JP2018160159A - Uttered sentence determining device, method, and program - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide an uttered sentence determining device that more properly determines a speech inadequate as a speech from a system.SOLUTION: In an arithmetic unit 20 of the uttered sentence determining device 100, a cue expression extracting unit 32 extracts, based on dictionaries and rules prepared in advance for extracting an expression of a speech and assigning a class thereto, an expression and cue expression information including a class of the expression from the speech. An inappropriate topic word label determining unit 36 extracts an inappropriate topic word expression from a speech, and determines an inappropriate topic word label based on the class of the topic word expression. An inappropriate speech label determining unit 40 extracts an inappropriate predicate expression from the speech, determines an inappropriate predicate label based on the class of the predicate expression, determines a target label based on the classes of case elements corresponding to the predicate expression, and links the determined inappropriate predicate label and target label and determines a result as an inappropriate speech label.SELECTED DRAWING: Figure 4

Description

  The present invention relates to an utterance sentence determination apparatus, method, and program, and more particularly, to an utterance sentence determination apparatus, method, and program for determining an utterance inappropriate as an utterance of a system.

  Non-patent document 1 and patent document 1 are examples of research for automatically acquiring utterance candidate sentences from a large number of user utterance sentences. These are methods for extracting high-quality utterance sentences by determining whether the utterances are correct as Japanese.

  As another related study, for example, Non-Patent Document 2 is a study for detecting “abusive language” such as hate speech. This is to determine whether a user comment is abusive or clean in the framework of supervised machine learning.

Japanese Patent No. 5676683

Michimasa Inaba, Ayaka Kamizono, Kenichi Takahashi, Acquisition of utterance candidate sentences for non-task-oriented dialogue system using Twitter, Journal of the Japanese Society for Artificial Intelligence Vol.29 No.1 2014. Chikashi Nobata, Joel Tetreault, Achint Thomas, Yashar Mehdad, Yi Chang, Abusive Language Detection in Online User Content.

  The conventional method determines correctness as Japanese as described above, and cannot filter utterances inappropriate as system utterances. For example, the utterance "Northern territory is Japanese" is a sentence with no problem in Japanese, but it is not appropriate as an utterance of the system because it may be a problem depending on the listener.

  Also, in related research to determine inappropriate comments, user comments are determined as binary values, abusive or clean, but whether or not a certain utterance should be inappropriate depends on the system character settings, For example, if a character is abusive, the expression of abuse is OK (but adult is NG), so it depends on the character. In binary classification, it is necessary to create correct data and a judgment model for each character, and it can be applied to various characters. The cost is high.

  The present invention has been made to solve the above problems, and provides an utterance sentence determination device, method, and program capable of more appropriately determining an inappropriate utterance as an utterance of a system. Objective.

  To achieve the above object, an utterance sentence determination device according to a first invention divides an inputted utterance into sentence units, and generates text analysis information by dividing the divided sentence into morpheme units. Based on an analysis unit, the text analysis information, and a dictionary and a rule created in advance for extracting an expression of the utterance and assigning a class, a method including the expression and the class of the expression from the utterance A clue expression extracting unit that extracts overexpression information, a stylistic label determining unit that determines a stylistic label indicating the type of the utterance based on the text analysis information, the text analysis information, and the overexpression expression information An inappropriate topic word expression is extracted from the utterance and an inappropriate topic word label is determined based on the topic word expression class; Based on the text analysis information and the clue expression information, an inappropriate predicate expression is extracted from the utterance, an inappropriate predicate label is determined based on the class of the predescript word expression, An inappropriate utterance label determination unit that determines an object label based on a class of corresponding case elements, and determines the inappropriate utterance label by linking the determined inappropriate predicate label and the object label; It is configured to include.

  Further, in the spoken sentence determination device according to the first aspect of the invention, the dictionaries include a body language that can be an inappropriate expression and a negative body language dictionary that includes the body language class, a target expression that can be an inappropriate expression, and the target expression. It consists of a target expression dictionary including classes, and a prescriptive dictionary including predicates that can be inappropriate expressions and the prescriptive class, and the rules extract target expressions that can be inappropriate expressions and assign classes. It is also possible to extract a predicate extraction rule for assigning a class by extracting a predicate expression that can be an inappropriate expression as well as an object extraction rule for the purpose.

  Further, in the utterance sentence determination device according to the first invention, the inappropriate utterance label determination unit corresponds to the class of the pre-description word expression included in the utterance and the pre-description word expression obtained from the clue expression information. An inappropriate predicate label determination unit that extracts an inappropriate predicate expression based on a class of case elements to be determined and determines an inappropriate predicate label based on a class of a previous descriptor expression, and obtained from the clue expression information. A target label determination unit that determines a target label based on a class of case elements corresponding to the pre-descriptor expression, the inappropriate predicate label, and the target label are connected to form the inappropriate utterance label. An inappropriate utterance label output unit.

  In the spoken sentence determination method according to the second invention, the text analysis unit divides the inputted utterance into sentence units, and generates the text analysis information by dividing the divided sentences into morpheme units; An expression extraction unit includes an expression and a class of the expression from the utterance based on the text analysis information and dictionaries and rules created in advance to extract the expression of the utterance and assign a class. A step of extracting handwriting expression information, a step of determining a stylistic label indicating a style of the utterance based on the text analysis information, and a step of determining an inappropriate topic word label determining unit Based on the information and the clue expression information, an inappropriate topic word expression is extracted from the utterance, and an inappropriate topic word label is extracted based on the topic word expression class. And an inappropriate utterance label determination unit extracts an inappropriate predicate expression from the utterance based on the text analysis information and the clue expression information, and based on a class of predescription expression The inappropriate predicate label is determined, the target label is determined based on the class of the case element corresponding to the previous descriptor expression, and the determined inappropriate predicate label and the target label are connected to be inappropriate. And determining the utterance label.

  Further, in the spoken sentence determination method according to the second invention, the dictionaries include a negative expression dictionary including an expression that can be an inappropriate expression and the expression class, an object expression dictionary including an object expression and the object expression class. , And a prescriptive dictionary including a predicate and a class of the predicates, and the rules extract a target expression that can be an inappropriate expression, an object extraction rule for assigning a class, and an inappropriate expression Predicate expressions that can be extracted are extracted, and predicate extraction rules for assigning classes may be included.

  Further, in the utterance sentence determination method according to the second invention, the step of determining the inappropriate utterance label determination unit includes a pre-description included in the utterance obtained by the inappropriate predicate label determination unit from the clue expression information. Extracting an inappropriate predicate expression based on a class of word expressions and a class of case elements corresponding to the preceding descriptor expression, and determining an inappropriate predicate label based on the class of the preceding descriptor expression; The target label determining unit determines a target label based on a class of case elements corresponding to the predescription word expression obtained from the clue expression information; and the inappropriate utterance label output unit includes the inappropriate predicate The method may include a step of connecting a label and the target label to form the inappropriate utterance label.

  A program according to the third invention is a program for causing a computer to function as each part of the spoken sentence determination device according to the first invention.

  According to the utterance sentence determination apparatus, method, and program of the present invention, an expression and a class of expression are extracted from an utterance based on a dictionary and rules created in advance to extract the expression of the utterance and assign a class. Extracting clue expression information including, extracting inappropriate topic word expressions from utterances, determining inappropriate topic word labels based on topic word expression classes, extracting inappropriate predicate expressions from utterances, The inappropriate predicate label is determined based on the predicate expression class, the target label is determined based on the class of the case element corresponding to the predicate expression, and the determined inappropriate predicate label and the target label are concatenated. By determining the inappropriate utterance label, it is possible to more appropriately determine an inappropriate utterance as the utterance of the system.

It is a figure which shows an example of an inappropriate topic word label. It is a figure which shows an example of an inappropriate utterance label. It is a figure which shows an example of a style label. It is a block diagram which shows the structure of the utterance sentence determination apparatus which concerns on embodiment of this invention. It is a figure which shows an example of text analysis information. It is a figure which shows an example of a negative word dictionary. It is a figure which shows an example of an object expression dictionary. It is a figure which shows an example of a precaution dictionary. It is a figure which shows an example of an object extraction rule. It is a figure which shows an example of a predicate extraction rule. It is a figure which shows an example of the class of predicate expression. It is a flowchart which shows the utterance sentence determination processing routine in the utterance sentence determination apparatus which concerns on embodiment of this invention.

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

<Outline according to Embodiment of the Present Invention>

  In the embodiment of the present invention, the above-mentioned problem is solved by means for giving a label indicating inappropriateness to an utterance that is not appropriate as an utterance of the system without using correct data.

  In the embodiment of the present invention, the input is an utterance sentence (text), and the output is <an inappropriate topic word label, an inappropriate utterance label, a style label>. Each label may be empty.

Here are some examples of input and output.

Example 1)
Input: T island is illegally occupied by T country Output: <POLITICAL_ISSUE, NEG_ACT / NATION,>

Example 2)
Input: The elderly don't squeeze Output: <PSET, / PSET, PROH / IMP>

Example 3)
Input: That is a different output: <, SWD, NOT_POLITE>

  For inappropriate topic word labels, inappropriate utterance labels, and style labels, a label set defined in advance by hand is used. In this configuration, the inappropriate topic word label shown in FIG. 1, the inappropriate utterance label shown in FIG. 2, and the style label shown in FIG. 3 are used, but other label sets may be defined and used.

<Configuration of Utterance Sentence Determination Device According to Embodiment of the Present Invention>

  Next, the configuration of the spoken sentence determination device according to the embodiment of the present invention will be described. As shown in FIG. 4, the utterance sentence determination device 100 according to the embodiment of the present invention includes a CPU, a RAM, a ROM that stores a program and various data for executing an utterance sentence determination processing routine described later, Can be configured with a computer including Functionally, the spoken sentence determination apparatus 100 includes an input unit 10, a calculation unit 20, and an output unit 48 as shown in FIG.

  The input unit 10 receives a set of utterances.

  The arithmetic unit 20 includes a negative body dictionary 21, a target expression dictionary 22, a predicate dictionary 23, a target extraction rule 24, a predicate extraction rule 25, a text analysis unit 30, a clue expression extraction unit 32, and a style label. The determination unit 34 includes an inappropriate topic word label determination unit 36, an inappropriate utterance label determination unit 40, and an output unit 48.

  For each utterance of the set of utterances received by the input unit 10, the text analysis unit 30 divides the utterance into sentence units, and divides the divided sentence into morpheme units to obtain text analysis information about each utterance. Generate. Sentence sentences (text) of divided utterances are input, and morphological information (part of speech, standard notation, reading, etc.) and dependency are obtained as text analysis information by morphological analysis, dependency analysis, and specific expression extraction, which are known techniques. Information and specific expression information.

  FIG. 5 shows text analysis information generated by the text analysis unit 30 when an utterance “T island is illegally occupied by K country” is received as an example. As shown in FIG. 5, the received utterance is divided into morpheme units, and text analysis information including appearance notation, part of speech, and thesaurus number (for example, a Japanese vocabulary system number) is generated. Also, “T island” and “K country” are extracted as the class “LOCATION” of the unique expression. The definition of the proper expression follows Reference Document 1. The phrase information includes information of “sentence ID, dependency clause ID / relationship type, main morpheme number / function word morpheme number”.

[Reference 1]: IREX (https://nlp.cs.nyu.edu/irex/index-j.html)

  The clue expression extraction unit 32 extracts the text analysis information of each utterance generated by the text analysis unit 30, the expression of the utterance, and the negative expression dictionary 21 and the target expression dictionary 22 which are created in advance to assign a class. Based on the prescriptive dictionary 23, the object extraction rule 24, and the predicate extraction rule 25, the clue expression information including the expression and the expression class is extracted from each utterance.

  As shown in FIG. 6, the negative word dictionary 21 is a dictionary including a word that can be an inappropriate expression, a standard notation of the word, and a word class.

  As shown in FIG. 7, the target expression dictionary 22 is a dictionary including the target expression, the standard notation of the target expression, and the class of the target expression.

  As shown in FIG. 8, the precaution dictionary 23 is a dictionary including prescriptions, standard notation of prescriptions, and prescriptive classes.

  As shown in FIG. 9, the target extraction rule 24 is a rule for extracting a target expression that can be an inappropriate expression and assigning a class. In addition, it supplements about description of rule. <> Represents a morpheme. The left of the colon (:) indicates the field to be matched, and the right indicates the field value. The slash (/) is an and condition. Others conform to the description of regular expressions.

  As shown in FIG. 10, the predicate extraction rule 25 is a rule for extracting a predicate expression that can be an inappropriate expression and assigning a class.

  A processing flow of the clue expression extraction unit 32 will be described. In the clue expression extraction unit 32, first, the text analysis information of the utterance analyzed by the text analysis unit 30 is input, and the negative phrasal dictionary 21, the target expression dictionary 22, and the predicate are applied to the morpheme sequence of the analyzed utterance sentence. If three dictionaries 23 are applied and there is an expression existing in the dictionary, the expression, standard notation, and class information are extracted as clue expression information together with the extraction source dictionary information. By adding dictionary information, it is possible to determine whether the extracted expression is a negative expression, a target expression, or a predicate. Further, the corresponding expression may be extracted using not only the dictionary but also a rule based on the regular expression of the object extraction rule 24. For example, “place name + person” is extracted as ETHNIC by applying rule A in FIG.

  Next, the predicate expression and its class (predicate class) are extracted using the extracted clue expression information and the predicate extraction rule 25. In the rule A shown in FIG. 10, when there is a notation “shi” after a morpheme that applies to the class “NEG_STATE” extracted by applying the prescriptive dictionary 23, the notation is designated as a predicate expression class “NEG_ACT”. It is extracted and used as clue expression information. For example, in the case of a sentence “Occupying”, “Occupying” is a predicate expression class and NEG_ACT is a predicate expression class.

  The rule B shown in FIG. 10 is an example of a rule that extracts a predicate expression representing a desire such as “I want to do XX / I want you to do XX” and gives a class “ACT_WANT” of the predicate expression. This is an example of a rule that extracts a predicate expression representing a desire for a negative action such as “I want to break” and assigns a class “NEG_WANT” of the predicate expression.

  For expressions that are extracted by the predicate dictionary 23 and do not apply to the predicate extraction rule 25, the expressions and classes extracted by the predicate dictionary 23 are classes of predicate expressions and predicate expressions. In addition, expressions corresponding to “ADLT” and “SWD” in negative expressions are also treated as predicate expressions and predicate expression classes.

  The style label determination unit 34 determines a style label indicating each style of the utterance based on the text analysis information of each utterance generated by the text analysis unit 30. Here, the style label is determined based on the morphological information at the end of the sentence in the text analysis information of the utterance. First, when the content word at the end of an utterance sentence is a verb, if the morpheme that follows is “should”, the style label is determined as “DEO”. In other cases, when the content word at the end of the utterance sentence is “verb continuous form” or “verb basic form” is followed by “na”, the style label is determined as “PROH / IMP”. Otherwise, the input utterance will match any of the following: “is”, “deshi (ta | te | cho)”, “mas”, “masashi (ta | te | yo)”, “not ($ | n)” Otherwise, the style label is determined as “NOT_POLITE”.

  The inappropriate topic word label determination unit 36, based on the text analysis information of each utterance generated by the text analysis unit 30 and the handcuff expression information extracted about each utterance by the handcuff expression extraction unit 32, An inappropriate topic word expression is extracted from each utterance, and an inappropriate topic word label is determined based on the topic word expression class.

  The inappropriate topic word label determination unit 36 first identifies a word (topic word) that is the topic of the utterance from the text analysis information of the utterance. Topic words can be identified using particles as clues. For example, it can be realized by extracting equivalents of noun phrases with the particles “ha”, “ga”, “te”, and “what”. When there is no corresponding word, the topic word is empty.

  In the case of the above-mentioned example “T island is illegally occupied by K country”, “T island”, which is a noun accompanied by the consecutive particle “ha”, is extracted as a topic word. Although topic word extraction based on rules has been described here, it may be realized using a chunking technique (CRF or the like) used in a general specific expression extraction technique (for example, Non-Patent Document 3).

[Non-patent literature 3]: Shinpei Kanno, Ikuo Kurohashi, Japanese proper expression recognition using global information, Journal of Information Processing Society of Japan Vol.49 No.11, 2008.

  Next, if the extracted topic word is a negative word or target expression, the inappropriate topic word label determination unit 36 outputs the class of the expression as an inappropriate topic word label. Otherwise, the inappropriate topic word label is empty.

  The inappropriate utterance label determination unit 40 utters speech based on the text analysis information of each utterance generated by the text analysis unit 30 and the cue expression information extracted for each utterance by the cue expression extraction unit 32. Inappropriate predicate expressions are extracted from each of them, an inappropriate predicate label is determined based on the class of the predicate expression, a target label is determined based on the class of the case element corresponding to the predicate expression, and the The appropriate predicate label and the target label are concatenated and determined as an inappropriate utterance label.

  The inappropriate utterance label determination unit 40 includes an inappropriate predicate label determination unit 42, a target label determination unit 44, and an inappropriate utterance label output unit 46.

  The inappropriate utterance label determination unit 40 first determines whether or not a predicate expression exists in the clue expression information. If the predicate expression exists, the following inappropriate predicate label determination unit 42 and target label determination unit 44 are processed. If not, the inappropriate utterance label output unit 46 outputs the inappropriate utterance label as empty. When there are a plurality of predicate expressions, processing is performed on the last predicate expression. The case element of the predicate is obtained by extracting an expression related to the predicate with a case particle such as “ga” and “wa” or a combined particle using the dependency result of the text analysis information.

  The inappropriate predicate label determination unit 42 determines each utterance based on the predicate expression class included in each utterance and the case element class corresponding to the predicate expression obtained from the clue expression information as follows. For example, an inappropriate predicate expression is extracted, and an inappropriate predicate label is determined based on the class of the predicate expression.

  If the class of the predicate expression matches “(PLUS | PROS)”, if there is a negative word in “G”, the inappropriate predicate label is “NEG_STATE”.

  For example, if the utterance is “war will increase”, the case element and predicate expression are “[G case: N_N] war, [class of predicate expression: PLUS]: increase”. Set to “NEG_STATE”.

  If the predicate expression class is “POS_EVAL” and there is a negative expression in “G”, the inappropriate predicate label is “POS_EVAL_NEG”.

  For example, if the utterance is “war is fun”, the case element and predicate expression are “[Case: N_N] war, [class of predicate expression: POS_EVAL]: fun”. POS_EVAL_NEG ”.

  If the predicate expression class is “WANT_ACT” and there is a negative word in “wo case” or “ni case”, the inappropriate predicate label is “NEG_WANT”.

  For example, if the utterance is “I want to do war,” the case element and predicate expression are “[Wo: N_N] war, [Predicate expression class: WANT_ACT]: I want to do”, so I ’ll give an inappropriate predicate label. ” NEG_WANT ”.

  If the predicate expression class is other than the above, the predicate expression class is set as an inappropriate predicate label. However, if the predicate expression class is either PLUS or MINUS, the inappropriate predicate label is empty.

  For example, if the utterance is “This idiot is”, [unsuitable predicate label: SWD]: idiot. On the other hand, if the utterance says “I'm hungry”, the inappropriate predicate label is empty.

  The target label determination unit 44 determines the target label based on the class of the case element corresponding to the predicate expression obtained from the clue expression information as follows.

  In the target label determination unit 44, the following processing is performed in the order of “ga”, “ha”, “wo”, “ni”, “de”, “other”, and the processing is stopped when the target label is determined. To do.

  First, if the “ga” of the predicate expression is any of a negative body expression, an object expression, or a specific expression, the class of the case element is set as the object label. If multiple classes exist in the same notation, the priority order is negative expression> target expression> specific expression.

  For example, if the utterance is “K island is illegally occupied on T island”, the target label and notation will be [NATION (K country)].

  If there is no case, or the target label is not determined, the same processing is performed in the order of “C”, “Wo”, “D”, “De” and “Other”. If the target label is not determined after checking all cases, the target label is empty. As an exception, when the inappropriate predicate label determined by the inappropriate predicate label determination unit 42 is “NEG_WANT”, the process is performed only for “Case”. If the target label is not determined despite the presence of either ga or c, the target label is empty, and if neither ga or c is present, the target label is “SELF”. To do.

  For example, if the utterance is “I want to crush country A”, the target label is [SELF]. On the other hand, if the utterance is “B country wants to crush A country”, the target label is [NATION].

  The inappropriate utterance label output unit 46 concatenates the inappropriate predicate label and the target label into an inappropriate utterance label. Here, an inappropriate predicate label and a target label connected by '/' are output as an inappropriate utterance label. If both are empty, '/' is not output. However, '/' may be output.

  For example, if the utterance is “I want to crush country A”, [NEG_WANT / SELF] is an inappropriate utterance label. Further, if the utterance is “B country wants to crush country A”, [NEG_WANT / NATION] is an inappropriate utterance label.

  If the inappropriate predicate label is PROS, CONS, or POS_EVAL, the following processing is performed to rewrite the inappropriate predicate label and output the inappropriate utterance label.

  If the target label is derived from a negative expression or target expression, if the inappropriate predicate label is PROS or POS_EVAL, POS_EVAL is the inappropriate predicate label, and if the inappropriate predicate label is CONS, NEG_EVAL is the inappropriate predicate label. .

  If the above is not true, the inappropriate predicate label is empty.

  For example, if the utterance is “Housewife is easy”, ([Target label: PSET (Housewife)], [Inappropriate predicate label: POS_EVAL (Easy)]), [POS_EVAL / PSET] is inappropriate. It becomes a label.

  If the utterance is “bicycle is easy”, it is ([target label: none], [inappropriate predicate label: POS_EVAL (easy)]), so the inappropriate predicate label is empty and [empty] is not. Appropriate utterance label.

  The output unit 48 uses the inappropriate topic word label determined by the inappropriate topic word label determination unit 36, the inappropriate utterance label output from the inappropriate utterance label output unit 46, and the style determined by the style label determination unit 34. Output the final result of combining the labels.

  For example, if the utterance is “K island is illegally occupied on T island”, <POLITICAL_ISSUE, NEG_ACT / NATION,> is output.

<Operation of Speech Sentence Determination Device According to Embodiment of the Present Invention>

  Next, the operation of the spoken sentence determination apparatus 100 according to the embodiment of the present invention will be described. When the input unit 10 receives a set of utterances, the utterance sentence determination device 100 executes an utterance sentence determination processing routine shown in FIG.

  First, in step S100, for each of the utterances of the set of utterances received by the input unit 10, the utterance is divided into sentence units, and the divided sentence is divided into morpheme units to obtain text analysis information about each utterance. Generate.

  Next, in step S102, the text analysis information of each utterance generated in step S100 and the expression of the utterance and the negative expression dictionary 21 and the target expression dictionary 22 created in advance for assigning classes are used. Based on the word dictionary 23, the object extraction rule 24, and the predicate extraction rule 25, the clue expression information including the expression and the expression class is extracted from each utterance.

  In step S104, a style label indicating each style of the utterance is determined based on the text analysis information of each utterance generated in step S100.

  In step S106, an inappropriate topic word expression is extracted from each utterance based on the text analysis information of each utterance generated in step S100 and the clue expression information extracted for each utterance in step S102. Then, an inappropriate topic word label is determined based on the topic word expression class.

  In step S108, it is determined whether or not the predicate expression exists in the clue expression information. If it exists, the process proceeds to step S112, and if not, the process proceeds to step S110.

  In step S110, the inappropriate utterance label is output as empty.

  In step S112, each of the utterances is determined based on the predicate expression class included in each of the utterances and the case element class corresponding to the predicate expression obtained from the clue expression information extracted for each of the utterances in step S102. For example, an inappropriate predicate expression is extracted, and an inappropriate predicate label is determined based on the class of the predicate expression.

  In step S114, a target label is determined based on the case element class corresponding to the predicate expression obtained from the clue expression information extracted for each utterance in step S102.

  In step S116, for each utterance, the inappropriate predicate label determined in step S112 and the target label determined in step S114 are concatenated to form an inappropriate utterance label.

  In step S118, for each utterance, the inappropriate topic word label determined in step S106, the inappropriate utterance label output from the inappropriate utterance label output unit 46, and the style label determined by the style label determination unit 34 are displayed. The combined final result is output and the process ends.

  As described above, according to the utterance sentence determination device according to the embodiment of the present invention, text analysis information is generated by dividing an inputted utterance into sentence units and dividing the divided sentence into morpheme units. Based on text analysis unit, text analysis information, and dictionary and rules created in advance to extract utterance expressions and assign classes, cue expression including expression and expression classes The information is extracted, the stylistic label indicating the style of the utterance is determined based on the text analysis information, and the inappropriate topic word expression is extracted from the utterance based on the text analysis information and the clue expression information. An inappropriate topic word label is determined based on the word expression class, and an inappropriate predicate expression is extracted from the utterance based on the text analysis information and the clue expression information. Inappropriate predicate label is determined, the target label is determined based on the class of the case element corresponding to the predicate expression, and the determined inappropriate predicate label and the target label are concatenated and determined as an inappropriate utterance label. By doing so, it is possible to more appropriately determine an utterance inappropriate as an utterance of the system.

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

  For example, in the above-described embodiment, the negative expression dictionary 21, the target expression dictionary 22, and the usage dictionary 23 are used as the dictionary for extracting the utterance expression and assigning the class, and the target extraction rule 24, and Although the predicate extraction rule 25 is given as an example, the present invention is not limited to this, and other dictionaries and other rules may be applied.

DESCRIPTION OF SYMBOLS 10 Input part 20 Operation part 21 Negative expression dictionary 22 Target expression dictionary 23 Prediction dictionary 24 Target extraction rule 25 Predicate extraction rule 30 Text analysis part 32 Clue expression extraction part 34 Stylistic label determination part 36 Inappropriate topic word label determination part 40 Not Appropriate utterance label determination unit 42 inappropriate predicate label determination unit 44 target label determination unit 46 inappropriate utterance label output unit 48 output unit 100 utterance sentence determination device

Claims (7)

A text analysis unit that generates text analysis information by dividing an input utterance into sentence units and dividing the divided sentences into morpheme units;
Based on the text analysis information and dictionaries and rules created in advance to extract the expression of the utterance and assign a class, clue expression information including the expression and the class of the expression is included from the utterance. A clue expression extraction unit to extract;
Based on the text analysis information, a stylistic label determining unit that determines a stylistic label indicating the stylistic of the utterance;
Inappropriate topic that extracts an inappropriate topic word expression from the utterance based on the text analysis information and the cue expression information, and determines an inappropriate topic word label based on the class of the topic word expression A word label determination unit;
Based on the text analysis information and the clue expression information, an inappropriate predicate expression is extracted from the utterance, an inappropriate predicate label is determined based on the class of the predescription expression, and the predescription expression An inappropriate utterance label determining unit that determines a target label based on a class of case elements corresponding to the above, and determines the inappropriate utterance label by linking the determined inappropriate predicate label and the target label;
Utterance sentence determination device including The dictionaries include an expression that can be an inappropriate expression and a negative expression dictionary including a class of the expression, an object expression dictionary including an object expression and the class of the object expression, and a statement including a predicate and the class of the expression. Consist of a dictionary,
The above rules extract target expressions that can be inappropriate expressions, extract target extraction rules for assigning classes, and predicate expression rules that can extract inappropriate predicate expressions that can be inappropriate expressions. The utterance sentence determination apparatus according to claim 1. The inappropriate utterance label determination unit
Extracting an inappropriate predicate expression based on the class of the predescription word expression included in the utterance and the class of the case element corresponding to the predescription word expression obtained from the clue expression information, An inappropriate predicate label determination unit that determines an inappropriate predicate label based on the class of
A target label determination unit for determining a target label based on a class of case elements corresponding to the pre-description word expression obtained from the clue expression information;
An inappropriate utterance label output unit that connects the inappropriate predicate label and the target label to form the inappropriate utterance label;
The utterance sentence judging device according to claim 2 containing. A step in which the text analysis unit divides the input utterance into sentence units, and generates the text analysis information by dividing the divided sentence into morpheme units;
The clue expression extraction unit extracts the expression and the class of the expression from the utterance based on the text analysis information and a dictionary and rules created in advance for extracting the expression of the utterance and assigning the class. Extracting handcuff expression information including;
A stylistic label determining unit determining a stylistic label indicating a stylistic style of the utterance based on the text analysis information;
The inappropriate topic word label determination unit extracts an inappropriate topic word expression from the utterance based on the text analysis information and the clue expression information, and inappropriate based on the topic word expression class. Determining a topic word label;
The inappropriate utterance label determination unit extracts an inappropriate predicate expression from the utterance based on the text analysis information and the clue expression information, and an inappropriate predicate label based on the class of the previous description word expression And determining the target label based on the class of the case element corresponding to the previous description expression, and determining the inappropriate utterance label by linking the determined inappropriate predicate label and the target label. Steps,
Utterance sentence determination method including The dictionaries include an expression that can be an inappropriate expression and a negative expression dictionary including a class of the expression, an object expression dictionary including an object expression and the class of the object expression, and a statement including a predicate and the class of the expression. Consist of a dictionary,
The above rules extract target expressions that can be inappropriate expressions, extract target extraction rules for assigning classes, and predicate expression rules that can extract inappropriate predicate expressions that can be inappropriate expressions. The spoken sentence determination method according to claim 4. The step of determining the inappropriate utterance label determination unit includes:
The inappropriate predicate label determination unit obtains an inappropriate predicate expression based on the class of the previous description word expression included in the utterance and the class of the case element corresponding to the previous description word expression obtained from the clue expression information. And determining an inappropriate predicate label based on the class of predescriptor expression;
A target label determining unit determining a target label based on a class of case elements corresponding to the previous description word expression obtained from the clue expression information;
An inappropriate utterance label output unit concatenating the inappropriate predicate label and the target label into the inappropriate utterance label;
The spoken sentence determination method according to claim 5, comprising:   The program for functioning a computer as each part of the utterance sentence determination apparatus of any one of Claims 1-3.