JP2014119988A - Synonym determination device, synonym learning device, and program - Google Patents

Abstract

PROBLEM TO BE SOLVED: To highly accurately perform synonym determination for a predicate.SOLUTION: A feature construction unit 222 extracts distribution similarity, a dictionary define statement feature, a semantic attribute feature and a functional expression feature for an "argument-predicate" pair input by an input unit 210. A synonym determination unit 230 determines whether or not a synonym determination model stored in a synonym determination model storage unit 232 for determining whether or not the "argument-predicate" pair is a synonym is synonymous with the "argument-predicate" pair on the basis of the extracted distribution similarity, dictionary define statement feature, semantic attribute feature and functional expression feature.

Description

  The present invention relates to a synonym determination device, a synonym learning device, and a program, and more particularly, to a synonym determination device, a synonym learning device, and a program for determining synonyms of predicate pairs.

  Currently, there is a demand for high-precision search technology for searching for desired information from a large amount of text information such as blogs and voice dialogue logs on the Web, and text mining technology for automatically extracting and counting only useful information. In order to realize these, it is necessary to understand the meaning of natural sentences by a computer.

  For example, if there are two sentences: (1) “I was satisfied with the lunch of XX” and (2) “I enjoyed the lunch of XX”, these were judged as “representing the same thing”. If it is not possible, the information required by the user cannot be searched correctly or the “same information gathering” necessary for text mining cannot be performed.

  Although the sentence (1) and the sentence (2) represent the same thing, it cannot be recognized from the character string that they have the same meaning (that is, synonyms). For this reason, a failure occurs in information retrieval and information extraction required by the user. Therefore, it is necessary to understand the meaning of the natural sentence using clues other than the surface character string.

  In particular, the predicate that indicates the “how” of a sentence such as “I was satisfied” or “I was satisfied” represents the core information of the sentence. High information retrieval and information extraction are possible.

  Conventional predicate synonym judgment method, predicate that is synonymous with the elements to be combined, such as “(the economy) gets cold” and “(the economy) gets worse” A synonym determination method is proposed that includes “what happened” (ie “cool” and “deteriorate”) as predicates, “what” and “what” parts (ie "Business" is called a term.) (Non-Patent Document 1).

  In this synonym determination method, a synonym calculation is performed using a pair of a term representing “what” or “what” and a predicate, and using a distribution similarity as in “economy is getting cold”.

  Distribution similarity is based on the idea that words with similar meanings are similar in the context in which the word appears. Based on how often the element appears, it is calculated whether the word appears in a similar context.

  In addition, the above synonym determination method has a feature of another “term-predicate” or “predicate” appearing in the vicinity of the target “term-predicate”, and distribution of elements appearing in the vicinity. The synonymity is determined using the similarity.

Toshihide Shibata and Ikuo Kurohashi (2010). Acquisition of synonym relations of context-dependent predicates. IPSJ SIG Notes 2010-NL-199 (13), 1-6.

  However, in the technique described in Non-Patent Document 1 above, the meaning of two sentences, which are expressions representing the opposite of (3) “remove the shelf” and (4) “install the shelf,” There is a problem that it is determined as “similar”.

  Further, regarding two sentences (hereinafter referred to as “time-elapsed relations”) representing “continuous operation” (5) “create text” and (6) “use text”. However, there is a problem that predicates give high similarity.

  These problems occur because the distribution similarity is “determined by the similarity of context”. For example, in the technique of Non-Patent Document 1, synonymity is determined using a word that appears together with a target “term-predicate” or “predicate” in one sentence. However, an annoyance predicate can appear in exactly the same context except for the predicate, and thus it is difficult to determine by distribution similarity. Similarly, there is a predicate representing the passage of time in one sentence, such as “Precursors cooperate to create a text and use the text in the class”. There are many things. In this case, both “create text-wo” and “use text-wo” share the word “teacher-gager cooperation”, resulting in high similarity.

  As described above, in the method of Non-Patent Document 1, by using the distribution similarity that measures the similarity of words using the appearance of the surrounding words as a feature, the “anonymity relationship” and “time passage” in which the surrounding words are similar are used. Since “relation” is also determined to be “similar”, there is a problem in that predicate synonym determination cannot be performed with high accuracy.

  A word whose part of speech is a verb, adjective, adjective verb, or noun is called a “content word”, and a sentence ending expression that appears after the predicate content word, such as a particle or auxiliary verb, is a “functional expression” (Non-patent Document 6: Shun Matsuyoshi, Sato) Rishi, Takehito Utsuro (2007) Compilation of Japanese functional expression dictionary Natural language processing, vol.14, No.5, 123-146.). Predicates that are subject to synonym determination are composed of, for example, “recruiting” predicates that are “content words” “recruiting” and “functional expressions” that represent the continuation of the operation “having”. ing.

  The function expression gives an important meaning to the predicate (for example, “represents continuation of operation”), and it is necessary for synonym determination to consider the predicate. For example, (7) “recruiting supporters”, (8) “recruiting supporters” (synonymous with (7)), (9) “maybe recruiting supporters” ((7) The “recruit” and “recruit” predicates may or may not be synonymous depending on which functional expression is combined.

  In addition, the appearance pattern of the functional expression itself may be an important feature indicating whether or not the predicates are synonymous. An explanation will be given by taking “clearing” which is the content word of the predicate “kitchen is tidy” and “smooth” being the content word of the predicate “kitchen is well prepared”. FIG. 2 shows an example of the frequency of appearance of functional expressions for the content words “tuck away” and “arrange” extracted from 8 million sentences in the blog. The functional expressions that appear after “tidy up” and “arrange” often have expressions such as “having” and “having” representing “continuation”. This is a verb in which the content words “keep away” and “arrange” are inanimate to each other (for example, “room”, “on the desk”, etc.). Therefore, by adding the expression “being”, it becomes easy to express the subject “state”. On the other hand, expression of desires such as “I want to get rid of” and “I want to put it in order” do not appear much. Thus, synonymous content words tend to appear with similar functional expressions. When the predicate synonym determination is intended, the appearance pattern of the functional expression in the predicate may be an important feature for determining whether or not it is synonymous. It is necessary to consider the functional expression of the predicate.

  However, in the method described in Non-Patent Document 1 above, in order to make a synonym determination in consideration of the function expression of the predicate, the above (7) (8) (9) Each of the distribution similarities must be calculated as three different predicates: “Yes”, “Recruiting”, and “May be recruited”. In that case, since the data of the distribution similarity calculation becomes sparse (that is, the appearance frequency of each predicate is greatly reduced from the content word), it is enormous to correctly calculate the distribution similarity. There is a problem of requiring a lot of data.

  The present invention provides a synonym determination device, a synonym learning device, and a program that are made to solve the above-described problems and can determine with high accuracy whether or not a predicate pair is synonymous. The purpose is to do.

  In order to achieve the above object, the synonym determination device of the first invention provides a predicate of an input predicate pair obtained from a definition sentence set consisting of definition sentences for each of a plurality of predicates prepared in advance. The first feature is whether or not there is a predicate that is paired in the definition statement of each predicate of the preceding description pair extracted based on each definition statement, and a plurality of prescriptions prepared in advance. The second attribute is a semantic attribute common to the preceding description part pair extracted based on the semantic attributes of each predicate of the input preceding description part pair, obtained from a semantic attribute set consisting of semantic attributes for each part. A feature construction unit that extracts at least one of the first feature and the second feature, a synonym determination model storage unit that stores a previously determined synonym determination model, and the feature construction unit The input based on the extracted features Predicate pairs are is configured to include, synonymous determination unit determines whether or not synonymous.

  According to the first invention, the feature construction unit obtains the definition statements of the predicates of the input predicate pair obtained from the definition statement set including the definition statements for each of the plurality of predicates prepared in advance. The first feature is whether or not there is a pair of predicates in the definition statement of each predicate of the predescription part pair extracted based on each of the predicates prepared in advance. A semantic attribute common to the preceding description part pair extracted based on the semantic attribute of each predicate of the input preceding description part pair, obtained from a semantic attribute set consisting of semantic attributes, is defined as a second feature, and A synonym in which at least one of the first feature and the second feature is extracted, and the synonym determination unit stores a previously determined synonym determination model for determining whether the predicate pair is synonymous. The judgment model storage unit and the feature construction unit The input predicate pair is synonymous based on the synonym determination model stored in the synonym determination model storage unit based on at least one of the extracted first and second features. It is determined whether or not there is.

  In this way, the first feature is whether or not there is a paired predicate in the definition statement of each predicate of the input predicate pair, and the semantic attribute of each predicate of the input predicate pair The semantic attribute common to the second feature is the second feature, at least one of the first feature and the second feature is extracted, and it is determined whether the input predicate pair is synonymous. It is possible to determine with high accuracy whether or not a pair of parts is synonymous.

  The synonym determination device according to the second aspect of the invention provides a definition statement for each predicate of the input “term-predicate” pair obtained from a definition statement set consisting of definition statements for each of a plurality of predicates prepared in advance. Whether or not there is a paired predicate in the definition statement of each of the predicates of the “term-predicate” pair, and the predicate of the “term-predicate” pair. The first feature is whether or not at least the previous description part exists among the paired “term-predicate” terms in each of the definition statements. In the “term-predicate” pair extracted based on the semantic attribute of each predicate of the inputted “term-predicate” pair, obtained from a semantic attribute set consisting of semantic attributes for each part A common semantic attribute is a second feature, and at least one of the first feature and the second feature is extracted. The “term-predicate” pair input based on the feature construction unit, the synonym judgment model storage unit storing the synonym judgment model obtained in advance, and the feature extracted by the feature construction unit are synonymous. And a synonym determination unit that determines whether or not there is.

  According to the second invention, each predicate of the inputted “term-predicate” pair obtained from the definition sentence set including the definition sentences for each of the plurality of predicates prepared in advance by the feature construction unit. Whether or not there is a pair of predicates in the definition statement of each of the predicates of the “term-predicate” pair, and the “term-predicate” pair The first feature is whether or not there is at least a predescription part out of whether or not there is a pair of “term-predicate” in the definition statement of each predicate. The “term-predicate” extracted based on the semantic attribute of each predicate of the inputted “term-predicate” pair obtained from a semantic attribute set consisting of semantic attributes for each of a plurality of predicates. The semantic attribute common to the pair is a second feature, and at least one of the first feature and the second feature A synonym determination model storage unit in which a synonym determination model obtained in advance for determining whether or not the “term-predicate” pair is synonymous is extracted by the synonym determination unit, and the feature construction Based on the synonym determination model stored in the synonym determination model storage unit based on at least one of the first feature and the second feature extracted by the unit, the inputted “term-description” It is determined whether the “part” pair is synonymous.

  Thus, whether or not there is a paired predicate in the definition statement of each predicate of the “term-predicate” pair input, and each of the predicates of the “term-predicate” pair Whether or not there is a pair of “term-predicate” terms in the definition statement, at least the preceding description portion exists as a first feature, and the input “term-predicate” The semantic attribute common to each predicate of the pair is the second feature, and at least one of the first feature and the second feature is extracted, and the input “term-predicate” pair is synonymous. It is possible to determine with high accuracy whether or not the predicate pair is synonymous.

  The synonym determination learning device of the third invention is extracted based on the definition sentence of each predicate of the predicate pair obtained from the definition sentence set including the definition sentences for each of the plurality of predicates prepared in advance. The first feature is whether or not there is a paired predicate in the definition statement of each predicate of the previous description part pair, and the meaning consists of semantic attributes for each of a plurality of predicates prepared in advance. A plurality of predicate pairs prepared in advance, with the second feature being a semantic attribute common to the semantic attributes of each predicate of the predescription part pair obtained from the attribute set, and information indicating whether or not they are synonymous. , A feature construction unit that extracts at least one of the first feature and the second feature, and the first feature extracted from the plurality of predicate pairs by the feature construction unit and the first feature At least one of the two features and the plurality of features And a synonym determination model learning unit that learns a synonym determination model for determining whether or not the previous description unit pair is synonymous based on the information on whether or not they are synonymous with respect to a part pair. Yes.

  According to the third aspect of the present invention, the feature construction unit extracts the predicates in the predicate pair of the predicate pair obtained from the definition statement set including the definition statements for each of the plurality of predicates prepared in advance. The first feature is whether or not there is a predicate as a pair in the definition statement of each predicate of the previous description part pair, and consists of semantic attributes for each of a plurality of predicates prepared in advance. A plurality of predicates that are obtained from the semantic attribute set and have a semantic attribute common to the semantic attributes of each predicate of the predescription part pair as a second feature, and are provided with information on whether or not they are synonymous, and prepared in advance For each pair, at least one of the first feature and the second feature is extracted, and the first feature and the second feature extracted by the feature construction unit for the plurality of predicate pairs And at least one of the plurality of predicate pairs For the same meaning whether the information before written portion pairs based on of learns synonymous judgment model for judging whether or not synonymous.

  In this way, the first feature is whether or not there is a pair of predicates in the definition statement of each predicate of the predicate pair, and the semantic attribute common to the semantic attributes of each predicate of the predicate pair Is a second feature, information on whether or not it is synonymous, and for each of a plurality of predicate pairs prepared in advance, extract at least one of the first feature and the second feature, By learning a synonym determination model for determining whether or not a predicate pair is synonymous, it is possible to determine with high accuracy whether or not a predicate pair is synonymous.

  A synonym determination learning device according to a fourth aspect of the present invention is based on a definition sentence of each predicate of a “term-predicate” pair obtained from a definition sentence set including a definition sentence for each of a plurality of predicates prepared in advance. Whether there is a paired predicate in the definition statement of each predicate of the “term-predicate” pair, and each of the predicates of the “term-predicate” pair. Of the plurality of predicates prepared in advance, the first feature is whether or not there is at least a predescription part out of whether or not there is a pair of “term-predicate” terms in the definition statement. The semantic attribute that is common to the semantic attributes of the predicates of the “term-predicate” pair obtained from the semantic attribute set consisting of the semantic attributes for each is the second feature, and information on whether or not they are synonymous is attached. For each of a plurality of “term-predicate” pairs prepared in advance, the first feature and the second feature A feature construction unit that extracts at least one, at least one of the first feature and the second feature extracted by the feature construction unit for the plurality of “term-predicate” pairs, and the plurality of “ A synonym determination model learning unit that learns a synonym determination model for determining whether or not the “term-predicate” pair is synonymous based on information on whether or not the “term-predicate” pair is synonymous. And.

  According to the fourth invention, the definition statements of the predicates of the “term-predicate” pair obtained from the definition sentence set including the definition sentences for each of the plurality of predicates prepared in advance by the feature construction unit Whether or not there is a paired predicate in the definition statement of each of the predicates of the “term-predicate” pair, and each of the predicates of the “term-predicate” pair. A plurality of predicates prepared in advance with a first feature as to whether or not there is at least a predescription part of whether or not there is a pair of “term-predicate” terms in the definition statement of The semantic attribute common to each of the predicates of the “term-predicate” pair obtained from the semantic attribute set consisting of semantic attributes for each of the two is a second feature, and information on whether or not they are synonymous is attached. And the first feature and the second feature for each of a plurality of “term-predicate” pairs prepared in advance. At least one of the first feature and the second feature extracted by the feature construction unit for the plurality of “term-predicate” pairs, and the plurality of “term- A synonym determination model for determining whether or not the “term-predicate” pair is synonymous is learned based on the information on whether or not the synonym is about the predicate pair.

  Thus, whether there is a paired predicate in the definition statement of each predicate of the “term-predicate” pair, and the definition of each predicate of the “term-predicate” pair. Each of the predicates of the “term-predicate” pair has the first feature whether or not there is at least a previous description part among the paired “term-predicate” terms. The semantic attribute that is common to the semantic attributes of the second attribute is the second feature, information on whether or not it is synonymous, and the first feature and the first feature for each of a plurality of “term-predicate” pairs prepared in advance. By extracting at least one of the two features and learning a synonym determination model for determining whether the “term-predicate” pair is synonymous, the “term-predicate” pair is synonymous. Whether or not there is can be determined with high accuracy.

  Further, the synonym determination device according to the first invention extracts at least one of the first feature and the second feature, and both of the definition sentences of each predicate of the preceding description pair. The number of vocabulary appearing in is a third feature, according to the degree of detail of the semantic attribute common to the previous description part pair extracted based on the semantic attribute of each predicate of the input previous description part pair The degree of overlap of the semantic attribute of the previous description part pair with the weight added is set as the fourth feature, and the words appearing around the previous description part in the text corpus are compared for each predicate of the input previous description part pair. A functional expression of each predicate of the input previous description part pair obtained from a semantic label set consisting of a semantic label of each functional expression of a plurality of predicates prepared in advance with a distribution similarity as a fifth feature Based on the meaning label The semantic label common to the previous description part pair extracted in this way is the sixth feature, the overlapping degree of the common semantic labels of the previous description part pair is the seventh feature, and the third to seventh features At least one feature is extracted, and the synonym determination unit can determine whether or not the previous description unit pair is synonymous based on the feature extracted by the feature construction unit.

  The synonym determination device according to the second aspect of the invention extracts at least one of the first feature and the second feature, and each of the predicates of the “term-predicate” pair. The number of vocabulary appearing in both of the definition sentences is a third feature, and the “term-predicate” pair extracted based on the semantic attributes of each of the inputted “term-predicate” pairs. The fourth attribute is the overlapping degree of the semantic attributes of the “term-predicate” pair to which a weight according to the degree of detail of the semantic attributes common to the four is added, and the “term-predicate” pair of “ For each of the “term-predicate”, a distribution corpus that compares words appearing around the “term-predicate” in the text corpus, and for each predicate of the “term-predicate” pair, the text corpus Compared to the words that appear around the predicate of "term-predicate" in A distribution similarity that compares at least words that appear in the vicinity of the “term-predicate” among the similarities is defined as a fifth feature, and a meaning that includes a semantic label of each functional expression of a plurality of predicates prepared in advance. A sixth semantic label common to the “term-predicate” pair extracted from the functional expression of each of the predicates of the input “term-predicate” pair obtained from the set of labels is obtained from the label set. The common semantic label of the “term-predicate” pair is defined as a seventh feature, and at least one of the third to seventh features is extracted and the determination is performed. The means may determine whether or not the “term-predicate” pair is synonymous based on the feature extracted by the feature extraction means.

  Moreover, the program of this invention is a program for functioning a computer as each means which comprises said synonym determination apparatus or synonym learning apparatus.

  As described above, according to the synonym determination device, the synonym learning device, and the program of the present invention, it is possible to determine with high accuracy whether or not the predicate pair is synonymous.

It is a block diagram which shows the functional structure of the feature vector construction apparatus of this Embodiment. It is an example of the appearance frequency of the functional expression with respect to the content word extracted from the blog 8 million sentences. It is a figure which shows the example of the analysis result in a basic analysis part. It is a figure which shows the example of the feature extracted in a feature extraction part. It is a figure which shows the example of the feature vector constructed | assembled in the feature vector construction part. It is a block diagram which shows the functional structure of the synonymous learning device of this Embodiment. It is a figure which shows the example of a correct answer corpus. It is a figure which shows the example of the distribution similarity calculated in a distribution similarity calculation part. It is a figure which shows the example of the definition sentence extracted in a definition sentence extraction part (definition sentence mutual complementarity). It is a figure which shows the example of the definition sentence extracted in a definition sentence extraction part (overlap of vocabulary). It is a block diagram which shows the functional structure of the dictionary definition sentence feature extraction part of this Embodiment. It is a figure which shows the list of the features extracted in a dictionary definition sentence feature extraction part. It is a figure which shows the example of the feature extracted in a dictionary definition sentence feature extraction part. It is a figure which shows the example of a precaution attribute. It is a figure which shows the example of the hierarchy of prescription attribute. It is a block diagram which shows the functional structure of the semantic attribute feature extraction part of this embodiment. It is a figure which shows the example of the feature extracted in a semantic attribute feature extraction part. It is a figure which shows the example of the semantic label provided in a semantic label provision part. It is a block diagram which shows the functional structure of the function expression feature extraction part of this embodiment. It is a figure which shows the example of function expression. It is a figure which shows the example of the feature produced in a synonym determination model learning part. It is a block diagram which shows the functional structure of the synonym determination apparatus of this Embodiment. It is a figure which shows the example of the feature produced in a feature construction part. It is a flowchart which shows the feature vector construction process routine in the synonym determination model learning process routine in the feature vector construction apparatus of this Embodiment. It is a flowchart which shows the synonym determination model learning process routine in the synonym learning apparatus of this Embodiment. It is a flowchart which shows the distribution similarity calculation process routine in the synonym determination model learning process routine in the synonym learning device of this Embodiment. It is a flowchart which shows the dictionary definition sentence feature extraction process routine in the synonym determination model learning process routine in the synonym learning device of this Embodiment. It is a flowchart which shows the semantic attribute feature extraction process routine in the synonym determination model learning process routine in the synonym learning device of this Embodiment. It is a flowchart which shows the function expression feature extraction process routine in the synonym determination model learning process routine in the synonym learning device of this Embodiment. It is a flowchart which shows the synonym determination processing routine in the synonym determination apparatus of this Embodiment. It is a figure which shows the example of the result of a synonym determination, and the input feature. It is a figure which shows the example of the result of a synonym determination, and the input feature. It is a figure which shows the example of the result of a synonym determination, and the input feature. It is a figure which shows the example of the result of a synonym determination, and the input feature.

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

<Configuration of feature vector construction device>
A feature vector construction device according to an embodiment of the present invention will be described. As shown in FIG. 1, a feature vector construction apparatus 100 according to an embodiment of the present invention includes a CPU, a RAM, and a ROM that stores programs and various data for executing a feature vector construction processing routine described later. Can be configured with a computer. Functionally, the feature vector construction device 100 includes an input unit 10, a calculation unit 20, and an output unit 30 as shown in FIG.

  The input unit 10 receives a plurality of sentences written in a natural language and digitized from an input device such as a keyboard. This set of sentences is called a text corpus. Note that the input unit 10 may accept input from the outside via a network or the like.

  The calculation unit 20 includes a basic analysis unit 24, a feature extraction unit 26, and a feature vector generation unit 28.

A text corpus received by the input unit 10 is input to the basic analysis unit 24. The basic analysis unit 24 performs morphological analysis and dependency analysis for each sentence of the input text corpus, and features an analysis result including at least morpheme notation, standard form and part of speech, and dependency information for each clause. The data is output to the extraction unit 26. FIG. 3 shows an example of the output of the basic analysis unit 24 for the sentence “planting flowers and flower beds completed”. Note that existing morphological analysis and dependency analysis may be used.

  The feature extraction unit 26 uses the analysis result of each sentence input from the basic analysis unit 24, and with respect to the “term-predicate” included in each sentence, Information (context information) is extracted and output as a term predicate feature. In addition, the feature extraction unit 26 uses the analysis result of each sentence input from the basic analysis unit 24, and for the predicate included in each sentence, information (word context information) of words appearing around the predicate. Extract and output as predicate features. In the present embodiment, for example, the term predicate feature and the predicate feature are extracted by the same method as in Non-Patent Document 1 described above. Specifically, another “term-predicate” and “predicate” related to the target “term-predicate” are extracted as term predicate features. Furthermore, a term (noun phrase having a case particle) related to a single “predicate” and another “predicate” are extracted as predicate features. An example of the output of the feature extraction unit 26 in this embodiment is shown in FIG. As shown in FIG. 4, “term-predicate” “planting” and “term-predicate” “flower-wo-planting” are different from “term-predicate” “flowerbed-ga-complete”. As a term predicate feature. For the predicate “complete”, the term “flowerbed-ga” and the predicate “plant” are extracted as predicate features.

  The feature vector generation unit 28 uses the analysis result of each sentence input from the basic analysis unit 24 to extract “term-predicate” included in each sentence as “term-predicate” of the feature vector construction target. . Then, the feature vector generation unit 28 uses, as elements, values calculated using the term predicate features input from the feature extraction unit 26 for each “term-predicate” of the extracted feature vector construction target. Is output to the output unit 30. Also, the feature vector generation unit 28 extracts predicates included in each sentence as feature vector construction target predicates using the analysis result of each sentence input from the basic analysis unit 24. Then, the feature vector generation unit 28 constructs a feature vector whose element is a value calculated using the predicate feature input from the feature extraction unit 26 for each extracted predicate of the feature vector construction target, and outputs the feature vector. To the unit 30. In the present embodiment, for example, a feature vector is constructed by the same method as in Non-Patent Document 1. Specifically, a feature vector having a weight value calculated on the basis of the mutual information (MI) of each term predicate feature and the “term-predicate” of the feature vector construction target is constructed. . In addition, a feature vector having a weight value calculated based on a predicate to be feature vector construction target and a mutual information (MI) of each predicate feature as an element value is constructed. The weight is calculated using the following formula (1). The mutual information (MI) is calculated using the following equation (2).

  A feature vector when the feature vector construction target is “term-predicate” is called a term predicate feature vector. When the feature vector construction target is “term-predicate”, u represents “term-predicate”, and f represents the term predicate feature. P (u) is the probability that the “term-predicate” of the feature vector construction target appears in the text corpus, P (f) is the probability that the term predicate feature appears in the text corpus, and P (u, f) is the feature Represents the probability that the “term-predicate” and term predicate features appear simultaneously. When MI is greater than 0, the value of weight is 1. When MI is 0 or less, the value of weight is 0. An example of the term predicate feature vector constructed in the table in FIG. 5 is shown. In the example, when u is “flowerbed-ga-complete” and f is “flower-wo-plant”, it indicates that MI is greater than 0 and the weight value is 1. Further, when u is “flowerbed-ga-finished” and f is “time-over”, it indicates that MI is 0 or less and the weight value is 0.

  A feature vector when a feature vector construction target is a predicate is called a predicate feature vector. When the feature vector construction target is a predicate, u represents a predicate, and f represents a predicate feature. P (u) is the probability that the predicate of the feature vector construction target appears in the text corpus, P (f) is the probability that the predicate feature appears in the text corpus, and P (u, f) is the description of the feature vector construction target. Represents the probability that a part and predicate feature will appear simultaneously. When MI is greater than 0, the value of weight is 1. When MI is 0 or less, the value of weight is 0. An example of the predicate feature vector constructed in the lower table of FIG. 5 is shown. In the example, when u is “complete” and f is “flowerbed-ga”, MI is greater than 0 and the weight value is 1. Further, when u is “completed” and f is “house-ga”, it indicates that MI is 0 or less and the weight value is 0.

As described above, in the present embodiment, the predicate feature vector in which each predicate included in the input text corpus is a construction target of the feature vector, and each “term-predicate” included in the input text corpus are included. Two types of term predicate feature vectors that are feature vectors to be constructed (hereinafter referred to as “feature vectors” together) are created.
In the present embodiment, both the predicate feature vector and the term predicate feature vector are created. However, only the feature vector used in the synonym learning device and synonym determination device described later may be created.

<Configuration of synonym learning device>
Next, the configuration of the synonym learning device according to the embodiment of the present invention will be described. As shown in FIG. 6, a synonymous learning apparatus 200 according to an embodiment of the present invention includes a CPU, a RAM, and a ROM that stores a program for executing a determination model learning process routine (to be described later) and various data. Can be configured. Functionally, the synonym learning device 200 includes an input unit 110, a calculation unit 120, and an output unit 150 as shown in FIG.

  The input unit 110 receives a plurality of predicate pairs and a plurality of “term-predicate” pairs to which information indicating whether or not synonyms are given from an input device such as a keyboard. A plurality of predicate pairs and a plurality of “term-predicate” pairs to which information indicating whether or not they are synonymous are provided are called correct corpus. Note that the input unit 110 may accept input from the outside via a network or the like.

  The calculation unit 120 includes a feature vector storage unit 124, a distribution similarity calculation unit 132, a definition sentence dictionary storage unit 134, a dictionary definition sentence feature extraction unit 136, a semantic attribute dictionary storage unit 138, and a function expression dictionary storage unit. 139, a semantic attribute feature extraction unit 140, a function expression feature extraction unit 141, a feature set storage unit 142, and a synonym determination model learning unit 144.

  The correct corpus 122 (described in FIG. 11) is a predicate pair and a “term-predicate” pair, which are received by the input unit 110 and are manually assigned semantic relationship information “synonymous” as illustrated in FIG. A set of correct data (positive example) and a set of correct answer data (negative example) that is a predicate pair and a “term-predicate” pair to which the semantic relationship information “annoyance” or “others” is attached.

  The feature vector storage unit 124 stores the feature vector output by the feature vector construction device 100.

  The distribution similarity calculation unit 132 uses the corresponding feature vectors obtained from the feature vector storage unit 124 for all the predicate pairs or “term-predicate” pairs of the input correct corpus 122, respectively. The distribution similarity is calculated, and the calculation result is output to the feature set storage unit 142 together with the predicate pair or “term-predicate” pair to be calculated. A distribution similarity calculated using the predicate feature vector is called a predicate distribution similarity, and a distribution similarity calculated using the term predicate feature vector is called a term predicate distribution similarity. When calculating the distribution similarity for a predicate pair, only the information about the predicate is used, so only the predicate distribution similarity is calculated, but the distribution similarity for the “term-predicate” pair. When calculating, it is possible to use both predicate distribution similarity and predicate distribution similarity because it is possible to use not only information on "term-predicate" but also information on predicates. . The calculation of the distribution similarity is a process of calculating the distribution similarity using at least one of the predicate feature vector and the term predicate feature vector obtained from the text corpus by the feature vector construction device 100. In this embodiment, the distribution similarity is calculated by the same method as in Non-Patent Document 1. Specifically, the distribution similarity is calculated using the following equations (3) to (5), and the measurement in equation (3) is the distribution similarity.

However, when calculating the distribution similarity using the term predicate feature vector, the numerator of the JACCARD coefficient in the above equation (4), when comparing two term predicate feature vectors element by element, This is the number of elements in which the value of the element of the term predicate feature vector is 1 and the value of the element of the other term predicate feature vector is also 1. Further, the denominator of the JACCARD coefficient is obtained by comparing the elements of one term predicate feature vector and the other term predicate feature vector when the two term predicate feature vectors of the pair used for calculating the distribution similarity are compared for each element. The number of elements in which the value of at least one of the elements is 1.

In addition, the numerator of the SIMPSON coefficient in the above equation (5) is the same as the numerator of the JACCARD coefficient, and the denominator of the SIMPSON coefficient is the element whose element value is 1 in the pair term predicate feature vector used for calculation. This is the smaller number of the number and the number of elements whose element value is 1 in the other term predicate feature vector.

However, the numerator of the JACCARD coefficient in the above equation (4), when calculating the distribution similarity using the predicate feature vector, when comparing two predicate feature vectors element by element, This is the number of elements whose element value of the feature vector is 1 and whose element value of the other predicate feature vector is also 1. In addition, the denominator of the JACCARD coefficient is obtained by comparing the elements of one predicate feature vector and the elements of the other predicate feature vector when the two predicate feature vectors of the pair used for calculating the distribution similarity are compared for each element. This is the number of elements in which the value of at least one element is 1.

In addition, the numerator of the SIMPSON coefficient in the above equation (5) is the same as the numerator of the JACCARD coefficient, and the denominator of the SIMPSON coefficient is the number of elements whose element value is 1 in the pair predicate feature vector used for calculation. And the smaller of the number of elements whose element value is 1 in the other predicate feature vector.

  In the present embodiment, both the predicate distribution similarity and the term predicate distribution similarity are used. FIG. 8 shows an example of the distribution similarity of “flowerbed-ga-complete” and “flowerbed-ga-complete”. The upper part of the table of FIG. 8 is an example of the term predicate distribution similarity, and the lower part is an example of the predicate distribution similarity. The predicate distribution similarity and the term predicate distribution similarity are collectively referred to as distribution similarity. The distribution similarity is an example of a fifth feature.

<Configuration of dictionary definition sentence feature extraction unit>
The dictionary definition sentence feature extraction unit 136 stores the content words of all predicate pairs in the input correct corpus or the content words of all “term-predicate” pairs in the definition sentence dictionary storage unit 134. Based on the definition sentence dictionary, a feature indicating “definition sentence mutual complementarity” and a feature indicating “overlapping vocabulary” are extracted and output to the feature set storage unit 142 together with a pair to be extracted. The definition sentence dictionary is composed of one or more definition sentences corresponding to the content words of the plurality of predicates, and each set of one or more definition sentences extracted from the definition sentence dictionary is called a definition sentence set. Note that a feature indicating definition sentence mutual complementarity is an example of a first feature, and a feature indicating overlapping of vocabularies is an example of a third feature. The first feature and the third feature are collectively referred to as a dictionary definition sentence feature.

  The dictionary definition sentence feature extraction unit 136 shows "definition sentence mutual complementarity" and "vocabulary overlap" from the feature that "synonymous predicates have similarities in mutual definition sentence sets that explain the meaning". By extracting the features, it is possible to avoid the problem of “determining a predicate that expresses the opposite in the same sense is synonymous” that was a problem in the conventional method.

  Here, “definition statement mutual complementarity” means that one's predicate appears in the definition statement set of the other's predicate, and is synonymous with “complete” and “complete” shown in FIG. For example, the predicate “complete” appears in the synonymous “finished” dictionary definition sentence set, and the predicate “completed” is synonymous with “completed”. Appears in the dictionary definition sentence set.

  Further, “overlapping vocabulary” means that vocabulary overlaps between definition sentence sets, and is a statement having two synonyms of “price-ga-high” and “price-ga-high” shown in FIG. For example, the term “price” appears in both definition sentence sets.

  In this way, there is a feature that “synonymous predicates have similarities in mutual definition sentence sets that explain the meaning”, and this feature is expressed in the form of “definition mutual complementarity” or “vocabulary overlap”. Can be expressed. For the sake of explanation, the first “term-predicate” predicate (for the “term-predicate” pair such as “flowerbed-ga-finish” and “flowerbed-ga-complete”) ( That is, “complete”) is Pred1, “term” (ie, “flowerbed”) is Arg1, the second “term-predicate” predicate (ie, “completed”) is Pred2, and term (ie, “complete”). “Flowerbed”) is designated Arg2. Similarly, for predicate pairs such as “completed” and “completed”, the first predicate is Pred1, and the second predicate is Pred2.

  The detailed configuration of the dictionary definition sentence feature extraction unit 136 is shown in FIG. The dictionary definition sentence feature extraction unit 136 includes a definition sentence extraction unit 1360, a definition sentence mutual complementarity extraction unit 1362, and a vocabulary overlap extraction unit 1364.

  The definition sentence extraction unit 1360 stores the definition stored in the definition sentence dictionary storage unit 134 for all the predicate pairs of the input correct corpus or the content words of each predicate of the “term-predicate” pair. The dictionary lookup of the sentence dictionary is performed, and the definition sentence set of each predicate is extracted for each predicate pair or “term-predicate” pair. The extracted definition sentence set is subjected to morphological analysis, and an analysis result including at least the notation, the standard form, the part of speech, and the reading for each morpheme is output to the definition sentence mutual complementarity extraction unit 1362 and the vocabulary overlap extraction unit 1364. .

The definition sentence mutual complementarity extraction unit 1362 defines, for each extraction target pair (predicate pair or “term-predicate” pair), from the result of the morphological analysis of the definition sentence set input from the definition sentence extraction unit 1360. A feature indicating sentence mutual complementarity is extracted and output to the feature set storage unit 142 together with the pair to be extracted. Specifically, whether Pred2 appears in the definition sentence set of Pred1 or whether Pred1 appears in the definition sentence set of Pred2 is extracted by character string matching. Whether Pred2 appears in the definition sentence set of Pred1 is Pred1Match, and whether Pred1 appears in the definition sentence set of Pred2 is Pred2Match. In the present embodiment, the feature value is set to 1 when it appears. Similarly, whether Arg2 has appeared in the definition sentence set of Pred1 or whether Arg1 has appeared in the definition sentence set of Pred2 is extracted. Arg1Match indicates whether Arg2 appears in the definition sentence set of Pred1, and Arg2Match indicates whether Arg1 appears in the definition sentence set of Pred2. Note that these values may include real numbers such as the number of overlaps and the value obtained by normalizing the number of overlaps with the total number of words in the definition sentence set.
In the present embodiment, all of Pred1Match, Pred2Match, Arg1Match, and Arg2Match are used as the first feature. However, when using the first feature, Arg1Match and Arg2Match may not be used.

  The vocabulary overlap extraction unit 1364 extracts, for each extraction target pair (predicate pair or “term-predicate” pair), the extraction target pair from the result of the morphological analysis of the definition sentence set input from the definition sentence extraction unit 1360. A feature indicating whether there is a vocabulary overlap between the definition sentence sets of the predicates is output to the feature set storage unit 142 together with the pair to be extracted. In the present embodiment, the number of vocabulary appearing in common in both definition sentence sets is used as a feature, and the vocabulary parts of speech (nouns, non-independence excluding independence, adjectives, adjective verbs and four types of parts of speech) are aggregated. . Here, NounMatch is the number of vocabularies with part-of-speech nouns that appear in both definition sets, VerbMatch is the number of vocabularies with part-of-speech parts, AdjMatch is the number of vocabularies with part-of-speech parts, and adjective verbs The number of vocabularies with parts of speech is called AdjNMatch (FIG. 12). In the present embodiment, all of NounMatch, VerbMatch, AdjMatch, and AdjNMatch are used as the third feature. However, at least one feature is sufficient. It should be noted that the presence or absence of vocabulary that appears in both definition sentence sets is not the number of vocabularies that appear in both definition sentence sets, and may be aggregated for each part of speech of the vocabulary. In this case, the value may be set to 1 and the value may be set to 0 if none. Furthermore, a value obtained by normalizing the number of vocabulary appearing in common in both definition sentence sets with the total number of words in the definition sentence set may be used. FIG. 13 shows an example of the feature list created.

  The definition sentence dictionary storage unit 134 stores a definition sentence dictionary that stores a definition sentence set corresponding to each of the plurality of predicates. As the definition sentence dictionary, an existing national language dictionary or a free dictionary on the Web that has been added and edited by a plurality of users may be used. The definition sentence dictionary is an example of a definition sentence set.

<Configuration of semantic attribute feature extraction unit>
The semantic attribute feature extraction unit 140 extracts the abstract semantic attributes of the content words of the predicates of the pair for each of all the pairs of correct corpus inputted (predicate pair or “term-predicate” pair). Are extracted and output to the feature set storage unit 142 together with the pair to be extracted. In the present embodiment, two features of “overlapping word attribute” and “separation ratio with semantic attribute weight”, which will be described later, are extracted as features indicating the overlapping of abstract semantic attributes. In this embodiment, a prescriptive attribute is used as an abstract semantic attribute of the predicate to be extracted. The semantic attribute dictionary is composed of one or more prescriptive attributes corresponding to each of the plurality of predicates, and each set of one or more prescriptive attributes extracted from the semantic attribute dictionary is called a prescriptive attribute set. A prescriptive attribute that appears in common in the prescriptive attribute set of both predicates is extracted as a feature of the “overlapping prescriptive attribute”. In addition, a “semantic attribute weighted overlap rate” calculated by assigning a weight to a hierarchy to which a prescriptive attribute that appears in common in both can belong can be extracted as a feature. The semantic attribute feature extraction unit 140 outputs these two features together with the extraction target pair to the feature set storage unit 142.

  The semantic attribute feature extraction unit 140 allows the abstract semantic attributes of the predicates themselves to overlap with the hierarchical overlap because of the feature that the synonymous predicates have similar abstract semantic attributes. By extracting “feature” in consideration of “depth”, it is possible to avoid the problem of “determining a predicate representing the passage of time erroneously as synonymous”, which was a problem in the conventional method.

  In this embodiment, as an abstract semantic attribute, “prescriptive attribute” (Non-patent Document 2: Satoru Ikehara, Masahiro Miyazaki, Satoshi Shirai, Akio Yokoo, Hiroaki Nakaiwa, Kentaro Ogura, Yoshifumi Oyama, Yoshihiko Hayashi (1999) Japanese vocabulary large CD-ROM version. Iwanami Shoten.) Is used. An example of the predicate attribute set is shown in FIG. 14 with an example where the predicate is “completed” and “completed”. As FIG. 15 shows, these attributes often have a hierarchical structure, and the attributes are further refined as the hierarchy progresses downward. For example, a more detailed attribute such as “ownership movement” is specified for an upper attribute “action” via an intermediate attribute “physical behavior”.

  A detailed configuration of the semantic attribute feature extraction unit 140 is shown in FIG. The semantic attribute feature extraction unit 140 includes a semantic attribute overlap extraction unit 1400 and a semantic attribute weighted overlap rate calculation unit 1402.

  The semantic attribute overlap extraction unit 1400, for each of all the input correct corpus pairs (predicate pair or “term-predicate” pair), is a prescription that is a semantic attribute of the content word of each predicate of the pair. The attribute set is extracted from the semantic attribute dictionary, the prescriptive attributes appearing in both of the paired prescriptive attribute sets are extracted as features, and are output to the semantic attribute weighted overlap rate calculating unit 1402 together with the extraction target pairs. In FIG. 17, there is an overlap from a prescriptive attribute set (attribute change, generation) for “completed” of “flowerbed-ga-complete” and a prescriptive attribute set (generation) for “completed” of “flowerbed-ga-finished”. An example in which “generation” is extracted as a prescriptive attribute is shown. Note that the overlap prescriptive attribute is an example of the second feature.

  The semantic attribute weighted overlap rate calculation unit 1402 uses all of the prescriptive attributes input from the semantic attribute overlap extraction unit 1400 and the semantic attribute dictionary for each pair (predicate pair or “term-predicate” pair). Based on the hierarchical information of the word attributes, a “semantic attribute weighted overlap rate” as a feature indicating the degree of overlapping of the prescriptive attributes is calculated according to the following expressions (6) and (7). Along with the input feature and extraction target pair, it is output to the feature set storage unit 142. The two features of overlapping prescriptive attributes and semantic attribute weighted overlap rates are characterized by the characteristics that "the more attributes are shared, the more predicates are similar" and "the more detailed attributes are shared , The predicates are similar to each other. The semantic attribute weighted overlap rate is an example of a fourth feature. The second feature and the fourth feature are collectively referred to as a semantic attribute feature.

  For example, when calculating the weighted overlap ratio of “flowerbed-ga-completed” and “flowerbed-ga-completed”, both “completed” predicate attribute set and “completed” predicate attribute set The prescriptive attribute appearing in is a prescriptive attribute of “generation”. Furthermore, since “generation” is the most detailed attribute of the hierarchy 4 as shown in FIG. 15, the weighted overlap ratio is calculated as in the following equation (8).

  The semantic attribute dictionary storage unit 138 stores a semantic attribute dictionary that stores semantic attributes corresponding to the content words of the plurality of predicates. In the present embodiment, a prescriptive attribute dictionary is stored as a semantic attribute dictionary. The semantic attribute dictionary is an example of a semantic attribute set.

<Configuration of functional expression feature extraction unit>
The functional expression feature extraction unit 141 extracts, for each of all pairs of the correct corpus (predicate pair or “term-predicate” pair), a feature indicating the overlapping of the meaning of the functional expression of the predicate of the pair, The result is output to the feature set storage unit 142 together with the pair to be extracted. In the present embodiment, two features of “overlapping semantic label” and “semantic label overlapping rate” to be described later are extracted as features indicating the overlapping of meanings of function expressions.

  The feature expression feature extraction unit 141 extracts features indicating the overlap of the function expressions of the predicates, thereby increasing the predicate synonym determination even when the distribution similarity is calculated using only the content words of the predicates. Can be done with precision. This requires a huge amount of data to calculate the distribution similarity, which is a combination of individual content words and functional expressions, which occurs when synonymous determination is performed using functional distribution only by using distribution similarity. Can be avoided.

  FIG. 19 shows a detailed configuration of the functional expression feature extraction unit 141. The functional expression feature extraction unit 141 includes a semantic label assignment unit 1500, an overlapping meaning label extraction unit 1502, and a semantic label overlap rate calculation unit 1504.

  The semantic label assigning unit 1500 obtains, from the functional expression dictionary, the functional label of the functional expression of each predicate of the pair for each of all the input correct corpus pairs (predicate pair or “term-predicate” pair). It is extracted and output to the overlapping meaning label extraction unit 1502 together with the pair. In this embodiment, a statistical meaning labeling method is used (Non-patent Document 7: Kenji Imamura, Atsuko Izumi, Genichiro Kikui, Satoshi Sato (2011). Semantic Label of Predicate Functional Representation Language Processing Society 17th Annual Tournament. 518-521). Specifically, morphological analysis is performed on each predicate of all pairs of correct corpus that have been input, and the most likely semantic label string is assigned using an analysis result including at least the notation for each morpheme and the standard form and part of speech. To do. FIG. 20 shows an example of a function expression dictionary. FIG. 18 shows an example in which a meaning label column is given when a “term-predicate” pair “flowerbed-ga-completed” and “flowerbed-ga-completed” is input. In the example shown in FIG. 18, the predicates are “completed” and “completed”, respectively, and the content word part and the function expression part are identified, and “completed” is added to “ta” of each function expression. The meaning label is given.

  In this embodiment, the content label part and the function expression part of the predicate are identified by the semantic label assigning unit 1500. However, it is also possible to perform preprocessing that leaves only the function expression that affects the meaning of the event. Good. (Non-Patent Document 8: Izumi T., Imamura K., Kikui G., & Sato S. (2010). Standardizing Complex Functional Expressions in JapansesePredicates: Applying Theoretically-Based Paraphrasing Rules. Proceedings of the Workshop on Multiword Exressions: From theory to applications (MWE 2010), 63-71)

  The overlapping meaning label extraction unit 1502 extracts, as overlapping meaning labels, meaning labels that appear in common in the meaning labels of both predicates from the extraction target pairs input from the meaning label assignment unit 1500 and the respective meaning labels. , And output to the semantic label overlap calculator 1504 together with the pair. In the example of the “term-predicate” pair “flowerbed-ga-completed” and “flowerbed-ga-completed”, the meaning label “completed” and predicate “completed” mean the predicate “completed”. From the label “completed”, “completed” is extracted as the overlapping meaning label. Note that the overlapping meaning label is an example of a sixth feature.

  The semantic label overlap rate calculation unit 1504 calculates a semantic label overlap rate according to the following equation (9) based on the extraction target pair and the overlap meaning label input from the overlap meaning label extraction unit 1502, and the pair and overlap meaning The calculation result together with the label is output to the feature set storage unit 142. The semantic label overlap rate is an example of a seventh feature. The sixth feature and the seventh feature are collectively called a function expression feature.

  The function expression dictionary storage unit 139 stores a function expression dictionary that stores semantic labels for function expressions of a plurality of predicates. In the present embodiment, the dictionary of Non-Patent Document 6 is used, but the present invention is not limited to this, and other sentence ending expression dictionaries and modality expression dictionaries may be used.

  In the feature set storage unit 142 of FIG. 6, each pair (predicate pair) obtained by the distribution similarity calculation unit 132, the dictionary definition sentence feature extraction unit 136, the semantic attribute feature extraction unit 140, and the function expression feature extraction unit 141 is stored. Alternatively, each feature of “term-predicate” pair) and a pair of feature extraction targets are input, and the feature set storage unit 142 stores each input feature for each pair.

  The synonym determination model learning unit 144 in FIG. 6 calculates each feature input from the feature set storage unit 142 for each of all the pairs of correct corpus that are input (predicate pair or “term-predicate” pair). Originally, the synonym determination model is learned, and the learned synonym determination model is output to the output unit 150. SVM is used for learning the synonym determination model (Non-patent Document 3: Chang, C.-C. and Lin, C.-J. (2011). LIBSVM: a library for support vector machines. ACM Transactions on Intelligent Systems and Technology, 2:27, 1-27.).

  Specifically, a predicate pair or “term-predicate” pair that is synonymous is “positive example (+1 in SVM)”, and a non-synonymous predicate pair or “term-predicate” pair is “negative”. As an example (-1 in SVM) ", the synonym determination model is learned using at least one of the predicate distribution similarity and the term predicate distribution similarity, and the dictionary definition sentence feature, semantic attribute feature, and functional expression feature. FIG. 21 shows a list of features for “flowerbed-ga-complete” and “flowerbed-ga-complete”.

<Configuration of synonym determination unit>
Next, the configuration of the synonym determination device 300 according to the embodiment of the present invention will be described in detail. As shown in FIG. 22, a synonym determination device 300 according to an embodiment of the present invention is a computer including a CPU, a RAM, and a ROM that stores a program for executing a synonym determination processing routine described later and various data. Can be configured. Functionally, the synonym determination device 300 includes an input unit 210, a calculation unit 220, and an output unit 250 as shown in FIG.

  The input unit 210 receives a synonym determination target pair (predicate pair or “term-predicate” pair) from an input device such as a keyboard. Note that the input unit 210 may accept input from the outside via a network or the like.

  The calculation unit 220 includes a feature vector storage unit 224, a definition sentence dictionary storage unit 226, a semantic attribute dictionary storage unit 228, a function expression dictionary storage unit 229, a feature construction unit 222, a synonym determination unit 230, and a synonym determination. It can be expressed by a configuration including the model storage unit 232.

  The feature vector storage unit 224 stores feature vectors output from the feature vector construction device 100.

  The feature construction unit 222 uses the same type of “distribution similarity” and “dictionary” as the synonym learning device 200 for the synonym determination target pair (predicate pair or “term-predicate” pair) received by the input unit 210. The “definition sentence feature”, “semantic attribute feature”, and “functional expression feature” are extracted and output to the synonym determination unit 230. FIG. 23 shows an example of the extracted features.

  The synonym determination unit 230 is output from the synonym learning device 200 based on the feature input from the feature construction unit 222 and stored in the synonym determination model storage unit 232 in which the synonym determination model obtained in advance is stored. Based on the determination model, the SVM classifier is used to determine whether the input synonym determination target pair (predicate pair or “term-predicate” pair) is “synonymous” and outputs the result to the output unit 250. .

<Operation of feature vector construction device>
Next, the operation of the feature vector construction device 100 according to the embodiment of the present invention will be described. First, a text corpus is input by the input unit 10. Then, when the CPU executes the program stored in the ROM of the feature vector construction device 100, the feature vector construction processing routine shown in FIG. 24 is executed.

  First, in step S100, a text corpus that is a set of a plurality of sentences is read.

  First, in step S102, morphological analysis is performed on one sentence among the plurality of sentences received in step S100.

  Next, in step S104, dependency analysis is performed on the sentence for which morphological analysis was performed in step S100.

  Next, in step S106, the “term-predicate” included in the sentence from the analysis result including at least the notation for each morpheme, the standard type, the part of speech, and the dependency information for each clause obtained in step S102 and step S104. Information (context information) of words appearing around "is extracted as a term predicate feature. In step S106, the morpheme notation obtained in steps S102 and S104, the standard type, the part of speech, and the analysis result including at least dependency information for each clause appear around the predicate included in the sentence. Extract word information (context information) as predicate features. In the present embodiment, for example, the term predicate feature and the predicate feature are extracted by the same method as in Non-Patent Document 1 described above. Specifically, another “term-predicate” and “predicate” related to the target “term-predicate” are extracted as term predicate features. Furthermore, a term (noun phrase having a case particle) related to a single “predicate” and another “predicate” are extracted as predicate features.

  Next, in step S108, it is determined whether or not the processing in steps S102 to S106 has been executed for all of the plurality of sentences received in step S100. When the processes of steps S102 to S106 are executed for all sentences, the process proceeds to step S110, and when there is a sentence that does not execute the processes of steps S102 to S106, the process proceeds to step S102. Repeat the process.

  Next, in step S110, a feature vector is constructed based on the features extracted in step S106 for the “term-predicate” included in at least one sentence. In step S110, a feature vector is constructed based on the features extracted in step S106 for the predicates included in at least one sentence.

  Next, in step S112, it is determined whether or not a term predicate feature vector has been constructed for each of all “term-predicate” included in at least one sentence read in step S100. In step S112, it is determined whether a predicate feature vector has been constructed for each of all predicates included in at least one sentence read in step S100. If a term predicate feature vector is constructed for each of all “term-predicates”, the process proceeds to step S114, and there is a “term-predicate” in which no term predicate feature vector is constructed. Moves to step S110 and repeats each process. If a predicate feature vector is constructed for each of all predicates, the process proceeds to step S114. If there is a predicate that does not construct a predicate feature vector, the process proceeds to step S110. Repeat each process.

  Next, in step S114, all the feature vectors constructed in step S110 are output by the output unit 30, and the process is terminated.

In the present embodiment, both the predicate feature vector and the term predicate feature vector are created. However, only the feature vector used in the synonym learning device and the synonym judgment device may be created.

<Operation of synonym learning device>
Next, the operation of the synonym learning device 200 according to the embodiment of the present invention will be described. First, the feature vector output by the feature vector construction device 100 is input by the input unit 110 and stored in the feature vector storage unit 124. A correct corpus is input by the input unit 110. Then, when the CPU executes the program stored in the ROM of the synonym learning device 200, the synonym determination model learning process routine shown in FIG. 25 is executed.

  First, in step S202, a correct corpus is read.

In the subsequent steps, when learning a synonym determination model for predicate pairs, processing is performed using only the correct answer data for the predicate pair in the read correct corpus, and the synonym for the “term-predicate” pair. When learning a decision model, processing is performed using only correct data of “term-predicate” pairs in the read correct corpus.

  Next, in step S204, for each of the plurality of correct answer data obtained in step S202, each of the correct answer data pair (predicate pair or “term-predicate” pair) stored in the feature vector storage unit 124. Based on the predicate feature vector, the predicate distribution similarity of the correct data pair is calculated. Further, in step S204, for each of the plurality of correct answer data obtained in step S202, based on each term predicate feature vector of the “term-predicate” pair of the correct answer data stored in the feature vector storage unit 124. Then, the term predicate distribution similarity of the correct answer data pair is calculated. In this embodiment, the predicate distribution similarity is calculated for the “term-predicate” pair, but only the term predicate distribution similarity may be calculated.

  Next, in step S206, for each of the plurality of correct answer data obtained in step S202, the dictionary definition sentence feature of the correct answer data pair (predicate pair or “term-predicate” pair) is extracted.

  Next, in step S208, for each of the plurality of correct answer data obtained in step S202, a semantic attribute feature of the correct answer data pair (predicate pair or “term-predicate” pair) is extracted.

  Next, in step S209, for each of the plurality of correct answer data obtained in step S202, the function expression feature of the correct answer data pair (predicate pair or “term-predicate” pair) is extracted.

  Next, in step S210, the distribution similarity obtained in step S204, the dictionary definition sentence feature obtained in step S206, the semantic attribute feature obtained in step S208, and the functional expression feature obtained in step S209. And a pair of correct data (predicate pair or “term-predicate” pair) for each of a plurality of correct data based on the information on whether or not the plurality of correct data are synonymous obtained in step S202. Learning data of positive examples or negative examples including all the features extracted for is generated.

  Next, in step S212, the synonym determination model is learned based on each learning data created in step S210, the synonym determination model is output by the output unit 150, and the process ends.

  Step S204 is realized by the distribution similarity calculation routine shown in FIG.

  First, in step S250, for each predicate pair of correct answer data to be calculated, each predicate feature vector of the pair stored in the feature vector storage unit 124 is read. In step S250, for the “term-predicate” pair of correct data to be calculated, the predicate feature vector and each term predicate feature vector of the pair stored in the feature vector storage unit 124 are read. In the present embodiment, the predicate feature vector is read for the “term-predicate” pair of the correct data to be calculated.

  Next, in step S252, based on the feature vector read in step S250, the distribution similarity is calculated for the correct data pair (predicate pair or “term-predicate” pair).

  Next, in step S254, it is determined whether or not the distribution similarity of the correct data pair has been calculated for each of the correct data read in step S202. If the distribution similarity is calculated for each pair of all correct data (predicate pair or “term-predicate” pair), the process is terminated, and the correct data pair for which distribution similarity is not calculated If there is a (predicate pair or “term-predicate” pair), the process proceeds to step S250, and each process is repeated using the correct answer data as correct answer data to be calculated.

  Step S206 is realized by the dictionary definition sentence feature extraction routine shown in FIG.

  First, in step S300, a definition sentence set of each predicate of a pair of correct data to be extracted (predicate pair or “term-predicate” pair) is extracted from the definition sentence dictionary storage unit 134.

  Next, in step S302, morphological analysis is performed on each definition sentence of the definition sentence set extracted in step S300.

  Next, in step S304, based on the result of the morpheme analysis including at least the notation, the standard form, the part of speech, and the reading for each morpheme obtained in step S302, the feature indicating the definition sentence mutual complementarity is extracted. Pred1Match and Pred2Match are extracted as features indicating definition sentence mutual complementarity. If the pair of correct data to be extracted is a “term-predicate” pair, Arg1Match and Arg2Match can be further extracted.

  Next, in step S306, based on the result of morpheme analysis including at least the notation, the standard form, the part of speech, and the reading for each morpheme obtained in step S302, a feature indicating vocabulary overlap is extracted. At least one of NounMatch, VerbMatch, AdjMatch, and AdjNMatch is extracted as a feature indicating vocabulary overlap.

  Next, in step S308, a dictionary definition sentence feature is constructed based on the feature indicating the mutual definition of the definition sentences extracted in step S304 and the feature indicating the overlap of the vocabulary extracted in step S306.

  Next, in step S310, it is determined whether a dictionary definition sentence feature has been constructed for each pair (predicate pair or “term-predicate” pair) of all correct data read in step S202. When the dictionary definition sentence features are constructed for all correct data pairs (predicate pairs or “term-predicate” pairs), the processing is terminated, and the correct data pairs that have not constructed dictionary definition sentence features ( If there is a predicate pair or “term-predicate” pair), the process proceeds to step S300, and each process is repeated with the correct answer data as the correct answer data to be extracted.

  The step S208 is realized by the semantic attribute feature extraction routine shown in FIG.

  First, in step S400, the overlapping prescriptive attributes of each predicate of a pair of correct data to be extracted (predicate pair or “term-predicate” pair) are extracted.

  Next, in step S402, based on the overlap prescriptive attributes obtained in step S400, the semantic attribute weighted overlap rate is calculated for the correct answer data pair (predicate pair or “term-predicate” pair) to be extracted. To do.

  Next, in step S404, based on the overlapping prescriptive attribute obtained in step S400 and the semantic attribute weighted overlapping rate obtained in step S402, the pair of correct data to be extracted (predicate pair or “term” -Build semantic attribute features for predicate pairs.

  Next, in step S406, it is determined whether or not a semantic attribute feature has been constructed for each pair (predicate pair or “term-predicate” pair) of all correct data read in step S202. When the semantic attribute feature is constructed for each pair of all correct data (predicate pair or “term-predicate” pair), the processing is terminated, and the correct data pair for which no semantic attribute feature is constructed If there is a (predicate pair or “term-predicate” pair), the process proceeds to step S400, and each process is repeated with the correct data as the correct data to be extracted.

  The step S209 is realized by the function expression feature extraction routine shown in FIG.

  First, in step S600, a morpheme analysis of each predicate of a pair of correct data to be extracted (predicate pair or “term-predicate” pair) is performed, and an analysis including at least a notation, a standard form, and a part of speech for each morpheme. The most likely meaning label string is assigned using the result.

  Next, in step S602, overlapping semantic labels that appear in common in both semantic labels of the predicates of each pair obtained in step S600 are extracted.

  Next, in step S604, a semantic label overlap rate is calculated using the paired overlapping semantic labels obtained in step S602.

  Next, in step S606, based on the overlapping semantic label obtained in step S602 and the semantic label overlapping rate obtained in step S604, the correct data pair (predicate pair or “term-predicate” to be extracted) is extracted. “Pairs” are constructed with functional expression features.

  Next, in step S608, it is determined whether or not a function representation feature has been constructed for each pair (predicate pair or “term-predicate” pair) of all correct data read in step S202. When the functional expression feature is constructed for each pair of all correct data (predicate pair or “term-predicate” pair), the processing is terminated, and the correct data pair for which no functional representation feature is constructed If there is a (predicate pair or “term-predicate” pair), the process proceeds to step S600, and each process is repeated with the correct data as the correct data to be extracted.

<Operation of synonym determination device>
Next, the effect | action of the synonym determination apparatus 300 which concerns on embodiment of this invention is demonstrated. First, the synonym determination model output by the synonym learning device 200 is input by the input unit 210 and stored in the synonym determination model storage unit 232. Further, when a synonym determination target pair (predicate pair or “term-predicate” pair) is input by the input unit 210, the CPU executes a program stored in the ROM of the synonym determination device 300. The synonym determination processing routine shown in FIG. 30 is executed.

  First, in step S500, the input synonym determination target pair (predicate pair or “term-predicate” pair) is received.

  Next, in step S502, similar to steps S250 and S252 above, the distribution similarity of the synonym determination target pair (predicate pair or “term-predicate” pair) is calculated.

  Next, in step S504, similar to steps S300, S302, S304, S306, and S308, a dictionary definition sentence feature of a synonym determination target pair (predicate pair or “term-predicate” pair) is extracted.

  Next, in step S506, the semantic attribute features of the synonym determination target pair (predicate pair or “term-predicate pair”) are extracted in the same manner as in steps S400, S402, and S404.

  Next, in step S507, as in the above-described steps S600, S602, S604, and S606, the function representation feature of the synonym determination target pair (predicate pair or “term-predicate pair”) is extracted.

  Next, in step S508, the distribution similarity obtained in step S502, the dictionary definition sentence feature obtained in step S504, the semantic attribute feature obtained in step S506, and the functional expression feature obtained in step S507. Based on the above, a feature of a synonym determination target pair (predicate pair or “term-predicate” pair) is created.

  Next, in step S 510, based on the feature created in step S 506 and the synonym determination model stored in the synonym determination model storage unit 232, a synonym determination target pair (predicate pair or “term-predicate”). It is determined whether or not (pair) is synonymous.

  Next, in step S512, the result determined synonymously in step S510 is output by the output unit 250, and the process ends.

<Example of synonym determination result>
An example in which the synonym determination process routine is executed will be described with reference to FIGS. 31 to 34. FIG. 31 shows the result of the synonym determination and the “term-predicate” pair when the “term-predicate” pair “shelf-wo-install” and “shelf-wo-remove” is input. The feature list about is shown.

  In the case of the example in FIG. 31, it is determined as “not synonymous”. This is a predicate of an aversive relationship that was difficult to determine when only the distribution similarity that is the method of Non-Patent Document 1 was used. As shown in the feature list created for the “term-predicate” pair, in the case of “shelf-install” and “shelf-remove”, the calculated predicate distribution similarity and Both the term predicate distribution similarities have high values (usually a distribution similarity of 0.2 or higher is a threshold value indicating “synonym”), but lexical overlap in the extracted dictionary definition sentence and Since there is no definition sentence mutual complementarity, it can be correctly judged as “not synonymous”.

  Next, FIG. 32 shows a list of features to be created when a “term-predicate” pair of “text-create” and “text-use” is input. In the “term-predicate” pair of FIG. 32, the calculated term predicate distribution similarity shows a relatively high value, but the predicate distribution similarity shows a low value and is extracted. Since there is no vocabulary overlap or definition sentence mutual complementarity in the dictionary definition sentence and there is no overlap in the extracted semantic attribute features, it can be correctly determined as “not synonymous”.

Next, FIG. 33 shows a list of features to be created when a “term-predicate” pair of “flowerbed-ga-finish” and “flowerbed-ga-complete” is input. In the case of FIG. 33, in addition to the calculated predicate distribution similarity and term predicate distribution similarity, the extracted dictionary definition sentence feature proposed in the present invention, the extracted semantic attribute feature, and the functional expression feature are characterized. Thus, it can be correctly determined as “synonymous”.

  Next, FIG. 34 shows a list of features to be created when the “term-predicate” pair of “supporter-wo-recruiting” and “supporter-wo-recruitment” is input. . In the case of FIG. 34, in addition to the calculated predicate distribution similarity, the extracted dictionary definition sentence feature, the extracted semantic attribute feature, and the functional representation feature proposed in the present invention are features, Can be determined. In the case of this example, when the synonym determination model is learned without using the functional representation feature and the synonym determination is not performed using the learned synonym determination model, it is erroneously determined as “not synonymous”.

  As described above, by incorporating a plurality of linguistic features related to “synonymous predicates”, it becomes possible to determine that synonymous predicates are correctly synonymous and other predicates are correctly “not synonymous”.

Although FIG. 33 has the same feature as the learning model, the determination result uses a synonym determination model learned with a correct corpus excluding the predicate pair. In the case of FIG. 33, in addition to the predicate distribution similarity and the term predicate distribution similarity, the dictionary definition sentence feature, the semantic attribute feature, and the function representation feature proposed in the present invention are features, and are correctly defined as “synonymous”. Can be judged.

As described above, according to the synonym determination device according to the embodiment of the present invention, by focusing on the predicate, a predicate pair or “term-predicate” that represents the same but different surface layers. The synonym determination can be automatically performed with high accuracy for the pair.

In addition, "lexical overlap" and "definition mutual complementarity" between dictionary definition sentence sets of predicates in synonym determination predicates or "term-predicate" pairs, and the abstract meaning of predicates Using the “overlapping word attribute” which is a feature indicating the overlap of attributes, and using the “semantic attribute weighted overlap rate” for reflecting in the synonym determination in consideration of the hierarchical information of the “overlapping word attribute” By performing synonym determination based on multiple linguistic features, synonym determination is automatically performed with high accuracy even for predicate pairs or "term-predicate" pairs that contain anomaly relationships and time-lapse relationships. Can be done.

In addition, the “overlapping semantic label” and “semantic label overlapping rate” of the functional expressions between the predicates of the synonym determination predicate pair or “term-predicate” pair are used as features, and “lexical overlap” and “definition sentence”. By using it in addition to “mutual complementarity”, “overlapping prescriptive attribute”, “semantic attribute weighted overlap rate” and “distribution similarity”, it is not necessary to calculate the distribution similarity with a huge amount of data. The synonym determination considering the meaning can be performed, and the synonym determination can be automatically performed with high accuracy.

  Further, according to the present embodiment, in the synonym determination method for determining whether or not the predicate pair and the “term-predicate” pair representing the same thing with different surface layers are synonymous, a plurality of linguistic expressions By using features as features, it becomes possible to determine synonyms of a more accurate predicate pair and “term-predicate” pair. As a result, in the text mining technology that extracts, summarizes, and presents only important information from a large amount of text, the same event can be correctly counted even when the surface layers are different.

  In addition, multiple linguistic features are used as features in a synonym determination method that uses a computer to determine whether or not a predicate pair and a “term-predicate” pair that represent the same thing but have different surface layers. This makes it possible to determine the synonyms of a more accurate predicate pair and “term-predicate” pair. As a result, in the search technique for searching for information requested by the user, it is possible to display text representing the same thing even if the search is performed with different expressions of character strings, and the accuracy of the search technique can be improved.

  Further, in the present embodiment, the linguistic feature that the meaning of the predicates of the predicate pair indicating the synonym and the predicate of the “term-predicate” pair is also similar to the “definition sentence mutual complementarity” of the dictionary definition sentence. Using the two features of "vocabulary overlap", it became a feature, and as a result, it became possible to correctly determine "anti-predicate parts" that were difficult to determine with existing distribution similarity methods.

  In addition, the semantic feature is a linguistic feature that the abstract semantic attributes of the predicate pairs that represent synonyms and the predicates of the “term-predicate” pair are similar. The more common the attributes, the more similar the predicates are, and the more shared the more detailed attributes, the more similar are the predicates. It is featured as a “semantic attribute weighted overlap rate”, and as a result, it is possible to correctly determine “not synonymous” for “predicate representing time course” that was difficult to determine with the existing distribution similarity method. I can do it now.

  Moreover, by learning a synonym determination model using a plurality of linguistic features related to “synonymous predicates” as features, synonyms can be determined more accurately than existing methods.

  Further, in the above-described embodiment, the distribution similarity, the feature indicating the definition sentence mutual complementarity, the feature indicating the overlap of the vocabulary, the overlapping word attribute, the semantic attribute weighted overlapping rate, the overlapping semantic label, Although all the seven features of the semantic label overlap rate have been extracted, the present invention is not limited to this, and only at least one of the feature indicating the definition sentence mutual complementarity and the feature indicating the overlap of the semantic attributes is extracted. It may be. In order to further improve accuracy, at least one of distribution similarity, vocabulary overlap feature, overlapping word attribute, semantic attribute weighted overlapping rate, overlapping semantic label, and semantic label overlapping rate is further extracted. May be.

  In the above embodiment, the feature vector construction apparatus 100 constructs feature vectors by the same method as in Non-Patent Document 1, but the present invention is not limited to this, and other vector construction methods are used. May be. (Non-Patent Document 4: Manning, C., & Schutze, H. (1999). Foundations of statistical natural language processing. Cambridge, MA: MIT Press.)

  In the above embodiment, the distribution similarity calculation unit 132 calculates the distribution similarity using the feature vector input from the feature vector storage unit 124. However, the present invention is not limited to this. Other similarity calculation methods may be used. (Non-Patent Document 4: Manning, C., & Schutze, H. (1999). Foundations of statistical natural language processing. Cambridge, MA: MIT Press.)

  In the above embodiment, the “predicate attribute” is used as an abstract semantic attribute. However, the present invention is not limited to this, and an LCS structure or the like may be used as a language resource. (Non-Patent Document 5: Koichi Takeuchi, Kentaro Inui, Atsushi Fujita (2006). Description of syntactic and semantic characteristics of Japanese verbs based on lexical concept structure, Lexicon Forum, No.2, pp.85-120.)

  In the above embodiment, SVM is used for learning the synonymous determination model. However, the present invention is not limited to this, and another determination model such as Decision Tree may be used.

  The present invention is not limited to the above-described embodiment, and various modifications and applications are possible without departing from the gist of the present invention.

  Moreover, although the above-described synonym determination device 300 includes a computer system, the “computer system” includes a homepage providing environment (or display environment) if a WWW system is used. .

  Further, in the present specification, the embodiment has been described in which the program is installed in advance. However, the program can be provided by being stored in a computer-readable recording medium or provided via a network. It is also possible to do. Moreover, you may comprise each part of the synonym determination apparatus 300 of this Embodiment with hardware. In addition, the database storing the correct corpus, text corpus, feature vector, definition sentence dictionary, semantic attribute dictionary, function expression dictionary, feature set, and judgment model is realized by a storage means exemplified by a hard disk device or a file server. The database may be provided inside the synonym determination device or may be provided in an external device.

10, 110, 210 Input unit 20, 120, 220 Operation unit 24 Basic analysis unit 26 Feature extraction unit 28 Feature vector generation unit 30, 150, 250 Output unit 100 Feature vector construction device 122 Correct corpus 124 Feature vector storage unit 132 Distribution similarity Degree calculation unit 134 Definition sentence dictionary storage unit 136 Dictionary definition sentence feature extraction unit 138 Semantic attribute dictionary storage unit 139 Functional expression dictionary storage unit 140 Semantic attribute feature extraction unit 141 Functional expression feature extraction unit 142 Feature set storage unit 144 Synonym determination model learning Unit 200 synonym learning device 222 feature construction unit 224 feature vector storage unit 226 definition sentence dictionary storage unit 228 semantic attribute dictionary storage unit 229 function expression dictionary storage unit 230 synonym determination unit 232 synonym determination model storage unit 300 synonym determination device 1360 definition sentence extraction Part 1362 definition sentence mutual complementarity extraction part 364 vocabulary of overlapping extraction unit 1400 meaning attribute overlap extraction unit 1402 meaning attribute weighted overlap ratio calculation unit 1500 means the label applying section 1502 overlap meaning label extraction unit 1504 means the label overlap ratio calculation unit

Claims (7)

Predicate of predescription part pair extracted based on definition sentence of each predicate of input predicate pair obtained from definition sentence set consisting of definition statements for each of a plurality of predicates prepared in advance The first feature is whether or not there is a pair of predicates in each definition statement,
Common to previous description part pairs extracted based on the semantic attributes of each predicate of the input previous description part pair, obtained from a semantic attribute set consisting of semantic attributes for each of a plurality of predicates prepared in advance A semantic attribute to be a second feature, and a feature construction unit that extracts at least one of the first feature and the second feature;
A synonym determination model storage unit in which a synonym determination model obtained in advance is stored;
A synonym determination unit that determines whether or not the input predicate pair is synonymous based on the synonym determination model stored in the synonym determination model storage unit based on the features extracted by the feature construction unit; ,
A synonym determination device. The feature construction unit extracts at least one feature from the first feature and the second feature,
The number of vocabulary appearing in both of the definition sentences of each predicate of the predescription part pair is a third feature,
Overlapping semantic attributes of previous description part pairs with weights added according to the degree of detail of semantic attributes common to the previous description part pairs extracted based on the semantic attributes of each predicate of the input previous description part pair The degree is the fourth feature,
For each predicate of the input previous description part pair, a distribution similarity that compares words appearing around the previous description part in the text corpus is set as a fifth feature,
Before extraction based on the semantic labels of the functional expressions of the predicates of the input predescription section pair obtained from the semantic label set consisting of the semantic labels of the functional expressions of the plurality of predicates prepared in advance The semantic label common to the description part pair is the sixth feature,
The overlapping degree of the common semantic labels of the previous description part pair is a seventh feature,
Extracting at least one of the third to seventh features;
The synonym determination device according to claim 1. The “term” is extracted based on the definition statement of each predicate of the input “term-predicate” pair obtained from a definition statement set including a definition statement for each of a plurality of predicates prepared in advance. -Whether or not there is a pairing predicate in the definition statement of each predicate of the "predicate" pair, and a pair in the definition statement of each of the predicates of the "term-predicate" pair The first feature is whether or not there is at least a predescription part of whether or not a term of “term-predicate” exists,
The “extracted” based on the semantic attribute of each predicate of the “term-predicate” pair input, obtained from a semantic attribute set consisting of semantic attributes for each of a plurality of predicates prepared in advance. A semantic attribute common to the term-predicate pair is a second feature, and a feature construction unit that extracts at least one of the first feature and the second feature;
A synonym determination model storage unit in which a synonym determination model obtained in advance is stored;
Based on the features extracted by the feature construction unit, it is determined whether or not the input “term-predicate” pair is synonymous based on the synonym determination model stored in the synonym determination model storage unit. A synonym determination unit;
A synonym determination device. The feature construction unit extracts at least one feature from the first feature and the second feature,
The number of vocabulary appearing in both of the definition sentences of each predicate of the “term-predicate” pair is a third feature
The weight is added according to the degree of detail of the semantic attribute common to the “term-predicate” pair extracted based on the semantic attribute of each predicate of the input “term-predicate” pair. The fourth feature is the overlapping degree of semantic attributes of the “term-predicate” pair,
For each of the “term-predicate” of the inputted “term-predicate” pair, a distribution similarity that compares words appearing around the “term-predicate” in the text corpus, and the “term- Each of the predicates in the “predicate” pair appears at least around the “term-predicate” among the distribution similarities obtained by comparing words appearing around the “term-predicate” predicate in the text corpus. The fifth feature is the distribution similarity between the words
Based on a semantic label of the functional expression of each predicate of the inputted “term-predicate” pair, obtained from a semantic label set consisting of semantic labels of functional expressions of each of a plurality of predicates prepared in advance. A semantic label common to the extracted “term-predicate” pair is a sixth feature,
The overlapping degree of the common semantic labels of the “term-predicate” pair is a seventh feature,
Extracting at least one of the third to seventh features;
The synonym determination device according to claim 3. Each of the predicates of the predescription part pair extracted based on the definition sentence of each predicate of the predicate pair obtained from the definition sentence set including the definition sentences for each of the plurality of predicates prepared in advance. The first feature is whether or not there is a pair of predicates in the definition statement, and a predescription pair pair obtained from a semantic attribute set consisting of semantic attributes for each of a plurality of predicates prepared in advance. A semantic attribute common to the semantic attributes of each predicate is a second feature, information on whether or not it is synonymous, and for each of a plurality of predicate pairs prepared in advance, the first feature and the A feature construction unit that extracts at least one of the second features;
Based on at least one of the first feature and the second feature extracted for the plurality of predicate pairs by the feature construction unit and information on whether or not they are synonymous with the plurality of predicate pairs. A synonym determination model learning unit for learning a synonym determination model;
A synonymous learning device. The “term-predicate” is extracted based on the definition statement of each predicate of the “term-predicate” pair, which is obtained from a definition statement set including definition statements for each of a plurality of predicates prepared in advance. "Whether there is a paired predicate in the definition statement of each of the predicates of the pair, and" pair- "in the definition statement of each of the predicates of the" term-predicate "pair The first feature is whether or not there is at least the previous description part of whether or not the term of “predicate” exists, and from a semantic attribute set consisting of semantic attributes for each of a plurality of predicates prepared in advance. A semantic attribute common to the semantic attributes of each predicate of the “term-predicate” pair obtained is set as a second feature, information on whether or not it is synonymous, and a plurality of “terms” prepared in advance. -For each predicate pair, a feature structure that extracts at least one of the first feature and the second feature And parts,
Whether at least one of the first feature and the second feature extracted for the plurality of “term-predicate” pairs by the feature construction unit and the synonym for the plurality of “term-predicate” pairs. A synonym determination model learning unit that learns a synonym determination model based on information on whether or not,
A synonymous learning device.   The program for functioning a computer as each means which comprises the synonym determination apparatus of any one of Claims 1-4, or the synonym learning apparatus of any one of Claims 5-6.