JP4437778B2 - Vertical relationship determination method, vertical relationship determination device, vertical relationship determination program, and recording medium - Google Patents

Description

  The present invention relates to a natural language processing system that performs machine translation and information retrieval, for example.

  Natural languages described in various languages such as Japanese and English often have abstract and highly ambiguous properties, but they can be processed computerically by handling sentences mathematically. As a result, various natural language processing systems such as a machine translation system, a dialogue system, a search system, and a question answering system are realized.

  By the way, honorifics consisting of respected words, humility words, polite words, etc. are used in natural language, especially Japanese, but it is not absolute which one of those honorific words is used, and human society related to sentences It is determined by relative measures such as target status, age, speaker's subjective familiarity, and gender. As an example, a case where honorific words are used for the sentence “A manager has come” will be described. For example, when a speaker speaks with an in-house colleague, a respected word is used to say “A manager has come”. Also, when the speaker shows respect for his colleague, the polite language is used to say, “The manager has come.” In addition, when the speaker speaks to the customer, the head of the department has come, using humility and polite language. Such honorifics can be used properly based on general common sense or the like of a person, but in a machine, it is performed based on a predetermined algorithm. Therefore, in order to correctly interpret and properly use honorifics in a natural language processing system, an algorithm for determining the human hierarchical relationship related to sentences is required. For this reason, although it is fragmentary, the proposal which considered honorific usage is made (for example, refer nonpatent literatures 1 and 2).

K. Dohsaka, Identifying the referents of zero-pronouns in Japanese dialogues, In Proc. Of the 9th European Conference on Artificial Intelligence. 240-245, 1990 M, Siegel, Japanese Honorification in an HPSG Frameworks, In Proc. Of the 14th Pacific Asia Conference on Language, Information and Computation, 289-300, Waseda University, Tokyo, 2000

  However, the conventional natural language processing system focuses on interactive sentences or first-person and second-person characters, and only reflects absolute and fragmentary honorific information from only the vocabulary, and is related to sentences. As a result, it has not been possible to improve the processing accuracy because the human vertical relations cannot be correctly determined.

  Accordingly, the present invention has been made to solve the above-described problems, and a hierarchical relationship determination method, a hierarchical relationship determination device, and a vertical relationship that can relatively determine the human hierarchical relationship related to a sentence. An object is to provide a determination program and a recording medium.

In order to solve the above-described problems, the hierarchical relation determination method according to the present invention analyzes a digitized input sentence, and talks about the speaker of the input sentence, the partner of the speaker, and the input sentence. The upper and lower relationship determination method for determining an upper limit relationship of an instruction target consisting of three persons with a target person , an analysis step in which an analysis unit performs morphological analysis and syntax analysis on the input sentence, and a detection unit , Referring to the honorific dictionary, a detection step of detecting a honorific and a verb that semantically restricts the relationship between the subject and the object in the sentence from the analysis result of the analysis unit, and the extraction unit analyzes the analysis unit An extraction step for extracting a specific noun that can be the instruction target from the result, and the instruction target specifying unit, the noun extracted by the extraction unit is the speaker when the first person is first person, and the partner when the second person is the second person In other cases The specific step of identifying the target person, and the honorific determination unit determines that the instruction object corresponding to the subject is higher than the other instruction objects when the honorific word detected by the detection unit is a respect expression, When the honorific detected by the detection unit is a humility expression, it is determined that the instruction object corresponding to the subject is lower than the other instruction objects, and the relationship between the subject and the object in the sentence detected by the detection unit is semantically A verb that semantically constrains the relationship between the subject and the object in the sentence detected by the detection unit, when the restricting verb is a respected expression, the instruction object corresponding to the subject is determined to be higher than the other instruction objects Is a modest expression, the instruction object corresponding to the subject is determined to be lower than the other instruction objects, and if the polite expression of the polite expression is not detected by the detection unit, the opponent is equal to the speaker or more than the speaker By judging the subordinate, Perform a honorific determining step of determining a vertical relationship dictates characterized Rukoto.

In the above-described hierarchical relationship determination method, when the honorific judgment unit cannot detect the hierarchical relationship of the target object from the honorific, the rank is a database that records the social, age, or generational hierarchical relationship between the nouns. You may make it perform further the rank determination step which refers to DB and determines the hierarchical relationship of an instruction | indication object .

In the above-described hierarchical relationship determination method, when the honorific determination unit detects the hierarchical relationship of the target object from honorifics, the rank determination unit is a database that records the social, age, or generational hierarchical relationship between each noun. You may make it further perform the rank confirmation step which confirms the hierarchical relationship of an instruction | indication object with reference to DB . In this rank confirmation step, a plurality of hierarchical relationships with different categories associated with each other are recorded in the rank DB, and the rank determination unit refers to the plurality of categories to confirm the hierarchical relationship of the instruction target. You may make it perform.

In the above hierarchical relationship determining method, specifying step, if the subject from the input sentence by the extraction unit is not extracted, it may be specific speaker as the subject.

Further, the hierarchical relationship determination device according to the present invention analyzes an input sentence that has been digitized, and includes three persons: a speaker of the input sentence, a partner of the speaker, and a target person who is talking about the input sentence. An upper / lower relationship determination device for determining an upper limit relationship of an instruction target, an analysis unit that performs morphological analysis and syntax analysis on an input sentence, and an honorific dictionary and sentence from an analysis result of the analysis unit with reference to an honorific dictionary That detects verbs that semantically constrain the relationship between the subject and object in the database, an extraction unit that extracts specific nouns that can be directed from the analysis results of the analysis unit, and nouns extracted by the extraction unit However, in the case of the first person, the speaker, in the case of the second person, the other party, in other cases the target object specifying part that specifies the target person, and the honorific expression detected by the detection part is the subject The target object corresponding to is higher than the other target objects. Determines that the detection unit is determined to lower than the referent otherwise indicated object corresponding to the subject in the case of honorific is humility expression detected, the relationship between the subject and the object in the sentence detecting unit detects The verb that constrains the relationship between the subject and the object in the sentence detected by the detection unit is determined that the instruction target whose semantically constrained verb corresponds to the subject of the respected expression is higher than the other target objects. In the case of humble expression, the instruction object corresponding to the subject is determined to be lower than the other instruction objects, and if the polite expression of the polite expression is not detected by the detection unit, the partner is equivalent to the speaker or lower than the speaker. And a honorific determination unit that determines the vertical relationship of the instruction target.

The hierarchical relationship determining program according to the present invention causes a computer to function as each unit of the hierarchical relationship determining device .

The recording medium according to the present invention is characterized in that the above-described hierarchical relationship determination program is recorded.

  According to the present invention, an electronic sentence is analyzed, a speaker, a partner, and a target person are identified, and a hierarchical relationship between the identified three parties is determined. It is possible to determine a more specific human relative vertical relationship related to the sentence of the partner and the subject.

[First Embodiment]
Hereinafter, a first embodiment of the present invention will be described with reference to the drawings. FIG. 1 is a block diagram showing the configuration of the vertical relationship determination apparatus according to the present embodiment. The hierarchical relationship detection apparatus 1 includes an input unit 11, an analysis unit 12, a grammar dictionary 13, a detection unit 14, an honorific dictionary 15, an extraction unit 16, a determination unit 17, and a rank DB (Data Base) 18. And the internal / external relation DB 19. Such an up-and-down relationship detection device 1 detects an input of information from the outside such as an arithmetic device such as a CPU, a storage device such as a memory and an HDD (Hard Disc Drive), and a keyboard, a mouse, a pointing device, a button, and a touch panel. Input devices, I / F devices that transmit and receive various types of information via communication lines such as the Internet, LAN (Local Area Network), and WAN (Wide Area Network), CRT (Cathode Ray Tube), LCD (Liquid Crystal) It comprises a computer equipped with a display device such as a display or FED (Field Emission Display) and a program installed in the computer. That is, the hardware device and the software cooperate to control the above hardware resources by a program, and the input unit 11, the analysis unit 12, the grammar dictionary 13, the detection unit 14, the honorific dictionary 15, and the extraction unit 16 described above. The determination unit 17 and the rank DB 18 are realized. Note that the program may be provided in a state of being recorded on a recording medium such as a flexible disk, a CD-ROM, a DVD-ROM, or a memory card.

  The input unit 11 is an interface device that accepts texts input from outside (hereinafter referred to as “input texts”).

  The analysis unit 12 is an arithmetic processing unit that performs morphological analysis on the input sentence input to the input unit 11 based on the grammatical rules stored in the grammar dictionary 13.

  The grammar dictionary 13 is a database in which grammatical rules for a natural language of interest are registered.

  The detection unit 14 is an arithmetic processing unit that detects a honorific from an input sentence that has been morphologically analyzed by the analysis unit 12 based on the honorific dictionary 15.

  As shown in FIG. 2, the honorific dictionary 15 is a database in which honorifics corresponding to each item of “predicate”, “sentence pattern”, “semantic restriction of verb / sa variable noun”, and “noun” are registered. The “respected expression” includes “respected words” as defined in the conventional Japanese language grammar, sentence patterns, semantic restrictions of verbs or sa-nouns, and the like. In addition, “modest expression” includes “respected words” as defined in the conventional Japanese language grammar, sentence patterns, semantic restrictions of verbs or sa-nouns, and the like. “Polite expression” includes “Polite language” defined in the conventional Japanese language grammar. Such honorific dictionary 15 is created in advance.

As an example, in the case of “predicate”, the basic form of each word of “verb”, “auxiliary verb”, “sa variable noun”, “adjective”, “adjective verb”, and the corresponding “respect expression” of the word that becomes this basic form, "Humility expression" and "Polite expression" are recorded.
Specifically, “verb” can be “respect expression” or “humility expression” but not “poor expression”. For example, the basic form “say” of a verb becomes “speak” in “respect expression” and “speak” in “modesty expression”.
Also, “auxiliary verbs” cannot be “modest expressions”. For example, the basic form “DA” of an auxiliary verb becomes “Is, Masu” in “Polite Expression”. In addition, the auxiliary verb “re” corresponds to “respect expression”.
For example, the basic form “research” and “visit” become “research” and “visit” in “respect expression”, respectively. Depending on the vocabulary and context, for example, the basic form “report” becomes “report”, and it is possible to interpret both “respect expression” and “humility expression”.
In addition, the “adjective”, for example, the basic form “early” becomes “early” in “respect expression”.
In addition, the “adjective verb” is, for example, the basic form “genki” becomes “good” in “respect expression”.
Each word can be both a “respect expression” and a “poor expression” when the subject is a second person.

In the case of “sentence pattern”, each word of “command sentence” and “use sentence” is recorded as “respect expression”, “humility expression”, or “poor expression”.
A “command sentence” cannot be a “poor expression”. For example, the imperative sentence “Please do” becomes “modesty expression”.
The “utility sentence” cannot be a “poor expression”. For example, the usage sentence “~ let” becomes “respect expression”.

  In the case of a verbal or sa-variant noun, for example, the basic form “Request” becomes “Look up” in “Humility”, the basic form “Enjoy” becomes “Instruct” in “Respectful expression”, and the basic form “ “To be present” becomes “To be seated” in “Humility”.

In the case of “noun”, the word “word”, “prefix”, “suffix” is recorded as to whether it can be “external relationship”, “internal relationship”, or “polite”. ing. Here, “inner relationship” means that the speaker determines that the speaker is intimately related to the target person, and “outer relationship” means that the speaker is not so close. Each “word” of the noun can be both “respect expression” and “poor expression” when the subject is the second person.
For example, the word “mother” in “inner relationship” becomes the word “mother” in “outer relationship”.

The “prefix” is, for example, the basic form “luggage” becomes “luggage” in “external relationship”. In addition, the basic form “company” is “your company” in “external relationship” and “our company” in “internal relationship”. The “prefix” may correspond to both “respect expression” and “poor expression” when the subject is the second person.
The “suffix” is, for example, “Sato-san” and “student” in “external relationship” are “Sato” and “Tanaka” in “internal relationship”.

  The extraction unit 16 includes three parties (hereinafter referred to as a speaker of the input sentence, a partner of the speaker, and a target subject of the topic of the sentence from the input sentence subjected to the morphological analysis by the analysis unit 12. (Referred to as “instruction target”).

  The determination unit 17 is an arithmetic processing unit that determines a social, age, or generational hierarchical relationship between input target instructions. The target target determination unit 17a, honorific determination unit 17b, rank determination unit 17c, The internal / external determination unit 17d, the closeness determination unit 17e, and the output unit 17f.

  The instruction target specifying unit 17a specifies which instruction target the instruction target extracted by the extraction unit 16 corresponds to. The honorific determination unit 17 b determines the vertical relationship of the instruction target based on the honorific extracted from the detection unit 14. The rank determination unit 17c determines the vertical relationship of the instruction target based on the rank DB 18. The internal / external determination unit 17d determines the vertical relationship of the instruction target based on the internal / external relationship of the words in the honorific dictionary 15. The closeness determination unit 17e determines the vertical relationship of the instruction target based on the determination results of the honorific determination unit 17b, the rank determination unit 17c, and the internal / external determination unit 17d. The output unit 17f outputs the determination results of the honorific determination unit 17b, the rank determination unit 17c, the inside / outside determination unit 17d, and the closeness determination unit 17e to the outside.

  The rank DB 18 is a database in which social, age or generational hierarchical relationships between nouns as shown in FIG. 3 are recorded. For example, in the social hierarchical relationship shown in FIG. 3A, in the case of “master” and “disciple”, it is recorded that “master” is socially higher than “disciple”. In addition, in the age / generational hierarchical relationship shown in FIG. 3B, it is recorded that “old man” and “adult” are higher in age than “adult” in the case of “old man” and “adult”. Yes. Such a rank DB 18 is created in advance.

  Note that the vertical relationship recorded in the rank DB 18 can also relate the vertical relationship of different categories. For example, as shown in FIG. 4, module 1 in which the hierarchical relationship between nouns in the category of education is recorded and module 2 in which the hierarchical relationship of noun relationships in the category of generation are recorded are related by the noun “Taro”. Has been made. In this case, Taro corresponds to “Student” in Module 1 and corresponds to “Principal” in Module 2. Module 1 “Student” is at the bottom, but Module 2 “Principal” is not at the bottom. In this way, the relative vertical relationship can be detected by associating the modules. In addition, “Grandparent” of “Taro” may be “Teacher” of “Minister of Education”. In such a case, a plurality of associations can be performed between modules.

  Next, the operation of the hierarchical relationship detection apparatus 1 according to the present embodiment will be described with reference to FIG. In the present embodiment, it is assumed that the hierarchical relationship detection apparatus 1 performs processing described below using Japanese as a language.

  When a sentence composed of text data or the like (hereinafter referred to as “input sentence”) is input from the outside through a user operation input, a recording medium, a communication line, etc., the input unit 11 accepts the input sentence, The data is sent to the analysis unit 12 (step S501). The input sentence may be one sentence or a plurality of sentences.

  When the input sentence is received, the analysis unit 12 analyzes the input sentence, that is, performs morphological analysis and syntax analysis (step S502). For example, when an input sentence “Proceed to the president” is received by the input unit 11, the analysis unit 12 reads “President (noun) / ni (particle) / advice (verb) / do (auxiliary verb)” as a word unit. And parts of speech are given to each word. Such morphological analysis and syntax analysis can be realized by a technique such as Head-Driven Phrase Structure Grammar (HPSG), for example. In this embodiment, each word is described by being separated by “/”.

  When the morphological analysis and the syntax analysis are performed, the detection unit 14 detects a honorific from the input sentence subjected to the morphological analysis based on the honorific dictionary 15 (step S503). Specifically, the detection unit 14 compares each word included in the input sentence with a word registered in the honorific dictionary 15 and extracts a word registered in the honorific dictionary 15. For example, in the case of the input sentence “Prompt to the president”, “Progress” is detected as honorific.

  When honorifics are not detected by the detection unit 14 (step S504: NO), the output unit 17f of the determination unit 17 outputs that no honorifics are detected (step S513). The output unit 17f displays on the display screen of the hierarchical relation detection apparatus 1 that the honorific word has not been detected, prints out, transmits it to the outside via a communication line, or records it on a recording medium.

  When an honorific is detected (step S504: YES), the extraction unit 16 extracts a specific noun that can be an instruction target from the input sentence analyzed by the analysis unit 12 (step S504). Here, as described above, the instruction target refers to the three speakers, that is, the speaker who issued the input sentence, the partner to whom the speaker spoke the input sentence, and the target person who is the subject of the topic of the input sentence. means. The extraction unit 16 extracts a specific noun that can be any one of such instructions, that is, a speaker, a partner, or a subject. At this time, the extraction unit 16 extracts a noun accompanied by a case particle because a noun that cannot be an instruction target may be extracted if only the noun is extracted. For example, in the case of the input sentence “advance to the president”, the noun “to the president” accompanied by the case particle is extracted. When the extraction unit 16 extracts a noun accompanied by a case particle, only the noun is output to the determination unit 17.

  When a specific noun that can be an instruction target is extracted, the instruction target specifying unit 17a of the determination unit 17 specifies whether the extracted noun corresponds to a speaker, a partner, or a target person (step S506). . Specifically, the instruction target specifying unit 17a specifies that the noun extracted by the extracting unit 16 is “speaker” when the first person is first person, “partner” when it is the second person, and “target person” otherwise. To do. For example, in the case of the input sentence “advance to the president”, the instruction target “president” extracted from “to the president” extracted by the extraction unit 16 is not the first person or the second person, and thus is specified as “target person”. . When there is no subject in the input sentence, the subject of the input sentence defaults to “speaker”. Further, when there is no noun corresponding to “partner” in the input sentence, “partner” is set as a default. Here, the default means a state in which a specific instruction target is not associated with an instruction target when there is no corresponding specific noun.

  Note that, depending on the input sentence, there may be a case where the instruction target is duplicated, for example, the other party is the same as the target person. In this case, each instruction object may be handled independently, or overlapping instruction objects may be handled in an integrated manner.

When the instruction target is specified, the honorific determination unit 17b of the determination unit 17 determines the vertical relationship of the instruction target based on the honorific detected by the detection unit 14 (step S507). Specifically, when the detected honorific is a respected expression, the subject is higher than the other instruction objects. In addition, when the detected honorific is a humility expression, the subject is in a lower order than the other instruction objects. In addition, in the case of corresponding to the respect expression of the semantically restricted verb of the detected honorific, the subject is higher than the other instruction target. In addition, in the case of corresponding to the modest expression of the semantically restricted verb of the detected honorific, the subject is in a lower level than the other instruction objects. If no polite honorific expression is detected, the other party is equal to or lower than the speaker.
For example, in the case of the input sentence “advance to the president”, the detected honorificial word “advancement” is the “modest expression” of the semantically restricted verb, so the speaker as the subject is subordinate to the “president”. In addition, since the polite expression honorific is not detected, the other party is equal to or lower than the speaker. Therefore, the hierarchical relationship “president> speaker ≧ partner” is detected.

  The honorific judgment unit 17b does not have a subject in the input sentence, and the subject cannot be specified from the context. However, when the honorific of the input sentence is a humility expression, it is specified by default that the speaker is the subject. Judgment of the upper and lower relation of the instruction target.

  Further, for example, the basic form “see” may have a plurality of stages in honorific expression, such as “seen”, “see”, and “become seen”. The honorific determination unit 17b may determine the distance between the instructions based on the level of honorific expression. For example, if honorifics such as “seen” are used, the vertical relationship distance between the target objects is relatively short, and if a more polite honorific such as “can be seen” is used Judge that the distance between objects is relatively long. Thereby, it is possible to determine the vertical relationship between the instruction targets in more detail.

  When the hierarchical relationship of the instruction target is detected by honorific or when the hierarchical relationship cannot be detected from honorific, the rank determining unit 17c confirms or determines the hierarchical relationship based on the rank DB 18 (step S508). For example, if the speaker of the input sentence “I gave a call to the president” is “director”, it is detected from the rank DB 18 that the “director” is lower than “president”. It is confirmed that the usage of “ta” is correct.

  By detecting the hierarchical relationship between the speaker and the target person, it is possible to simultaneously detect the vertical relationship between the subject and the object. Therefore, not only the subject but also the object can be used to detect the vertical relationship between the target objects.

  When the rank determination unit 17c does not detect the vertical relationship (step S509: NO), the internal / external determination unit 17d determines the vertical relationship of the instruction target based on the internal / external relationship of the words included in the honorific dictionary 15 (step S510). ). For example, in the case of the input sentence “husband promoted to Mr. Tanaka”, the speaker uses the subject “husband” having an internal relationship, and is therefore lower than the target person. Therefore, the hierarchical relationship “speaker <subject” is detected.

  Further, when the inside / outside determination unit 17d does not detect the vertical relationship (step S511: NO), the closeness determination unit 17e determines that the closeness of the three target objects is low, and the speaker is more than the other target objects. It is determined that the position is lower (step S512). Therefore, the inside / outside determination unit 17d detects a hierarchical relationship of “subject> speaker, partner ≧ speaker”.

  When the vertical relationship is detected (step S509: YES, S511: YES, S512), the output unit 17f outputs the detected vertical relationship (step S513). The output unit 17f displays the detected hierarchical relationship on the display screen of the hierarchical relationship detection device 1, prints it out, transmits it to the outside via a communication line, and records it on a recording medium. Thereby, the user can recognize the vertical relationship of the instruction target derived from the input sentence.

  As described above, according to the present embodiment, the hierarchical relationship of the instruction target is determined based on the type of honorific, the social, age or generational hierarchical relationship between the subjects, the internal / external relationship, etc. Thus, it is possible to more specifically determine the human vertical relationship related to the sentence.

[Second Embodiment]
Next, the second embodiment of the present invention will be described in detail. FIG. 7 is a block diagram showing the configuration of the sentence generation device 2 according to the present embodiment. In the present embodiment, components equivalent to those in the first embodiment are denoted by the same names and reference numerals, and description thereof will be omitted as appropriate.

  The sentence generation device 2 includes an input unit 11, an analysis unit 12, a grammar dictionary 13, an honorific dictionary 15, an extraction unit 16, a determination unit 17, a rank DB 18, a selection unit 21, a generation unit 22, The machine translation device 23 is configured. Such a sentence generation device 2 includes an arithmetic device such as a CPU, a storage device such as a memory and an HDD, an input device that detects input of information from the outside, such as a keyboard, a mouse, a pointing device, a button, and a touch panel, and the Internet. It consists of an I / F device that transmits and receives various types of information via a communication line such as a LAN or WAN, a computer having a display device such as a CRT, LCD, or FED, and a program installed in the computer . That is, the hardware device and software cooperate to control the above hardware resources by a program, and the above-described input unit 11, analysis unit 12, grammar dictionary 13, honorific dictionary 15, extraction unit 16, and determination unit 17 described above. The rank DB 18, the selection unit 21, the generation unit 22, and the machine translation device 23 are realized. Note that the program may be provided in a state of being recorded on a recording medium such as a flexible disk, a CD-ROM, a DVD-ROM, or a memory card.

  The determination unit 17 is an arithmetic processing unit that determines the vertical relationship of the instruction target of the input sentence. The instruction target specifying unit 17a, the rank determination unit 17c, the inside / outside determination unit 17d, the closeness determination unit 17e, and an output unit 17f. The instruction target specifying unit 17a specifies which instruction target the instruction target extracted by the extraction unit 16 corresponds to. The rank determination unit 17c determines the vertical relationship of the instruction target based on the rank DB 18. The internal / external determination unit 17d determines the vertical relationship of the instruction target based on the internal / external relationship of the words in the honorific dictionary 15. The closeness determination unit 17e determines the vertical relationship of the instruction target based on the determination results of the honorific determination unit 17b, the rank determination unit 17c, and the internal / external determination unit 17d. The output unit 17f outputs the determination results of the honorific determination unit 17b, the rank determination unit 17c, the inside / outside determination unit 17d, and the closeness determination unit 17e to the selection unit 21.

  The accepting unit 20 is an arithmetic processing unit that accepts an input sentence from the outside and translates the input sentence by the machine translation device 23 to generate a translated sentence. The accepting unit 20 not only translates the input sentence, but also summarizes the input sentence, extracts a sentence from the input sentence, or generates a sentence based on a question and answer for the input sentence. You may do it.

  The selection unit 21 is an arithmetic processing unit that selects the type of honorific to be used for the translated sentence based on the analysis result by the analysis unit 12 and the detection result of the hierarchical relationship of the instruction target by the determination unit 17.

  The generation unit 22 extracts a honorific to be used for the translated sentence from the honorific dictionary 15 based on the calculation result of the selection unit 21, and generates a translated sentence using the honorific.

  The machine translation device 23 is a known machine translation system that translates one language into another language using a foreign language dictionary, a bilingual dictionary, or the like. As the other language, not only one language but also a plurality of languages can be set.

  Next, operation | movement of the sentence generation apparatus 2 concerning this Embodiment is demonstrated with reference to FIG. In the following, an input sentence in English is input to the reception unit 20, and the machine translation device 23 will be described by taking as an example a case where English is set as the first language and Japanese is set as the other language. To do.

  First, when an English input sentence made up of text data or the like is input from the outside via a user operation input, a recording medium, a communication line, etc., the accepting unit 20 converts the input sentence into Japanese in the machine translation device 23. Translation is performed (step S701). For example, when the input sentence “I called the president” is input, the reception unit 20 causes the machine translation device 23 to translate the input sentence into Japanese, and generates a translated sentence “I called the president”.

  When the translated sentence is generated, the analysis unit 12 analyzes the translated sentence, that is, performs morphological analysis and syntactic analysis (step S702). For example, when the translated text “I called the president” is generated by the reception unit 20, the analysis unit 12 reads “I (noun) / ha (particle) / president (noun) / ni (particle) / phone. (Noun) / is divided into (words) / do (verbs) / ta (auxiliary verbs) ”and part of speech is given to each word.

  When the analysis is performed, the extraction unit 16 extracts a specific noun that can be an instruction target from the translated sentence that has been analyzed (step S703). For example, in the case of the translated sentence “I called the president”, the nouns accompanied by the case particles are extracted, “I am”, “to the president”. When the extraction unit 16 extracts a noun accompanied by a case particle, only the noun is output to the determination unit 17.

  When a specific noun that can be an instruction target is extracted, the instruction target specifying unit 17a of the determination unit 17 specifies whether the extracted translation noun is a speaker, a partner, or a target person (step) S704). For example, in the case of the translated sentence “I called the president”, “I” extracted by the extraction unit 16 is a first person, so “speaker” and “president” are not first person and second person, so “target” Identifies it. The subject is also identified as “I” from the result of the case particle of morphological analysis.

  When the instruction target is specified, the rank determination unit 17c of the determination unit 17 determines the vertical relationship of the instruction target based on the rank DB 18 (step S705). For example, if it is detected from the context of the translated sentence “I called the president” that the speaker is “section manager”, the other party is “manager”, and the target person is “president”, FIG. From the rank DB 18 as shown, a vertical relationship of “President> Manager> Manager” is detected.

  If the rank determination unit 17c does not detect the vertical relationship (step S706: NO), the internal / external determination unit 17d determines the vertical relationship of the instruction target based on the internal / external relationship of the words in the honorific dictionary 15 (step S707). For example, in the case of a translated sentence “husband calls Tanaka”, the subject “husband” is an internal relationship. Therefore, the hierarchical relationship of “speaker <target person (Tanaka), speaker = target person (husband)” is detected.

  Also, when the vertical relationship is not detected by the inside / outside determination unit 17d (step S708: NO), the closeness determination unit 17e determines that the closeness of the three target objects is low, and the speaker is lower than the other target objects. It is determined that it is a lower order (step S709). Therefore, the inside / outside determination unit 17d detects a hierarchical relationship of “subject> speaker, partner ≧ speaker”.

  When the vertical relationship is detected (step S706: YES, S708: YES, S709), the output unit 17f outputs the detected vertical relationship to the selection unit 21. The selection unit 21 selects one of “respect expression”, “humility expression”, and “poor expression” based on the hierarchical relationship of the instruction target input from the output unit 17 f and the translation sentence analyzed by the analysis unit 12. Which one is to be used is selected (step S710). Specifically, when the object is higher than the subject, the selection unit 21 determines to use “humility expression” for the translated sentence, based on the structure of the translated sentence and the hierarchical relationship of the instruction target. If the other party is higher than the speaker, the selection unit 21 determines to use “a polite expression” for the translated sentence. For example, in the case of the translated sentence “I called the president”, the object is “President” who is the target, and the subject is “Section Manager (I)”. Use "expression". In addition, the “director” who is the other party is higher than the “section manager (I)” who is the speaker, and therefore “polite expression” is used in the translated sentence.

  When the type of honorific is selected, the generation unit 22 extracts the honorific to be used for the translated sentence from the honorific dictionary 15 and generates a translated sentence using the honorific (step S711). Specifically, the generation unit 22 first detects a basic form of a word to be replaced with honorific from the translated sentence. This is done by searching for an appropriate word from the semantic constraints of the predicate, sentence pattern, verb or sa variable noun from the translated sentence. Next, the generation unit 22 extracts honorific words that can replace the detected word from the honorific dictionary 15 based on the selection result of the selection unit 21. Finally, the generation unit 22 generates a translated sentence in which the detected word is replaced with the extracted honorific. The generated translation is displayed on the display screen of the sentence generation device 2, printed out, transmitted to the outside via a communication line, or recorded on a recording medium. For example, in the case of a translated sentence “I called the president”, the generation unit 22 detects “shi” as a replacement word, and uses a humble expression “shi” and a polite expression “masashi” as honorifics to replace it. It is extracted from the honorific dictionary 15 and finally the translated sentence “I called the president” is output.

  Note that the generation unit 22 may extract honorifics to be used in the translation based on the hierarchical relationship determination result using the rank DB 18 by the rank determination unit 17c. For example, the basic form “see” has a plurality of stages of honorific expression such as “seen”, “see”, and “become seen”. Therefore, for example, when the distance between the target objects is close, such as “section manager” and “employee”, use a honorific such as “seen”, and the distance between the target objects such as “chairman” and “employee” If you are far away, use a more polite honorific word such as “Look at it”. In this way, honorifics can be used in more detail.

  Thus, according to the present embodiment, based on the social, age or generational hierarchical relationship between each target object, the internal / external relationship, etc., the vertical relationship of the target object is determined, and the determination result is By selecting the honorific of the translated sentence based on it, it is possible to generate a more appropriate sentence.

  In the present embodiment, the above-described processes are performed for a translated sentence obtained by translating an English input sentence into Japanese. However, the language of the input sentence and the language of the translated sentence are not limited to English or Japanese, respectively. It can be set freely as appropriate.

  In addition, the sentence generation device according to the present embodiment generates a sentence that is semantically correct for a translation sentence obtained by machine translation, but a sentence that can generate a grammatically appropriate sentence is not limited to a translation sentence. For example, it is possible to generate semantically correct sentences for various sentences such as a summary sentence that mechanically summarizes sentences, an extracted sentence that is mechanically extracted from sentences, and a sentence generated by a question and answer system.

  The present invention can be used in a natural language processing system. It can also be used for a language learning apparatus for learning a language such as a foreign language, a learning apparatus for learning correct sentences in a new employee training for a company or the like.

1 is a block diagram illustrating a configuration of a vertical relationship determination device 1. FIG. It is a figure which shows the structure of a honorific dictionary. It is a figure which shows the structure of rank DB. It is the figure which linked | related each module of rank DB. 3 is a flowchart showing the operation of the vertical relationship determination device 1. 3 is a block diagram illustrating a configuration of a sentence generation device 2. FIG. 4 is a flowchart showing the operation of the sentence generation device 2.

Explanation of symbols

DESCRIPTION OF SYMBOLS 1 ... Vertical relationship determination apparatus, 2 ... Sentence generation apparatus, 11 ... Input part, 12 ... Analysis part, 13 ... Grammar dictionary, 14 ... Detection part, 15 ... Honorific dictionary, 16 ... Extraction part, 17 ... Determination part, 17a ... Reference target specifying unit, 17b ... honorifical word judging unit, 17c ... rank judging unit, 17d ... inside / outside judging unit, 17e ... closeness judging unit, 17f ... output unit, 18 ... rank DB, 19a ... predicate obtaining unit, 19b ... translation unit , 19c ... determination means, 19d ... registration means, 20 ... reception part, 21 ... selection part, 22 ... generation part, 23 ... machine translation device.

Claims (8)

The upper and lower relations which analyze the digitized input sentence and determine the upper limit relation of the instruction target composed of three persons, the speaker of the input sentence, the other party of the speaker, and the target person who is talking about the input sentence A determination method comprising:
An analysis step in which an analysis unit performs morphological analysis and syntax analysis on the input sentence;
A detecting step of detecting a honorific word and a verb that semantically restricts a relationship between a subject and an object in a sentence from an analysis result of the analyzing unit with reference to the honorific dictionary;
An extraction step for extracting a specific noun that can be the instruction target from the analysis result of the analysis unit;
A specifying step for specifying that the noun extracted by the extracting unit is the speaker when the first person is first person, the partner when the second person is the second person, and the target person in other cases; ,
The honorific judgment unit determines that the instruction object corresponding to the subject is higher than the other instruction objects when the honorific word detected by the detection unit is a respect expression, and the honorific word detected by the detection unit is a humble expression Determines that the instruction object corresponding to the subject is lower than the other instruction objects, and if the verb that semantically restricts the relationship between the subject and the object in the sentence detected by the detection unit is a respected expression , If the verb that restricts the relationship between the subject and the object in the sentence detected by the detection unit is a humility expression, the instruction corresponding to the subject is determined. If the target is determined to be lower than the other instruction target and the polite expression of the polite expression is not detected by the detection unit, the opponent is determined to be equal to or lower than the speaker, thereby Honorifics judgment step to judge the relationship Vertical relationship determination method characterized by executing and. When the honorific judgment unit cannot detect the hierarchical relationship of the instruction target from the honorific, the rank determination unit refers to the rank DB that is a database that records the social, age, or generational hierarchical relationship between the nouns. The rank determination step according to claim 1, further comprising a rank determination step of determining the vertical relationship of the instruction target. When the honorific determination unit detects the hierarchical relationship of the instruction target from the honorific, the rank determination unit refers to the rank DB, which is a database that records the social, age, or generational hierarchical relationship between each noun. The method of claim 1, further comprising: a rank confirmation step of confirming the vertical relationship of the instruction target. The specifying step, if the subject from the input sentence by the extraction unit is not extracted, the vertical relationship determination method according to claim 1, wherein the identifying the speaker as the subject. In the rank confirmation step, a plurality of vertical relationships in which categories associated with each other by words included in the rank DB are recorded are recorded, and the rank determination unit refers to the plurality of categories and determines whether the indication target 4. The hierarchical relation determination method according to claim 3, wherein the relation is confirmed . The upper and lower relations which analyze the digitized input sentence and determine the upper limit relation of the instruction target composed of three persons, the speaker of the input sentence, the other party of the speaker, and the target person who is talking about the input sentence A determination device,
An analysis unit that performs morphological analysis and syntax analysis on the input sentence;
With reference to the honorific dictionary, the detection unit that detects the honorific and the verb that semantically restricts the relationship between the subject and the object in the sentence from the analysis result of the analysis unit;
An extraction unit for extracting a specific noun that can be the instruction target from the analysis result of the analysis unit;
When the noun extracted by the extraction unit is first person, the speaker, in the case of the second person, the partner, in other cases, the instruction target specifying unit for specifying the target person,
When the honorific detected by the detection unit is a respected expression, the instruction object corresponding to the subject is determined to be higher than the other instruction objects, and when the honorific detected by the detection unit is a humble expression, it corresponds to the subject. If the verb that restricts the relationship between the subject and the object in the sentence detected by the detection unit is a respect expression, the instruction target corresponding to the subject is When the verb that semantically constrains the relationship between the subject and the object in the sentence detected by the detection unit is higher than the other instruction target, the instruction target corresponding to the subject is another instruction. Honorifics for determining the hierarchical relationship of the indicated object by determining that the polite expression of polite expression is not detected by the detection unit, and determining that the other party is equivalent to the speaker or lower than the speaker And a determination unit. A vertical relation judging device. A program for determining an upper / lower relationship for causing a computer to function as each means of the apparatus for determining an upper / lower relationship according to claim 6 . A computer-readable recording medium in which the hierarchical relationship determination program according to claim 7 is recorded.

Images