JPH0795323B2 - Natural language processor - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION [Object of the Invention] (Industrial field of application) The present invention aims to effectively support sentence proofreading by finding the logical structure of a sentence structure of a text (sentence) composed of natural language. The present invention relates to a natural language processing device.

(Prior Art) Recently, various researches on natural language processing have been advanced, and for example, it has been attempted to use it as a means for proofreading a prepared sentence (text). However, in the conventional device that supports proofreading using natural language processing,
Support grammatical correction by checking lexical and morphological checks such as typographical errors and omissions in texts, and fluctuations in sending kana notation, or by checking for grammatical errors and exceeding specified values for sentence length. It is simply composed of. For this reason, matters that can be judged as good or bad by the relative relationship between multiple sentences that make up a sentence using natural language processing, such as detection and indication of a part where the development of the theory is unclear Was very difficult to do.

By the way, for example, "Information Processing Society of Japan Study Group Report V01.89, No.
6 89-NL-70 ”, the rhetorical expressions such as conjunctions appearing in a sentence are used as clues, and the rhetorical expressions are used to refer to the connection rules for the series permitted as the connection relation between multiple sentences. It is attempted to find the context structure by investigating. However, the above-mentioned document merely discloses a method of extracting the argument structure (context structure) from a sentence (text). Moreover, because the sentences that make up a text do not always have a connective expression such as a connective,
Even using rhetorical expressions such as those shown in this document, some problems remain for accurately determining the context structure. For this reason, the fact is that the proofreading has not been supported by detecting the unclear part of the theory development.

(Problems to be solved by the invention) In the case of supporting document proofreading by using conventional natural language processing in this manner, a noun or verb appearing exclusively in a text is used as a clue to examine a context structure close to the text content, Moreover, it only performs lexical and morphological checks and grammatical checks on the words that appear in the text. Therefore, there is a limit in supporting document proofreading only by using the conventional natural language processing technology. For example, it is hoped that an unclear part of the development of the theory can be detected to efficiently support grammar proofreading. There wasn't.

The present invention has been made in consideration of such circumstances, and the purpose thereof is to examine the theory expansion structure of a sentence,
To effectively detect matters (sentences) whose quality is judged by the relative relationship between multiple sentences, and to effectively support proofreading of unclear parts such as the development of the theory. It is to provide a natural language processing device capable of performing.

[Structure of the Invention] (Means for Solving the Problems) A text input section for inputting a document in natural language, and the entire document based on the formal information of chapters / sections included in the document input from the text input section A format structure processing unit that analyzes the formal structure of the, a morphological analysis unit that performs a morphological analysis on the document input from the text input unit, and a syntax analysis unit that parses the morphological analysis result of the morphological analysis unit,
Correspondence table storage that stores correspondences between rhetorical expressions of connectives and their connection relations based on rhetorical expressions in the document input from the text input unit according to the analysis result of the formal structure, morpheme, and syntax From the connection relation extraction unit that obtains the connection relation between the sentences in the document using the section, and the series of the connection relations between the sentences obtained by the connection relation extraction unit,
An inappropriate combination used in the theory structure evaluation unit that generates a candidate for the excluded structure structure that is inappropriate as the structure structure of a sentence and evaluates the likelihood of each candidate structure structure is used in advance. A relation table storage unit that stores the connection prohibition rule and the negative connection rule stored in association with each other;
Context for performing a context process for obtaining a context structure for each argument structure according to the syntactic analysis result obtained by the syntactic analysis unit, based on the evaluation value for each argument structure candidate that has been likelihood evaluated by the argument structure evaluation unit. From the processing unit, the context structure obtained by the context processing unit, and the likelihood evaluation result obtained by the argument structure evaluation unit, refer to the knowledge indicating the connection between sentences stored in the knowledge storage unit in advance. Then, the likelihood structure determination unit that determines the predetermined logical structure from the logical structure candidates and the likelihood evaluation result obtained by the logical structure evaluation unit with respect to the predetermined logical structure determination result by the logical structure determination unit are determined. , Refer to a rule stored in the relation table storage section for an inappropriate part detection unit that detects a logical structure determined to be inappropriate and an alternative expression equivalent to the logical structure detected by the inappropriate part detection unit Then asked And structure modifying unit, characterized in that anda alternative representation presentation section for presenting the alternative representation obtained by the structure modification unit.

(Operation) According to the present invention, paying attention to the rhetorical expression in the text, the connection relation between the sentence parts is examined from the rhetorical expression, and the sentence structure is evaluated according to the relative connection rule of the series of connection relations. Since the theoretical structure that can be represented by the sentence structure is evaluated, the unclear part of the theoretical development can be effectively detected. As a result, it is possible to effectively evaluate and present the theory structure of the sentence, and to output the difficult-to-understand part of the theory development and alternative expressions, etc., so that lexical and morphological check results and In combination with the presentation of check results for grammatical errors, it becomes possible to effectively support the proofreading.

(Embodiment) Hereinafter, a context processing device according to an embodiment of the present invention will be described with reference to the drawings.

FIG. 1 is a schematic configuration diagram of a main part of the embodiment apparatus, and 1 is a text input unit for inputting text such as text data or recorded data to be analyzed. A natural language text (sentence) input via the text input unit 1 is given to the sentence analysis unit 2 and subjected to format structure analysis and morpheme analysis. The format structure analysis unit 2a that constitutes the sentence analysis unit 2 analyzes the format structure of the entire text based on the format information such as chapters and sections included in the text input from the text input unit 1 and the structure thereof. Is to determine. Further, the morpheme analysis unit 2b constituting the sentence analysis unit 2 performs morpheme analysis of the sentence included in the paragraph, for example, in paragraph units, with respect to the text whose formal structure is determined by the analysis by the formal structure analysis unit 2a. To execute. By this morphological analysis, morphemes such as nouns and verbs in the text are individually obtained.

The processing functions of the sentence analysis unit 2 described above have the same functions as those of the conventional context processing device, and can be realized by appropriately using various conventionally proposed methods.

Here, the first feature of the present device is that the sentence analysis unit 2 described above.
According to the result of the formal structure analysis of the text and the result of the morpheme analysis, the connection relation extracting unit 3 for obtaining the connection relation between a plurality of sentence parts (for example, paragraphs) forming the text using the rhetorical expression in the text as a clue. The point is that the logical structure evaluation unit 4 and the logical structure determination unit 5 described later are provided.

The connection relation extraction unit 3 stores in the correspondence table storage unit 7 a syntactic analysis result obtained by the syntactic analysis unit 6 that analyzes the syntactic structure according to, for example, the information of the part of speech based on the analysis result of the sentence analysis unit 2. The connection relation between a plurality of sentences forming a sentence (input text) is individually obtained by using the stored information. The correspondence table stored in the correspondence table storage unit 7 shows correspondence between rhetorical expressions such as conjunctions and their connection relations, as shown in FIG. 2, for example. The rhetorical expression in the text is checked with reference to the table storage unit 7, and the connection relation between the paragraphs connected by the rhetorical expression is obtained.

For example, when a rhetorical expression "But" is requested, the connection relation extraction unit 3 "captures" the paragraph existing before the rhetorical expression and the paragraph with the rhetorical expression.
We are looking for a connection relationship that indicates that The connection relation extracting unit 3 sequentially finds all rhetorical expressions that appear in the text, and individually finds the connection relations between the sentence parts (sentences) indicated by the rhetorical expressions.

Then, the argument structure evaluation unit 4 is, for example, shown in FIG.
As shown in (b), the relation table storage unit 8 that stores a prohibition rule regarding the connection of the connection relation indicated by the rhetorical expression as a relation table is referred to, and the connection between the sentence parts obtained by the connection relation extraction unit 3 Structure of the theory from the series of relations to the paragraph and between paragraphs
Is to seek.

Specifically, the relation table stored in the relation table storage unit 8 has information indicating structural connection prohibition rules of connection relations represented by rhetorical expressions as shown in FIG. 3 (a), and FIG. A table in which information indicating a negative connection rule of the above-mentioned connection relation as shown in b) is registered, and the structure between the connection relations which is inappropriate as the textual structure of the text is shown as a rule. In particular, the connection prohibition rule shown in FIG. 3 (a) shows a combination structure that does not actually appear in the connection relation between paragraphs connected by rhetorical expressions. The negative connection rule shown in FIG. 3 (b) shows a combination structure that is rarely used in general, if not as much as the above connection prohibition rule.

The argument structure evaluation unit 4 refers to the relation table storage unit 8 having such information, and from the series showing the connection relation between the paragraphs connected by the rhetorical expression obtained by the connection relation extraction unit 3, the argument structure of the text As the combination structure is excluded, the combination structure is obtained, candidates for the argument structure of the text are created, and the likelihood evaluation is performed on these candidates.

In this way, the context processing unit 9 follows the evaluation value for the argument candidate whose likelihood is evaluated by the argument structure evaluating unit 4 and the syntactic analysis result obtained by the syntactic analyzing unit 6 according to the argument structure candidate of the text. Context processing is performed for, for example, processing such as reference resolution for anaphoric expressions is performed. In this context processing, for example, according to the analysis result of the text obtained by the sentence analysis unit 2, for example, the words appearing in the text (sentence) are examined, and the relationship between a plurality of sentences is determined from the relationship between the words appearing in each sentence. Is being determined.

That is, the context processing unit 9 checks words appearing in each sentence of a plurality of sentences forming an input sentence (text). For example, the topic expression extracting unit determines a word as a topic presentation expression in the sentence. Looking for. The word as the topic presentation expression is extracted, for example, on the condition that the particle attached to the word is “ha” or “mo”. Specifically, when the expression "Xwa" exists, the word "X" is extracted as a word as a topic presentation expression. Also, "Y of X
When the expression "ha" exists, both "X" and "Y" are extracted as words that present a topic.

It should be noted that not all of the words extracted in this way represent the topic presentation. For example, among the words extracted using the particle as a clue, for example, "recently"
It cannot be said that it is a topic presentation expression such as "Fig. X is" or "in the case of ...", and there is something that does not serve as a clue for checking the connection relation between sentences. For such words, for example, these are determined beforehand as unnecessary words and stored in the unnecessary word table storage unit, and by referring to this unnecessary word table storage unit, they are excluded from words as topic presentation expressions. You can leave it.

Further, the context processing unit 9 basically checks whether or not the same word is repeatedly used between a plurality of sentences.
When a word as a topic presentation expression is sought in a certain sentence, it is checked whether the word is also used in another sentence, and if the word is used in another sentence, the sentence is identified. The information (sentence management number) is requested. Then, according to the word as the topic presentation expression and the information of the appearing word obtained by the word repetitive extraction processing, the relationship between a plurality of sentences forming the input text is determined according to the word information.

Specifically, when a word as a topic presentation expression is extracted from a certain sentence by the topic expression extraction process, the information of the appearing word in each sentence obtained by the word repetition extraction process for this word is checked, A sentence in which the same word as the word appears in a sentence preceding the sentence in which the word as the presentation expression is extracted is obtained, and this sentence is associated with the sentence in which the topic presentation expression is extracted. If the word extracted as a topic presentation expression is a demonstrative or a pronoun, the sentence relevance judgment processing determines that the word as a topic presentation expression in the sentence preceding the sentence including the topic presentation expression composed of this demonstrative or pronoun. A sentence having a word is examined, and sentences subsequent to this sentence are associated with the sentence for which the above-mentioned demonstrative or pronoun is obtained.

By such a determination process, the relationship between a plurality of sentences forming the input text and each other is examined with the word appearing in each sentence as a clue, and the context process is executed.

The context processing result thus obtained, the argument structure candidate obtained by the argument structure evaluation unit 4 and the likelihood evaluation result thereof are given to the argument structure determination unit 5.

The argument structure determining unit 5 includes information about the relevance between sentences (the Kist argument structure candidate and its likelihood evaluation result) obtained by the argument structure evaluating unit 4 and the context of the input text obtained by the context processing unit 9. From the structure, the knowledge structure of the sentence structure stored in the knowledge storage unit 10 is referred to determine the logical structure.

That is, the logical structure determining unit 5 uses the knowledge stored in the knowledge storage unit 10, for example, the knowledge indicating the connection between the sentences of the parts of speech appearing in the text, and the logical structure of the logical structure is unique while following the logical structure of the text described above. To decide. At this time, the logical structure determining unit 5 activates the user input / output unit 11 as necessary, interactively takes in information necessary for determining the logical structure from the user, and proceeds with the processing for determining the logical structure.

Therefore, the inappropriate portion detection unit 12 determines the likelihood evaluation result obtained by the above-mentioned theory structure evaluation unit 4 with respect to the theme structure determined by the above-mentioned theory structure determination unit 5. Then, when the evaluation is low, the sequence part of the connection expression that is the cause is detected as an unsuitable (unclear) part due to the theoretical structure. The information on the inappropriate (unclear) portion of the argument structure thus detected is presented and output to the user via the user input / output unit 11.

The structure correction unit 13 is equivalent to the argument structure determined by the argument structure determination unit 5 for the inappropriate portion detected by the inappropriate portion detection unit 12, and is stored in the relationship table storage unit 8. A sequence of connection relationships that does not conflict with the above-mentioned connection prohibition rule is generated by rearranging the sentences. The information on the new connection relation structure rearranged and generated in this way is the user input / output unit.
At 11, the information is presented to the user and the confirmation is prompted. Then, the alternative expression presenting unit 14 searches the structure modified by the structure modifying unit 13 for the rhetorical expression for each connection appearing in the structure with reference to the above-mentioned correspondence table, and also as described above. Based on the results of various analyses, we investigate whether there is a more appropriate alternative expression. When an appropriate alternative expression is found, this is used as the user input / output unit.
Presented from 11.

The above is the main configuration of the apparatus and its function.

Here, the above-mentioned rhetorical expression-based structure determination processing will be described in more detail with reference to FIGS. 4 to 6.

FIG. 4 is a diagram showing an example of text to be subjected to context processing. When such text is input, the format structure analysis unit 2a divides the text part constituting the text into texts or paragraphs based on the textual information such as indentation, punctuation marks, commas, periods, etc. Is to seek. Here, the input text shown in FIG. 4 is divided into a plurality of sentences, and here, for example, sentences, ... Are respectively obtained. The morphological analysis unit 2b executes the morphological analysis on each sentence thus obtained.

For this morphological analysis, the analysis method shown in, for example, "Makoto Nagao," Japanese Information Processing "published by The Institute of Electronics, Information and Communication Engineers, (1984)" may be used. According to such a Japanese sentence analysis method, for example, when a sentence “The following refers to drawings.” Is given, “below (noun) + (particle) + is (particle) + drawing (noun ) + To (particle) + reference (noun) + to (particle) + (verb). The result of morphological analysis will be required.

Then, the connection relation extraction unit 3 searches for the rhetorical expression in each sentence by referring to the correspondence table storage unit 7 based on the morphological analysis result obtained for each sentence constituting the input text, and detects the detected rhetorical expression. Information about the connection relationship that the rhetorical expression means is obtained from the correspondence table storage unit 7.

In the example shown in FIG. 4, a rhetorical expression "following" is obtained from a sentence and its connection relation is "junction". In addition, the rhetorical expression "Mata" and its connection relation "parallel" are obtained from the sentence, and the rhetorical expression "Following" and its connection relation "junction" are similarly obtained from the sentence.

From such a connection relation, the connection relation extracting unit 3 determines the fourth relation.
As shown in A of FIG. 5, the text shown in the figure requests that a plurality of sentences, ...

However, such a connection relation series merely shows connection relations between a plurality of sentences individually, and does not show how the argument structures are combined.

Then, considering the combination of the connection relation series forming the argument structure, if there are these four paragraphs, theoretically, there are five possible combinations as shown in FIG. 5B. When there are two sentences, the combination structure can take one form, and when there are three sentences, the combination structure can take two forms. In the case of 5 sentences, there are 14 types, and the number of forms that the combination structure can take is theoretically determined by the number of sentences.

The combination structure taken by the series of connection relations indicates where and in what relation each sentence in the text is connected. For example, [[[ordering] parallel] ordering] (Structure of sentence combination) indicates that a sentence is derived from a sentence, a sentence is present in parallel by describing another thing, and the sentence is derived based on these.

Similarly, for example, in the theory structure of [[junction] parallel [junction]], sentences are derived from sentences and sentences are derived from sentences. And, as a whole, a logical structure that these are described in parallel is shown.

However, the combination structure taken by the series of connection relations shown in FIG. 5B merely represents the structural form of the combination, and some of them are unsuitable as the argument structure in the sentence structure. Therefore, the argument structure evaluation unit 4 refers to the relationship table storage unit 8 described above, finds a structure having an inappropriate rhetorical relationship in the combination structure, and excludes the structure from candidates for the argument structure.

For example, the connection prohibition rule "forward connection [... [X forward connection" (where X is one sentence or a plurality of sentences connected by a connection relation is located in the relational table shown in FIG. 3A). If any of the above combination structures includes a combination structure of "ordered [X-ordered]", it is regarded as inappropriate as an argument structure, and it becomes [BAD] as a candidate for the argument structure. Give a rating. Specifically, since the underlined portion of the second candidate structure structure [ forward [[parallel] forward ]] shown in B of FIG. 5 corresponds to this, the likelihood evaluation is performed on this candidate structure structure. Give [BAD].

Therefore, the first argument structure treasure candidate [forward [parallel [forward]]] and the third theoretical structure candidate [[forward [parallel]] forward] and the fourth shown in FIG. 5B. For the candidate structure structure [[forward] parallel [forward]], the negative connection rule "forward [... [X parallel]" parallel [... [] shown in the relationship table shown in FIG. Therefore, the likelihood evaluation [NG] is given to these argument structure candidates. Since only the fifth argument structure candidate [[[ordering] parallel] ordering] does not conflict with the above-mentioned connection prohibition rule or negative connection rule, [OK] is given as its likelihood evaluation.

Therefore, the context processing unit 9 executes context processing such as reference resolution of the anaphoric expression on the syntactic analysis result obtained from the syntactic analysis unit based on the likelihood evaluation result of the above-mentioned argument structure candidate.

For example, when the likelihood evaluation of the theory structure candidate [[ordering] parallel [ordering]] described above is high, the topic presentation expression of the sentence “Theorem 3” is the target of the anaphoric expression “kore” shown in the sentence in the text. ] Is given priority and context processing is performed. Such a context processing result is given to the argument structure determining unit 5.

Then, the logical structure determination unit 5 refers to the knowledge information registered in the knowledge storage unit 10 according to the context processing result and the likelihood evaluation result of the logical structure candidate, and determines the logical structure among the candidates. To decide from. Specifically, in order from the argument structure candidate with the highest likelihood evaluation, whether or not the interpretation is consistent with the content (structure of the content) referred to in the text is examined with reference to the knowledge about the content.

For example, in this example, the knowledge storage unit is provided with knowledge indicating the connection between sentences such as the target indicated by the words “Theorem 1”, “Theorem 2”, “Proof”, etc. that appear in the text and which represent the topic. 10 is searched and checked, whether or not there is a contradiction in the connection structure of the information contents possessed by these is checked based on the above-mentioned context analysis result, and the argument structure is determined. In this example, the contents of "Theorem 1" and "Theorem 2" mentioned in sentences and sentences and the contents of "Theorem 3" and "Theorem 4" mentioned in sentences and sentences are independent. Is obtained by referring to knowledge information. If knowledge about such text contents is obtained, then the logical structure determining unit 5 determines the fifth logical structure candidate (logical structure candidate) that has received the highest evaluation by the likelihood evaluation described above.
Examine the inconsistency of the argument structure. Then, in the fifth candidate structure structure, it is shown that the sentence is a structure having a content dependency not only with the sentence but also with the sentence, so that it contradicts the knowledge information described above. It turns out. As a result, although the fifth argument structure candidate received the highest likelihood evaluation, it is rejected as being unsuitable as the argument structure for the text because the development of the argument content is inconsistent. .

Thereafter, similarly, the inconsistency of the argument structure candidate having the next highest likelihood evaluation is determined according to the knowledge information about the word in the sentence obtained by referring to the knowledge storage unit 10. In this example, [NG] is the first, 3
Examine whether there is any contradiction in the contents of the second and fourth candidate structure structures.

Then, among these argument structure candidates, sentences and sentences,
It is judged that the structure showing that the sentences are independent from each other in content is only the fourth argument structure. As a result, the argument structure determining unit 5 determines that the fourth argument structure candidate shows the development structure of the argument in the text in content, and uniquely determines the argument structure as the argument structure of the text.

Thus, here, the analysis result of the input text given in natural language (theoretical structure), in which the logical structure of the input text is determined without causing a contradiction in the content of the logical expansion, and the text expansion structure is clarified Will be required.

Now, when the logical structure for the input text is determined in this way, the inappropriate portion detection unit 12 checks the likelihood evaluation result obtained for the determined logical structure. Then, when the evaluation result is not the maximum among the evaluation results for the plurality of candidates described above, that is, when the evaluation result is low, it is determined that there is a part in the text that is difficult to understand. In this case, refer to the connection relation prohibition rule and the negative connection rule used in the above-mentioned likelihood evaluation to find out which connection relation part is the cause of difficulty in understanding the theory development. Find out.

In the case of the example shown here, the connection expression (lethal expression) “also” placed between sentences and the connection expression (letter expression) “according to between sentences” ”Is the part that obscures the structure of the development of the theory. This is because, although the logical content here is derived only from sentences, sentences are derived both from the sentence and from the sentence. Can be interpreted as:
From such ambiguity, it is judged that the above two rhetorical expressions obscure the theory development. The inappropriate portion detection unit 12 presents and outputs the ambiguous text portion thus detected to the user via the user input / output unit 11.

Therefore, the structure correction unit 13 is equivalent to the argument structure uniquely determined by the argument structure determination unit 5 for each inappropriate portion detected as the ambiguous sentence portion, and the relation table storage A partial structure that does not conflict with the connection prohibition rule and the like stored in the unit 8 and a series of connection relationships for the part are obtained as an alternative expression structure. That is, the sentence structure is rearranged to obtain a sentence structure that satisfies the above-mentioned conditions.

Specifically, from the above-mentioned fourth argument structure, the inappropriate part is obtained as "parallel [forward]", and an alternative structure is obtained as "parallel [reason]". As a result, the argument structure of the entire text is obtained as [[junction] parallel [reason]], and the argument structure of an alternative expression that can clearly express the argument development of the input text is required.

The alternative expression presentation unit 14 obtains rhetorical expressions corresponding to each connection relation appearing in the structure from the correspondence table storage unit 7, uses information such as the above-mentioned syntactic analysis result, and from the knowledge storage unit 10. Using the obtained knowledge memory, a better alternative expression for the inappropriate portion is obtained, and this is presented and output to the user via the user input unit 11 as shown in FIG. 6, for example.

Thus, according to the present apparatus thus configured, not only the nouns and verbs of the text are analyzed, but also the discourse structure of the text indicated by the rhetorical expressions appearing in the text is analyzed, and the discourse structure is evaluated. Since the logical structure of the above text is determined, even if information about nouns and verbs appearing in the text is not registered in the dictionary,
It is possible to perform the determination process of the logical structure by appropriate natural language processing without being influenced by the unregistered word.
Moreover, since the natural language processing is performed by using the theoretical structure analyzed from the rhetorical expression, it is possible to accurately grasp the theoretical structure such as the premise, the conclusion, the judgment, the illustration, etc., and accurately grasp the context. It is possible to evaluate and interpret the argument structure.

After that, the unclear part of the theory development is detected, the corresponding part is pointed out and displayed, and an alternative expression to replace the unclear sentence structure of the theoretical structure is sought and output to the user.
It is possible to effectively find out which part of the textual development of the text is unclear and how it can be corrected to eliminate the unclear textual development.

As a result, when the proofreading process is performed on the text, the natural language process described above can be effectively utilized to effectively support the proofreading of the text.

It should be noted that the present invention is not limited to the above-described embodiment, and the analysis of the format structure, the morphological analysis, etc. may be performed by appropriately using various algorithms conventionally proposed. Further, the information registered in the correspondence table or the relation table is not limited to the above example. The method of evaluating the argument structure candidate and the algorithm for uniquely determining the argument structure can be variously modified.

Furthermore, although the Japanese processing has been described here, the context processing for other languages such as English can be similarly applied. In addition, the present invention can be variously modified and implemented without departing from the scope of the invention.

[Effects of the Invention] As described above, according to the present invention, the rhetorical expressions appearing in a text are focused on to find candidates for the argument structure, and the candidates for the argument structure are evaluated to evaluate the arguments of the text. Since the structure is examined and the unclear part is searched for as the structure of the theory development, it is possible to effectively advance the natural language processing on the text and accurately grasp the theory structure. Then, unclear parts are detected as the development of the argument structure, the corresponding parts are pointed out and displayed, or alternative expressions thereof are provided and output, so that natural language processing can effectively support proofreading. And so on, a great effect is practically achieved.

[Brief description of drawings]

FIG. 1 shows a context processing device according to an embodiment of the present invention. FIG. 1 is a schematic configuration diagram of a main part of the embodiment device, FIG. 2 is a diagram showing a configuration example of a correspondence table storage part, and FIG. Is a diagram showing a configuration example of a relation table storage unit, FIG. 4 is a diagram showing an example of input text to be context-processed, FIG. 5 is a diagram schematically showing a form of text argument structure analysis processing, FIG. FIG. 6 is a diagram showing a modification example of text shown by an alternative expression. 1 ... Text input unit, 2 ... Sentence analysis unit, 2a ... Format structure analysis unit, 2b ... Morphological analysis unit, 3 ... Connection relation extraction unit, 4 ... Logical relation evaluation unit, 5 ... Logical structure determination unit, 6 ... Syntactic analysis Division, 7
... correspondence table storage unit, 8 ... relation table storage unit, 9 ...
Context processing unit, 10 ... Knowledge storage unit, 11 ... User input / output unit, 12
… Inappropriate part detection unit, 13… Structure correction unit, 14… Alternative expression presentation unit.

Claims (1)

[Claims] 1. A text input unit for inputting a document in natural language, and chapters included in the document input from the text input unit.
A format structure processing unit that analyzes the format structure of the entire document from the format information of the section, a morphological analysis unit that performs a morphological analysis on the document input from the text input unit, and a morphological analysis result of the morphological analysis unit And a correspondence between the rhetorical expression of the conjunction and its connection relationship based on the rhetorical expression in the document input from the text input unit according to the analysis result of the formal structure, morpheme and syntax. From the connection relation extraction unit that obtains the connection relation between the sentences in the document using the correspondence table storage unit that stores the attachment, and the sequence of the connection relations between the sentences obtained by the connection relation extraction unit, Inappropriate combinations as the logical structure are generated by generating excluded candidate logical structure candidates and performing likelihood evaluation of each candidate of the logical structure and the inappropriate combinations used in the logical structure evaluating unit. A relation table storage unit that stores the relation between the stored connection prohibition rule and the negative connection rule, and an evaluation value for each candidate structure structure that has been likelihood evaluated by the structure structure evaluation unit. According to the syntactic analysis result obtained by the analysis unit, a context processing unit that performs a context process for obtaining a context structure for each argument structure, a context structure obtained by the context processing unit, and an argument structure evaluation unit. From the likelihood evaluation result, the knowledge indicating the connection between the sentences stored in advance in the knowledge storage unit is referred to,
It is inappropriate to judge the likelihood evaluation result obtained by the theoretical structure evaluation unit with respect to the predetermined logical structure determination result by the logical structure determination unit, which determines a predetermined logical structure from candidate logical structures. An inappropriate portion detection unit that detects the argument structure determined to be, and an alternative expression equivalent to the argument structure detected by the inappropriate portion detection unit are obtained by referring to the rules stored in the relationship table storage unit. A natural language processing apparatus comprising: a structure correction unit; and an alternative expression presentation unit that presents the alternative expression obtained by the structure correction unit.