JP2009193448A - Dialog system, method, and program - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a dialog system capable of executing a natural dialog (response) which can sufficiently perform rendition of a sense of closeness to human beings. <P>SOLUTION: The dialog system for giving and receiving data of natural language sentences between human beings and a device performing interface to talk with the human beings includes: an analysis part for analyzing utterances of human beings; a target part approval part for approving a target part so as to extract from human utterances elements for the system to utter by using a result of analysis; and an extraction part for extracting elements from the human utterances so that system utterances become appropriately long on the basis of the approved target part. <P>COPYRIGHT: (C)2009,JPO&INPIT

Description

  The present invention relates to an interactive system, method, and program, and can be applied to, for example, an interactive information search apparatus.

2. Description of the Related Art Conventionally, there is a response generation device described in Patent Document 1 as a device that analyzes a human utterance, identifies (extracts) a predicate and a case element corresponding to the predicate, and responds using them. This conventional device realizes a response of the system (device) utterance “Did you make a sideboard?” To the user's utterance “… I made all the sideboards in the living room”. In the apparatus described in Patent Document 1, since a plurality of utterance candidates for the system are prepared, the utterance candidates can be selected at random, or the priority can be freely set (for the group classified by the utterance candidate generation method). Can be selected.
JP 2007-206888 A

  By the way, as an interactive information search device, the patent applicant has been researching and developing a laddering type search device. In other words, the device researches and develops a device that draws out the needs and values of the user and searches for services and contents that match the extracted information by repeating questions that are gradually delved into the dialogue with the user. Yes. In order to appropriately draw out the user's needs and values, it is required that the user have a close feeling (familiarity) with natural dialogue.

  However, the above-described conventional apparatus is a method of identifying (extracting) predicates and corresponding case elements and responding using them, and the method of generating responses is limited, effectively producing a close feeling. Can not do it.

  In the conventional apparatus, the predicate and case element leave only the central word, and the modifier is not used for the response. Only one case element can be combined with a predicate. For this reason, the naturalness of dialogue cannot be sufficiently guaranteed.

  Furthermore, in the ladder type search device, it is necessary to prepare several utterances (a kind of questions to the user) for the purpose of obtaining information from the user, and the system must take the initiative to turn the story. The utterance from the system is “response received from the user” or “simple interaction”, and does not disclose or suggest how the system leads the conversation.

  Furthermore, in the conventional apparatus, only the vocabulary used by the user can be used, and the response becomes monotonous.

  In addition, in the conventional apparatus, when neither a predicate nor a case element exists, a simple conflict ("hee" or "yes") appears only, and it is not strong as an effect of closeness.

  Therefore, there is a demand for a dialogue system, method, and program that can realize a natural dialogue (response) that can sufficiently produce a sense of closeness to humans.

  The first aspect of the present invention uses an analysis unit for analyzing a human utterance and a result of the analysis in a dialog system in which a natural language sentence data is exchanged with a device that interfaces with a human to interact with the human. Then, from the human utterance so that the system utterance becomes an appropriate length based on the target location, and the target location recognition unit for identifying the target location for extracting the elements for the system to utter from the human utterance And an extraction unit for extracting elements.

  According to a second aspect of the present invention, there is provided a dialogue method in which a device for interfacing with a human being and a dialogue system exchange natural language sentence data and interact with a human, wherein the dialogue system includes an analysis unit, a target location recognition unit, and an extraction unit. The analysis unit analyzes a human utterance, the target location recognition unit uses the analysis result, and the target location for extracting elements for the system to utter from the human utterance. The extraction unit extracts the elements from the human utterance so that the system utterance becomes an appropriate length based on the target portion.

  According to a third aspect of the present invention, there is provided a computer program for constructing a dialog system for interacting with humans by transferring natural language sentence data to and from a device that interfaces with humans, and an analysis unit for analyzing human speech. And using the above analysis results, the system utterance has an appropriate length based on the target location recognition unit for identifying the target location for extracting the elements for the system to utter from the human utterance and the target location. It is made to function as an extraction part which extracts an element from human speech so that it may become.

  According to the present invention, the selection of the location to be used for parrot return and the length (number of words) for the parrot return are changed according to the expression in human speech, etc., so that a sense of closeness to humans can be sufficiently produced. It is possible to provide a dialogue system, method and program capable of realizing a natural dialogue (response).

(A) First Embodiment Hereinafter, a first embodiment of an interactive system, method and program according to the present invention will be described in detail with reference to the drawings.

(A-1) Configuration of the First Embodiment FIG. 1 is a functional block diagram showing the main configuration of the dialogue system 10 according to the first embodiment. FIG. 1 shows a component that generates a system utterance using a user utterance as an input.

  The dialogue system 10 may be constructed as a part of a larger device such as a laddering search device. Further, the dialogue system 10 may be constructed by installing a dialogue program (including fixed data) in a general-purpose information processing apparatus such as a PC (personal computer) or a server. In any case, functionally, it can be represented by the configuration of FIG. The installation of the interactive program is not limited to the method by downloading via a communication network, but may be a method via a computer-readable recording medium. For example, when used as a part of a ladder type search device having a function of searching and providing a job change destination, the dialogue system 10 is mounted on a Web server that provides a site for job change destination provision. It will be.

  In FIG. 1, the dialogue system 10 according to the first embodiment includes an analysis unit 11, a target location recognition unit 12, an extraction unit 13, and a deformation unit 14. The target location recognition unit 12, the extraction unit 13, and the deformation unit 14 constitute a parrot return response generation unit 15.

  The dialogue system 10 receives user utterances composed of natural language sentences. For example, a natural language sentence (text) entered in a field for speech sentence input on a Web page displayed on a personal computer as a user terminal by the user is input to the dialogue system 10. Further, for example, the apparatus on which the dialogue system 10 is mounted may be provided with an input device such as a keyboard to capture user utterances. Further, for example, the user's utterance may be captured by recognizing a voice (voice signal) captured by the microphone of the user terminal or the microphone of the apparatus on which the interactive system 10 is mounted.

  The analysis unit 11 performs morphological analysis and syntax analysis on the user utterance, divides it into words (morphemes), and clarifies the sentence structure. Existing analysis methods can be applied to morphological analysis and syntax analysis.

  The target part recognition unit 12 recognizes a part of the user utterance that can be determined to be suitable for returning a parrot. The determination criteria and determination method will be described in the operation description to be described later. Some of the determination criteria are as follows. The “predicate + its object or subject” near the end in the user's utterance is set as the target location (candidate). A “noun + its modifier” near the end in the user utterance is set as a candidate (candidate). Several words including intention / subjective expressions such as “unreasonable”, “problem”, “want”, “cannot” in the user's utterance, or similar expressions are set as candidates (candidates). Instead of the intention / subjective expression itself, the part that describes the content of “worried” or “want” is the target part (candidate). When there are a plurality of target locations (candidates), the target location recognition unit 12 narrows down to one target location according to a predetermined rule. The specific method will be clarified in the operation section.

  The extraction unit 13 extracts (selects) a portion (subtree in the syntax tree) that has a natural length by returning a parrot from the target location in the user utterance certified by the target location certification unit 12 and its vicinity. As will be described later in the operation section, any number of words (4 words or 2 words) that does not deviate significantly from the standard length (for example, 3 words) determined by the type of expression is allowed. It should be noted that an upper limit number of words may be set instead of the standard length so as to ensure a natural length in the parrot return (do not forcibly increase when short). When the extracted part (subtree in the syntax tree) has an unacceptable length, the extraction unit 13 also shortens the extracted part. In shortening, an object or subject is dropped or a modifier is dropped according to a predetermined rule.

  The transformation unit 14 transforms (or shapes) the expression when the parrot return portion obtained by the extraction unit 13 corresponds to a predetermined rule. For example, convert tense or honorific expression. If nouns (phrases) are extracted, “Is it?” Or “Is it?” Is added.

  The parrot return part (natural language sentence) obtained by the processing of the deforming unit 14 becomes the system utterance. For example, the system utterance is included and displayed on a Web page displayed on a personal computer which is a user terminal. Further, for example, the device on which the dialogue system 10 is mounted may include a display device to display a system utterance. Furthermore, for example, voice synthesis may be performed on a system utterance made up of text data, and a voice (voice signal) of the system utterance may be uttered from the speaker of the user terminal or the speaker of the apparatus equipped with the dialogue system 10.

  The analysis unit 11, the target location recognition unit 12, the extraction unit 13, and the transformation unit 14 store, for example, a dedicated control device or a processor (CPU) that executes a program, and a program and data that is executed by the processor. It is realized by hardware resources including storage devices such as RAM (Random Access Memory), ROM, and HDD.

  In the above description, each function is described. However, the physical configuration of the hardware to be realized is clearly separated for each part and does not need to be prepared independently. For example, the HDD that stores the program of the target location recognition unit 12 may be the same as the HDD that stores the analysis dictionary data of the analysis unit 11, and further, a part of a device that realizes other functions may be used. May be used. Further, a part of the dialogue system 10 may be arranged in another place connected by a network.

(A-2) Operation of the First Embodiment Next, the operation of the dialogue system 10 according to the first embodiment having the above-described units (the dialogue method according to the first embodiment) will be described with reference to the drawings. explain. FIG. 2 is a flowchart showing the operation of the interactive system 10 according to the first embodiment.

  The dialogue system 10 according to the first embodiment starts the processing shown in FIG. 2 when a user utterance is input, and performs morphological analysis / syntactic analysis (S100) by the analysis unit 11 and target location by the target location recognition unit 12. Authorization (S101), extraction by the extraction unit 13 (S102), and transformation (shaping) by the transformation unit 14 (S103) are sequentially executed to form a system utterance. Below, the processing of each step S100, S101, S102, S103 will be described in detail.

  The analysis unit 11 performs morphological analysis and syntax analysis by a known analysis method (S100). FIG. 3 shows a morphological analysis result for a user utterance “I want to do a job that allows me to grow as a human while touching a person”, and FIG. 4 shows a parsing result (syntax tree) for the morphological analysis result. Yes.

  The target location authorization unit 12 authorizes the target location for the parrot return while using the authorization special expression list shown in FIG. 5 (S101).

  As shown in FIG. 5, the special expression list for authorization defines a group name, a specific special expression, and an extraction center.

  The first line L11 is a special expression such as “I want”, “I want”,... In the user utterance (the list is described in the present form, the original form, etc., but is included in the user utterance in other forms. Is included; the same applies to other lines), it belongs to the group of “intention expressions” and is the main noun of the main part whose predicate is “I want” or “I want” Represents the center of extraction. Since the example of the analysis result in FIG. 4 includes “tai”, this corresponds to this case, and “work”, which is the main noun of the main part, is set as the center of extraction.

  The second row L12 belongs to the “subjective expression” group when the user utterance includes special expressions such as “problematic”, “impossible”, “can do”, and so on. This indicates that the core noun in the immediately preceding dependency element such as “” or “can” is placed at the center of extraction. The example of the analysis result in FIG. 4 includes “can”, and thus corresponds to this case, and “growth” that is the core noun in the immediately preceding dependency element is the center of extraction. Also, if the user utterance is “... I am in trouble if there is little overtime”, “Place” will be the center of extraction.

  The third row L13 belongs to the group of “activity expression” when a special expression such as “notice”, “experience”, etc. is included in the user utterance, and “notice”, “experience”, etc. This means that the special expression (core vocabulary) itself is placed at the center of extraction. For example, if the user's utterance is “… I noticed that it is a work that can be utilized,” “notice” will be the center of extraction.

  The fourth row L14 belongs to the “presence / absence expression” group when the user utterance includes special expressions such as “Yes”, “No”, etc., and the special utterances such as “Yes”, “No”, etc. This means that the main core nouns whose expressions are predicates are placed at the center of extraction. For example, if the user's utterance is “... the place where it is growing seems to have momentum in the company”, “momentum” will be the center of extraction.

  The target location authorizing unit 12 confirms whether or not the expression described in the “specific special expression” in the authorization special expression list in FIG. 5 exists in the analysis result described above. If it exists, the target location recognition unit 12 recognizes the location of the analysis result (user utterance) corresponding to the “extraction center” of the corresponding line in the special expression list as the target location. If there are a plurality of recognized target parts, the one closest to the predicate of the main sentence is selected in the syntax analysis result. In other words, the selection is made in consideration of the distance in the syntax analysis result, not the distance as the appearing character string. In the parsing result of FIG. 4, in the collation with the special expression list, there are the target part “work” related to the special expression “I want” and the target part “growth” related to the special expression “capable”. As is clear from FIG. 4, the target part having a short distance to the part “I want to do” is “work”, and therefore, the target part “work” is placed at the center of extraction.

  The extraction unit 13 extracts a user utterance part to be used for parrot return while using the extraction special expression list shown in FIG. 6 (S102). The extraction unit 13 basically takes out a subtree (a group of trees that can be parsed as a result of the syntax analysis) rooted by the word (group) of the extraction center of the target location, which is recognized by the target location recognition unit 12. Extract user utterances used for parrot return. The extraction unit 13 determines whether or not the extracted user utterance part is within the upper limit number of words determined based on a word number guide to be described later in the extraction special expression list. If the part is extracted as it is and it does not fit, the expression is reduced from the extracted user utterance part according to the extraction (element selection) rule to be described later in the extraction special expression list to reduce it to the maximum number of words or less. The subsequent user utterance part is taken as the extraction result.

  As shown in FIG. 6, the special expression list for extraction defines group names, word count guidelines, and extraction rules.

  In the first row L21, the number of words in a predetermined part belonging to the “intention expression” group is 5, and the number of extracted user utterance parts (subtrees) is larger than the upper limit number determined based on this word number guide Describes how to reduce the number of words as follows. The first reduction is to delete the modification elements for the center of extraction (see FIG. 5) from the (distant) far side. Second, if the distances are the same, the case elements such as the main case and the target case Remove anything other than. Third, when the case elements have the same viewpoint, the method of deleting the far side on the appearance character string is applied. The deletion here is performed by using a word or a sub-tree branch as a minimum unit, and repeating the reduction according to the first to third rules until it falls within the upper limit number. For example, if the upper limit number is the word count guideline + 1 word, the upper limit word count for the “intention expression” group is 6.

  In the parsing result of FIG. 4, as described above, “work” is the extraction center and becomes the root. Therefore, when extracting the user utterance part (subtree) for parrot return, “work that can grow as a human being” "(7 words) is taken out. Since the upper limit of 6 has been exceeded, some deletions are made. Here, “I am” and “As a human being” have the same level of reliance on “work”, but the second rule of deleting all but the case element leaves the “I” of the case , Delete "as a human". As a result, the number of words becomes 5 and is within the upper limit number (6), and the stage where this deletion is performed is taken as the extraction result. In other words, “work that I can grow” is the extraction result. This extraction result is not a continuous part of the user's utterance but a connection between two parts.

  In the above description, the deviation from the word count guideline is not allowed only when the number of words is large, but when the word count guideline is small, it may be added (the same applies to any row in FIG. 6). The added rule may be symmetric with the rule for deletion. For example, firstly, add the closest one of the modification elements for the extraction center (see FIG. 5). Second, if the proximity is the same, add case elements such as the main case and the purpose case. Third, when the case elements have the same viewpoint, the method of adding the closest one on the appearance character string is applied.

  The second row L22 of the special expression list for extraction has a word number guideline of a predetermined location belonging to the “subjective expression” group, and the extracted user utterance portion (subtree) is extracted from the upper limit number determined based on this word number guideline. ) Is deleted from the farther (relative) of the modifiers for the extraction center (see FIG. 5). However, there are two exceptions. In cases where there is a precaution in the modifier, the elements corresponding to the principal and purpose are not deleted. As a result of deletion, if it becomes only the core noun, it is not deleted, but the extraction result is left as it is.

  If the user's utterance is “... I am in trouble if there is not too much overtime,” as described above, “Place” will be the center of extraction, and “Location with less overtime” (4 words) will be the user utterance part for returning parrots (4 words) First, it is extracted as a subtree. Since the number of words in the sub-tree exceeds the upper limit, “Overtime” is far away, so I want to delete it, but there is a saying that “Overtime” is “less”, so “Overtime” is not deleted. . In order to fit in the number of words, all you have to do is delete all `` less overtime '', but then it will only be the core noun `` Place '', so accept the word number over and finally extract `` Place with less overtime '' As a result.

  The third row L23 of the special expression list for extraction is the result of extracting only the core words that are recognized as the extraction center (see FIG. 5) and the word count guideline of the predetermined part belonging to the “activity expression” group is 1. Is described.

  For example, if the user ’s utterance is “… I noticed that it was a work that can be utilized,” and “Awareness” was the extraction center, “Awareness” cannot be divided in particular. Do not execute.

  The fourth row L23 of the special expression list for extraction has a word number guideline of a predetermined location belonging to the “presence / absence expression” group, and the extracted user utterance portion (subtree) is larger than the upper limit number determined based on this word number guideline. ) Is deleted from the farther (relative) of the modifiers for the extraction center (see FIG. 5).

  For example, if the user's utterance is “... the place where the growth is likely to have internal momentum” and “momentum” is the center of extraction, “internal momentum” (3 words) is for returning parrots. It is extracted as a user utterance part (subtree). Since “in-house momentum” (3 words) is within the upper limit number (3 words), “in-house momentum” becomes the extraction result as it is. Note that the syntax tree is not listed, but “the one that is growing” is directly related to “probably ...”, so when viewed from the extraction center “momentum”, the target of extraction It is outside.

  The transformation unit 14 transforms (shapes) the character string extracted by the extraction unit 13 in accordance with the following internally recorded transformation rules (S103). For example, a correspondence table for conversion to honorific is prepared and transformed (converted) when a heading matches. For example, convert “Awareness” to “Awareness” (respected word), or prepare a correspondence table that converts “You” to “You” (a polite language) and apply it. . In addition, for example, if the extraction result ends with a noun (phrase), “Is it?” Or “Is it?” Is added. Finally, general morpheme generation (process opposite to morpheme analysis) is performed and output in a form that is not amusing as Japanese.

  In addition, when the process of the object location recognition part 12 and the extraction part 13 is complete | finished, when it cannot extract, the system utterance of a parrot return response will not be performed.

(A-3) Effects of the First Embodiment According to the first embodiment, a special expression list for authorization is prepared, the intention / subjective expression in the user's utterance is searched, and the intention / subjective expression or its periphery Since this element is preferentially used for a response (system utterance), it is possible to effectively produce a sense of closeness to the user.

  In addition, according to the first embodiment, unlike the conventional apparatus, not only the predicate and the central word of the case element are extracted, but also the preferential word is used and the word or its periphery is used for the response. After deciding whether to do so, the elements were trimmed (added) to match the length of the guideline set in advance, so the length of the system response became a natural length and dialogue Can guarantee the naturalness.

  Furthermore, according to the first embodiment, the transformation unit 14 applies transformation (paraphrase processing) to the part extracted from the user utterance so as to obtain the final parrot return system utterance. , It is possible to prevent the response from becoming monotonous or unnatural.

  As described above, as a result of effectively producing a close-in feeling and assuring the naturalness of the dialogue, the dialogue is lost and it is easy to collect information from the user.

(B) Second Embodiment Next, a second embodiment of the interactive system, method and program according to the present invention will be described in detail with reference to the drawings.

  FIG. 7 is a functional block diagram showing the main configuration of the interactive system 10A according to the second embodiment, in which the same and corresponding parts as those in FIG. 1 according to the first embodiment are denoted by the same reference numerals. Yes.

  The dialogue system 10A according to the second embodiment includes a next topic selection unit 20 and a topic database 21 in addition to the configuration of the dialogue system 10 according to the first embodiment.

  The topic database (topic DB) 21 stores dialog scenario information, system utterances, and the like. For example, if the dialogue system 10A is incorporated in a search device for introducing a job change destination, there are many items such as desired work location, desired annual income, working hours (including acceptable overtime hours), working days, etc. The system utterances are stored in a hierarchical manner for each item (for example, system utterances that ask the desired work location, and if the user utters the Kanto region, system utterances that draw out hope in a smaller area) The system utterance transition method within a certain item (dialogue scenario), and when the collection of information about a certain item is completed, the transition method (dialogue scenario) of which item the system utterance will be transferred to is memorized. is doing.

  In the second embodiment, if the target location recognition unit 12 cannot recognize the target location, the target location determination unit 12 notifies the next topic selection unit 20 of the fact. The topic selection unit 20 is notified. The next topic selection unit 20 extracts and outputs a system utterance (next topic) according to the stored contents of the topic database 21 when the target location recognition or extraction is not successful.

  According to the second embodiment, the same effect as that of the first embodiment can be obtained, and further, the effect that the topic can be changed by the initiative of the system is also achieved. In other words, there is a possibility that it will not shift to another topic only with a parrot return response, but this can be avoided.

(C) Third Embodiment Next, a third embodiment of the interactive system, method and program according to the present invention will be described in detail with reference to the drawings.

  FIG. 8 is a functional block diagram showing the main configuration of the dialogue system 10B according to the third embodiment. The same reference numerals are given to the same and corresponding parts as those in FIG. 1 according to the first embodiment. Yes.

  The dialogue system 10B according to the third embodiment includes a paraphrase unit 30 in the parrot return response generation unit 15B in addition to the configuration of the dialogue system 10 of the first embodiment.

  The paraphrasing unit 30 has a built-in synonym phrase dictionary and replaces all or part of the extracted user utterance part if it can be replaced with another expression, and the extracted user utterance part has the same content. It replaces with other expressions. The synonym phrase dictionary is, for example, a database in which a certain phrase and its paraphrase are paired. For example, it is a database that can acquire “smooth turn” as a headline and “go smoothly” as a replacement word. By pulling this database, when the phrase “work is turning well” is extracted from the user utterance, it can be rephrased as “work goes smoothly”.

  When the paraphrase unit 30 does not operate, the transformation unit 14 performs a transformation process on the extraction result by the extraction unit 13. When the paraphrase unit 30 operates, the transformation unit 30 outputs the paraphrase unit 30. The transformation process is executed on the paraphrase character string as the extraction result.

  According to the third embodiment, the same effects as those of the first embodiment can be obtained, and further, the expression used by the user can be rephrased to form a system utterance, and the monotonousness of the parrot return response can be reduced. Can be prevented.

(D) Fourth Embodiment Next, a fourth embodiment of the interactive system, method and program according to the present invention will be described in detail with reference to the drawings.

  FIG. 9 is a functional block diagram showing the main configuration of the interactive system 10C according to the fourth embodiment. The same reference numerals are given to the same and corresponding parts as those in FIG. 1 according to the first embodiment. Yes.

  The dialogue system 10C according to the fourth embodiment includes a frame addition unit 40 in the parrot return response generation unit 15C in addition to the configuration of the dialogue system 10 of the first embodiment.

  The frame addition unit 40 has a built-in database for taking out additional frames (considerations), and according to the contents of the extraction result of the extraction unit 13 (or the original user utterance content), Choose the appropriate phrase from the additional phrases such as “It ’s hard”, “It was hard”, “It was good”. For example, “Is that so?” Is used as a general-purpose frame that is added without considering emotions. For example, a database in which a headword such as “cannot be” or “cannot be” and a frame such as “is hard” is stored, and the headword of the database is included in the result extracted by the extraction unit 13. If there is, the frame paired with it is selected and sent to the deformation unit 14C. Further, for example, a subgroup is provided in the group in the special expression list for authorization shown in FIG. 5 (in the case of subjective expression, a positive subjective expression corresponding to “Yes”, or a negative subjective expression corresponding to “Uncomfortable” or “Unreasonable”) The subgroup name may be used on the heading side of the database for extracting the additional frame (consideration).

  In the database for extracting the addition frame (consideration) in the frame addition unit 40, additional position information that defines whether to add before the addition target parrot return response or at the end of the addition target parrot return response The frame adding unit 40 sends the selected frame and the additional position information to the deforming unit 14C. For example, it is specified that “is it?” Is added before the parrot return response, and “is hard” is added at the end of the parrot return response.

  The deforming unit 13C adds the frame (consideration) sent from the phrase adding unit 40 to the front or end of the parrot return response after the deformation process, and outputs it as a system utterance.

  According to the fourth embodiment, the same effects as those of the first embodiment can be obtained, and furthermore, the frame (contrast) selected from a plurality of types is included in the parrot return response. A feeling can be exhibited more strongly.

(E) Fifth Embodiment Next, a fifth embodiment of the interactive system, method and program according to the present invention will be described in detail with reference to the drawings.

  FIG. 10 is a functional block diagram showing the main configuration of the interactive system 10D according to the fifth embodiment. The same reference numerals are given to the same and corresponding parts as those in FIG. 1 according to the first embodiment. Yes.

  The dialogue system 10D according to the fifth embodiment includes a system utterance confirmation unit 50 in the parrot return response generation unit 15D in addition to the configuration of the dialogue system 10 of the first embodiment. Further, in the dialogue system 10D according to the fifth embodiment, a system utterance history database (system utterance history DB) 51 is also a constituent element.

  The system utterance history database 51 stores at least the immediately preceding system utterance. For example, a database storing dialogue (system utterances and user utterances) history can be used as the system utterance history database 51 of the fifth embodiment.

  The system utterance confirmation unit 50 is provided with information on the element word (see the extraction center in FIG. 5) that is to be recognized as the target location from the target location recognition unit 12D. The system utterance confirmation unit 50 confirms whether or not the element word to be recognized as the target location matches the word included in the system utterance immediately before existing in the system utterance history database 51. When the immediately preceding system utterance includes an element word to be recognized as a target location, the system utterance confirmation unit 50 notifies the target location determination unit 12D and excludes it from the target location authorization candidates.

  For example, if there is one element word candidate to be recognized as the target location and it is excluded from the candidate candidates for the target location, a parrot return response will not be made for the current user utterance. In addition, for example, when there are a plurality of candidate element words to be recognized as target locations, when some are excluded from the candidate candidates for the target location, one of the remaining recognition candidates is selected.

  According to the fifth embodiment, the same effects as those of the first embodiment can be obtained, and furthermore, the parrot return response candidates are compared with the past system utterances. It is possible to prevent the system utterances from overlapping, and to realize a natural dialogue.

(F) Other Embodiments In the description of the above-described embodiment, various modified embodiments have been referred to. However, modified embodiments as exemplified below can be cited.

  The technical features of the above embodiments may be applied in combination if applicable in combination.

  In the first embodiment, while the special expression list for authorization including the specific special expressions as shown in FIG. 5 is used, the object portion for recognizing the parrot is shown. You may make it authorize the target part of a parrot return using attribute values. For example, a parrot return target location may be recognized using expressions belonging to the time attribute or the regional attribute. Regardless of whether the user's utterance is "Please do overtime within 2 hours" or "Overtime within 30 minutes is good", "Within 2 hours" or "Within 30 minutes" is a candidate for the parrot return target candidate As described above, the target portion may be authorized by the time attribute. As for the attribute value, the extraction center as shown in FIG. 5, the word number standard and the extraction rule as shown in FIG. 6 are determined.

  In the second embodiment, the system utterance is switched to the next topic when the parrot cannot be returned. However, the system utterance may be switched to the next topic even when the parrot can be returned. . For example, the number of consecutive parrot return responses may be counted, and when the number of consecutive times reaches a predetermined number, the next system utterance may be switched to the next topic. In this case, a parrot return response may be added before the next topic to make a system utterance.

  In the third embodiment, the case where there is one paraphrase destination candidate is shown, but a plurality of paraphrase destination candidates may be prepared for the same paraphrase source. In this case, the oldest paraphrase destination candidate may be applied.

  In the fourth embodiment, the frame is always added when the frame addition condition is satisfied. However, whether the frame is added even if the frame addition condition is satisfied according to the number of consecutive additions or the addition ratio. It may be determined whether or not. For example, the frame is not added in the next system utterance in which the addition of the frame is continued twice.

  In the fifth embodiment, when the element word candidate at the target location is included in the previous system utterance, it is excluded from the candidates, but is included in the most recent system utterances of the last few times. In this case, it may be excluded from the candidates.

It is a functional block diagram which shows the main structures of the dialogue system which concerns on 1st Embodiment. It is a flowchart which shows operation | movement of the dialogue system which concerns on 1st Embodiment. It is explanatory drawing which shows the morphological analysis result with respect to user utterance "I want to do the job which can grow up as a person while contacting a person." It is explanatory drawing which shows the syntax analysis result (syntax tree) with respect to user utterance "I want to do the job which can grow up as a person while contacting a person." It is explanatory drawing which shows the special expression list for authorization which the object location certification | authentication part in 1st Embodiment incorporates. It is explanatory drawing which shows the special expression list for extraction which the extraction part in 1st Embodiment incorporates. It is a functional block diagram which shows the main structures of the dialogue system which concerns on 2nd Embodiment. It is a functional block diagram which shows the main structures of the dialogue system which concerns on 3rd Embodiment. It is a functional block diagram which shows the main structures of the dialogue system which concerns on 4th Embodiment. It is a functional block diagram which shows the main structures of the dialogue system which concerns on 5th Embodiment.

Explanation of symbols

  DESCRIPTION OF SYMBOLS 10, 10A, 10B, 10C, 10D ... Dialog system, 11 ... Analysis part, 12, 12D ... Target location recognition part, 13 ... Extraction part, 14, 14C ... Deformation part, 15, 15B, 15C, 15D ... Parrot return response Generation unit, 20: Next topic selection unit, 21 ... Topic database, 30 ... Paraphrase unit, 40 ... Frame addition unit, 50 ... System utterance confirmation unit, 51 ... System utterance history database.

Claims (10)

In a dialogue system that interacts with humans by transferring natural language sentence data to and from devices that interface with humans,
An analysis unit that analyzes human speech,
Using the above analysis results, the target location authorizing unit for authorizing the target location for extracting the elements for the system to utter from human speech,
An interactive system comprising: an extraction unit that extracts an element from a human utterance so that the system utterance has an appropriate length based on the target portion.   The dialogue system according to claim 1, further comprising a deforming unit that transforms the extracted human speech element into a natural form as a system speech.   The dialogue system according to claim 1 or 2, wherein the target part recognition unit is different in the target part to be authenticated according to the type of specific expression used during human speech.   The dialogue system according to any one of claims 1 to 3, wherein the extraction unit has a different extraction method according to the type of specific expression used during human speech.   Next topic selection that causes the system utterance of the next topic to be extracted from the topic database and output when the target location recognition unit does not succeed in target location recognition or when the extraction unit does not succeed in extraction The interactive system according to claim 1, further comprising a unit.   The dialogue system according to claim 1, further comprising a paraphrase unit that converts the element word extracted by the extraction unit into another expression. When the element word extracted by the extraction unit or a word that triggers a conflict is included in a human utterance, the phrase addition unit further generates a frame of a match corresponding to the phrase,
The dialogue system according to any one of claims 1 to 6, wherein a final system response is obtained by adding a compatibility frame to a system response determined according to an extraction result of the extraction unit. The system further includes a system utterance confirmation unit that confirms whether or not the word of the target part that the target part authorization unit is trying to recognize matches the word included in the system utterances of the previous few times.
The target part certifying part makes an inquiry to the system utterance confirmation part when certifying the target part, and if it matches the word included in the system utterance several times before, it shall not be considered as the target part. 8. The interactive system according to claim 1, wherein In a dialogue method in which a device that interacts with humans and a dialogue system exchange natural language sentence data and interact with humans,
The dialogue system includes an analysis unit, a target location recognition unit, and an extraction unit.
The analysis unit analyzes human speech,
The target location recognition unit uses the analysis result to identify the target location for extracting the elements for the system to utter from human utterances,
A dialogue method, wherein the extraction unit extracts elements from human utterances so that a system utterance has an appropriate length based on the target portion. A computer that builds a dialogue system that interacts with humans by sending and receiving natural language sentence data to and from devices that interface with humans.
An analysis unit that analyzes human speech,
Using the above analysis results, the target location authorizing unit for authorizing the target location for extracting the elements for the system to utter from human speech,
An interactive program that functions as an extraction unit that extracts elements from a human utterance so that a system utterance has an appropriate length based on the target portion.

Images