JP2018195331A - Speech sentence generation device, and method and program thereof - Google Patents

Abstract

To provide a speech sentence generation device capable of creating various dialogue speech sentences in a dialogue system by extracting a word representing a conversation subject and a modification structure from a morpheme string of speech sentence.SOLUTION: The speech sentence generation device comprises: a conversation subject extraction part configured to extract a modification structure which represents a content of a speech sentence using a morpheme string of the speech sentence as the input; a related conversation subject extraction part configured to output a first related modification structure which is different from the modification structure, and which includes one of the phrases constituting the modification structure using the modification structure as the input; and a speech sentence generation part configured to create a dialogue speech sentence by substituting the modification structure and the first related modification structure into a template using the modification structure and the first related modification structure as the input.SELECTED DRAWING: Figure 11

Description

  The present invention relates to an utterance sentence generation apparatus, a method thereof, and a program in a system (hereinafter referred to as a dialogue system) that interacts with a user (user) using a natural language.

  In recent years, there is an increasing need for a chat dialogue system that performs open domain chat without specific tasks. These chat conversations are very important because they may have therapeutic properties and can be applied to the manifestation of requests that users themselves are not aware of in task-oriented dialog systems. . However, in the open domain chat dialogue system, since the variation of user utterances is greatly spread, it is extremely difficult to manually construct a knowledge source for appropriately responding in advance.

  Conventional approaches to this problem include a method of describing response patterns in advance for topics that people might be interested in, or a method of repeating safe utterances and questions that match any user utterance (non- Patent document 1) etc. are known.

  However, it is difficult to deal with new words using the rules. Another problem is that since the playback is only performed under certain conditions as in a tape recorder, the dialogue is easy to end with a single question and there is no development. In the context-independent approach described in Non-Patent Document 1, the user's utterance tends to be overtaken, and the user is quickly bored.

  In order to overcome the difficulty and monotony of dealing with such new words, in recent years, there has been a widespread movement to use large-scale sentences on the Web. For example, Non-Patent Document 2 or 3 discloses a method of selecting a sentence similar to a user utterance in an article on the Web or a microblog and making it an utterance sentence.

  However, since the context in which the similar sentence appears is different from the context in which the user utterance appears, there is a problem including an unnecessary utterance sentence. In order to solve this problem, a method of generating a spoken sentence by estimating a focal word representing a focal point of a topic from words included in a user's utterance and substituting the focal word into a template has been studied (non-patent document). Reference 4).

J. Weizenbaum, “ELIZA-A Computer Program For the Study of Natural Language Communication Between Man and Machine”, Communications of the ACM. ACM 9 [1] 36-45 (1966). Masahiro Shibata et al., “Method for selecting appropriate candidate sentences from WWW documents for realizing free chat conversation”, Journal of the Japanese Society for Artificial Intelligence, vol.24, no.6, pp.507-519, 2009. Alan Ritter, Colin Cherry, and William.B. Dolan. 2011.Data-Driven Response Generation in Social Media.In Proceedings of the 20111 Conference on Empirical Methods in Natural Language Processing, pages 588-593. Yuka Kobayashi et al., “Aged Dialogue Interface-Voice Dialogue Interface for User Listeners”, Information Processing Society of Japan Interaction 2011.

  However, in the conventional technique using the focal word, the focal word is limited to nouns, and the number of the focal words is one, and there is a problem that it is not possible to generate a utterance sentence with abundant variations.

  The present invention has been made in view of this problem, and it is an object of the present invention to provide an utterance sentence generation apparatus, a method thereof, and a program that enable generation of an abundance of utterance sentences.

  One aspect of the present invention is a topic extraction unit that extracts a dependency structure representing the content of a spoken sentence using a morpheme string of the spoken sentence as an input, and a clause that constitutes the dependency structure using the dependency structure as an input A related topic extraction unit that outputs a first related dependency structure that is different from the dependency structure, and receives the dependency structure and the first related dependency structure as inputs, and the dependency structure An utterance sentence generation unit that generates a dialog utterance sentence by substituting the first related dependency structure into a template.

  One aspect of the present invention is a topic extraction unit that extracts a dependency structure representing the content of an utterance sentence using a morpheme string of the utterance sentence as an input, N is an integer of 1 or more, and the dependency structure is a zero relation First dependency relationship that is different from the 0th related dependency structure, including any of the clauses constituting the 0th related dependency structure, with the 0th related dependency structure as an input. A receiving structure is obtained, and for i = 1,..., N, any of the clauses constituting the i-th related dependency structure and any of the clauses constituting the i-1th related dependency structure are included. A related topic extraction unit that obtains an i + 1th related dependency structure and outputs a kth related dependency structure (k = 1,..., N + 1), and the jth related dependency structure (j = 0, ..., N + 1) as input, and substituting the above j-th related dependency structure (j = 0, ..., N + 1) into the template Includes a spoken sentence generator for forming, the.

  According to the utterance sentence generation device of the present invention, it is possible to generate an appropriate utterance sentence using the topic of the user utterance. Since dependency structures related to topics are collected from a large amount of natural sentences, it becomes possible to generate utterance sentences for user utterances of a wide range of topics.

The figure which shows the function structural example of the utterance sentence production | generation apparatus 100 of this invention. The figure which shows the operation | movement flow of the utterance sentence production | generation apparatus 100. FIG. The figure which shows the function structural example of the utterance sentence production | generation apparatus 200 of this invention. The figure which shows the operation | movement flow of the utterance sentence production | generation apparatus 200. FIG. The figure which shows the function structural example of the utterance sentence production | generation apparatus 300 of this invention. The figure which shows the operation | movement flow of the utterance sentence production | generation apparatus 300. FIG. The figure which shows the relationship between the morpheme sequence of a user speech sentence, and a dependency. The figure which shows the function structural example of the utterance sentence production | generation apparatus 400 of this invention. The figure which shows the function structural example of the utterance sentence production | generation apparatus 500 of this invention. The figure which shows the function structural example of the utterance sentence production | generation apparatus 600 of this invention. The figure which shows the function structural example of the utterance sentence production | generation apparatus 700 of this invention. The figure which shows an example of the result of having performed the morphological analysis of the speech sentence. The figure which illustrates the state in which at least 1 of the clauses in a dependency structure has a dependency relationship with the clause in another dependency structure. The figure which shows the function structural example of the utterance sentence production | generation apparatus 800 of this invention. The figure which illustrates the dependency structure obtained by searching the dependency relationship database 890. The figure which shows the example of a related word and a related dependency structure. The figure which shows the example of a dialog utterance sentence. The figure which shows the outline | summary which searches a related dependency structure. The figure which shows the function structural example of the utterance sentence production | generation apparatus 900 of this invention.

  Embodiments of the present invention will be described below with reference to the drawings. The same reference numerals are given to the same components in a plurality of drawings, and the description will not be repeated.

  FIG. 1 shows a functional configuration example of an utterance sentence generation apparatus 100 of the present invention. The operation flow is shown in FIG. The utterance sentence generation device 100 includes a topic extraction unit 110, an utterance sentence generation unit 120, and a control unit 130. The utterance sentence generation device 100 is realized by reading a predetermined program into a computer composed of, for example, a ROM, a RAM, a CPU, and the like, and executing the program by the CPU. The same applies to each device described below.

  The utterance sentence generation apparatus 100 receives a morpheme string obtained by voice recognition of a user utterance or a morpheme string obtained by morphological analysis of a text sentence of a user utterance by a morpheme analysis unit 140. A morphological analysis unit 140 indicated by a broken line in FIG.

  The topic extraction unit 110 receives a morpheme string of an utterance sentence as an input, and extracts a word representing the content of the utterance sentence or a dependency structure with the word (step S110). That is, the topic is a word that represents the content of an utterance and a dependency structure. The utterance sentence is a user utterance in the dialogue system, and may be a user's utterance voice itself or a text sentence obtained as a result of voice recognition of the user utterance. The utterance sentence is a relatively short sentence composed of about 1 to 3 sentences.

  In the following description, a word representing the content of an utterance will be defined as a focal word. For example, Reference 1 (Barbara J. Grosz, Scott Weinstein, and Aravind K. Joshi. 1995. Centering: A Framework for Modeling the Local Coherence of Discourse. Computational linguistics, 21 (2): 203- 225.) and Reference 2 (Marilyn A. Walker, 1998. Centering, Anaphora Resolution, and Discourse Structure Oxford University Press on Demand.). In addition, the Example which extracts the word showing the content of the utterance sentence, and a dependency structure, and produces | generates a dialog utterance sentence is demonstrated in Example 5 or later.

  The utterance sentence generation unit 120 generates the dialog utterance sentence by substituting the focus word into the template with the focus word extracted by the topic extraction unit 110 as an input (step S120). A template is a fixed phrase in which a focus word is incorporated (assigned). Specific examples will be described later.

  For example, if the user's utterance is “I watched movie A in Toyosu movie theater today”, the words representing the focus of the topic are two proper nouns “Toyosu movie theater” and “Movie A”. Is a word. “Toyosu movie theater” is a phrase consisting of a proper noun representing the place name “Toyosu”, the particle “no”, and the general noun “movie theater”, but in this example it is treated as a proper noun. .

  For example, it is assumed that “I like you” or “Do you like it” is prepared as a template. A template is prepared by classifying a plurality of types for each utterance intention for each focal word. The utterance intention includes “question”, “self-disclosure”, “consideration”, and the like. The above “Like” is classified as a companion, and “Do you like it” as a question. By classifying the templates for each utterance intention in this way, the prospect of the relationship between the templates is improved. That is, unnecessary duplication of templates can be prevented.

  Based on the premise, the utterance sentence generation unit 120 has “the number of proper nouns × the number of templates” “Toyosu movie theater” + “I like it”, “Toyosu movie theater” + “Do you like it?”, “Movies” Four dialogue utterances of “A” + “Like”, “Movie A” + “Do you like it” are generated. The control unit 130 controls the process of repeating the generation of the dialog utterance. The control unit 130 is a general unit that controls the time-series operation of each unit of the utterance sentence generation device 100 and does not perform a special process. The same applies to other embodiments, and the description of the control unit will be omitted hereinafter. In the dialogue system, one of the plurality of dialogue utterances corresponding to the user utterance is selected and used by an utterance sentence selection device (not shown).

  As described above, according to the utterance sentence generation device 100, a plurality of focal words representing the focal point of a topic are extracted from words included in a morpheme sequence of a user utterance sentence, and the conversation utterance sentence is obtained by substituting the focal word into a template. Therefore, it is possible to generate dialogue utterances rich in variations.

  FIG. 3 shows a functional configuration example of the utterance generation device 200 of the present invention. The operation flow is shown in FIG. The utterance generation device 200 is different from the utterance generation device 100 (FIG. 1) in that it includes a related word estimation unit 250 and the operation of the utterance sentence generation unit 220.

  The related word estimation unit 250 receives the focal word extracted by the focal word extraction unit 110 as an input, and estimates a synonym of the focal word as a related word (step S250). As a method for estimating a synonym of a focal word, a method using a thesaurus dictionary or LDA (Reference 3: David M. Blei, et al., Latent Dirichlet Allocation, the Journal of Machine Learning Research, vol. 3, pp. 993-1022, 2003), etc., a method using a topic model (Topic Model) that estimates synonyms by co-occurrence relationships between words is well known.

  The related word estimation unit 250 receives the focal word “movie A” as an input, and uses a topic model, for example, a word related to movie A such as “XXXXX” or “ΔΔΔΔ”, or “movie B”. A similar genre movie such as, and abbreviations and notations such as “OO” are extracted as related words.

  The spoken sentence generation unit 220 receives the focused word extracted by the focused word extraction unit 110 and the related word estimated by the related word estimation unit 250 as input, and substitutes the focused word and the related word into a template to generate a dialogue uttered sentence. Generate (step S220). In this example, “○○○○” + “I like you”, “△△△△” + “I like you”, “○○” + “I like you” etc. Added as.

  Since the related words estimated by the related word estimation unit 250 are added to the utterance generation device 100, the utterance generation device 200 can further enrich the variation of the dialog utterance sentence.

  FIG. 5 shows a functional configuration example of the utterance sentence generation apparatus 300 of the present invention. The operation flow is shown in FIG. The utterance sentence generation device 300 is different from the utterance sentence generation device 200 (FIG. 3) in that it includes a dependency relationship analysis unit 360 and the operation of the focal word extraction unit 310. The related word estimation unit 250 and the utterance sentence generation unit 220 are the same as the utterance sentence generation device 200 (FIG. 3) as apparent from the reference numerals.

  The dependency relationship analysis unit 360 receives a morpheme sequence, performs dependency analysis on the morpheme sequence, and outputs a dependency relationship of phrases (step S360). For dependency analysis, a general Japanese dependency analysis method is used.

  For example, FIG. 7 shows the relationship between the morpheme sequence and the dependency when the user utterance is “I watched movie A in Toyosu movie theater today”. The first line in FIG. 7 is a morpheme string, and the second line is a dependency relation between phrases. Table 1 shows the dependency relationship.

  The focal word extraction unit 310 receives, as input, the morpheme sequence of the user utterance and the dependency relationship analyzed by the dependency relationship analysis unit 360, a proper noun, a general noun, a predicate representing the focus of the topic included in the morpheme sequence, Are extracted (step S310). A predicate is a situational noun, verb, adjective, adjective verb. Situation nouns are nouns that evoke a specific situation, and there are at least the following four types (Reference 4: “Kuroda Kou,“ Integration of Situation Noun Term Structure and Verb Term Structure Analysis and consensus of Japanese supportive verb syntax used, "Annual Conference of the Language Processing Society of Japan, 2008"). A: Verb conjunctive form, B: Sa variant noun, C: Non-conjunctive / non-abstract abstract noun (requires support verb), D: Non-conjunctive / non-sainous concrete noun (combined with a specific verb to make a situation noun ).

  For example, when the user utterance sentence is “I watched a movie today,” the focal word extraction unit 310 selects the proper noun “◯ △ □ ○” included in the user utterance sentence as the focal word. Extract as In the case of a user utterance sentence including a plurality of proper nouns, the focal word extraction unit 310 extracts any N proper nouns (N is a natural number) from the words closest to the end of the utterance as focal words.

  Further, the focused word extraction unit 310 extracts, as a focused word, common nouns included in the user uttered sentence with a small number of appearances. The reason why the focus words are those that do not appear frequently is to prevent extraction of nouns such as “ko” that are general and do not express the topic. In this example, “movie” is extracted.

  In addition, the focal word extraction unit 310 extracts the predicate that is associated with the highest rank as the focal word. Note that in Japanese, there are many cases related to words from the front to the back, so the predicate closest to the end of the sentence may be extracted as the focus word.

  When the focus word is a proper noun, the utterance sentence generation unit 220 often uses a proper noun as a related adjective / verb as a dialogue utterance, so the template is “[proprietary noun] seems to be [adjective]”. If it is "Neah", it will generate "Sounds are interesting" as a dialogue utterance. The adjectives / verbs used here are included in the related words estimated by the related word estimation unit 250 using, for example, a topic model.

  When the focus word is a general noun, adjectives and verbs related to the general noun often appear depending on the context. Therefore, the utterance sentence generation unit 220 generates a dialog utterance sentence using the related adjectives / verbs as they are. For example, if the related adjectives / verbs are assumed to be "interesting", "fun", etc., and the template is "what [general noun] is [adjective / verb]?" ? "As an utterance sentence. In this way, when two different words are substituted into one template to generate a dialog utterance, dialog utterances of all combinations of the focus word and the related word are generated (FIG. 6: Step S330).

  Note that using relevant adjectives and verbs as they are may result in inappropriate dialogue utterances that do not match the conversation context. In that case, the positive / negative adjectives are estimated using the Japanese evaluation expression dictionary, and the corresponding evaluation expressions can be applied regardless of the topic, such as “Like” and “I do n’t like”. A sentence may be generated. For example, “what [general noun] do you like?” Or “what [general noun] do you like?” May be prepared as a template, and the general noun may be applied.

  When the focal word is a predicate, the utterance sentence generation unit 220 generates a dialog utterance sentence using a noun and a surface case related to the predicate (situation noun / verb). If you use a noun that has a dependency relationship as it is, it will be a dialogue utterance that you answer with Yes / No, and the topic will not spread, so you can estimate the meaning of the noun from a vocabulary system like Wordnet (Wordnet) A 5W1H type question corresponding to “Where” and “Time” is generated. However, tense matching is difficult to handle, so we specifically generate Where to ask for "location". For example, if the template is “Where [surface class] [predicate]?”, “Where did you see it?” Is generated as a dialog utterance.

  According to the utterance sentence generation device 300, a group of words having a dependency relationship is substituted into a template, so that it is possible to generate a conversation utterance sentence that can deal with a wide range of topics and has a meaningful meaning. The configuration of the dependency relationship analysis unit 360 has been described as being added to the utterance sentence generation device 200, but the configuration in which the dependency relationship analysis unit 360 is added to the utterance sentence generation device 100, that is, the related word estimation unit 250 is omitted. An utterance sentence generation device having the above-described configuration is also conceivable.

  FIG. 8 shows a functional configuration example of the utterance sentence generation apparatus 400 of the present invention. The utterance sentence generation device 400 is different from the utterance sentence generation device 200 (FIG. 3) in that it includes a dependency relation dictionary 470 and the operation of the related word estimation unit 450. The topic extraction unit 110 and the utterance sentence generation unit 220 are the same as the utterance sentence generation device 200 (FIG. 3) as apparent from the reference numerals. The operation flow of the utterance sentence generation device 400 is the same as the operation flow (FIG. 4) of the utterance sentence generation device 200.

  The dependency relationship dictionary 470 records a word group that appears as a dependency relationship with respect to a predetermined word from a large amount of natural sentences together with the number of times. The dependency relationship dictionary 470 is described in, for example, Reference 3 (Kouichi Takeuchi et al., “Subclassification of Verb Term Structure Based on Inclusive Relationship of Meaning”, Annual Conference of the Language Processing Society of Japan, 2008.) It was constructed separately.

  The dependency relation dictionary 470 is constructed by collecting natural sentences from articles such as microblogs that contain a large amount of colloquial expressions and extracting relationships between words. Since microblogs contain a large amount of subjective sentences, it is expected that subjective expressions for certain words will be extracted. It is possible to provide a dependency relation dictionary that is more suitable for a dialogue system than extracting relations between words from written sentences such as newspaper articles. For example, it is assumed that an adjective such as “interesting”, “like”, or “terrible” can be extracted as a related word from a microblog including “movie A”.

  The related word estimation unit 450 receives the focal word extracted by the topic extraction unit 110 as an input and extracts a related word of the focal word with reference to the dependency relation dictionary 470. For example, when the focus word is “movie A”, the related word estimation unit 450 extracts adjectives such as “interesting”, “like”, and “terrible” from the dependency relation dictionary 470 as related words.

  The utterance sentence generation unit 220 generates a dialog utterance sentence by substituting the focus word and the related word into a template, using the focus word extracted by the topic extraction unit 110 and the related word extracted by the related word estimation unit 250 as inputs. (Step S220). For example, if the template is “[proprietary noun] seems to be [adjective]?”, The utterance sentence generation unit 220 generates a dialog utterance sentence “movie A seems to be interesting”.

  The utterance sentence generation device 400 has been described with the configuration in which the dependency relation dictionary 470 is added to the utterance sentence generation apparatus 200 (FIG. 3). However, the utterance sentence generation device 400 adds the dependency relation dictionary 470 to the utterance sentence generation device 300 (FIG. 5). A sentence generation device may be configured.

  Further, although the dependency relationship dictionary 470 has been described as an example using a separately constructed one, the content of the dependency relationship dictionary may be sequentially updated. FIG. 9 shows a functional configuration example of the utterance sentence generation device 500 configured to sequentially update the dependency relation dictionary using the morpheme string input to the topic extraction unit 110.

  The dependency relationship dictionary 570 of the utterance sentence generation device 500 records the dependency relationship and the surface case of the phrase output by the dependency relationship analysis unit 360, and updates the number of appearances of the dependency relationship and the surface case of the same type. Record. In this manner, the dependency relationship dictionary 670 may be configured to be updated with morpheme strings that are sequentially input.

  In addition, a dependency relation dictionary may be created in the utterance sentence generation device. FIG. 10 shows a functional configuration example of an utterance sentence generation device 600 configured to construct a dependency relation dictionary inside the utterance sentence generation device.

  The dependency relationship dictionary 670 of the utterance sentence generation device 600 includes a natural sentence temporary storage unit 671, a morpheme analysis unit 672, a dependency relationship analysis unit 673, and a dependency relationship recording unit 674. The natural sentence temporary storage unit 671 stores natural sentences collected from the outside. The outside is a network environment such as the Internet, and regularly receives and stores blog articles on the Web.

  The morpheme analyzer 672 analyzes the text information stored in the natural sentence temporary storage 671 and outputs a morpheme string. The dependency relationship analysis unit 673 estimates the dependency relationship between words from the morpheme string output by the morpheme analysis unit 672, and extracts the dependency relationship and the surface case. For example, “today” of the dependency relationship shown in FIG. 7 is output together with “ha” case and “verb” “saw”.

  The dependency relationship recording unit 674 records the dependency relationship and the surface case output by the dependency relationship analysis unit 673. At this time, the same dependency relationship and surface case are updated and recorded. In this manner, the dependency relationship dictionary 670 may be automatically constructed.

  FIG. 11 shows a functional configuration example of the utterance sentence generation apparatus 700 of the present invention. The utterance sentence generation apparatus 700 includes a topic extraction unit 710, a dependency relationship analysis unit 360, a related topic extraction unit 780, and an utterance sentence generation unit 720. The dependency relationship analysis unit 360 is the same as the utterance sentence generation device 300 (FIG. 5), as is clear from the reference numerals.

  The topic extraction unit 710 uses the morpheme string and the dependency relationship output by the dependency relationship analysis unit 360 as inputs, and extracts a word representing the topic of the utterance sentence included in the morpheme sequence and the dependency structure. Here, the dependency structure is a set of two phrases having a dependency relationship.

  For example, when the utterance sentence is “I am very hungry”, the result of morphological analysis of the utterance sentence is shown in FIG. The first line shows the utterance sentence, the second line shows the morphological analysis result, the third line shows the dependency analysis result, and the dependency structure after the fourth line.

  Of the dependency relationships output by the dependency relationship analysis unit 360, all of the dependency relationships that do not include a stop word are used as topics. Words use only proper nouns as topics. Stop words include pronouns, auxiliary verbs that are used without specific meanings such as “do, say, become, is,”, abstract nouns such as “koto, no”, and words related to time. For example, words such as “today”, “the other day”, and “○ hour ○ minute” are used. Stop words are words that are frequently used and have no specific meaning. For example, endings that are unique to stories such as “I like” are also included in the stop word.

  That is, the utterance sentence extraction unit 710 performs a process of excluding a dependency structure including a stop word and a word from the topic of the utterance sentence. However, in addition to the method of determining by the meaning of the word as described above, a method of filtering by the number of appearances is also useful. A filter is to introduce an idea such as TFIDF.

  When extracting words and dependency structures (topics), the standard form of the first word of a phrase, a POS tag, a phrase ID indicating the uniqueness of the phrase, a simple semantic attribute (location, action, question, ...), phrase The content word / phrase notation, the standard form of the content word / phrase, and the case information may be extracted simultaneously.

  The related topic extraction unit 780 receives the word representing the topic of the utterance and the dependency structure, and outputs a related word and a related dependency structure that are closely related to the word and the dependency structure. Here, deeply related means that the same or similar words co-occur between clauses, that there is a dependency relationship between clauses, and a pair of words that have a strong co-occurrence relationship in the corpus. It means that it is included one by one in two phrases. The dependency relationship between clauses is a state in which at least one of the clauses in the dependency structure A is in a dependency relationship with the clauses in the dependency structure B. FIG. 13 illustrates the state. “Daibu / Tight”, which is a phrase related to “Tight” of “Hidden / Tight” related to “Hungry” of “Hungry / Hungry”, is a dependency structure related to the word and the dependency structure.

  Here, the related word and the related dependency structure are related topics. The related dependency structure of “Daibu / Tight” is generated by applying a predetermined rule with reference to the above-described definition to a word representing a topic of an utterance and a dependency structure. As the simplest method, a phrase “big” or “tight” corresponding to the dependency structure of “hungry / hungry” is prepared in the related topic extraction unit 780. In addition to the rules, as a method of generating a related dependency structure, a clause having a co-occurrence relationship of words, a clause including a synonym, a clause having a dependency relationship, and a clause having a word having a strong co-occurrence relationship in the corpus are used. It may be a related topic.

  The utterance sentence generation unit 720 receives words and dependency structures representing the topic of the utterance sentence output from the topic extraction unit 710, and related words and association dependency structures output from the related topic extraction unit 780, and inputs those words. A dialogue utterance is generated by inputting a dependency structure into a template. The template is the same as described above. For example, by preparing “Is it?” Or “Is?” As a template and substituting the extracted dependency structure, dialogue utterances such as “Is it pretty?” Or “Is it pretty?” Is generated.

  According to the utterance sentence generation device 700, a word representing a topic of an inputted utterance sentence, a dependency structure (topic), a related word having a dependency relation with the dependency structure, and a related dependency structure (related topic), Therefore, it is possible to generate a meaningful dialog utterance corresponding to a wider topic than the utterance generation device 300.

  FIG. 14 shows a functional configuration example of the utterance sentence generation apparatus 800 of the present invention. The utterance sentence generation device 800 differs from the utterance sentence generation device 700 in that it includes a dependency relation database 890 and the function of the related topic extraction unit 880.

  The dependency relationship database 890 is a database that can search for a dependency structure related to a dependency structure when a certain dependency structure is given. For example, the database shown in FIG. 15 is obtained when a dependency structure related to this structure is searched from the structure of “hungry / empty”. Also, if this search can be performed in multiple stages, it is possible to search for “rice / eat” related to “hungry” and further to a dependency structure related to “rice / eat”. . In this embodiment, the dependency relationship database 890 is constructed by storing the dependency structure that appears from a large amount of natural sentences together with a unique number (sentence number) representing the natural sentence in which the dependency structure appears. A specific example of the construction method will be described later.

  The related topic extraction unit 880 receives a word representing a topic of an utterance and a dependency structure, and extracts and outputs the related word and the related dependency structure with reference to the dependency relationship database 890. When extracting, if you do not know what part of speech the word and dependency structure (topic) that represents the topic of the spoken sentence has, you will not know what the dependency structure related to the related word represents, and Related words that do not match well and related dependency structures can be extracted. Therefore, a condition restricted by the topic part of speech input for each type of topic to be extracted and the part of speech of the topic to be extracted is set. By setting conditions, it is possible to extract a topic whose type of topic is clear and easily matches the template.

  As a condition example, it is assumed that the input dependency structure is composed of (word A / case F / word B) and the related dependency structure to be extracted is composed of (word C / case G / word D). explain. Furthermore, “predicate” refers to a part of speech that is likely to be a predicate, such as a verb, adjective, verb noun, and adjective verb, and “verb” refers to a part of speech that expresses a motion such as a verb / verb, “adjective / adjective verb” A part of speech that is likely to be an evaluation expression such as is defined as an “evaluation word”.

  In order to extract a related word including an evaluation expression and a related dependency structure, restrictions on the word A: general / proprietary noun, word B: verb, word D: evaluation word, dependency on word B → word D, The dependency relationship database 890 is searched under the condition. In the following, capital letters mean words, but the word “word” may be omitted. For example, if the input dependency structure is word A: “abdomen” → B: “free”, the related word and the related dependency structure shown in FIG. 16A can be extracted. The related words having the dependency relationship (B: free → D: tight) and the related dependency structure (C: considerable → D tight) are extracted with at least one of the clauses in the dependency structure input in this way. be able to.

  A dependency structure composed of (noun, F, verb) is assumed to have a structure similar to a so-called predicate term structure and express some event. With this constraint condition, topics (related words and related dependency structures) including evaluation expressions for the event can be extracted.

  To extract a related word including a cause expression and a related dependency structure, the word A: general / proprietary noun, the word B: verb, the word D → B (D · H = “so de kara” · B) The dependency relationship database 890 is searched with the constraint condition of the dependency having the following structure: word D: verb. Assuming that the input dependency structure is the above-described example, as shown in FIG. 16B, a related dependency structure of (C · G · D) + “from” + (A · F · B) is taken out. The reason why (A · F · B) occurs can be extracted.

  In order to extract a related word including an interrogative expression and a related dependency structure, a constraint condition of word A: general / proprietary noun, word B: verb, word D = word B, word C: interrogative . Assuming that the input dependency structure is the above-described example, as shown in FIG. 16C, the relationship dependency structure of interrogative + (A · F · B) can be extracted, and (A · F · B ) Can be extracted.

  In order to extract related words and related dependency structures including self-disclosure expressions, word A: general / proprietary noun, case F: “ha”, word B: noun, “self”, → dependency on word A It is assumed that the constraint condition is large, word C = word A, case G: “ha”, word D: noun, word D ≠ word B. Here, the large number of dependency is a number narrowed down to the top three, for example. If the dependency structure to be inputted is the above example, as shown in FIG. 16 (d), the corresponding “own” + (A · ha · D) with respect to the opponent (A · has B). ) Related dependency structure can be extracted.

  A description will be given of how to prepare a preferable template of the utterance sentence generation unit 720 in the case of generating a dialog utterance sentence using the related dependency structure extracted by providing the constraint condition as described above.

  By creating the template of the above-mentioned spoken sentence generation unit 720 for each constraint condition when extracting the related words and the related dependency structure from the dependency relationship database 890, it is possible to improve the prospect of the relationship between the templates. it can. When the evaluation expression is extracted, for example, a template of “word C + case G + word D” or “word C + case G + word D?” Is prepared, and it is word C + case G + word D. Is it word C + case G + word D? Generate a dialogue utterance. An example is shown in FIG. It is possible to generate a dialogue utterance sentence “Self-disclosure_evaluation” dialogue utterance “It's quite tight” and a (question_evaluation) dialogue utterance sentence “Is it pretty tight?”

  When the cause expression is extracted, for example, a template of “word C + case G + word D?” Is prepared, and an utterance intention (question_facts) “If you eat anything” as shown in FIG. You can generate dialogue utterances such as "Isn't there?" Or "Did you eat anything?"

  When the interrogative expression is extracted, for example, a template of “word C + case G?” Or “word C + case G + word B?” Is prepared, and the intention of utterance (question_fact is shown in FIG. 17C). ) Dialogue utterances “Why?” And “Why are you free?” Dialogue utterances can be generated. However, if the word C is an interrogative questioning the reason such as “why”, the dialogue utterance may become inappropriate, so the template is changed as follows, for example. If a template such as “word C + case G + komo word A + word B kana” is prepared, a dialogue utterance sentence such as “why I am hungry like this” can be generated.

  When the self-disclosure expression is extracted, for example, a template of “My (word C) is (word D)” or {I am (word D) is (word A)} is prepared as shown in FIG. As shown in d), it is possible to generate a dialogue utterance sentence “My stomach is a black hole” with an utterance intention (self-disclosure_facts).

  You may make it generate | occur | produce a dialog speech sentence by substituting the thing with the higher appearance number of the related dependency structure extracted without the constraint conditions to a template. Of course, the input dependency structure for extraction is also substituted into the template.

  When an utterance is generated using only words included in these dependency structures, it is necessary to appropriately determine what expression should be used for each word. Therefore, a dialog utterance sentence is generated using the example used in the sentence to which the retrieved related dependency structure belongs. For example, when generating a dialogue utterance sentence from the dependency relation in the backward direction, the notation of the related dependency structure z in the subsequent stage of the dependency structure x retrieved directly from the input dependency structure (for example, “I was hungry”) (For example, “eat rice”) The number of occurrences is examined as a unit, and the highest number of occurrences are converted so that only the last part matches a simple template such as “It ’s OO” By connecting, a dialogue utterance is generated like "I'm hungry because I'm hungry".

  If the number of occurrences is one, it is often excluded because it is often a context-specific expression. By the above method, it is possible to obtain a dialogue utterance that is difficult to become grammatically unnatural by making use of the connection between the input dependency structure and the related dependency structure and the function expressions included in each.

A method of creating the dependency relationship database 890 described above will be described with reference to FIG. In the method shown in FIG. 18, from a database in which a flat dependency structure consisting of one dependency source and a dependency destination is recorded, first, from two clauses s ¹ _i and s ² _{i in} the input dependency structure i. Then, the standard form of the first word of each phrase is taken out, and a dependency structure group X that appears in the order included in the input dependency structure is searched.

Next, for each of the obtained dependency structures xεX, a dependency structure y including any one of the constituent clauses s ¹ _x and s ² _x εs _x is searched using the sentence ID and the clause ID. . Since the dependency structure y includes any of the clauses s ¹ _x and s ² _{x in x} , y can be considered as a related dependency structure in which x and some clauses overlap.

Further, if a phrase that includes _y including s ¹ _y , s ² _y ∈ s _y and does not include s _x is searched in the same manner, z becomes a phrase such as “rice → eat” for “hungry → empty”. It is possible to obtain a related dependency structure that does not overlap. The related dependency structure thus obtained has specific related properties such as reasons, results, and limitations with respect to the input dependency structure. It is considered that the nature differs depending on the appearance position and the case with respect to the input dependency structure.
A dependency relationship database 890 is a database of the related dependency structures thus obtained. The utterance sentence generation device 800 provided with the dependency relation database 890 uses a database in which the related dependency structure related to the dependency structure is systematically extracted from the flat dependency structure. Can generate a wider range of dialogue utterances. The flat dependency structure is a dependency structure composed of one dependency source and a dependency destination.

  The dependency relationship database 890 is created from the dependency relationship database 890 'in which a flat dependency structure composed of one dependency source and a dependency destination is recorded by the method shown in FIG. A configuration is also conceivable in which the processing for retrieving the receiving structure is performed each time. In other words, the related dependency structure may be searched every time, instead of creating the database in advance.

  In this case, the related topic extraction unit 880 ′ searches for related words and related dependency structures by the method described with reference to FIG. Part of speech information, case, word information, etc. may be used for the search. This method of searching for the related dependency structure every time increases the amount of calculation, but the depth of the dependency can be freely changed, and thus is an advantageous method for generating various dialogue utterances.

  FIG. 19 shows a functional configuration example of the utterance sentence generation apparatus 900 of the present invention. The utterance sentence generation device 900 differs from the utterance sentence generation device 800 in that it includes a natural sentence storage unit 995 and the function of the utterance sentence generation unit 920.

  The natural sentence storage unit 995 stores a natural sentence corresponding to a sentence number corresponding to the dependency structure and case stored in the dependency relation database 890. The utterance sentence generation unit 920 includes a sentence number indicating a word input from the topic extraction unit 310 and a dependency number, and a sentence number indicating a related word input from the related topic extraction unit 880 and a related dependency structure. A natural sentence corresponding to the number is read from the natural sentence storage unit 995 to generate a dialog utterance sentence. The dialogue utterance sentence may be a natural sentence read out from the natural sentence storage unit 995 or may be changed such that the end of the sentence is changed to “Da” or “Masara”.

  According to the utterance sentence generation device 900, since a dialog utterance sentence is generated from a large amount of natural sentences without using a template, it is possible to generate a conversation utterance sentence for utterance sentences of a wide range of topics. The natural sentences stored in the natural sentence storage unit 995 are preferably collected from a microblog containing a large amount of subjective utterances for the purpose of generating spoken dialogue conversation sentences as in the case of the dependency relation dictionary 470 described above.

  The natural sentence storage unit 995 may be provided in all of the above-described embodiments. For example, the utterance sentence generation device 100 ′ provided with the natural sentence storage unit 995 generates a dialogue utterance sentence by searching for a synonym from the natural sentence storage unit 995 using the focus word output from the topic extraction unit 110 as a query. May be.

  According to the utterance sentence generation device 100 described above, a plurality of focal words representing the focal point of a topic are extracted from the user utterance sentence, and the conversation utterance sentence is generated by substituting the plural focal words into a template. A simple dialogue utterance can be generated. Moreover, according to the utterance sentence generation device 200, the related words that are synonyms of the focus word are estimated, and the utterance sentence is generated using the focus word and the related word, so that the dialog utterance sentence that can deal with a wider range of topics is generated. Is possible.

  Furthermore, according to the utterance sentence generation device 300, a group of words having a dependency relationship is substituted into the template, so that it is possible to generate a conversation utterance sentence that can deal with a wide range of topics and has meaning. Moreover, since the utterance sentence generators 300, 400, 500, and 600 use dependency relations included in a large amount of natural sentences such as microblogs for estimating the relation between the focus word and the related word, the user utterance sentence You can enrich the variety of dialogue utterances.

  In addition, the utterance sentence generation device 700 substitutes a word representing the content of the utterance sentence, a dependency structure, a related word related to the word and the dependency structure, and a related dependency structure into the template, so that it can deal with a wider range of topics. A meaningful dialogue utterance can be generated. Further, since the utterance sentence generation device 800 uses the dependency relation database 890, it is possible to generate a wider conversation utterance sentence. Further, since the utterance sentence generation device 900 generates a dialog utterance sentence from a large amount of natural sentences without using a template, it is possible to generate a dialog utterance sentence with a natural expression corresponding to a wide range of topics.

  When the processing means in the above apparatus is realized by a computer, the processing contents of the functions that each apparatus should have are described by a program. Then, by executing this program on the computer, the processing means in each apparatus is realized on the computer.

  The program is distributed by selling, transferring, or lending a portable recording medium such as a DVD or CD-ROM in which the program is recorded. Further, the program may be distributed by storing the program in a recording device of a server computer and transferring the program from the server computer to another computer via a network.

  Each means may be configured by executing a predetermined program on a computer, or at least a part of these processing contents may be realized by hardware.

Claims (6)

A topic extraction unit that extracts a dependency structure representing the content of the utterance sentence, using the morpheme string of the utterance sentence as an input;
With the dependency structure as an input, a related topic extraction unit that outputs any first related dependency structure that is different from the dependency structure, including any of the clauses that configure the dependency structure;
Using the dependency structure and the first related dependency structure as input, an utterance sentence generation unit that generates a dialog utterance sentence by substituting the dependency structure and the first related dependency structure into a template;
An utterance sentence generation device comprising: A topic extraction unit that extracts a dependency structure representing the content of the utterance sentence, using the morpheme string of the utterance sentence as an input;
N is an integer of 1 or more, and the dependency structure is the 0th related dependency structure,
With the 0th related dependency structure as an input, obtain a first related dependency structure that includes any of the clauses constituting the 0th related dependency structure and is different from the 0th related dependency structure, i = 1,..., N includes any clause that constitutes the i th related dependency structure and i + 1 that does not contain any clause that constitutes the i−1 th related dependency structure. A related topic extraction unit that obtains a related dependency structure of and outputs a k-th related dependency structure (k = 1,..., N + 1),
Dialogue by substituting the jth related dependency structure (j = 0, ..., N + 1) into the template with the jth related dependency structure (j = 0, ..., N + 1) as input An utterance sentence generator for generating an utterance sentence;
An utterance sentence generation device comprising: In the utterance sentence generation device according to claim 1 or 2,
In addition, there is a dependency relationship database,
The dependency relationship database is a database that can search for a dependency structure related to the dependency structure when a certain dependency structure is given.
The related topic extraction unit receives the dependency structure as an input, refers to the dependency relationship database, and stores the first related dependency structure or the kth related dependency structure (k = 1,..., N + 1) output,
An utterance sentence generation device characterized by the above. The topic extraction unit receives a morpheme sequence of an utterance sentence, and extracts a dependency structure representing the content of the utterance sentence;
The related topic extraction unit receives the dependency structure as an input, and outputs a first related dependency structure that is different from the dependency structure including any of the clauses constituting the dependency structure; and
The utterance generation unit generates the dialog utterance by substituting the dependency structure and the first related dependency structure into a template, using the dependency structure and the first related dependency structure as inputs. Sentence generation process,
An utterance sentence generation method comprising: The topic extraction unit receives a morpheme sequence of an utterance sentence, and extracts a dependency structure representing the content of the utterance sentence;
N is an integer of 1 or more, and the dependency structure is the 0th related dependency structure,
The related topic extraction unit receives the 0th related dependency structure as an input and includes any one of the clauses constituting the 0th related dependency structure, which is different from the 0th related dependency structure. A dependency structure is obtained, and for i = 1,..., N, any of the clauses constituting the i-th related dependency structure and including any of the clauses constituting the i-1th related dependency structure A related topic extraction process of obtaining an i + 1th related dependency structure not included and outputting a kth related dependency structure (k = 1,..., N + 1);
The utterance sentence generation unit receives the jth related dependency structure (j = 0,..., N + 1) as an input, and uses the jth related dependency structure (j = 0,..., N + 1) as a template. Utterance sentence generation process to generate dialogue utterance sentence by substituting into,
An utterance sentence generation method comprising:   The program for functioning a computer as an utterance sentence production | generation apparatus of any one of Claim 1 thru | or 3.

Images