JP2019036171A - System for assisting in creation of interaction scenario corpus - Google Patents

Abstract

To provide a system for assisting in creation of an interaction scenario corpus which does not need to constrain operators by allowing the same scenario to be created by a plurality of persons and can improve work efficiency.SOLUTION: At a time t1, attribute information of a worker in charge of an inputting operation of an utterance is registered to a scenario creation assistance device 1. At a time t2, the scenario creation assistance device 1 selects and extracts an incomplete scenario for requesting an input of an utterance to workers W1 and W2 from a scenario DB2. At a time t3, the incomplete scenario is provided to each of the workers W1 and W2, and the input of utterances is requested. At times t4 and t5, the workers W1 and W2 input utterances to be added into the incomplete scenario to an utterance input field 31. At times t6 and t7, when the workers W1 and W2 click a transmission button 32, the utterance input by each of the workers W1 and W2 is notified to the scenario creation assistance device 1, and is additionally registered to the scenario DB2 as the utterance of each incomplete scenario.SELECTED DRAWING: Figure 3

Description

  The present invention relates to a dialogue scenario corpus creation support system, and in particular, presents an incomplete scenario in which the number of registered utterances is insufficient to a worker who is responsible for utterance input work, and inputs and updates the utterance as the next speaker. The present invention relates to a dialogue scenario corpus creation support system that completes a dialogue scenario by repeating registration.

  In order to realize a high level of dialogue in the dialogue agent system, it is necessary to create large-scale text data (dialog scenario corpora) describing the expected dialogue exchange (dialog scenario). A dialogue scenario corpus can be created from existing chat or SNS data, or manually entered, but it is generally better to manually create a higher quality that suits the application. It often becomes a thing.

  Patent Document 1 discloses a method of generating a dialogue corpus by performing natural language processing on input sentences in text communication by a plurality of users such as chat, and acquiring and accumulating relation information between sentences. .

  Non-Patent Document 1 recruits two workers who perform text dialogue using crowdsourcing, and supports dialogue input in chat format and annotation work that assigns its role to the entered text. A system is disclosed.

  Non-Patent Document 2 collects communications on SNS such as Twitter as a corpus, and finds posts on SNS that have a high degree of similarity to user utterances. A method for realizing a dialogue system by using a response is disclosed. If there is no post similar to the user utterance, Non-Patent Document 2 requests a worker on crowdsourcing to create a reply.

JP 2008-299754A

"Conversation corpus construction method using open platform and crowdsourcing" Hiroshi Tsukahara, Kei Utsumi, Language Processing Society Annual Conference (March 2015) "Dialogue system using real-time crowdsourcing and Twitter large-scale corpus" Fumihiro Bessho, Tatsuya Harada, Yasuo Kuniyoshi, IPSJ Research Report (May 2012)

  In Patent Document 1 and Non-Patent Document 2, on the assumption that text communication data such as chat and SNS already exists, these relationships are acquired and used as a corpus. It is difficult to use it as it is because the information is transmitted from the standpoint of. For example, men's posts and women's posts are mixed, and the combined corpus destroys the character of the dialogue agent. In order to improve the quality of dialogue agents, it is necessary to manually create a corpus that takes into account the personality.

  Non-Patent Document 1 proposes a method of collecting a plurality of workers using crowdsourcing and allowing two worker pairs to chat. In this method, since it is necessary to gather two workers at the same time for chatting, time constraints are increased. In addition, while the other party is inputting a sentence, the other worker is in a waiting state, so a lot of time is wasted and efficiency is lowered. Furthermore, there is also a problem that the content of conversations overlaps with a plurality of worker pairs and is wasted.

  An object of the present invention is to solve the above technical problem, and by enabling the creation of the same scenario by a plurality of people, there is no need to restrain the operator, and an interactive scenario corpus that can improve work efficiency is provided. The purpose is to provide a creation support system.

  In order to achieve the above object, the present invention is characterized in that a dialogue scenario corpus creation support system that completes a dialogue scenario by repeating utterance input from the standpoint of each uttering speaker has the following configuration: There is.

  (1) A database for storing incomplete scenarios with insufficient number of utterances registered, means for selecting incomplete scenarios for requesting utterance input to workers responsible for utterance input, and presenting selected incomplete scenarios to workers And means for registering an utterance input to the presented incomplete scenario in the database, and selecting, presenting and registering the input utterance is repeated.

  (2) It has means for storing the attribute information of each speaker and means for acquiring the attribute information of each worker, and the means for selection is given priority to an incomplete scenario in which the attribute information of the speaker is similar to that of the worker. To choose.

  (3) The selecting means preferentially selects an incomplete scenario having a history of speech input by the worker.

  (4) It has means for discriminating the topic of the incomplete scenario, and means for calculating the relationship between the topic of the incomplete scenario and the worker identification information, and the means for selecting is the relationship with the worker identification information. Priority is given to the incomplete scenario of a higher topic.

  (5) The means for presenting further includes means for presenting speaker attribute information to a worker.

  (6) A means for allowing workers to report incomplete scenarios including inappropriate utterances is further provided, and incomplete scenarios whose number of reports exceeds the reference value are deleted.

  (7) The words included in each input utterance are clustered, and based on whether the number of utterances reaches a predetermined number, the predetermined number when judging whether or not the dialogue scenario is completed is set as the number of clusters. It was made variable accordingly.

  (8) The database has a tree structure composed of a plurality of dialogue scenarios that relay at least one node utterance from one root utterance to a leaf utterance.

  (9) A means for aggregating nodes with similar utterance contents for each layer is provided.

  (10) There is provided means for prohibiting utterance input including a word / phrase when the frequency of appearance of the word / phrase included in the input utterance exceeds a predetermined value in the same hierarchy as the input utterance.

  (11) The presenting means presents the incomplete scenario in which the utterances up to the nth and the utterances after the n + 2th are registered to the worker, and requests filling-in input of the n + 1th utterance. I did it.

  According to the present invention, the following effects are achieved.

  (1) Since multiple workers can independently create multiple scenarios in parallel, it is possible to construct efficient scenarios and easily construct scenarios that include various topics. .

  (2) Since it repeats inputting each speaker's utterance alternately, it becomes possible to construct a structured scenario and improve the quality of conversation in the dialogue system.

  (3) Multiple workers can input utterances as either one utterer or the other utterer according to the incomplete scenario requested for utterance input. Since an uncompleted scenario is presented, it is possible to create an efficient scenario without waiting for the dialogue partner's utterance input and without being restricted in time.

  (4) Since the virtual attribute information of the speaker is presented to the worker together with the unfinished scenario, the utterance can be input even when different workers input the same speaker's utterance or when the same worker inputs the utterance of the different speaker. Maintain consistency in content.

  (5) Since utterances with similar contents are integrated into one, it is possible to suppress an explosive increase in the number of branches while leaving various utterances.

  (6) Since the scenario completion conditions are variable according to the number of clusters when the words included in each input utterance are clustered, a scenario with a large number of hierarchies can be constructed for a topic rich scenario, The number of hierarchies can be reduced for scenarios with few topics. Therefore, it is possible to construct an efficient scenario with little waste.

  (7) Scenarios containing inappropriate utterances are reported to each worker, and scenarios with a large number of reports are deleted from the database, eliminating the waste of continuing to create inappropriate scenarios and eliminating scenarios. Quality can be maintained.

  (8) Since the incomplete scenario in which the utterances up to the nth and the utterances after the n + 2th are registered is presented to the worker and the + 1st filling speech is requested, the existing utterance is requested. The tree structure can be expanded, and multiple consistent scenarios can be easily added.

1 is a block diagram of a first embodiment of a dialogue scenario corpus creation support system to which the present invention is applied. FIG. It is the figure which showed the example which made the dialog scenario tree structure. This is a sequence flow showing a procedure in which a plurality of workers W1 and W2 complete the scenario by alternately inputting the utterances of the respective speakers. It is the figure which showed the input example (the 1) of the speech by each worker. It is the figure which showed the input example (the 2) of the speech by each worker. It is the figure which showed the input example (the 3) of the speech by each worker. It is a functional block diagram of the dialogue scenario corpus creation assistance system which concerns on 2nd Embodiment of this invention. It is the figure which showed the example of presentation of the virtual attribute information of speaker P2. It is a functional block diagram of the dialogue scenario corpus creation assistance system which concerns on 3rd Embodiment of this invention. It is the figure which showed an example of the speech input screen in 3rd Embodiment. It is a functional block diagram of the dialogue scenario corpus creation assistance system which concerns on 4th Embodiment of this invention. It is the figure which showed the example in which the input content of an utterance overlaps. It is a functional block diagram of the dialogue scenario corpus creation assistance system which concerns on 5th Embodiment of this invention. It is the figure which showed the example in which an input prohibition word is set. It is a functional block diagram (the 1) of the dialogue scenario corpus creation assistance system which concerns on 6th Embodiment of this invention. It is a functional block diagram (the 2) of the dialogue scenario corpus creation assistance system which concerns on 6th Embodiment of this invention. It is a functional block diagram of the dialogue scenario corpus creation assistance system which concerns on 7th Embodiment of this invention. It is the figure which showed a mode that a tree structure was expanded by inputting speech utterly.

  Hereinafter, embodiments of the present invention will be described in detail with reference to the drawings. FIG. 1 is a block diagram showing the configuration of a dialogue scenario corpus creation support system to which the present invention is applied. A scenario creation support device 1, a scenario database (DB) 2 and a worker terminal 3 are connected to each other via a network. Configured. The creation support system of the present invention repeats utterance input from the standpoint of each utterer who interacts to complete a dialogue scenario.

  In the scenario DB 2, many unfinished scenarios in which the number of registered utterances is insufficient are registered. FIG. 2 is a diagram showing a configuration example of a scenario in the present embodiment. The branch from the first utterance corresponding to the root to a plurality of utterances corresponding to the nodes is repeated, and a plurality of final utterances corresponding to the leaves are obtained. A plurality of interactive scenarios are registered in a tree structure, and an utterance is registered in each node.

  In the scenario creation support apparatus 1, the worker assignment unit 101 accepts participation registration of worker Wis responsible for utterance input work by crowdsourcing, for example, and assigns a scenario creation task to each worker Wi registered for participation. Each worker Wi may be assigned a speech input task related to a specific speaker, or may be assigned a speech input task without specifying a speaker.

  The scenario selection unit 102 selects, for each worker Wi, an incomplete scenario for which input of an utterance is requested. In the present embodiment, the scenario selection unit 102 includes an attribute evaluation unit 102a and an utterance evaluation unit 102b. The attribute evaluation unit 102a compares the virtual attribute information of the speakers P1 and P2 registered in advance with the attribute information of each worker Wi, and gives priority to the incomplete scenario whose speaker attribute is similar to the worker Wi. To choose.

  The utterance evaluation unit 102b compares the topic of the incomplete scenario with the attribute information of each worker Wi, and preferentially selects the incomplete scenario of the topic that is more relevant to the attribute information of the worker Wi. The topic of an incomplete scenario can be estimated by causing a classifier to learn a document with a known topic as machine learning learning data and using the current scenario as a topic discrimination target.

  Or you may make it preferentially select the incomplete scenario with the log | history which the worker Wi input utterance in the past. For example, if the priority is set higher for an incomplete scenario where the number of utterances input by the worker Wi in the past is higher, the worker can work on the conversation consistently, so that it becomes possible to reduce the contradiction and breakdown of the dialogue.

  The scenario presentation unit 103 presents the selected incomplete scenario to the worker terminal 3 of each worker Wi and requests input of an utterance. The scenario registration unit 104 additionally registers the utterance input from the worker terminal 3 by the worker to the corresponding node of the corresponding incomplete scenario on the scenario DB 2.

  In the present embodiment, the scenario selection unit 102 selects an incomplete scenario registered up to the nth utterance from the scenario DB 2, and the scenario presentation unit 103 presents each worker Wi to the n + 1th to the incomplete scenario. Ask for utterance input. Each worker Wi operates the worker terminal 3 to input the (n + 1) th utterance in the incomplete scenario. The scenario registration unit 104 additionally registers the input n + 1-th utterance to the corresponding node of the incomplete scenario. Such selection of an incomplete scenario, presentation to each worker Wi, utterance input and utterance registration by each worker Wi are repeated until the input of the final utterance is completed and the dialogue scenario is completed.

  FIG. 3 is a sequence flow showing a procedure for completing the scenario by alternately inputting the utterances of the respective speakers P1 and P2 by a plurality of worker Wis, and FIG. 4 shows an example of utterance input by each worker Wi. It is a figure. Here, a case where two workers W1 and W2 are requested to input an utterance will be described as an example.

  At time t1, for example, a scenario creation worker is recruited by crowdsourcing, and a profile or the like is registered in the support apparatus 1 as attribute information of the applicant. At time t2, the scenario selection unit 102 of the scenario creation support apparatus 1 selects and extracts from the scenario DB 2 unfinished scenarios requested to be created by the workers W1 and W2. At time t3, the scenario presenting unit 103 presents the incomplete scenario in which only the theme and the first utterance are registered to each of the workers W1 and W2, and requests the input of the utterance.

  This embodiment assumes support for creating a dialogue scenario in which virtual conversations between two speakers P1 and P2 are repeated alternately for each theme such as “How to spend holidays” and “Eating and drinking”. In the initial state of the scenario, only the theme and the first utterance of the one-side speaker P1 for the theme are registered. In this embodiment, as shown in FIG. 4 (a), only the theme “how to spend holidays” and the first utterance of the speaker P1 for the theme “what are you doing on holidays?” Are registered. The completed incomplete scenario is presented to each of workers W1 and W2 and displayed on the display of each worker terminal 3.

  At times t4 and t5, the workers W1 and W2 input the next utterance for the immediately preceding utterance of the presented incomplete scenario from the standpoint of the other speaker P2. Here, since the incomplete scenario has been registered up to the first utterance of the speaker P1, as shown in FIG. 4 (b), each worker W1, W2 has the key in the utterance input field 31 in the position of the speaker P2. The second utterance is input by appropriate means such as input or voice input.

  This embodiment shows an example in which “I play tennis in a circle” is input. When the workers W1 and W2 click the send button 32 at time t6 and t7 after inputting the utterance, as shown in FIG. 4 (c), the second utterance input by each worker W1 and W2 is the scenario support device 1. Is registered in the scenario DB 2 as the second utterance of each incomplete scenario.

  At time t8, an incomplete scenario requested to worker W1 and an incomplete scenario requested to worker W2 are selected and extracted from scenario DB 2 again. At time t9, the extracted unfinished scenarios are presented to the workers W1 and W2, and a scenario creation is requested.

  At times t10 and t11, workers W1 and W2 input next utterances to the presented incomplete scenarios, respectively. Here, as shown in FIG. 5 (a), since each incomplete scenario has been registered up to the second utterance by the speaker P2, as shown in FIG. 5 (b), each worker W1, W2 Enter the third utterance "How many years have you continued?" From the viewpoint of speaker P1. At times t12 and t13, as shown in FIG. 5 (c), the third utterance input by the workers W1 and W2 is registered in the scenario DB2.

  At time t14, the incomplete scenario requested to worker W1 and the incomplete scenario requested to worker W2 are selected and extracted from the scenario DB 2 again. At time t15, the extracted incomplete scenario is presented to each of the workers W1 and W2, and a scenario creation is requested.

  At times t16 and t17, workers W1 and W2 input next utterances to the presented incomplete scenarios, respectively. Here, as shown in FIG. 6 (a), since each incomplete scenario has been registered up to the third utterance by the speaker P1, each worker W1, W2 is shown in FIG. 6 (b). Enter the 4th utterance "From university, 10 years" as the speaker P2. At times t18 and t19, as shown in FIG. 6C, the fourth utterance input by the workers W1 and W2 is registered in the scenario DB2. Similarly, the workers W1 and W2 repeatedly input the (n + 1) th utterance if the nth utterance is registered in the presented incomplete scenario.

  According to the present embodiment, each worker W can input an utterance as either one of the one-side Japanese speaker P1 and the other-side speaker P2 according to the incomplete scenario requested to input the utterance, and one utterance input Since the next unfinished scenario is presented after finishing the process, it is possible to create an efficient scenario without waiting for the input of the other party's utterance and without being restricted in time.

  FIG. 7 is a functional block diagram showing the configuration of the dialogue scenario corpus creation support system according to the second embodiment of the present invention. The same reference numerals as those described above represent the same or equivalent parts, and the description thereof is omitted. To do. In the present embodiment, the scenario presenting unit 103 also presents the persona storage unit 103a for storing the persona as virtual attribute information of each speaker, and the persona of the speaker when presenting the incomplete scenario to the worker Wi. The persona presenting unit 103b is characterized.

  When the scenario presenting unit 103 presents the incomplete scenario to the worker Wi and requests the input of the utterance, as shown in FIG. 8, the scenario presenting unit 103 presents the virtual persona of the speaker together with the worker Wi, Is characterized in that it can input utterances considering the persona. In the present embodiment, virtual personas of the speakers P1 and P2 are registered in advance, and the name, nickname, age, gender, date of birth, etc. are presented.

  According to this embodiment, since the speaker's virtual persona is presented together with the incomplete scenario, when different worker Wis input the same speaker's speech, or the same worker Wi inputs the same speaker's speech Even in this case, the consistency of the utterance contents can be maintained.

  FIG. 9 is a functional block diagram showing the configuration of the dialogue scenario corpus creation support system according to the third embodiment of the present invention. The same reference numerals as those described above represent the same or equivalent parts. In the present embodiment, the scenario presenting unit 103 includes an inappropriate report requesting unit 103c that causes the worker Wi to report an incomplete scenario including an inappropriate utterance, and the scenario registration unit 104 determines that the inappropriate report count is a reference value. It is characterized in that a scenario deletion unit 104a for deleting unfinished scenarios exceeding the above from the scenario DB 2 is provided.

  When the uncompleted scenario is presented to the worker Wi and the input of the utterance is requested, the inappropriate report requesting unit 103c, in addition to the utterance input field 31 and the send button 32, as shown in FIG. An inappropriate report button 33 is displayed for reporting to the system side that an inappropriate expression or content is included in the utterance.

  In the illustrated example, the "Hello Nice to meet you." The last speech, because it is unnatural from the history of the utterances of up to it, the worker Wi can click the incorrect report button 33 without speech input, it is presented The scenario creation support apparatus 1 is notified that an inappropriate utterance is included in the incomplete scenario. In the scenario registration unit 104, the scenario deletion unit 104a counts the number of inappropriate reports for each incomplete scenario, and deletes the uncompleted scenarios with inappropriate reports exceeding a predetermined number from the scenario DB2.

  According to the present embodiment, since an incomplete scenario including an inappropriate utterance can be eliminated, waste that continues to create an inappropriate scenario is eliminated, and the quality of the scenario can be maintained high.

  FIG. 11 is a functional block diagram showing a configuration of a dialogue scenario corpus creation support system according to the fourth exemplary embodiment of the present invention. The same reference numerals as those described above represent the same or equivalent parts. In this embodiment, for a plurality of unfinished scenarios with the same theme, clustering based on utterance contents is performed for each layer, and scenarios are aggregated by replacing utterances classified into the same cluster with one representative utterance. This is characterized in that a scenario aggregation unit 105 is provided.

  If the dialogue scenario has a tree structure, the scenario branches as many as the number of child nodes, so that the scenario pattern increases explosively as the dialogue proceeds. For example, if a tree structure is constructed by inputting 10 utterances for one utterance, the 10th stage becomes 1 billion ways, which requires enormous input. The scenario aggregator 105 groups utterances with similar contents using a clustering method such as k-means or X-means, and if the k-means method is used, the utterance close to the center of the cluster, X-means is optimal. Continue to speak only for utterances.

  In the example of FIG. 12, “I like beer”, “I often drink beer in the summer”, “I like local beer and it ’s one of the fun of traveling” are classified into the same cluster, “I like beer.” Is selected as the representative utterance. Therefore, all utterances belonging to the same cluster as “I like beer” are collected in the node “I like beer”. According to the present embodiment, an explosive increase in the number of branches can be suppressed while leaving various utterances.

  FIG. 13 is a functional block diagram showing a configuration of a dialogue scenario corpus creation support system according to the fifth exemplary embodiment of the present invention. The same reference numerals as those described above represent the same or equivalent parts. The present embodiment is characterized in that the scenario registration unit 104 includes an input prohibited word processing unit 104b.

  The input-prohibited word processing unit 104b calculates the frequency of appearance of a phrase within an utterance for an incomplete scenario with the same title, and sets frequent words that appear beyond a predetermined frequency as input-prohibited words. If an input prohibited word is included in the input utterance, the worker Wi is notified that the input prohibited word is included, and is prompted to input another utterance.

  In the example shown in FIG. 14, in response to the previous utterance “What is your favorite liquor?”, Another worker Wi has already “I like beer.” “Inventors often drink beer.” I like it and it is one of the fun of traveling. ”“ Beer ”is set as a prohibited word because there are many beer related topics such as“ I drink a little beer at home ”.

  As a result, when an utterance including “beer” is input, the input prohibition word processing unit 104b grays out the transmission button 32 and includes a message that “beer” is an input prohibition word and an “input prohibition word”. Presents a message saying “I am present” and prompts other utterances to be input. According to the present embodiment, an explosive increase in the number of branches can be suppressed while leaving various utterances.

  FIGS. 15 and 16 are functional block diagrams showing the configuration of a dialogue scenario corpus creation support system according to the sixth embodiment of the present invention. The same reference numerals as those described above represent the same or equivalent parts. In this embodiment, when evaluating a dialogue scenario in which the number of registered utterances reaches a predetermined number as completion, the predetermined value is dynamically changed based on the appearance frequency of words included in the utterance and the time required for utterance input. A feature is that a completion condition setting unit 106 is provided.

  In the example of FIG. 15, the completion condition setting unit 106 includes a clustering unit 106 a, which extracts words / phrases included in each utterance for each incomplete scenario and performs clustering. If the number of clusters is large, it is determined that there is a high possibility that the topic will be expanded, and the predetermined number is set to a large value. On the other hand, if the number of clusters is small, the appearance frequency of similar words is high, and the topic may be expanded. Therefore, the predetermined number is set to a small value.

  In the example of FIG. 16, the completion condition setting unit 106 includes an input time counting unit 106b, and measures the time required for speech input after the input prohibited word is presented to the worker Wi, for example. And when utterances are input in a short time, it is determined that the topic is abundant and the predetermined number is set to a large value, while when utterance input is not performed in a short time, it is determined that the topic is insufficient and the predetermined number Set to a small value.

  According to the present embodiment, a scenario with a large number of layers can be constructed for a scenario with abundant topics, while a number of layers can be reduced for a scenario with few topics, so that an efficient scenario can be constructed according to the topic. This makes it possible to improve the efficiency of speech input.

  FIG. 17 is a functional block diagram showing a configuration of a dialogue scenario corpus creation support system according to the seventh exemplary embodiment of the present invention. The same reference numerals as those described above represent the same or equivalent parts. The present embodiment is characterized in that the scenario creation support apparatus 1 includes a scenario expansion unit 107.

  In each of the above embodiments, when requesting an input of an utterance by presenting an incomplete scenario to worker Wi, an utterance history up to that point is presented, and an input of the next utterance is requested in relation to the previous utterance Was. On the other hand, in this embodiment, the scenario expansion unit 107 presents the incomplete scenario in which the utterances up to the nth and the utterances after the n + 2 are registered to the worker Wi, and fills the n + 1th embedment Requests utterance input.

  In the input example shown in FIG. 18 (a), “What are you doing during the holidays” is registered as the first utterance, “How many years have you been continuing” as the third utterance, and “University” as the fourth utterance? Because it is almost 10 years from now, "unfinished scenarios each registered" are presented to the worker. In this case, it is possible to input utterances such as “golf”, “grass baseball”, and “drive” as the second utterance. According to this embodiment, it is possible to expand an existing tree structure as shown in FIG. 18 (b) by inputting a single utterance in a filling manner, and it is possible to easily create a plurality of consistent scenarios. Can be added.

  In each of the above embodiments, the case where the dialogue scenario has a tree structure has been described as an example. However, the present invention is not limited to this, and each dialogue scenario may be independent of each other.

  DESCRIPTION OF SYMBOLS 1 ... Scenario creation assistance apparatus, 2 ... Scenario database, 3 ... Worker terminal, 31 ... Speech input column, 32 ... Send button, 33 ... Inappropriate report button, 101 ... Worker allocation part, 102 ... Scenario selection part, 102a ... Attribute evaluation unit, 102b ... utterance evaluation unit, 103 ... scenario presentation unit, 103a ... persona storage unit, 103b ... persona presentation unit, 103c ... inappropriate report request unit, 104 ... scenario registration unit, 104a ... scenario deletion unit, 104b ... Input prohibition word processing unit, 105 ... scenario aggregation unit, 106 ... completion condition setting unit, 106a ... clustering unit, 106b ... input time counting unit, 107 ... scenario expansion unit

Claims (11)

In the dialogue scenario corpus creation support system that completes the dialogue scenario by repeating the utterance input from the standpoint of each talking speaker,
A database that stores incomplete scenarios with insufficient utterance registrations,
Means for selecting an incomplete scenario for requesting utterance input to a worker who is in charge of utterance input work;
Means for presenting the selected incomplete scenario to a worker;
Means for registering, in the database, utterances input for the presented incomplete scenario;
A dialogue scenario corpus creation support system, characterized in that selection of the incomplete scenario, presentation, and registration of an inputted utterance are repeated. Means for storing attribute information of each speaker;
Means for obtaining attribute information of each worker,
2. The dialogue scenario corpus creation support system according to claim 1, wherein the selecting means preferentially selects an incomplete scenario whose speaker attribute information is similar to that of a worker.   2. The dialogue scenario corpus creation support system according to claim 1, wherein the selecting means preferentially selects an incomplete scenario having an utterance input history by the worker. A means of determining the topic of an incomplete scenario;
A means for calculating the relationship between the topic of the incomplete scenario and the identification information of the worker,
2. The dialogue scenario corpus creation support system according to claim 1, wherein the selecting means preferentially selects an incomplete scenario of a topic having a higher relevance with worker identification information.   5. The dialogue scenario corpus creation support system according to claim 1, wherein the presenting means further comprises means for presenting speaker attribute information to a worker. Further comprising means for allowing workers to report incomplete scenarios containing inappropriate utterances,
6. The interactive scenario corpus creation support system according to claim 1, wherein an incomplete scenario whose number of reports exceeds a reference value is deleted.   The words and phrases included in each input utterance are clustered, and the predetermined number when judging whether or not the dialogue scenario is completed based on whether or not the number of utterances reaches a predetermined number is variable according to the number of clusters. The dialogue scenario corpus creation support system according to any one of claims 1 to 6, wherein   The database according to any one of claims 1 to 7, wherein the database has a tree structure including a plurality of dialogue scenarios that relay at least one node utterance from one root utterance to reach a leaf utterance. Dialog scenario corpus creation support system.   9. The dialogue scenario corpus creation support system according to claim 8, further comprising means for aggregating nodes having similar utterance contents for each hierarchy.   9. The dialogue scenario according to claim 8, further comprising means for prohibiting an utterance input including the word / phrase when an appearance frequency of the word / phrase included in the input utterance exceeds a predetermined value in the same hierarchy as the input utterance. Corpus creation support system.   The presenting means presents an incomplete scenario in which up to nth utterances and n + 2 and subsequent utterances are registered to a worker, and requests n + 1th utterance input. 8. A dialogue scenario corpus creation support system according to 8.

Images