JP6586055B2 - Deep case analysis device, deep case learning device, deep case estimation device, method, and program - Google Patents

Description

  The present invention relates to a deep case analysis device, a deep case learning device, a deep case estimation device, a method, and a program.

  As a conventional deep case analysis technique, there is a rule-based method based on a case frame as described in Non-Patent Document 1. For each verb, a possible noun semantic category and case particle pair and deep case information (referred to as case frame information) corresponding to the pair are determined in advance. For example, for the verb “eat”, information such as (animal, but, main case), (food, and target case) is defined as (noun semantic category, case particle, deep case). When an input (noun, case particle, verb) is given, a deep case corresponding to the noun semantic category and case particle pair is acquired from the case frame information of the verb. For example, when (cake, eat) is entered, the meaning category of “cake” is food, so from the case frame information of “eat”, a deep case corresponding to (food, eat) Get a case.

Nagao, edited by “Natural Language Processing”, Iwanami Shoten, Iwanami Lecture Software Science 15, 1996.

  In the rule-based method based on the case frame, there is a problem that the construction cost is high in constructing the case frame information and assigning a semantic category to a new word. In addition, a set of deep cases that cannot be matched with the semantic category system and case frame information once defined (nouns, case particles, verbs) and the deep case cannot be estimated, which is necessary for deep case estimation. There is a problem of lack of robustness. For example, a certain (noun semantic category X, case particle, deep case) pair is not in the case frame information, but for a specific noun whose noun semantic category is X, the pair is included in the case frame information. There is a need, or conversely, a certain (noun semantic category X, case particle, deep case) pair is in the case frame information, but for a specific noun whose noun semantic category is X, the pair There are some cases where the deep case is not applicable. For specific nouns, it is necessary to create a semantic category X that does not fit into existing semantic categories, and to include a set of (noun semantic category X, case particle, deep case) in case frame information. Sometimes it comes.

  An object of the present invention is to solve the above problems, and to provide a deep case analysis device, a deep case learning device, a deep case estimation device, a method, and a program for robustly estimating a deep case. is there.

  In order to solve the above-described problem, the deep case analysis apparatus according to the first aspect of the present invention, for a body phrase clause and a prescriptive phrase in a dependency relationship, which deep case is used for the prescriptive phrase of the body phrase. This is a deep case analysis device that estimates whether or not the answer is a set of correct answer data, which is a set of body phrase clause and use phrase clause data in a dependency relationship and a correct deep case corresponding to the data. A correct feature-added feature vector set generation unit for generating a correct feature-added feature vector set by generating a feature vector that is a set of features and feature values from the data for each correct-attached data, and the correct answer-added feature A classification model generation unit for generating a classification model for classifying a deep case from a vector set, and a body phrase clause and a use phrase clause data A having a dependency relation are input, and a feature vector B is generated from the data A. A feature vector generating unit which, from said feature vector B and the classification model is configured to include a classification unit configured to calculate a score the data A corresponds to the deep cases, the.

  The deep case learning device according to the second invention has an input of a set of correct data with a set of data of a body phrase clause and a prescriptive phrase in a dependency relationship and a correct deep case corresponding to the data, A correct feature-added feature vector set generation unit that generates a feature-added feature vector set by generating a feature vector that is a set of features and feature values from the data for each correct-attached data, and the correct-added feature vector A classification model generation unit that generates a classification model for classifying the deep case from the set.

  The deep case estimation apparatus according to the third aspect of the present invention receives, as input, a body phrase clause and a prescriptive phrase data A that are in a dependency relationship, and generates a feature vector B that is a set of features and feature values from the data A. For each correct answer data included in a set of correct answer data, which is a set of a feature vector generation unit, data of body phrase clauses and use phrase clauses having dependency relations, and a correct deep case corresponding to the data A score corresponding to each deep case from the classification model for classifying a deep case, which is a set of feature vectors to be generated, and is generated in advance from the feature vector set with correct answer and the feature vector B And a classifying unit for calculating.

  The deep case analysis method according to the fourth invention includes a correct feature-added feature vector set generation unit, a classification model generation unit, a feature vector generation unit, and a classification unit. On the other hand, a deep case analysis method in a deep case analysis apparatus for estimating which deep case corresponds to the prescription of the prescriptive phrase, wherein the correct feature vector set generation unit includes a dependency relationship A set of correct data with a correct answer corresponding to the data of the body phrase clause and prescriptive phrase data and the correct deep case corresponding to the data is input, and for each correct answer data, a feature and a feature value are obtained from the data. Generating a correct feature-added feature vector set by generating a feature vector that is a set of sets, and the classification model generating unit for classifying a deep case from the correct answer-added feature vector set A feature model generating step, the feature vector generating unit receiving data A of a body phrase clause and a prescriptive phrase in a dependency relationship, generating a feature vector B from the data A, and the classifying unit The data A includes a step of calculating a score corresponding to each deep case from the feature vector B and the classification model.

  Further, the deep case learning method according to the fifth invention is a deep case learning method in a deep case learning device including a correct answer feature vector set generation unit and a classification model generation unit, wherein the correct answer feature vector set generation unit Is a set of data with correct answers, which is a set of data of body phrase clauses and use phrase clauses in a dependency relationship and a deep case of a correct answer corresponding to the data, and for each correct answer data, A step of generating a feature vector set with correct answers by generating a feature vector that is a set of features and feature values, and the classification model generating unit for classifying a deep case from the correct feature-added feature vector set Generating a classification model.

  A deep case estimation method according to a sixth aspect of the present invention is a deep case estimation method in a deep case estimation apparatus including a feature vector generation unit and a classification unit, wherein the feature vector generation unit has a dependency relationship. A step of generating a feature vector B which is a set of features and feature values from the data A as input of clause A and phrase clause data A, and the classification unit and a body phrase clause having a dependency relationship From the correct feature-added feature vector set, which is a set of feature vectors generated for each correct-attached data included in the correct-attached data set that is a set of clause data and a correct deep case corresponding to the data The classification model for classifying the deep case generated in advance and the step of calculating a score corresponding to each deep case from the feature vector B are included in the data A.

  The program of the present invention causes a computer to function as each part of the deep case analysis device, the deep case learning device, or the deep case estimation device, or causes the computer to perform the deep case analysis method. This is a program for executing the steps of the deep case learning method or the deep case estimation method.

  Further, as a feature of the present invention, a notation character string, a part of speech, or a semantic category existing in data may be taken.

  In addition, as a set of features and feature values according to the present invention, the body phrase clause appendices and usages taken from the body clauses and prescriptive clauses that have a dependency relationship in the corpus with respect to the body language in the corpus. Take a set of pairs of phrases or predicates and their frequency, or, for the predicates in the data, the body phrases and their frequencies that are dependent on the prescriptive phrase containing the predicates in the corpus Taking a set of pairs, or taking a set of body phrase clauses and their frequencies in the corpus for the use clauses in the data, or taking any set of the above In the set of, the meaning category of the prescriptive phrase in the prescriptive phrase and the body phrase in the prescriptive phrase is the same, and the other notation information is the same and the frequency is added. Good.

  Moreover, you may make it take the concept vector of each morpheme in data as a set of the set of the feature and feature value of this invention.

  In the present invention, a classification model reflecting a large amount of data is derived from a large amount of learning data by a statistical method. For this reason, even if there is some noise in the learning data, such as an incomplete feature value, the classification model becomes accurate, so the deep case can be estimated robustly. In addition to the semantic categories assigned to each word in advance, there are other features corresponding to the meaning of a word such as a semantic category and features that can be automatically acquired. For this reason, even if the semantic category assignment is not complete, it is possible to accurately estimate the deep case based on the information of other features, and the construction cost can be reduced as compared with the conventional method.

  According to the present invention, a deep case can be estimated robustly.

It is a block diagram which shows the functional structure of the deep case analysis apparatus which concerns on embodiment of this invention. It is a figure which shows an example of the collection of data with a correct answer. It is a figure which shows an example of a correct answer addition feature vector set. It is a figure which shows an example of the data A which is the estimation object of a deep case. It is a figure which shows an example of the feature vector B produced | generated from the data A. It is a figure which shows an example of a data set with a correct answer at the time of fixing a body phrase phrase attachment part to "de". It is a figure which shows an example of the data A at the time of fixing a body phrase phrase attachment part to "de". It is a figure which shows an example of the set of the group of the expression "pencil" and a dependency relation (an attachment part of a body sentence, a pretext (end form)), and its frequency. It is a figure which shows an example of the set of a set of the frequency which is in dependency relation with the predicate "write" (a body part, a body sentence clause attachment part), and its frequency. It is a figure which shows an example of the co-occurrence vector produced | generated using the semantic category. It is a flowchart figure which shows the learning process routine in the learning part of the deep case analysis apparatus which concerns on embodiment of this invention. It is a flowchart figure which shows the estimation process routine in the estimation part of the deep case analysis apparatus which concerns on embodiment of this invention.

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

<Outline of Embodiment of the Present Invention>
An embodiment of the present invention relates to a deep case analysis apparatus and method for estimating, for a body phrase clause and a prescriptive phrase in a dependency relationship, a deep case corresponding to the prescriptive phrase of the body phrase And the program.

  The deep case referred to in the embodiment of the present invention represents the semantic role of nouns for verbs and other predicates. For example, in “eating with chopsticks in the room”, the phrase phrases “in the room” and “with chopsticks” have a dependency relationship with the phrase “eating”, and the surface layer of the phrases “room” and “chopsticks” in the phrase phrase Although the case is a de-case, the deep case for the predicate “eating” in the preaching phrase “eating” is a place case and a tool case, respectively. In general, various types of deep cases have been proposed. Examples include a main case, a target case, a tool case, a source case, a target case, a place case, a time case, and an experienced case. The embodiment of the present invention relates to a deep case analysis technique for estimating a corresponding deep case for a body phrase clause and a use phrase clause having a dependency relationship, after finite types of deep case types are determined in advance. is there. Note that the phrase used in the embodiment of the present invention includes “symbol +” such as “student”.

<Configuration of deep case analyzer>
A configuration of the deep case analysis apparatus according to the embodiment of the present invention will be described. FIG. 1 is a configuration example of a deep case analysis apparatus according to claim 1 of the present invention. As shown in FIG. 1, a deep case analysis 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 processing routines to be described later. Can be configured with a computer. Functionally, the deep case analysis apparatus 100 includes an input unit 10, an arithmetic unit 20, and an output unit 30 as shown in FIG.

  The input unit 10 receives as input a set of correct data with a set of body phrase clause and use phrase clause data having a dependency relationship and a correct deep case corresponding to the data. In addition, the input unit 10 receives data A of the body phrase clause and the prescriptive phrase having a dependency relationship as input. A set of correct data and data A will be described later.

  The computing unit 20 includes a learning unit 22, a classification model storage unit 24, and an estimation unit 26. The learning unit 22 receives a set of correct data with learning data and generates a classification model for classifying the deep case. After the processing of the learning unit 22 is completed, when the input unit 10 inputs the data A of the body phrase clause and the prescriptive phrase having a dependency relationship in the estimation unit 26, the data A is referred to by referring to the classification model. The deep case corresponding to is estimated.

  The learning unit 22 includes a correct feature-added feature vector set generation unit 220 and a classification model generation unit 222.

  The correct feature-added feature vector set generation unit 220 receives correct answer-added data, which is a set of the dependency phrase and prescriptive phrase data received by the input unit 10 and the correct deep case corresponding to the data. Takes a set as input. FIG. 2 shows an example of a set of data with correct answers. Each correct data includes a body part that is an independent part in a body phrase, a body part appendage, a word that is a self-supporting part in a prescriptive phrase, a part phrase appendage, and a deep case. The body language may take only the last constituent morpheme. In FIG. 2, the prescription is shown in an end form. When the prescriptive phrase is “body word +”, this word is used as a word. The sixth data is taken from “Eat the mouse”, and the predicate is “Eat”, the end of the self-supporting part “Eat” in the prescriptive phrase “Eat”. For the predicate “eat”, the word “mouse” corresponds to the subject case.

  The correct answer-added data is created, for example, by assigning the corresponding correct deep case to the body phrase clause and the prescriptive phrase in the dependency relationship extracted by dependency analysis of the text corpus.

  The correct feature-added feature vector set generation unit 220 generates correct feature-added feature vectors for each correct-attached data by generating a feature vector that is a set of features and feature values from the body phrase clause and the prescriptive phrase as the data. Create a set. FIG. 3 shows an example of a correct feature-added feature vector set. The number of dimensions of the feature vector is N, and each feature value takes a real value.

  The classification model generation unit 222 generates a classification model for classifying the deep case from the correct answer feature vector set generated by the correct answer feature vector set generation unit 220. Specifically, for each deep case, the correct-added feature vector set is divided into a feature vector group of the deep case and a feature vector group that is not the deep case, and the binary classification of whether or not the deep case is to be solved Are generated by a machine learning method such as a support vector machine. In this way, a corresponding classification model is generated for each deep case.

  The classification model storage unit 24 stores the classification model generated for each deep case by the classification model generation unit 222.

  The estimation unit 26 includes a feature vector generation unit 260 and a classification unit 262.

  The feature vector generation unit 260 receives the data A of the body phrase clause and the prescriptive phrase that are received by the input unit 10 and have a dependency relationship. The content of the data A is the same as the data of the data set with correct answer that is the input of the feature vector set generating unit 220 with correct answer. FIG. 4 shows an example of the data A. The data A is composed of a body part which is a self-supporting part in a body phrase clause, a body part clause attachment part, a word which is a self-supporting part in a prescriptive phrase part, and a word phrase attachment part. ing.

  The feature vector generation unit 260 generates the feature vector B from the data A by the same algorithm as the algorithm for generating the feature vector from the data of the correct-added feature vector set generation unit 220. The content of the feature vector B is the same as the feature vector of the correct-added feature vector set output from the correct-added feature vector set generation unit 220. FIG. 5 shows an example of the feature vector B. The number of dimensions of the feature vector is N, and each feature value takes a real value.

  The classification unit 262 calculates a score corresponding to each deep case from the feature vector B generated by the feature vector generation unit 260 and the classification model of each deep case stored in the classification model storage unit 24. To do. Specifically, for each deep case, a score corresponding to the deep case is calculated from the feature vector B and the classification model corresponding to the deep case. A deep case having a score equal to or higher than a certain threshold is output as an estimated deep case. The data A in FIG. 4 is converted into the feature vector B in FIG. 5 by the feature vector generation unit 260, and the deep case is estimated to be a tool case by the classification unit 262.

  As mentioned above, although the example of a structure of the deep case analysis apparatus of Claim 1 of this invention was described, you may perform learning and presumption, after fixing a body phrase phrase attachment part to one. That is, learning is performed after the data set with the correct answer is limited to only data having a fixed phrasal phrase appendix. The estimation is also performed by using the data A having the body phrase phrase appendage as an input. Since the feature related to the body phrase clause attachment is common to all data and does not make sense as a classification feature, the feature-added feature vector set generation unit 220 and the feature vector generation unit 260 do not extract the feature related to the body phrase clause attachment.

  For example, if the body phrase clause attachment is fixed to “de”, the data set with the correct answer in FIG. 2 becomes FIG. 6 in which the body phrase clause attachment is limited to data with “de” only. The estimation is performed by using the data A in which the body phrase clause appendage as shown in FIG. 7 is “de”. Since the feature related to the body phrase clause attachment is not extracted, the body clause clause attachment is not described in FIGS. 6 and 7.

  Next, features and feature values effective for classification of deep cases extracted from data in the feature vector set generation unit 220 and the feature vector generation unit 260 will be described in detail below.

  Since the deep case is the semantic role of nouns for predicates, the semantic information of each morpheme can be an effective feature for classification. Moreover, since the deep case determination depends on the body phrase clause appendix and the use phrase clause appendage, these notations can be effective features for classification.

  Therefore, a notation character string, part of speech or semantic category existing in the data can be taken as the feature. These features are handled as different features even if they are the same character string, if the type of extraction source (a body part, a body phrase clause attachment part, a predicate, a word phrase attachment part) is different.

  Examples of the notation include a body text, a body phrase appendage part, a script (end form), and a text string of the text phrase appendage. Moreover, the notation of each constituent morpheme is also mentioned. In this case, regarding the body language, only the last constituent morpheme may be used as a feature. In the case of the sixth data in FIG. 2, the body phrase clause attachment is “ga” and the prescriptive phrase attachment is “enabled”, but from this feature, it is identified that the data is passive or possible expression It is possible to learn and estimate the deep case reflecting that.

  The part of speech includes the part of speech of the last constituent morpheme of the body and the part of speech of the predicate. Depending on the morphological analyzer, the part of speech may be composed of a plurality of fine parts of speech. In such a case, the whole part of speech can be taken as a feature, or each fine part of speech can be taken as a feature. Some of the fine parts of speech indicate that they correspond to personal names, place names, etc., and such information is also an effective feature for the classification of deep cases.

  The semantic category means a group of similar words as one category. By assigning a semantic category to each word in the word dictionary for morphological analysis, the semantic category of the last constituent morpheme or the semantic category of the prescription can be taken as a feature.

  All the differences in word notation, part of speech, and semantic category that exist in the learning data are features. When the target data includes the feature, the feature value of the feature is 1, and when it does not include the feature value of the feature is 0. .

  In addition, as a set of features and feature values represented by a feature vector, the body phrase included in the corpus, including the body phrase in the corpus, and from the body phrase clause and the use phrase clause that are in a dependency relationship It can include a set of pairs of parts and prescription clauses or predicates and their frequencies. Here, the word may be the last constituent morpheme. FIG. 8 shows the word “pencil” in the corpus, which includes “pencil” and is taken from the word phrases and prescriptive phrases that are in a dependency relationship (an appendix to the word phrases, the word (end form)). It is a set of pairs of the frequencies. This is a set of features and feature value sets extracted from the word “pencil”.

  In addition, as a set of a set of features and feature values represented by the feature vector, a set of a set of a combination of a body phrase clause and its frequency in a dependency relationship with a prescriptive clause including the prescription in the corpus is used for the prescription in the data. Alternatively, for a prescriptive phrase in the data, it is possible to take a set of a set of body phrase clauses and their frequencies in a dependency relationship with the prescriptive phrase in the corpus. The body language in the body language clause may be the last constituent morpheme. FIG. 9 shows a set of combinations of the frequency of the phrase “writing”, which is taken from the body phrase clause that is dependent on the phrase clause including “writing” in the corpus (body language, body language phrase appendix). is there. This is a set of features and feature values extracted from the predicate “write”.

  A set of the features and feature values listed above is called a co-occurrence vector. A phrase having a close co-occurrence vector adopts the co-occurrence vector based on the property that it is semantically close.

  In addition, as a set of a set of features and feature values represented by a feature vector, in any of the set of the above-mentioned sets, the semantic category of the prescriptive phrase in the prescriptive phrase or the prescriptive phrase is the same, and other notation The same information can be regarded as the same and the frequency added. The semantic category of the body may be the semantic category of the last constituent morpheme. FIG. 10 shows that in the co-occurrence vector of FIG. 9, the meaning category of “student” and “teacher” is the same for [person] for (student, ga) and (teacher, ga), and other notation information is Since “g” is the same, the feature is identified as ([person], g), and the frequency is set to 100. In addition, in the co-occurrence vector of FIG. 9, the meaning category of “book” and “novel” is the same in [book] for (book,) and (novel,), and other notation information is “ The feature is identified as ([book],), and the frequency is set to 374. Further, in the co-occurrence vector of FIG. 9, since the semantic category of “Yokohama” is (place name) with respect to (Yokohama,), the frequency 27 is taken as (with [place name],).

  Moreover, the concept vector of each morpheme in data can be taken as a set of a set of features and feature values represented by the feature vector. A word concept vector generated by the method of Non-Patent Document 2 is an example of a concept vector, and each concept vector of a word pair that is semantically close is close.

[Non-patent document 2] Katsuto Bessho, Toshiro Uchiyama, Kei Uchiyama, Ryoji Kataoka, Masahiro Oku, “Corpus concept-based generation method based on co-occurrence between words and semantic attributes,” IPSJ Journal, Dec. 2008, Vol.49, No.12, pp.3997-4006.

  For example, as a set of a set of features and feature values represented by the feature vector, the concept vector of the last constituent morpheme of the body word or the concept vector of the predicate is taken. Alternatively, a concept vector normalized to length 1 by adding the concept vectors of the constituent morphemes may be taken.

  FIG. 11 is an example of a processing flow of the learning unit 22. When the input unit 10 receives a set of correct data, the learning process routine shown in FIG. 11 is executed.

  First, in step S <b> 100, the correct feature-added feature vector set generation unit 220 acquires a set of correct answer-added data received by the input unit 10.

  In step S102, the correct feature-added feature vector set generation unit 220, for each correct-attached data in the correct answer-added data set accepted in step S100, from the body phrase and the prescriptive phrase as the data, and the feature and feature value. By generating a feature vector that is a set of groups, a feature vector set with correct answers is generated.

  In step S <b> 104, the classification model generation unit 222 classifies each deep case from the correct case feature vector set generated by the correct feature vector set generation unit 220 to classify whether or not the deep case is the deep case. Generate a model. Then, the classification model generation unit 222 stores the classification model in the classification model storage unit 24, and ends the learning processing routine.

  FIG. 12 is an example of a processing flow of the estimation unit 26. When the input unit 10 receives the data A, which is a deep case estimation target, an estimation processing routine shown in FIG. 12 is executed.

  First, in step S <b> 200, the feature vector generation unit 260 acquires data A received by the input unit 10.

  Next, in step S202, the feature vector generation unit 260 obtains the feature vector B from the data A acquired in step S200 using the same algorithm as the algorithm for generating the feature vector from the data of the correct feature-added feature vector set generation unit 220. Generate.

  Next, in step S204, the classification unit 262 uses the data A acquired in step S200 from the feature vector B generated in step S202 and the classification model of each deep case stored in the classification model storage unit 24. Calculates a score corresponding to each deep case. Then, a deep case having a score equal to or higher than a certain threshold is set as an estimated deep case.

  In step S206, the classification unit 262 outputs the estimated deep case obtained in step S204 as the estimation result, and ends the estimation processing routine.

  As described above, according to the deep case analysis apparatus of the present embodiment, the text is converted into a semantic structure by estimating an accurate deep case, and then matching (searching) between the texts at the semantic structure level is performed. Etc.) and conversion (generation, summarization, translation, etc.) processing accuracy can be improved.

  The deep case analysis apparatus according to the present embodiment can reduce the construction cost as compared with the conventional method and has robustness necessary for estimating an accurate deep case.

  It is possible to construct the processing described so far as a program, install the program from a communication line or a recording medium, and implement it by means such as a CPU.

  The present invention is not limited to the above-described embodiments, and various modifications and applications are possible within the scope of the claims.

  For example, in the present embodiment, the case where the classification model learning process and the deep case estimation process are configured as one apparatus has been described as an example, but the present invention is not limited to this. For example, the classification model learning process and the deep case estimation process may be configured as separate devices. In this case, you may comprise as a deep case learning apparatus provided with the learning part 22, and a deep case estimation apparatus provided with the estimation part 26. FIG.

  The present invention can be applied to a language processing technology that performs matching (searching, etc.) and conversion (generation, summarization, translation, etc.) at a semantic structure level between texts after converting the text into a semantic structure.

DESCRIPTION OF SYMBOLS 10 Input part 20 Operation part 22 Learning part 24 Classification model memory | storage part 26 Estimation part 30 Output part 100 Deep case analysis apparatus 220 Correct answer feature vector set generation part 222 Classification model generation part 260 Feature vector generation part 262 Classification part

Claims (8)

A set of correct data with a combination of body phrase clause and use phrase data in a dependency relationship and a correct deep case corresponding to the data is input, and for each correct data, A correct feature-added feature vector set generation unit that generates a correct feature-added feature vector set by generating a feature vector that is a set of feature values;
A classification model generation unit for generating a classification model for classifying a deep case from the correct feature vector set;
Only including,
As a set of sets of the features and the feature values,
Using the notation character strings that exist in the body phrase clause and prescriptive phrase in the data,
And,
A pair of the body phrase clause attachment part of the body phrase clause and the phrase clause or the phrase taken from a phrase clause having a dependency relationship with the body clause including the body language in the corpus for the body phrase in the data, A set of pairs with the frequency of a pair, or a set of pairs of a body phrase clause appendage of the body phrase clause and the prescriptive phrase or prescription, and the frequency of the pair, When counting the frequency of the pair, whether the meaning category of the prescription in the pair of prescriptive clauses is the same and the other notation information is the same, and the frequency of the pair is counted ,
For a pretext in the data, take a set of a set of a body phrase clause that is in a dependency relationship with a prescriptive phrase including the prescription in the corpus and the frequency of the body phrase clause, or A set of body phrase clauses having a dependency relationship with a phrase and the frequency of the body phrase clauses, and when counting the frequency of the body phrase clauses, the semantic categories of the body phrases in the body phrase clause are the same, and , The other notation information is the same, and the frequency of the body phrase is counted, or
For a prescriptive phrase in the data, take a set of a set of a body phrase phrase that is in a dependency relationship with the prescriptive phrase in the corpus and a frequency of the body phrase phrase, or a dependency relationship with the prescriptive phrase It is a set of a set phrase phrase and the frequency of the set phrase phrase, and when counting the frequency of the set phrase phrase, the notation semantic category in the set phrase phrase is the same and other notation information is the same Take the same frequency and count the frequency of the body phrase,
A deep case learning device characterized by that. A feature vector generation unit that receives data A of the body phrase clause and the use clause clause in a dependency relationship, and generates a feature vector B that is a set of features and feature values from the data A;
The feature vector generated for each correct answer data included in a set of correct answer data that is a set of data of the body phrase clause and the prescriptive phrase data in the dependency relation and the correct deep case corresponding to the data. A classification model for classifying a deep case, which is a set, and is generated in advance from a correct feature vector set, and a classification unit for calculating a score corresponding to each deep case from the feature vector B; ,
Only including,
As a set of sets of the features and the feature values,
Using the notation character strings that exist in the body phrase clause and prescriptive phrase in the data,
And,
A pair of the body phrase clause attachment part of the body phrase clause and the phrase clause or the phrase taken from a phrase clause having a dependency relationship with the body clause including the body language in the corpus for the body phrase in the data, A set of pairs with the frequency of a pair, or a set of pairs of a body phrase clause appendage of the body phrase clause and the prescriptive phrase or prescription, and the frequency of the pair, When counting the frequency of the pair, whether the meaning category of the prescription in the pair of prescriptive clauses is the same and the other notation information is the same, and the frequency of the pair is counted ,
For a pretext in the data, take a set of a set of a body phrase clause that is in a dependency relationship with a prescriptive phrase including the prescription in the corpus and the frequency of the body phrase clause, or A set of body phrase clauses having a dependency relationship with a phrase and the frequency of the body phrase clauses, and when counting the frequency of the body phrase clauses, the semantic categories of the body phrases in the body phrase clause are the same, and , The other notation information is the same, and the frequency of the body phrase is counted, or
For a prescriptive phrase in the data, take a set of a set of a body phrase phrase that is in a dependency relationship with the prescriptive phrase in the corpus and a frequency of the body phrase phrase, or a dependency relationship with the prescriptive phrase It is a set of a set phrase phrase and the frequency of the set phrase phrase, and when counting the frequency of the set phrase phrase, the notation semantic category in the set phrase phrase is the same and other notation information is the same Take the same frequency and count the frequency of the body phrase,
A deep case estimation apparatus characterized by that. A feature vector generation unit that receives data A of the body phrase clause and the use clause clause in a dependency relationship, and generates a feature vector B that is a set of features and feature values from the data A;
The feature vector generated for each correct answer data included in a set of correct answer data that is a set of data of the body phrase clause and the prescriptive phrase data in the dependency relation and the correct deep case corresponding to the data. A classification model for classifying a deep case, which is a set, and is generated in advance from a correct feature vector set, and a classification unit for calculating a score corresponding to each deep case from the feature vector B; ,
Only including,
As a set of sets of the features and the feature values,
Using the notation character strings that exist in the body phrase clause and prescriptive phrase in the data,
And,
A pair of the body phrase clause attachment part of the body phrase clause and the phrase clause or the phrase taken from a phrase clause having a dependency relationship with the body clause including the body language in the corpus for the body phrase in the data, A set of pairs with the frequency of a pair, or a set of pairs of a body phrase clause appendage of the body phrase clause and the prescriptive phrase or prescription, and the frequency of the pair, When counting the frequency of the pair, whether the meaning category of the prescription in the pair of prescriptive clauses is the same and the other notation information is the same, and the frequency of the pair is counted ,
For a pretext in the data, take a set of a set of a body phrase clause that is in a dependency relationship with a prescriptive phrase including the prescription in the corpus and the frequency of the body phrase clause, or A set of body phrase clauses having a dependency relationship with a phrase and the frequency of the body phrase clauses, and when counting the frequency of the body phrase clauses, the semantic categories of the body phrases in the body phrase clause are the same, and , The other notation information is the same, and the frequency of the body phrase is counted, or
For a prescriptive phrase in the data, take a set of a set of a body phrase phrase that is in a dependency relationship with the prescriptive phrase in the corpus and a frequency of the body phrase phrase, or a dependency relationship with the prescriptive phrase It is a set of a set phrase phrase and the frequency of the set phrase phrase, and when counting the frequency of the set phrase phrase, the notation semantic category in the set phrase phrase is the same and other notation information is the same Take the same thing and count the frequency of the body phrase,
And, as the set of the feature and the feature value set, the notation of the body phrase clause appendix and the notation of the phrase clause appendage, which are present in the data, are used.
A deep case estimation apparatus characterized by that. A feature vector generation unit that receives data A of the body phrase clause and the use clause clause in a dependency relationship, and generates a feature vector B that is a set of features and feature values from the data A;
The feature vector generated for each correct answer data included in a set of correct answer data that is a set of data of the body phrase clause and the prescriptive phrase data in the dependency relation and the correct deep case corresponding to the data. A classification model for classifying a deep case, which is a set, and is generated in advance from a correct feature vector set, and a classification unit for calculating a score corresponding to each deep case from the feature vector B; ,
Only including,
As a set of sets of the features and the feature values,
Using the notation character strings that exist in the body phrase clause and prescriptive phrase in the data,
And,
A pair of the body phrase clause attachment part of the body phrase clause and the phrase clause or the phrase taken from a phrase clause having a dependency relationship with the body clause including the body language in the corpus for the body phrase in the data, A set of pairs with the frequency of a pair, or a set of pairs of a body phrase clause appendage of the body phrase clause and the prescriptive phrase or prescription, and the frequency of the pair, When counting the frequency of the pair, whether the meaning category of the prescription in the pair of prescriptive clauses is the same and the other notation information is the same, and the frequency of the pair is counted ,
For a pretext in the data, take a set of a set of a body phrase clause that is in a dependency relationship with a prescriptive phrase including the prescription in the corpus and the frequency of the body phrase clause, or A set of body phrase clauses having a dependency relationship with a phrase and the frequency of the body phrase clauses, and when counting the frequency of the body phrase clauses, the semantic categories of the body phrases in the body phrase clause are the same, and , The other notation information is the same, and the frequency of the body phrase is counted, or
For a prescriptive phrase in the data, take a set of a set of a body phrase phrase that is in a dependency relationship with the prescriptive phrase in the corpus and a frequency of the body phrase phrase, or a dependency relationship with the prescriptive phrase It is a set of a set phrase phrase and the frequency of the set phrase phrase, and when counting the frequency of the set phrase phrase, the notation semantic category in the set phrase phrase is the same and other notation information is the same Take the same thing and count the frequency of the body phrase,
And, as a set of the feature and the set of feature values, using the notation of the body phrase clause appendix and the notation of the phrase clause appendage existing in the data,
And, as a set of the feature and the feature value set, for the feature from which the feature in the data is extracted, the information of the last constituent morpheme part of the statement is used.
A deep case estimation apparatus characterized by that. A deep case learning method in a deep case learning device including a correct feature vector set generation unit and a classification model generation unit,
The correct feature-added feature vector set generation unit receives as input a set of correct answer-added data, which is a set of data of the body phrase clause and prescriptive phrase clause having a dependency relationship and a correct deep case corresponding to the data. Generating a correct feature-added feature vector set by generating a feature vector that is a set of features and feature values from the data for the attached data;
The classification model generation unit generating a classification model for classifying a deep case from the correct feature vector set;
Only including,
As a set of sets of the features and the feature values,
Using the notation character strings that exist in the body phrase clause and prescriptive phrase in the data,
And,
A pair of the body phrase clause attachment part of the body phrase clause and the phrase clause or the phrase taken from a phrase clause having a dependency relationship with the body clause including the body language in the corpus for the body phrase in the data, A set of pairs with the frequency of a pair, or a set of pairs of a body phrase clause appendage of the body phrase clause and the prescriptive phrase or prescription, and the frequency of the pair, When counting the frequency of the pair, whether the meaning category of the prescription in the pair of prescriptive clauses is the same and the other notation information is the same, and the frequency of the pair is counted ,
For a pretext in the data, take a set of a set of a body phrase clause that is in a dependency relationship with a prescriptive phrase including the prescription in the corpus and the frequency of the body phrase clause, or A set of body phrase clauses having a dependency relationship with a phrase and the frequency of the body phrase clauses, and when counting the frequency of the body phrase clauses, the semantic categories of the body phrases in the body phrase clause are the same, and , The other notation information is the same, and the frequency of the body phrase is counted, or
For a prescriptive phrase in the data, take a set of a set of a body phrase phrase that is in a dependency relationship with the prescriptive phrase in the corpus and a frequency of the body phrase phrase, or a dependency relationship with the prescriptive phrase It is a set of a set phrase phrase and the frequency of the set phrase phrase, and when counting the frequency of the set phrase phrase, the notation semantic category in the set phrase phrase is the same and other notation information is the same Take the same frequency and count the frequency of the body phrase,
A deep case learning method characterized by that. A deep case estimation method in a deep case estimation apparatus including a feature vector generation unit and a classification unit,
The feature vector generation unit receives the data A of the body phrase clause and the use phrase clause in a dependency relationship, and generates a feature vector B that is a set of features and feature values from the data A;
The classification unit generates for each correct answer data included in a set of correct answer data, which is a set of body phrase clause and prescriptive phrase data in a dependency relationship and a correct deep case corresponding to the data. From the classification model for classifying the deep case, which is generated in advance from the correct feature vector set, and the feature vector B, a score corresponding to each deep case is obtained. A calculating step;
Only including,
As a set of sets of the features and the feature values,
Using the notation character strings that exist in the body phrase clause and prescriptive phrase in the data,
And,
A pair of the body phrase clause attachment part of the body phrase clause and the phrase clause or the phrase taken from a phrase clause having a dependency relationship with the body clause including the body language in the corpus for the body phrase in the data, A set of pairs with the frequency of a pair, or a set of pairs of a body phrase clause appendage of the body phrase clause and the prescriptive phrase or prescription, and the frequency of the pair, When counting the frequency of the pair, whether the meaning category of the prescription in the pair of prescriptive clauses is the same and the other notation information is the same, and the frequency of the pair is counted ,
For a pretext in the data, take a set of a set of a body phrase clause that is in a dependency relationship with a prescriptive phrase including the prescription in the corpus and the frequency of the body phrase clause, or A set of body phrase clauses having a dependency relationship with a phrase and the frequency of the body phrase clauses, and when counting the frequency of the body phrase clauses, the semantic categories of the body phrases in the body phrase clause are the same, and , The other notation information is the same, and the frequency of the body phrase is counted, or
For a prescriptive phrase in the data, take a set of a set of a body phrase phrase that is in a dependency relationship with the prescriptive phrase in the corpus and a frequency of the body phrase phrase, or a dependency relationship with the prescriptive phrase It is a set of a set phrase phrase and the frequency of the set phrase phrase, and when counting the frequency of the set phrase phrase, the notation semantic category in the set phrase phrase is the same and other notation information is the same Take the same frequency and count the frequency of the body phrase,
A deep case estimation method characterized by that. Program for a computer to function as each unit of the deep case learning equipment according to claim 1, wherein. Program for causing a computer to function as each section of the deep case estimation apparatus according to any one of 請 Motomeko 2 to claim 4.

Images