire the dependency relationship between words and the type of the dependency relationship; Regarding the program.

  More particularly, the present invention relates to a natural language processing system, a natural language processing method, and a computer program that divide long sentences to reduce the amount of computation required for syntactic and semantic analysis, and in particular, to syntax each element of a divided sentence. The present invention relates to a natural language processing system, a natural language processing method, and a computer program that obtain a syntactic and semantic analysis result of an original sentence by accurately connecting the syntax analysis results of each element after semantic analysis.

  Words that humans use for everyday communication, such as Japanese and English, are called “natural languages”. Many natural languages have a naturally occurring origin and have evolved with the history of mankind, people and society. Of course, people can communicate with each other by gestures and hand gestures, but natural language can realize the most natural and advanced communication.

  On the other hand, with the development of information technology, computers have become established in human society and have deeply penetrated into various industries and daily life. Now, not only computer data, but almost all information content such as images and sounds are handled on the computer, making it possible to perform advanced processing such as editing / processing, storage, management, transmission and sharing of information. .

  For example, a natural language written in various languages such as Japanese and English is inherently abstract and highly ambiguous, but can be processed computerically by handling sentences mathematically. . As a result, various applications / services related to natural language are realized by automated processing such as machine translation, dialogue system, search system, and question answering system.

  Natural language processing is generally divided into processing phases of morphological analysis, syntax analysis, semantic analysis, and context analysis. In morpheme analysis, a sentence is segmented into morphemes which are the smallest semantic units, and part-of-speech recognition processing is performed. In syntax analysis, sentence structure such as phrase structure is analyzed based on grammatical rules. Since the grammatical rule is a tree structure, the parsing result generally has a tree structure in which individual morphemes are joined based on a dependency relationship. In semantic analysis, a semantic structure that expresses the meaning conveyed by a sentence is obtained based on the meaning (concept) of the words in the sentence and the semantic relationship between words, and the semantic structure is synthesized. In context analysis, a sentence series (discourse) is regarded as a basic unit of analysis, and a discourse structure is constructed by obtaining a semantic group between sentences.

  A morpheme is an element that is the smallest grammatical unit, such as a word or affix, in linguistics. Therefore, in morphological analysis, in order to identify the grammatical attributes (part of speech, utilization, etc.) of the morpheme, the word is divided and part of speech is added. That is, morphological analysis is a basic process in kana-kanji conversion, information retrieval, machine translation, and the like in order to identify a word in an input sentence and analyze the change in the word form (for example, Non-Patent Document 1).

  The syntactic and semantic analysis process is a process that receives a natural language sentence and identifies the dependency relationship between words and the type of relationship (case frame or modification type of subject, object, etc.) based on grammatical rules. It is.

  With regard to syntax analysis that extracts only the former dependency relationship, it is possible to obtain an analysis result in real time using statistical processing or the like. Regarding syntax analysis, systems such as Cabocha (for example, see Non-Patent Document 2) and KNP (for example, see Non-Patent Document 3) are already known in the art.

  Here, when considering a practical application, the part-of-speech information for each morpheme and the dependency relationship between morphemes are not sufficient as analysis results, and semantic analysis corresponding to the processing of the next stage of syntax analysis is not sufficient. It is thought that it is important to obtain information such as practice and case frames.

  However, since this semantic analysis processing is very expensive in terms of computational complexity, it is difficult to obtain an analysis result in real time as compared to syntax analysis. Since the amount of calculation increases exponentially as the number of morphemes in a sentence increases, it is particularly a challenge to realize semantic analysis for long sentences.

  As a method for solving this problem, pre-processing is considered in which a long sentence is divided and the amount of calculation required for the analysis can be reduced before the sentence is input to the syntactic and semantic analysis system.

  For example, for a long Japanese sentence, search for a set of division candidate character strings prepared in advance, determine whether there is a division candidate character string in this sentence, divide it, perform morphological analysis on the divided sentence, The end of a sentence can be adjusted (for example, refer to patent documents 1).

  Furthermore, it is generally understood in the art that when a sentence is divided, it is preferable to set a position where a connection particle is present in the sentence as a division position. For example, there is a method of obtaining a connection particle or a linear combination rank after function from a corpus and using it for prioritizing a parsing floor (for example, see Non-Patent Document 4).

  In addition, the technique of dividing sentences to reduce the cost or calculation time required for language processing is adopted not only in syntax semantic analysis but also in machine translation and morphological analysis.

  For example, when the length of the original sentence exceeds the limit, the original text data is divided into word strings of a length that can be analyzed, and after partial translation, the translated text data of each word string is combined into one sentence. Can be output as a translation result (see, for example, Patent Document 2).

  Moreover, in the translation for long sentences, after translating each element of the original sentence divided by the preprocessing, a natural translation sentence can be obtained by inserting words such as connectives and relative pronouns between the elements ( For example, see Patent Document 3).

Japanese Examined Patent Publication No. 7-21803 Japanese Patent Laid-Open No. 10-21239 JP 7-182346 A Uchimoto Kiyotaka and Ma Ao, “Humanities and Information Processing” (Study Publishing, pp. 13-15, 1999) Taku Kudo and Yuji Matsumoto "Japanese dependency analysis by applying the chunking stage" (Journal of Information Processing Society of Japan, 43 (6), 1834-1842) Kurohashi Ikuo, "Do not do well, KNP" (Information Processing, 41 (11), 1215-1220) Co-authored by Natsuki Ichimaru and Hiroyuki Himatsu “Syntax Analysis of Compound Sentences Based on Joining Order of Connected Particles” (2003-NL-158)

  The computational complexity of language processing, including syntactic and semantic analysis, increases exponentially as the number of morphemes in a sentence increases. For this reason, a method is adopted in which long sentences are divided to reduce the amount of calculation required for syntactic and semantic analysis. However, in this case, there is still a problem that after the syntax / semantic analysis of each element of the divided sentence, the syntax / semantic analysis result of each element must be accurately connected to obtain the analysis result of the entire sentence.

  On the other hand, in each of the methods described above, the individual processing results are simply linked when the divided elements are combined again, but this cannot be applied to the target syntactic and semantic analysis system. It is. This is because it is not sufficient to relate individual processing results in the syntactic and semantic analysis system, and it is necessary to specify the type of the relationship. For example, Patent Document 3 adopts a method of inserting connection words between them. However, since the method of linking the analysis results may differ depending on the part of the division and the result of syntactic and semantic analysis of each element after the division, a correct result cannot be obtained simply by inserting words.

  The present invention has been made in view of the technical problems as described above, and its main purpose is to perform a syntactic and semantic analysis process on the input sentence, and to determine the dependency relationship between words and their dependency relationship. An object of the present invention is to provide an excellent natural language processing system, natural language processing method, and computer program capable of obtaining an accurate analysis result regarding a type.

  A further object of the present invention is to provide an excellent natural language processing system, natural language processing method, and computer program capable of reducing the amount of calculation required for syntactic and semantic analysis by dividing long sentences.

  A further object of the present invention is to obtain an accurate syntactic and semantic analysis result for the entire original sentence by correctly connecting the parsing results of each element after syntactic and semantic analysis of each element of the divided sentence. Another object of the present invention is to provide a natural language processing system, a natural language processing method, and a computer program.

  The present invention has been made in consideration of the above problems, and a first aspect of the present invention is a natural language processing system that analyzes an input natural language sentence, and includes one morpheme or 2 included in the input sentence. Based on the morpheme information of the above continuous morphemes, the division position in the input sentence is determined, the sentence division processing means for dividing the input sentence into the previous element and the rear element of the division position, and divided by the division position Syntactic and semantic analysis means that performs syntactic and semantic analysis on each element of the input sentence and obtains the dependency relationship between words and words and the type of the relationship for each element, and the relationship between words and words in each divided element Based on the receiving relationship, the type of the relationship, and the morpheme information at the split position of the input sentence, the syntactic analysis result of the element is determined by determining the connecting position and the dependency relationship in the syntactic and semantic analysis result of each divided element. A natural language processing system characterized by comprising the text elements connecting processing means for coupling the Judges.

  The computational complexity of language processing, including syntactic and semantic analysis, increases exponentially as the number of morphemes in a sentence increases. For this reason, a method is adopted in which long sentences are divided to reduce the amount of calculation required for syntactic and semantic analysis. In this case, a syntactic and semantic analysis result for each element obtained by division is obtained. The result of syntactic and semantic analysis of each element is expressed as a tree structure joined based on the dependency relationship between words, for example, and the type of dependency relationship is added to the branch connecting words.

  However, in this case, after syntactic and semantic analysis of each element of the divided sentence, there is a problem that the syntactic and semantic analysis result of each element must be accurately connected to obtain an analysis result as the entire sentence. In other words, it is unclear which words in the syntactic and semantic analysis tree obtained for each element should be joined with what kind of dependency relationship, and each semantic element of the sentence once divided is syntactically analyzed. It is difficult to connect correctly later.

  On the other hand, according to the present invention, based on the morpheme information at the divided position of the input sentence, the position to be linked and the dependency relationship in the syntactic and semantic analysis result of each divided element are determined, and the elements are linked. Therefore, the syntactic and semantic analysis of each element of the divided sentence is performed, and then the syntactic analysis result of each element is accurately connected to obtain an accurate syntactic and semantic analysis result for the entire original sentence.

  Here, the division position determined based on the morpheme information of one morpheme or two or more consecutive morphemes included in the sentence, and the syntax meaning of each element determined based on the morpheme information at the division position of the input sentence There is a certain correspondence between the connection position and the dependency relationship in the analysis result.

  Therefore, the morpheme information of one morpheme or two or more consecutive morphemes at the sentence division position is described as a division processing rule, and the syntactic analysis results of each divided element corresponding to this division processing rule are The position for connection and the dependency relationship may be described as a connection processing rule, and a plurality of division processing rules and a connection processing rule corresponding to each of them may be held in advance. In such a case, the sentence element concatenation processing means uses the concatenation processing rule corresponding to the division processing rule used for the division of the input sentence to obtain the semantic analysis result for each element obtained from the syntax semantic analysis means. Can be linked.

  Therefore, according to the present invention, it is possible to analyze a long sentence that has not been possible to perform syntactic and semantic analysis. By holding the sentence split processing rules and the corresponding link processing rules in association with each other, it is possible to connect the appropriate partial analysis results according to the way of splitting. Despite the fact that the sentence is divided into a plurality of elements, it is finally possible to correctly output the semantic analysis result for the entire sentence.

  Further, when a plurality of division processing rules for determining the division position of the input sentence are prepared, priority may be assigned to each division processing rule. Then, the division processing rules are selectively applied according to the priority order, and the input sentence is divided at the obtained division position, thereby determining a more appropriate division position and efficiently obtaining a more accurate syntactic and semantic analysis result. Can be obtained.

  FIG. 1 schematically shows an example of a natural language processing system according to the present invention. The illustrated natural language processing apparatus 10 includes a sentence input unit 11, a division / connection rule holding unit 12, a sentence division processing unit 13, a syntax and semantic analysis unit 14, a sentence element connection unit 15, and an analysis result output unit 16. Consists of.

  The sentence input unit 11 receives a natural language sentence described in Japanese or other languages, and performs a morpheme analysis process that divides the sentence into words, that is, morphemes, performs part-of-speech assignment, and identifies a utilization form. Perform word breaks and parts of speech.

  The division / connection rule holding unit 12 holds a division rule composed of a condition of a division position and a priority order to be applied, and a connection rule composed of a connection position and a relation name associated with each division rule.

  The sentence division processing unit 13 inquires the division / concatenation rule holding unit 12, sequentially applies the division rules to the input sentence in order of priority, determines the division position using the matching division rule, and divides the input sentence Split into elements before and after the position.

  The syntactic and semantic analysis unit 14 performs a process of extracting a dependency relationship and specifying a relationship type such as a case frame for each element of the divided sentence. By this syntactic and semantic analysis processing, the dependency relationship between words and the analysis result describing the relationship are output for each element of the sentence.

  The sentence element connection unit 15 inquires of the division / connection rule holding unit 12 and combines the analysis results of the sentence elements based on the connection rule corresponding to the division rule applied by the sentence division processing unit 13. Then, the analysis result output unit 16 outputs the combined analysis result as the analysis result of the entire original sentence.

  In addition, the natural language processing system according to the present invention may further include an analysis result evaluation unit that evaluates the processing result of the syntax semantic analysis for each element obtained by dividing the input sentence.

  The analysis result evaluation means can determine whether or not to further divide the element in response to the fact that the syntax and semantic analysis means cannot execute the analysis on each element of the input sentence within a predetermined time. In such a case, the sentence division processing unit divides the element again according to the determination by the analysis result evaluation unit, and the syntax-semantic analysis unit further performs syntax-semantic analysis for each element obtained by dividing the element again. By performing the processing, it is possible to perform more accurate syntactic and semantic analysis even if the input sentence is a long sentence.

  Further, the analysis result evaluation means can evaluate a syntax-semantic analysis result for each element obtained by dividing the input sentence according to a certain division processing rule, and determine whether the input sentence should be divided again. . In such a case, the sentence division processing unit applies another division processing rule to the input sentence again according to the determination by the analysis result evaluation unit, and further the syntax-separation analysis unit again processes the input sentence. By performing the syntactic and semantic analysis processing on each element obtained by dividing, it is possible to perform more accurate syntactic and semantic analysis even if the input sentence is a long sentence.

  The sentence division processing means may select a division processing rule based on a machine learning method and divide the input sentence.

  For example, when the sentence division processing unit divides a sentence at a division position where a punctuation point is connected to a connection particle A class, the connection processing unit converts an element before the division position to an element after the division position. By connecting the syntactic and semantic analysis results of each element as a continuous modification component for the verb, it is possible to obtain the syntactic and semantic analysis results for the entire original sentence.

  In addition, when the sentence division processing unit divides the sentence at the division position where the punctuation marks are connected to the connection particle B class, the connection processing unit inserts a synonym for bundling each element, and before the division position, The syntactic and semantic analysis results of the entire original sentence can be obtained by linking the syntactic and semantic analysis results of the element and the subsequent element by a juxtaposition relationship via a pronoun.

  In addition, when the sentence division processing unit divides the sentence at the division position where the reading marks are connected to the continuous form after use, the connection processing unit inserts a synonym for bundling each element and precedes the division position. The syntactic and semantic analysis results of the entire original sentence can be obtained by linking the syntactic and semantic analysis results of the element and the subsequent element by a juxtaposition relationship via a pronoun.

  In addition, when the sentence division processing unit divides the sentence at the division position where the reading mark is connected to the case particle, the connection processing unit determines from the division position that the verb of the element after the division position is a citation verb. Concatenate the results of syntactic and semantic analysis of each element using the previous element as the citation of the verb, otherwise connect the results of syntactic and semantic analysis of each element using the element preceding the split position as the continuous modification component Thus, the syntactic and semantic analysis result can be obtained for the entire original sentence.

  Further, when the sentence division processing unit divides the sentence at a division position where the noun, the counsel and the reading mark are continuous, the connection processing unit has a case element omitted in the verb of the element after the division position. If there is an element before the division position, the syntactic analysis results of each element are connected to each other with the corresponding case as the case. Can be used to obtain a syntactic and semantic analysis result for the entire original sentence.

  Further, when the sentence division processing unit divides the sentence at the division position where the punctuation marks are connected to the adverb nouns, the connection processing unit applies the element before the division position to the verb of the element after the division position. By connecting the syntactic and semantic analysis results of each element as the continuous modification component, it is possible to obtain the syntactic and semantic analysis results for the entire original sentence.

  In addition, when the sentence division processing unit divides a sentence with a reading point as a division position, the connection processing unit uses each element as a continuous modification component for the verb of the element after the division position. By connecting the syntactic and semantic analysis results of, the syntactic and semantic analysis results can be obtained for the entire original sentence.

  In addition, when the sentence division processing unit divides the sentence using the connection particle A class as a division position, the connection processing unit uses an element that modifies the element preceding the division position as the component modification verb for the element after the division position. By synthesizing the syntax and semantic analysis results of each element, it is possible to obtain the syntax and semantic analysis results for the entire original sentence.

  Further, when the sentence division processing unit divides the sentence using the connection particle B class as the division position, the connection processing unit inserts a pronoun for binding the elements, and the elements before and after the division position By synthesizing each syntactic and semantic analysis result of the elements in a side-by-side relationship via a synonym, a syntactic and semantic analysis result can be obtained for the entire original sentence.

  Further, when the sentence division processing means divides the sentence with the continuous form after use as a division position, the connection processing means inserts a pronoun for binding each element, and the elements before and after the division position. By synthesizing each syntactic and semantic analysis result of the elements in a side-by-side relationship via a synonym, a syntactic and semantic analysis result can be obtained for the entire original sentence.

  In addition, when the sentence division processing unit divides the sentence using the case particle as a division position, the connection processing unit selects an element before the division position if the verb of the element after the division position is a citation verb. By linking the syntactic analysis results of each element as the citation of the verb, or by linking the syntactic analysis results of each element with the element before the split position as the continuous modification component, Syntactic and semantic analysis results can be obtained for the entire sentence.

  In addition, when the sentence division processing unit divides the sentence with the auxiliary particle as a division position, the connection processing unit determines that if there is an abbreviated case element in the verb of the element after the division position, By concatenating the syntactic and semantic analysis results of each element with the previous element as the corresponding case, or by concatenating the syntactic and semantic analysis results of each element with the element preceding the split position as the continuous modifier component, Syntactic and semantic analysis results can be obtained for the entire original sentence.

  Further, when the sentence division processing unit divides the sentence using the adverb noun as a division position, the connection processing unit uses the element before the division position as a continuous modification component for the verb of the element after the division position. By connecting the syntax and semantic analysis results of each element, it is possible to obtain the syntax and semantic analysis results for the entire original sentence.

  According to a second aspect of the present invention, there is provided a computer program written in a computer-readable format so that a process for analyzing an input natural language sentence is executed on a computer system. Based on the morpheme information possessed, a division position in the input sentence is determined, a sentence division processing step for dividing the input sentence into a preceding element and a subsequent element of the division position, and each element of the input sentence divided by the division position Each of the divided elements based on the morphological information at the division position of the input sentence, and the syntactic and semantic analysis step for obtaining the element-to-element dependency relation and the type of the relation for each element. A sentence element linking process step for linking element syntactic analysis results by determining positions to be connected and dependency relations in the syntax semantic analysis results. It is a computer program that.

  The computer program according to the second aspect of the present invention defines a computer program described in a computer-readable format so as to realize predetermined processing on a computer system. In other words, by installing the computer program according to the second aspect of the present invention in the computer system, a cooperative action is exhibited on the computer system, and the natural language according to the first aspect of the present invention. The same effects as the processing system can be obtained.

  According to the present invention, it is possible to perform syntactic and semantic analysis processing on an input sentence, and to obtain an accurate analysis result regarding a dependency relationship between words and a type of the dependency relationship. A system, a natural language processing method, and a computer program can be provided.

  Furthermore, according to the present invention, it is possible to provide an excellent natural language processing system, natural language processing method, and computer program that can reduce the amount of calculation required for syntactic and semantic analysis by dividing a long sentence.

  In addition, according to the present invention, after syntactic and semantic analysis of each element of the divided sentence, it is possible to obtain an accurate syntactic and semantic analysis result of the original sentence by accurately connecting the syntax analysis results of each element. A natural language processing system, a natural language processing method, and a computer program can be provided.

  According to the present invention, after dividing a long sentence that is difficult to analyze by using a sentence-separation rule and a concatenation rule associated with each other and a syntactic and semantic analysis result for each element obtained by dividing the sentence, By performing the syntactic and semantic analysis processing on and connecting the solution results, it is possible to obtain a more accurate analysis result for the entire original sentence.

  That is, according to the present invention, it is possible to analyze a long sentence that has not been possible to perform syntactic and semantic analysis. By having a sentence splitting rule and a corresponding connection rule in pairs, it is possible to link the appropriate partial analysis results according to the way of splitting. It is possible to output the analysis result correctly.

  Other objects, features, and advantages of the present invention will become apparent from more detailed description based on embodiments of the present invention described later and the accompanying drawings.

  Hereinafter, embodiments of the present invention will be described in detail with reference to the drawings.

  FIG. 2 schematically shows a functional configuration of the natural language processing apparatus 100 according to an embodiment of the present invention. The illustrated natural language processing apparatus 100 includes a sentence input unit 101, a division / connection rule holding unit 102, a sentence division processing unit 103, a syntax and semantic analysis unit 104, an analysis result evaluation unit 107, and a sentence element connection unit 105. And an analysis result output unit 106.

  The sentence input unit 101 receives a natural language sentence described in Japanese or other languages, divides the sentence into words, that is, morphemes, and performs morpheme analysis processing for identifying part-of-speech and using forms, thereby separating words. And part of speech.

  The division / connection rule holding unit 102 holds a division rule composed of a condition of a division position and a priority order to be applied, and a connection rule composed of a connection position and a relation name associated with each division rule. The division rule is described as morpheme information of one morpheme or two or more continuous morphemes for determining a division position in the input sentence. The connection rule describes the position and dependency relationship for connecting the syntax and semantic analysis results of each divided element. However, details of the division rule and the connection rule will be described later.

  The sentence division processing unit 103 inquires the division / concatenation rule holding unit 102, sequentially applies the division rules in order of priority, determines the division position using the matching division rules, and determines the input sentence before and after the division position. Split into elements.

  The syntactic and semantic analysis unit 104 performs processing for extracting dependency relationships and specifying types of relationships such as case frames for each element of the divided sentence. By this syntactic and semantic analysis processing, the dependency relationship between words and the analysis result describing the relationship are output for each element of the sentence.

  The sentence element linking unit 105 makes an inquiry to the division / connection rule holding unit 102 and combines the analysis results of the sentence elements based on the linking rule corresponding to the division rule applied by the sentence division processing unit 103. Then, the analysis result output unit 106 outputs the combined analysis result as the analysis result of the entire original sentence.

  The analysis result evaluation unit 107 receives the processing result from the syntax and semantic analysis unit 104 as input, and evaluates the processing result of the syntax and semantic analysis for each element obtained by dividing the input sentence. If it is determined that a normal analysis result has not been obtained from the syntax and semantic analysis unit 104, an instruction is output to the sentence division processing unit 102, the input sentence is recursively divided, and a more appropriate division position is obtained. To improve the accuracy of input sentence division and syntactic analysis.

  For example, the analysis result evaluation unit 107 determines that the syntax / semantic analysis unit 105 cannot execute analysis on each element of the input sentence within a predetermined time, that is, in response to occurrence of a timeout error, Decide whether to split further. When it is determined that further division is required, the sentence division processing unit 102 recursively performs division processing on the elements, and the syntax semantic analysis unit 105 performs syntax semantic analysis processing on each element obtained by dividing the element again. Apply.

  Further, the analysis result evaluation unit 107 evaluates the syntactic and semantic analysis results for each element obtained by dividing the input sentence according to a certain division processing rule, and determines whether or not the input sentence should be divided again. When it is determined that the input sentence should be re-divided, the sentence division processing unit 102 applies another division processing rule to the input sentence and divides it again, and the syntax-semantic analysis unit 105 determines the input sentence. A syntactic and semantic analysis process is performed again for each element obtained by division again.

  The details of the recursive division processing of the input sentence by evaluating the analysis result will be described later.

  The natural language processing according to the present invention can be implemented by being incorporated into a syntax and semantic analysis unit in the syntax and semantic analysis processing system. For example, Lexical Functional Grammar (LFG) can be cited as a typical example of grammar theory for performing syntactic / semantic analysis. In LFG, linguistic knowledge, that is, grammar of native speakers is configured as a component separated from computer processing and other non-grammatical processing parameters that affect the processing operation of the computer.

  FIG. 3 schematically shows a functional configuration of the syntactic and semantic analysis processing system 200 based on the LFG grammar theory.

  The morphological analysis unit 202 performs processing corresponding to the sentence input unit 101 in the natural language processing apparatus 100 shown in FIG. In the example shown in the figure, the morpheme analysis unit 202 has a morpheme rule 202A and a morpheme dictionary 202B relating to a specific language such as Japanese, and performs a part-of-speech recognition process by segmenting an input sentence into morphemes that are the smallest semantic units. As the morphological analysis system, for example, a Japanese morphological analysis system such as “Chasen” can be applied, but the gist of the present invention is not limited to this. For example, Yuji Matsumoto, Kei Kitauchi, Tatsuo Yamashita, Yoshitaka Hirano, Hiroshi Matsuda, Kazuma Takaoka, Masayuki Asahara, “Corporation version 2.2.1 Instruction Manual for Japanese Morphological Analysis System” (Nara Institute of Science and Technology, 2000).

  For example, when a sentence “My dog eats apple” is input by the morphological analysis processing, the {morphological analysis result} of {I {noun} {joint particle}} dog {noun} is {subject} The apple {noun} {case particle} eat {verb}. {Phrase}} is output.

  Such a morphological analysis result is then input to the syntax and semantic analysis unit 203. The sentence division processing unit 102, the syntax and semantic analysis unit 103, the analysis result evaluation unit 105, and the sentence element coupling unit 106 in the natural language processing apparatus 100 illustrated in FIG.

  In the example shown in FIG. 3, the syntax / semantic analysis unit 203 has dictionaries such as a grammar rule 203A and a case frame dictionary 203B, analyzes a phrase structure based on a grammar rule, and the meaning of words in a sentence and between words. Analyzes the semantic structure that expresses the meaning that the sentence conveys based on the semantic relations of words (the case frame dictionary describes the relationship between verbs and other components in the sentence, such as the subject, predicates and The semantic relationship of such words can be extracted). As a result of parsing, “c-structure (constituent structure)” representing a phrase structure of a sentence including words and morphemes as a tree structure, and an input sentence based on a case structure such as a subject and an object are questioned. “F-structure (functional structure)” is output as a result of semantic and functional analysis such as sentences, past tense, and polite sentences.

  FIG. 4 and FIG. 5 respectively show c-structure and f-structure obtained as a result of processing the input sentence “My dog eats apple” by the syntax and semantic analysis unit 203.

  c-structure represents the structure of words and phrases in a sentence in a tree structure format, and is defined by a syntax category. For example, phonological interpretation for generating a phoneme string can be performed based on c-structure. On the other hand, f-structure clearly expresses a grammatical function, and includes a grammatical function name, a semantic form, and a feature symbol. By referring to such f-structure, it is possible to obtain an understanding of the meaning of a subject, a subject, an complement, a modifier, and an adjunct. The f-structure is a set of features associated with each node of the c-structure, and is expressed in the form of an attribute-attribute value matrix as shown in FIG. 5, for example. That is, in the example shown in FIG. 5, the left side in [] is a feature (attribute) name, and the right side is a feature value (attribute value).

  Also, f-structure can be expressed as a tree structure that is joined based on the dependency relationship between words, that is, a “dependency tree”. A relationship type is added. 6 represents the f-structure shown in FIG. 5 in the form of a dependency tree. As illustrated, the dependency tree includes a node indicating a word, a link indicating dependency, and a link label indicating a relationship name.

  For details of LFG, see, for example, R.A. M.M. Kaplan and J.H. Bresnan co-author of the paper. "Lexical-Functional Grammar: A Formal System for Grammatical Representation" (The MIT Press, Cambridge (1982) Reprinted in Formal Issues in Lexical-Functional Grammar, pp.29-130.CSLI publications, Stanford University (1995 ).) Etc.

  Subsequently, an operation example of natural language processing in the natural language processing apparatus 100 according to the present embodiment will be specifically described. In the following description, Japanese will be described, but it should be understood that the same effect can be obtained in any language that can be applied to the syntax and semantic analysis processing.

  First, it is assumed that a natural language sentence “Enkakuji is very well-known as a building in the Kamakura period and widely known” is input to the sentence input unit 101. This sentence is converted into a morpheme string with part-of-speech information as shown in FIG. 7 by morpheme analysis processing.

  Next, the sentence division processing unit 102 sequentially applies the division rules in order of priority, determines the division position using the matching division rules, and divides the input sentence into elements before and after the division position. FIG. 8 shows an example of the division rule. The division rule is described as morpheme information of one morpheme or two or more continuous morphemes for determining a division position in the input sentence. FIG. 8 shows an example of the division rule.

  In the example shown in FIG. 8, as the division rule A with priority 1, the location where the reading mark is connected to the connection particle A class is determined as the division position. Further, as a division rule B with a priority of 2, a portion where a reading point is connected to the connection particle B class is determined as a division position. Further, as a division rule C of priority 3, a place where a reading point is connected to a continuous form after use is decided as a division position. Further, as a division rule D having a priority level 4, a position where a reading mark is connected to a case particle is determined as a division position. Further, as a division rule E of priority 5, a place where a noun, an auxiliary particle and a reading point continue is determined as a division position. Further, as a division rule F having a priority level 6, a position where a reading point is connected to an adverb noun can be determined as a division position. Further, as a division rule G with a priority of 7, a place where a reading point appears is decided as a division position. In addition, as a division rule A ′ having a priority of 8, a place where a connection particle A class appears is determined as a division position. In addition, as a division rule B ′ having a priority of 9, a location where a connection particle B class appears is determined as a division position. In addition, as a division rule C ′ having a priority of 10, a portion where a continuous form after use appears is determined as a division position. In addition, as a division rule D ′ having a priority of 11, a location where a case particle appears is determined as a division position. In addition, as a division rule E ′ having a priority of 12, a part where a coordinator appears is determined as a division position. In addition, as a division rule F ′ having a priority of 13, a location where an adverb noun appears appears as a division position.

  When the sentence division processing unit 102 inquires of the division / concatenation rule holding unit 103 for an input sentence converted into a morpheme string, in the input sentence, there is “so,” which is a connected particle A class + reading mark. In FIG. 8, rule A given priority 1 is applied to determine the division position. As a result, the above-mentioned input sentence is divided into two sentence elements, that is, the front element “Enkakuji is very famous as a building in the Kamakura period” and the rear element “widely known”. The

  Next, the syntax and semantic analysis unit 104 performs syntax and semantic analysis on each of the previous element and the subsequent element obtained by dividing the input sentence. FIG. 9 shows the syntax and semantic analysis results for each element before and after the division position in the form of a dependency tree. As illustrated, the dependency tree includes a node indicating a word, a link indicating dependency, and a link label indicating a relationship name.

  Next, the sentence element coupling unit 105 concatenates the syntax and semantic analysis results of the previous element and the subsequent element shown in FIG. In the present embodiment, the sentence element linking unit 105 determines whether each of the divided elements is based on the word-to-word dependency relationship in each divided element, the type of the relationship, and the morpheme information at the input sentence division position. In the syntactic and semantic analysis results, the connecting positions and dependency relationships are determined, and the syntactic and semantic analysis results of the elements are connected.

  The connection rule describes the position and dependency relationship for connecting the syntactic and semantic analysis results of each divided element. As described above, the connection rule is determined based on the morpheme information at the division position of the input sentence, which means that there is a correspondence between the division rule and the connection rule. Therefore, in this embodiment, the division / connection rule holding unit 103 holds a connection rule in association with the division rule. FIG. 10 shows an example of a connection rule.

  In the example shown in FIG. 10, as the connection rule A corresponding to the division rule A, the elements before the division position are connected as syntactic modification components for the verbs of the elements after the division position, and the syntactic analysis results of the respective elements are connected. Is negotiating. In addition, as a connection rule B corresponding to the division rule B, a dummy node (a synonym for bundling each element) is inserted, and each syntactic and semantic analysis result of the element before and after the division position is used as a synonym. It is negotiated to be connected by juxtaposition. Also, as a concatenation rule C corresponding to the division rule C, a synonym for bundling each element is inserted, and the syntactic and semantic analysis results of the element before and after the division position are placed side by side via the synonym. It is decided to connect by. In addition, as the connection rule D corresponding to the division rule D, if the verb of the element after the division position is a citation verb, the syntactic analysis result of each element is obtained with the element before the division position as the citation of the verb. Otherwise, it is agreed to connect the results of syntactic analysis of each element using the elements before the division position as the continuous modification component. In addition, as a connection rule E corresponding to the division rule E, if there is a case element omitted in the verb of the element after the division position, the result of the syntax and semantic analysis of each element with the element before the division position as the corresponding case Otherwise, it is decided to connect the results of syntactic and semantic analysis of each element by using the element before the division position as a continuous modification component. Further, as a connection rule F corresponding to the division rule F, it is decided to connect the results of syntactic and semantic analysis of each element using the element before the division position as a continuous modification component for the verb of the element after the division position. . In addition, as a connection rule G corresponding to the division rule G, it is decided that the elements before the division position are connected to each other as the consecutive modification component for the verb of the element after the division position, and the syntactic analysis results of each element are connected. . In addition, as a connection rule A ′ corresponding to the division rule A ′, it is decided to connect the results of syntactic analysis of each element using the element before the division position as a continuous modification component for the verb of the element after the division position. ing. Further, as a concatenation rule B ′ corresponding to the division rule B ′, a synonym for bundling each element is inserted, and each syntactic and semantic analysis result of the element before and after the division position is passed through the synonym. Arrange to connect by juxtaposition. Further, as a connection rule C ′ corresponding to the division rule C ′, a synonym for bundling each element is inserted, and the syntactic and semantic analysis results of the element before and after the division position are passed through the synonym. Arrange to connect by juxtaposition. In addition, as the connection rule D ′ corresponding to the division rule D ′, if the verb of the element after the division position is a citation verb, the result of syntactic analysis of each element with the element before the division position as the citation of the verb Otherwise, it is decided to connect the results of syntactic and semantic analysis of each element by using the element before the division position as a continuous modification component. Further, as a connection rule E ′ corresponding to the division rule E ′, if there is a case element omitted in the verb of the element after the division position, the syntax meaning of each element is assumed to be the case before the division position. It is decided that the analysis results are connected to each other, and otherwise, the syntax-semantic analysis results of the respective elements are connected to each other with the element preceding the division position as the continuous modification component. Also, as a connection rule F ′ corresponding to the division rule F ′, it is decided to connect the results of syntactic analysis of each element using the element before the division position as a continuous modification component for the verb of the element after the division position. ing.

  The sentence element linking unit 105 inquires of the division / connection rule holding unit 103 about a connection rule in order to connect the syntax and semantic analysis results of the previous element and the subsequent element. Since the input sentence taken up as an example here is divided into the above elements by the rule A, the connection rule A corresponding to the rule A is applied. Therefore, a link is established from the front element to the rear element, the link label ADJUNCT of the continuous modification component is given, and the analysis results of the two elements are connected. FIG. 11 shows the syntactic and semantic analysis result of the entire original sentence obtained as a result of concatenating the analysis results of the two elements obtained by this procedure.

  Finally, in the analysis result output unit 106, the analysis result shown in FIG. 11 is the analysis result of the original input sentence “Engonji is very well known as a building in the Kamakura period”. Is output.

  Next, the division rules and connection rules other than the above are used in the case of using the natural language sentence "The House of Representatives objected to the establishment of the Japan-US Investment Promotion Organization decided upon President Clinton's visit to Japan." An example of the operation of the syntactic and semantic analysis process when is applied will be described.

  First, in the sentence input unit 101, this sentence is input and converted into a morpheme string with part-of-speech information in the same manner as described above by morphological analysis processing.

  Next, the sentence division processing unit 102 divides the input sentence. That is, when the sentence division processing unit 102 inquires of the division / concatenation rule holding unit 103 about an input sentence converted into a morpheme string, the “establishment” is a noun + an auxiliary particle + reading mark in the input sentence. Therefore, rule E having priority 5 in FIG. 8 is applied. As a result, the input sentence is divided into two sentence elements, the former element “Establishment of the Japan-US Investment Promotion Organization decided when President Clinton visited Japan” and the latter element “The House of Representatives objected.” The

  Next, the syntax and semantic analysis unit 104 performs syntax and semantic analysis for each divided element. FIG. 12 shows the syntax and semantic analysis results for each element before and after the dividing position in the form of a dependency tree. As illustrated, the dependency tree includes a node indicating a word, a link indicating dependency, and a link label indicating a relationship name.

  Next, the sentence element connecting unit 105 connects the analysis results of the two elements before and after shown in FIG. That is, the sentence element connection unit 105 inquires of the division / connection rule holding unit 103 about a connection rule. Since these elements are obtained by dividing the input sentence by the division rule E, the connection rule E corresponding to the division rule E is applied. Therefore, since the case element OBJ omitted is present in the rear element, the front element is connected to the rear element as OBJ. FIG. 13 shows a syntactic and semantic analysis result for the entire original sentence obtained by concatenating the parsing results of two elements obtained by this procedure.

  Finally, in the analysis result output unit 106, FIG. 13 shows an analysis result of the original input sentence “The House of Representatives objected to the establishment of the Japan-US Investment Promotion Organization decided when President Clinton visited Japan”. The analysis result is output.

  The analysis result evaluation unit 107 evaluates the processing result of the syntax and semantic analysis, and depending on the evaluation, recursively performs the split processing on the input sentence, and improves the accuracy of the input sentence split and the syntax and semantic analysis processing at a more appropriate split position. Can be achieved (described above). Although it is not necessary to perform recursive division processing of the input sentence in the above-described two input sentence processing examples, a description will be given below of a processing procedure of syntactic and semantic analysis in a case involving recursive division processing of the input sentence.

  First, in the sentence input unit 101, a natural language sentence “This problem is closely related to the progress of the Chinese investigation, and it is said that the Japanese investigation authorities will discuss at the joint meeting on the 17th”. It is accepted as a sentence and converted into a morpheme string with part-of-speech information in the same manner as described above by morpheme analysis processing.

  Next, the sentence division processing unit 102 divides this input sentence. That is, when an inquiry is made to the split / concatenation rule holding unit 103 for an input sentence converted to a morpheme string, the auxiliary sentence combination form + punctuation mark “Ori” exists in the input sentence. A division rule C of degree 3 is applied. As a result, the above-mentioned input sentence has two sentence elements, namely the former element “This issue is closely related to the progress of the Chinese investigation,” and the latter element “the Japanese investigation authorities It is said that they will talk with each other. "

  Next, the syntax and semantic analysis unit 104 performs syntax and semantic analysis on each element obtained by division. Here, it is assumed that the syntactic and semantic analysis has failed due to a time-out error in each of the preceding and succeeding elements.

  Next, the analysis result evaluation unit 107 evaluates the syntax and semantic analysis results for each element, and determines the next process to be performed. FIG. 14 shows a processing procedure in the analysis result evaluation unit 107 in the form of a flowchart. Hereinafter, the flow of processing based on the evaluation of the syntactic and semantic analysis results for each element will be described according to this flowchart.

  First, it is determined whether or not the analysis result is successful (step S1). Here, since the analysis has failed, it is subsequently determined whether or not the cause of the failure is a timeout (step S2). Here, since the syntax and semantic analysis process has failed due to a timeout error, the next process is determined to be "splitting sentence element", and in response to this, the previous element and the subsequent element are divided again. (Step S4).

  In response to the request for division processing again, the sentence division processing unit 102 performs recursive division processing for the element. In other words, when the sentence converted to the morpheme sequence of the previous element “This problem is closely related to the progress of the Chinese investigation” is inquired of the division / connection rule holding unit 103, the sentence of the element contains Since there is a “problem is” which is a noun + ancillary particle + a reading mark, the division rule E of priority 5 in FIG. 8 is applied. As a result, the element is further divided into two elements, namely the former element “This problem is” and the latter element “which is closely related to the progress of the Chinese investigation”.

  Next, the syntax and semantic analysis unit 104 performs syntax and semantic analysis. Here, it is assumed that the processing of the two elements is successful. FIG. 15 shows these analysis results. In the analysis result evaluation unit 107, since each element has been successfully analyzed, no further processing is performed.

  Next, the sentence division processing unit 102 performs a recursive division process on the post-element of the input sentence “It is said that the Japanese investigation authorities will discuss at the joint meeting on the 17th”. In other words, when a sentence converted to a morpheme string is inquired to the division / concatenation rule holding unit, since there is a punctuation mark in the sentence, the rule G of priority 7 in FIG. 8 is applied. As a result, it is divided into two sentence elements, that is, the former element “The Japanese investigative authorities are on the 17th” and the latter element “It is said that they will discuss at the joint meeting.”

  Next, the syntactic and semantic analysis unit 104 performs syntactic and semantic analysis on each element that has been re-divided. Here, it is assumed that the analysis of the previous element has failed due to a syntax error.

  Next, the analysis result evaluation unit 107 evaluates the syntax and semantic analysis result for each element, and determines the processing to be performed in the next step. The flow of processing when a syntax error in syntax semantic analysis occurs will be described with reference to the flowchart shown in FIG.

  First, it is determined whether or not the syntax and semantic analysis result for each element in the syntax and semantic analysis unit 104 is successful (step S1). Here, since the analysis has failed, it is further determined whether or not the cause of the failure is a timeout (step S2). Here, since the failure is due to a syntax error instead of a timeout error, the next process is determined to be “re-divide by applying the rule of the current priority n + 1 to the original sentence (element)” (step S3). ). In other words, by performing the input sentence division process again using a different division rule from this time, a syntactic and semantic analysis process is tried for each element obtained at a different division position.

  In the example used here, since the priority of the current division rule is 7, the rule after priority 8 is said to be the original sentence element “The Japanese investigative authorities will discuss at the joint meeting on the 17th. . ”And the division process is performed again.

  The sentence division processing unit 102 performs subdivision on the sentence element “It is said that the Japanese investigation authorities will discuss at the joint meeting on the 17th”. That is, the division / connection rule holding unit 103 inquires of the division rule of priority 8 or later about the element converted into the morpheme string. Here, since the case particle “to” exists in the sentence, the rule D ′ having the priority 11 in FIG. 8 is applied. As a result, the element is further divided into two sentence elements, namely, the former element “The Japanese investigative authorities will discuss at the joint meeting on the 17th” and the latter element “It is said.”

  Next, the syntactic and semantic analysis unit 104 performs syntactic and semantic analysis on each of these elements recursively divided. Here, it is assumed that the processing of the two elements is successful. FIG. 16 shows the results of these syntax analysis. In the analysis result evaluation unit 107, since each element has been successfully analyzed, no further processing is performed.

  Through the above-described process, the division of the sentence “It is said that this issue is closely related to the progress of the Chinese investigation and that the Japanese investigative authorities will discuss at the joint meeting on the 17th.” . FIG. 17 shows a division rule that is applied when the input sentence is divided and the elements of the further divided sentence are further recursively divided.

  Next, the sentence element connecting unit 105 connects the analysis results of the respective elements. First, the two analysis results shown in FIG. 15 are connected. Since these elements are divided by the division rule E, the connection rule E in FIG. 10 corresponding to the division rule E is applied. Here, since the case element OBJ omitted is present in the rear element, the front element is connected to the rear element as OBJ. FIG. 18 shows the result obtained by concatenating the syntactic and semantic analysis results of the two elements obtained by the above-described procedure.

  Next, the two analysis results shown in FIG. 16 are connected. Since these elements are divided by the division rule D ′, the connection rule D ′ in FIG. 10 corresponding to the division rule D ′ is applied. In this case, since the reference element OBL_to omitted is present in the rear element, the front element is connected to the rear element as OBL_to. FIG. 19 shows a result obtained by concatenating the syntactic and semantic analysis results of two elements obtained by the above-described procedure.

  Further, the results of syntactic and semantic analysis shown in FIGS. 18 and 19 are further connected to obtain the result of syntactic and semantic analysis for the entire original input sentence. Since these elements are obtained by dividing the original input sentence by applying the division rule C, the connection rule C corresponding to the division rule C is applied. Therefore, a dummy node (a synonym for bundling each element) is inserted, a link is established between the front element and the rear element, a link label SET as a juxtaposed element is given, and the analysis result of the two elements is Connected. FIG. 20 shows a syntactic and semantic analysis result for the entire original input sentence, which is obtained by connecting two elements obtained by this procedure.

  Finally, in the analysis result output unit 106, it is said that the original input sentence “This problem is closely related to the progress of the Chinese investigation, and the Japanese investigation authorities will discuss at the joint meeting on the 17th. The analysis result shown in FIG. 20 is output.

  The present invention has been described in detail above with reference to specific embodiments. However, it is obvious that those skilled in the art can make modifications and substitutions of the embodiment without departing from the gist of the present invention.

  In the present specification, the embodiment of the present invention and the operation and effect thereof have been described with a focus on the case where the natural language processing according to the present invention is applied to an input sentence composed of Japanese. It is not limited, and it goes without saying that even if the input sentence is described in another language, the present invention can be similarly applied to obtain the advantageous effects.

  In short, the present invention has been disclosed in the form of exemplification, and the description of the present specification should not be interpreted in a limited manner. In order to determine the gist of the present invention, the claims section described at the beginning should be considered.

FIG. 1 is a diagram schematically showing an example of an implementation form of a natural language processing system according to the present invention. FIG. 2 is a block diagram schematically showing a functional configuration of the natural language processing apparatus 100 according to an embodiment of the present invention. FIG. 3 is a diagram schematically showing a functional configuration of the syntactic and semantic analysis processing system 200 based on the LFG grammar theory. FIG. 4 is a diagram showing c-structure obtained as a result of processing the input sentence “My dog eats apple” by the syntax and semantic analysis unit 203. FIG. 5 is a diagram showing f-structure obtained as a result of processing the input sentence “My dog eats apple” by the syntax and semantic analysis unit 203. FIG. 6 is a diagram expressing the f-structure shown in FIG. 5 in the form of a dependency tree. FIG. 7 shows a morpheme sequence with part-of-speech information obtained by morphological analysis processing by the sentence input unit 101 of the natural language sentence “Enkakuji is well known as a building in the Kamakura period”. FIG. FIG. 8 is a diagram illustrating an example of the division rule. Fig. 9 shows the dependency tree format for the results of syntactic and semantic analysis for each element obtained by dividing the input sentence "Enkakuji is very well known as a building in the Kamakura period." It is the figure shown by. FIG. 10 is a diagram illustrating an example of a connection rule. FIG. 11 is a diagram showing the syntax and semantic analysis result of the original sentence as a whole obtained by concatenating the syntax and semantic analysis results of each element shown in FIG. Figure 12 shows the result of syntactic and semantic analysis for each element obtained by dividing the input sentence "The US House of Representatives was opposed to the establishment of the Japan-US Investment Promotion Organization when President Clinton visited Japan." It is the figure shown in the form. FIG. 13 is a diagram showing the syntax and semantic analysis result of the original sentence as a whole obtained by concatenating the syntax and semantic analysis results of each element shown in FIG. FIG. 14 is a flowchart showing a processing procedure in the analysis result evaluation unit 107. FIG. 15 is a diagram illustrating an example of an analysis result for a sentence element in the syntax and semantic analysis unit 104. FIG. 16 is a diagram illustrating an example of an analysis result of a sentence element in the syntax and semantic analysis unit 104. FIG. 17 is a diagram illustrating a division rule that is applied when the input sentence is divided and the elements of the further divided sentence are further recursively divided. FIG. 18 is a diagram illustrating a result obtained by concatenating the syntax and semantic analysis results of the two elements illustrated in FIG. FIG. 19 is a diagram illustrating a result obtained by concatenating the syntax and semantic analysis results of the two elements illustrated in FIG. FIG. 20 is a diagram showing a syntax and semantic analysis result of the entire original input sentence obtained by connecting two elements.

Explanation of symbols

DESCRIPTION OF SYMBOLS 10 ... Natural language processing apparatus 11 ... Sentence input part 12 ... Division | segmentation and connection rule holding part 13 ... Sentence division | segmentation processing part 14 ... Syntax and semantic analysis part 15 ... Sentence element connection part 16 ... Analysis result output part 100 ... Natural language processing apparatus 101 ... sentence input unit 102 ... division / concatenation rule holding unit 103 ... sentence division processing unit 104 ... syntax and semantic analysis unit 105 ... sentence element connection unit 106 ... analysis result output unit 107 ... analysis result evaluation unit

Claims (43)

A natural language processing system for analyzing an input natural language sentence,
Sentence division processing means for determining a division position in the input sentence based on morpheme information of the input sentence, and dividing the input sentence into a front element and a rear element of the division position;
Syntactic and semantic analysis for each element of the input sentence divided by the division position, and to obtain a word-word dependency relationship and the type of the relationship for each element;
A sentence element connection processing means for determining a connection position and a dependency relationship in a syntax-semantic analysis result of each divided element based on morpheme information at a division position of an input sentence, and connecting element syntax analysis results;
A natural language processing system comprising: One or more division processing rules describing morpheme information of one morpheme or two or more consecutive morphemes related to a sentence division position;
One or more connection processing rules that describe the position and dependency relationship for connecting the syntactic and semantic analysis results of each divided element corresponding to the division processing rule,
The sentence element connection processing means connects the syntactic and semantic analysis results of the elements divided by using a connection processing rule corresponding to the division processing rule applied to the input sentence division processing by the sentence division processing means.
The natural language processing system according to claim 1. A natural language processing system for analyzing an input natural language sentence,
A division / concatenation processing rule holding means for associating and holding a division processing rule for determining a position to divide a sentence and a concatenation processing rule for concatenating a plurality of syntax semantic analysis results into one;
A sentence division processing means for dividing the input sentence into elements by referring to the division processing rules held in the division / concatenation processing rule holding means;
Syntax semantic analysis means for performing syntax semantic analysis on each element of the input sentence divided by the sentence division processing means;
Sentence element concatenation processing means for concatenating the semantic analysis results for each element obtained from the syntactic and semantic analysis means using a concatenation processing rule corresponding to the division processing rule used for dividing the input sentence by the sentence division processing means. When,
A natural language processing system comprising: Multiple split processing rules with priorities are prepared,
The sentence division processing means applies a division processing rule in accordance with the priority order and divides the input sentence at the obtained division position.
The natural language processing system according to claim 2, wherein the system is a natural language processing system. An analysis result evaluation means for evaluating the processing result of the syntax-semantic analysis for each element obtained by dividing the input sentence;
The natural language processing system according to claim 2, wherein the system is a natural language processing system. The analysis result evaluation unit determines that the element should be further divided in response to the fact that the syntax-semantic analysis unit cannot perform analysis on each element of the input sentence within a predetermined time period.
The sentence division processing unit divides the element again according to the determination by the analysis result evaluation unit,
The syntactic and semantic analysis means performs a syntactic and semantic analysis process on each element obtained by dividing the element again.
The natural language processing system according to claim 5. The analysis result evaluation means evaluates the syntactic and semantic analysis results for each element obtained by dividing the input sentence according to a certain division processing rule, and determines whether or not to divide the input sentence.
The sentence division processing unit applies another division processing rule to the input sentence according to the determination by the analysis result evaluation unit, and divides again,
The syntactic and semantic analysis means performs a syntactic and semantic analysis process on each element obtained by dividing the input sentence again.
The natural language processing system according to claim 5. The sentence division processing means selects a division processing rule based on a machine learning method and divides an input sentence.
The natural language processing system according to claim 2, wherein the system is a natural language processing system. When the sentence division processing unit divides the input sentence at the division position where the reading marks are connected to the connection particle A class, the connection processing unit applies the element before the division position to the verb of the element after the division position. Concatenate the syntactic and semantic analysis results of each element as a continuous modification component,
The natural language processing system according to claim 1, wherein the natural language processing system is a natural language processing system. When the sentence division processing means divides the input sentence at the division position where the reading mark is connected to the connection particle B class, the connection processing means inserts a synonym for bundling each element, and before the division position, Concatenates the results of syntactic and semantic analysis of elements and subsequent elements by means of juxtaposition relations via pronouns;
The natural language processing system according to claim 1, wherein the natural language processing system is a natural language processing system. When the input sentence is divided at the dividing position where the reading marks are connected to the continuous form after use by the sentence dividing processing means, the connecting processing means inserts a synonym for bundling each element, and before the dividing position, Concatenates the results of syntactic and semantic analysis of elements and subsequent elements by means of juxtaposition relations via pronouns;
The natural language processing system according to claim 1, wherein the natural language processing system is a natural language processing system. When the sentence division processing unit divides the input sentence at a division position where a punctuation point is connected to a case particle, the connection processing unit determines that the verb of the element after the division position is a citation verb before the division position. Connecting the syntactic analysis results of each element with the element of the verb as the quotation of the verb, otherwise connecting the syntactic analysis results of each element with the element before the division position as the continuous modification component,
The natural language processing system according to claim 1, wherein the natural language processing system is a natural language processing system. When the sentence division processing unit divides the input sentence at the division position where the noun, the counsel and the reading point are continuous, the connection processing unit may include a case element omitted in the verb of the element after the division position. For example, the elements before the division position are connected to each other as the corresponding case, and the syntax-semantic analysis results of the elements are connected to each other. Link,
The natural language processing system according to claim 1, wherein the natural language processing system is a natural language processing system. When the sentence division processing unit divides the input sentence at the division position where the punctuation point is connected to the adverb noun, the connection processing unit uses the element before the division position for the verb of the element after the division position. Link the syntactic and semantic analysis results of each element as a modifier.
The natural language processing system according to claim 1, wherein the natural language processing system is a natural language processing system. When the sentence division processing unit divides the input sentence using the reading point as a division position, the connection processing unit uses the element before the division position as a continuous modification component for the verb of the element after the division position. Concatenate syntactic and semantic analysis results,
The natural language processing system according to claim 1, wherein the natural language processing system is a natural language processing system. When the sentence division processing unit divides the input sentence using the connection particle A class as a division position, the connection processing unit uses the element before the division position as a continuous modification component for the verb of the element after the division position. Concatenate the syntactic and semantic analysis results of each element,
The natural language processing system according to claim 1, wherein the natural language processing system is a natural language processing system. When the sentence division processing unit divides the input sentence using the connection particle B class as a division position, the connection processing unit inserts a pronoun for binding each element, and the elements before and after the division position are inserted. Link each result of syntactic and semantic analysis of elements by juxtaposition via pronouns,
The natural language processing system according to claim 1, wherein the natural language processing system is a natural language processing system. When the sentence division processing means divides the input sentence using the continuous form after use as a division position, the connection processing means inserts a pronoun for binding the elements, and the elements before and after the division position Link each result of syntactic and semantic analysis of elements by juxtaposition via pronouns,
The natural language processing system according to claim 1, wherein the natural language processing system is a natural language processing system. When the sentence division processing unit divides the input sentence using the case particle as a division position, the connection processing unit selects the element before the division position if the verb of the element after the division position is a citation verb. Concatenating the results of syntactic and semantic analysis of each element as a verbal citation, otherwise concatenating the results of syntactic and semantic analysis of each element using the element before the split position as a continuous modification component,
The natural language processing system according to claim 1, wherein the natural language processing system is a natural language processing system. When the sentence division processing means divides the input sentence using the particle as a division position, the connection processing means, if there is an abbreviated case element in the verb of the element after the division position, Concatenating the results of syntactic and semantic analysis of each element with the corresponding element as the case, otherwise concatenating the results of syntactic and semantic analysis of each element with the element before the split position as a continuous modification component,
The natural language processing system according to claim 1, wherein the natural language processing system is a natural language processing system. When the sentence division processing unit divides the input sentence using the adverb noun as a division position, the connection processing unit uses each element before the division position as a continuous modification component for the verb of the element after the division position. Concatenates the results of syntactic analysis of elements,
The natural language processing system according to claim 1, wherein the natural language processing system is a natural language processing system. A natural language processing method for analyzing an input natural language sentence,
A sentence division processing step for determining a division position in the input sentence based on morpheme information included in the input sentence, and dividing the input sentence into a front element and a rear element of the division position;
A syntax and semantic analysis step is performed for each element of the input sentence divided by the division position, and a dependency relationship between words and words and a type of the relationship are obtained for each element;
A statement element connection processing step for determining a connection position and a dependency relationship in a syntax and semantic analysis result of each divided element based on morpheme information at a division position of an input sentence, and connecting the syntax and semantic analysis results of elements; ,
A natural language processing method comprising: Describes the division processing rules that describe the morpheme information of one morpheme or two or more consecutive morphemes related to the sentence division position, and the position and dependency relationship for linking the results of syntactic analysis of each divided element Is associated with one or more concatenation processing rules,
In the sentence element connection processing step, the syntactic and semantic analysis results of the elements divided using the connection processing rule corresponding to the division processing rule applied to the input sentence division processing in the sentence division processing step are connected.
The natural language processing method according to claim 22. A natural language processing system for analyzing an input natural language sentence,
A division processing rule for determining a position to divide a sentence is associated with a concatenation processing rule for concatenating a plurality of syntactic and semantic analysis results into one,
A sentence division processing step for dividing an input sentence into elements by referring to any of the division processing rules;
A syntax and semantic analysis step for performing syntax and semantic analysis on each element of the input sentence divided in the sentence division processing step;
Sentence element concatenation processing step of concatenating the semantic analysis results for each element obtained by the syntactic and semantic analysis step using a concatenation processing rule corresponding to the division processing rule used for the division of the input sentence in the sentence division processing step When,
A natural language processing method comprising: Multiple split processing rules with priorities are prepared,
In the sentence division processing step, the division processing rule is applied according to the priority order, and the input sentence is divided at the obtained division position;
The natural language processing method according to claim 23 or 24, wherein An analysis result evaluation step for evaluating the processing result of the syntactic and semantic analysis for each element obtained by dividing the input sentence;
The natural language processing method according to claim 23 or 24, wherein In the analysis result evaluation step, in response to failure to execute analysis for each element of the input sentence in the syntax-semantic analysis step within a predetermined time, it is determined that the element should be further divided,
In the sentence division processing step, the element is divided again according to the determination in the analysis result evaluation step,
In the syntactic and semantic analysis step, a syntactic and semantic analysis process is performed on each element obtained by dividing the element again.
27. The natural language processing method according to claim 26. In the analysis result evaluation step, the syntax-semantic analysis result for each element obtained by dividing the input sentence according to a certain division processing rule is evaluated, and it is determined whether or not the input sentence should be divided again.
In the sentence division processing step, another division processing rule is applied to the input sentence according to the determination in the analysis result evaluation step, and the division is performed again.
In the syntactic and semantic analysis step, a syntactic and semantic analysis process is performed on each element obtained by dividing the input sentence again.
27. The natural language processing method according to claim 26. In the sentence division processing step, a division processing rule is selected based on a machine learning method, and an input sentence is divided.
The natural language processing method according to claim 23 or 24, wherein When the input sentence is divided at the division position where the reading marks are connected to the connected particle A class in the sentence division processing step, in the connection processing step, the element preceding the division position is changed to the verb of the element after the division position. Concatenate the results of syntactic analysis of each element as a continuous modification component for
The natural language processing method according to claim 22 or 24, wherein the natural language processing method is used. When the input sentence is divided at the dividing position where the punctuation marks are connected to the connected particle B class in the sentence dividing process step, a synonym for bundling each element is inserted in the connecting process step, and before the dividing position, Link the results of syntactic and semantic analysis of the elements of and the following elements by a juxtaposition relationship via pronouns.
The natural language processing method according to claim 22 or 24, wherein the natural language processing method is used. When the input sentence is divided at the division position where the punctuation marks are connected to the continuous form after utilization in the sentence division processing step, in the connection processing step, a substitute term for bundling each element is inserted, and the sentence before the division position is inserted. Link the results of syntactic and semantic analysis of the elements of and the following elements by a juxtaposition relationship via pronouns.
The natural language processing method according to claim 22 or 24, wherein the natural language processing method is used. When the input sentence is divided at the division position where the punctuation marks are connected to the case particles in the sentence division processing step, in the connection processing step, if the verb of the element after the division position is a citation verb, from the division position Concatenate the results of syntactic and semantic analysis of each element with the previous element as the citation of the verb, otherwise concatenate the results of syntactic and semantic analysis of each element with the element preceding the split position as the continuous modification component,
The natural language processing method according to claim 22 or 24, wherein the natural language processing method is used. In the sentence division process, when the input sentence is divided at the division position where the noun, the counsel and the punctuation mark are continuous, in the connection processing step, there is a case element omitted in the verb of the element after the division position. For example, the elements before the division position are connected to each other as the corresponding case, and the syntax-semantic analysis results of the elements are connected to each other. Link,
The natural language processing method according to claim 22 or 24, wherein the natural language processing method is used. When the input sentence is divided at the division position where the punctuation marks are connected to the adverb nouns in the sentence division processing step, in the connection processing step, the element before the division position is assigned to the verb of the element after the division position. Concatenate the syntactic and semantic analysis results of each element as a continuous modification component,
The natural language processing method according to claim 22 or 24, wherein the natural language processing method is used. When the input sentence is divided by using the reading point as the division position in the sentence division processing step, in the connection processing step, the element before the division position is used as a continuous modification component for the verb of the element after the division position. Concatenate syntactic and semantic analysis results,
The natural language processing method according to claim 22 or 24, wherein the natural language processing method is used. When the input sentence is divided using the connection particle A class as the division position in the sentence division processing step, in the connection processing step, the element before the division position is used as a continuous modification component for the verb of the element after the division position. Concatenate the syntactic and semantic analysis results of each element,
The natural language processing method according to claim 22 or 24, wherein the natural language processing method is used. In the sentence division processing step, when the input sentence is divided using the connection particle B class as a division position, in the connection processing step, a synonym for bundling each element is inserted, and the elements before and after the division position are inserted. Link each result of syntactic and semantic analysis of elements by juxtaposition via pronouns,
The natural language processing method according to claim 22 or 24, wherein the natural language processing method is used. When the input sentence is divided using the continuous form after utilization in the sentence division processing step as a division position, in the connection processing step, a pronoun for binding each element is inserted, and the elements before and after the division position are inserted. Link each result of syntactic and semantic analysis of elements by juxtaposition via pronouns,
The natural language processing method according to claim 22 or 24, wherein the natural language processing method is used. When the input sentence is divided using the case particles as the division positions in the sentence division processing step, in the connection processing step, if the verb of the element after the division position is a citation verb, the element before the division position is Concatenating the results of syntactic and semantic analysis of each element as a verbal citation, otherwise concatenating the results of syntactic and semantic analysis of each element using the element before the split position as a continuous modification component,
The natural language processing method according to claim 22 or 24, wherein the natural language processing method is used. In the sentence division processing step, when the input sentence is divided by using the auxiliary particle as a division position, in the connection processing step, if there is a case element omitted in the verb of the element after the division position, the input sentence is preceded by the division position. Concatenating the results of syntactic and semantic analysis of each element with the corresponding element as the case, otherwise concatenating the results of syntactic and semantic analysis of each element with the element before the split position as a continuous modification component,
The natural language processing method according to claim 22 or 24, wherein the natural language processing method is used. When the input sentence is divided by using the adverb noun as a dividing position in the sentence dividing process step, each of the elements before the dividing position is used as a continuous modification component for the verb of the element after the dividing position. Concatenates the results of syntactic analysis of elements,
The natural language processing method according to claim 22 or 24, wherein the natural language processing method is used. A computer program written in a computer-readable format so as to execute processing for analyzing an input natural language sentence on a computer system,
A sentence division processing step for determining a division position in the input sentence based on morpheme information included in the input sentence, and dividing the input sentence into a front element and a rear element of the division position;
A syntax and semantic analysis step is performed for each element of the input sentence divided by the division position, and a dependency relationship between words and words and a type of the relationship are obtained for each element;
A statement element linking process step for linking element syntactic analysis results of the elements by determining a linking position and a dependency relationship in a syntax semantic analysis result of each divided element based on morpheme information at the division position of the input sentence; ,
A computer program comprising:

Images