JP2019191646A - Registered word management device, voice interactive system, registered word management method and program - Google Patents

Abstract

To provide a registered word management device, a voice interactive system, a registered word management method and a program which can manage integratedly an automatically generated different name and a different name inputted by a system manager, including whether or not those registration is validated.SOLUTION: A registered word management device comprises: a different name generation unit which automatically generates a different name which is a phrase obtained by replacing an official name registered in a phrase database, and records the generated different name on the phrase database as a registration candidate; and an UI control unit which displays a registered word management screen which presents the different name recorded on the phrase database to a user, in a manner which can identify whether or not the different name is automatically generated or inputted by the user, and in the manner which can change whether or not to validate the registration of the different name, and accepts a user operation to the registered word management screen to reflect contents of the operation to the phrase database.SELECTED DRAWING: Figure 1

Description

  Embodiments described herein relate generally to a registered word management device, a spoken dialogue system, a registered word management method, and a program.

  The spoken dialogue system acquires information necessary to achieve a predetermined purpose through dialogue with an end user. For example, in a slot-fill type spoken dialogue system, referring to a phrase database in which various phrases are registered, the phrases that are applicable as item information necessary for the execution of the target application are extracted from the end user's utterance, and each item is extracted. Fill in the slots. At this time, if not only the official name of the target pointed by the phrase but also an alias (common name, abbreviation, etc.) in which the formal name is reworded is registered in the phrase database in association with the official name, the end user can use the alias. Even if you utter utterances, you can replace the alias with the official name and fill it in the slot, realizing a smooth conversation.

  A phrase database used in a spoken dialogue system is generally configured such that a system administrator can customize a general-purpose database using a dedicated tool. Using this dedicated tool, the system administrator can associate and register the official name of the word and its alias that are required according to the application in the word database. As for the alias, a technique for automatically generating an abbreviation from an official name according to a predetermined rule has been proposed.

  However, the alias registered in the phrase database is not always effective in improving the performance of the system. For example, if the registered alias reading is similar to the reading of another official name, when the system is actually run, this alias will be confused with another official name and the performance of the system will decrease. Sometimes. In addition, the automatically generated alias may be a thing that the end user hardly utters in practice. From such a point of view, there is a demand for the construction of a mechanism that can manage automatically generated aliases and aliases entered by the system administrator, including whether or not to enable registration thereof.

Japanese Patent Laying-Open No. 2005-181442 JP 2008-77335 A

  The problem to be solved by the present invention is that a registered word management device and a spoken dialogue system capable of managing automatically generated aliases and aliases inputted by a system administrator, including whether or not to enable registration thereof, in an integrated manner It is to provide a registered word management method and program.

  The registered word management apparatus according to the embodiment is a registered word management apparatus that manages registered words in a phrase database used in the spoken dialogue system, and includes an alias generation unit and a UI control unit. The alias generation unit automatically generates an alias that is a phrase obtained by rephrasing the formal name registered in the phrase database, and records the generated alias as a registration candidate in the phrase database. The UI control unit can identify whether the alias recorded in the phrase database is automatically generated or input by the user, and enables registration of the alias. A registered word management screen that is presented to the user in a switchable manner is displayed, a user operation on the registered word management screen is accepted, and the operation content is reflected in the phrase database.

FIG. 1 is a block diagram illustrating an example of a functional configuration of the voice interaction system according to the embodiment. FIG. 2 is a diagram illustrating a configuration example of the phrase database. FIG. 3 is a flowchart showing a processing procedure of the generation function by log analysis of the alias generation unit. FIG. 4 is a flowchart showing the processing procedure of the generation function based on the rules of the alias generation unit. FIG. 5 is a flowchart showing a processing procedure of the generation function by language analysis of the alias generation unit. FIG. 6 is a diagram illustrating an example of a registered word management screen displayed by the UI control unit. FIG. 7 is a flowchart illustrating a processing procedure of the UI control unit. FIG. 8 is a block diagram illustrating a hardware configuration example of the voice interaction system according to the embodiment.

  Hereinafter, a registered word management device, a spoken dialogue system, a registered word management method, and a program according to embodiments will be described in detail with reference to the drawings. The embodiment shown below is an application example to a slot-fill type speech dialogue system. However, the applicable speech dialogue system is not limited to this, and a phrase database (hereinafter referred to as a phrase database) in which phrases necessary for dialogue processing are registered. The present invention can be applied to any type of speech dialogue system that performs dialogue processing using a database (abbreviated as “DB”). Further, in the embodiment described below, the dialogue with the end user is performed by voice. However, the response from the system to the end user may be performed by combining voice output and text display. It may be performed only by display.

  FIG. 1 is a block diagram illustrating an example of a functional configuration of a voice interaction system 1 according to the present embodiment. As shown in FIG. 1, the voice dialogue system 1 includes a dialogue processing unit 10, a dialogue knowledge DB 20, a dialogue log DB 30, and a phrase DB 40. The spoken dialogue system 1 also includes an alias generation unit 110 and a UI (user interface) control unit 120 as functions for managing phrases (registered words) registered in the phrase DB. The alias generation unit 110 and the UI control unit 120 constitute the registered word management apparatus 100 according to the present embodiment. The registered word management device 100 is incorporated in the spoken dialogue system 1 as a dedicated management tool used by a system administrator who manages the spoken dialogue system 1, for example.

  The dialogue processing unit 10 includes various engines that execute various tasks required for voice dialogue such as a speech recognition engine, an intention understanding engine, a dialogue control engine, and a voice synthesis engine, and appropriately stores the dialogue knowledge DB 20 and the phrase DB 40. While using it, voice interaction with end users is performed. Then, the dialogue processing unit 10 acquires information on items necessary for executing the target application through dialogue with the end user. For example, if the target application is “negotiation record”, the information of each item such as “visit date / time”, “visit destination”, “next scheduled date”, “special notes” is information to be acquired. In addition, if the target application is “accommodation reservation”, information on each item such as “estimated check-in date / time”, “scheduled check-out date / time”, “number of guests”, “accommodation plan”, etc. Become.

  Information acquired by the dialog processing unit 10 through the dialog with the end user is filled in the slot 51 of the corresponding item in the slot table 50. The slot table 50 has a slot 51 in which information (item value) for each item necessary for executing the target application is filled. When each slot 51 of the slot table 50 is filled, a target application is executed based on the slot table 50. For example, if the target application is “negotiation record”, the negotiation record is registered in an external system, for example, and if the target application is “accommodation reservation”, an accommodation reservation query is executed.

  The slot table 50 is created based on a template corresponding to an application in response to the identification of the target application through interaction with the end user. The template defines an item number (slot number), an item name, and a type of the slot 51 of the item for each item necessary for executing the application. The types of the slot 51 include, for example, a date / time type in which date / time information is filled, an option type in which a word / phrase selected from predetermined options is filled, and a free description type in which a natural sentence is filled. For items of which the type of the slot 51 is an option type, the options of the words and phrases to be filled in the slot 51 are further defined by the template. For example, the template is appropriately created by the system administrator, and is stored in the dialogue knowledge DB 20 as one piece of dialogue knowledge.

  The dialogue processing unit 10 controls the dialogue with the end user based on, for example, a dialogue scenario stored in the dialogue knowledge DB 20 as one piece of dialogue knowledge. Then, information to be filled in each slot 51 of the slot table 50 is acquired from the recognition result text obtained by voice recognition of the utterance of the end user. For example, for the option type slot 51, a phrase that matches one of the options defined in the template is extracted from the recognition result text. For example, the recognition score of each word included in the recognition result text, the order of appearance of each word and the relationship with the item name in the slot 51, the reading of the word included in the recognition slot and the reading of the word defined as each option The word / phrase to be filled in the option type slot 51 can be extracted from the recognition result text using the degree of coincidence as a clue.

  A phrase defined by a template as a phrase option to be filled in the option type slot 51 is registered as an official name in the phrase DB 40. In the phrase DB 40, various words and phrases handled by the voice interaction system 1 (in the present embodiment, these are collectively referred to as “phrases”) are registered. The phrase DB 40 can register, in addition to the formal name of the target indicated by each phrase, an alias that is a phrase that is a paraphrase of the formal name in association with the formal name. The alias registered in the phrase DB 40 includes an alias automatically generated by the alias generation unit 110 described later and an alias input by the system administrator (user) through a registered word management screen described later. If an alias of a phrase (official name) defined by the template as an option described above is registered in the phrase DB 40, the alias is extracted from the speech recognition text even if the end user utters the alias. It can be replaced with an official name and filled in an option type slot 51.

  Note that the phrase DB 40 can be used not only for extracting a phrase that fills the slot 51 from a recognition result text obtained by voice recognition of an end user's utterance, but also as a dictionary for voice recognition of an end user's utterance.

  The dialogue processing unit 10 utters (recognizes) an end user's utterance (recognition of information of each item necessary for execution of a target application (a word / phrase embedded in each slot 51 of the slot table 50) through a series of dialogues with the end user. Result text). At this time, if the information of a plurality of items is included in one utterance of the end user, the dialogue processing unit 10 once fills the information filled in the plurality of slots 51 of the slot table 50 from the one utterance of the end user. It is possible to win. If there is an unconfirmed slot 51, the dialogue processing unit 10 makes an inquiry about information to be filled in the slot 51 based on the dialogue scenario, and sequentially acquires information to be filled in each slot 51 of the slot table 50. I will do it. The dialogue history with the end user by the dialogue processing unit 10 is recorded in the dialogue log DB 30 as a dialogue log.

  In addition to the above-described dialog scenarios and templates, the dialog knowledge DB 20 stores various dialog knowledge useful for the dialog processing unit 10 to perform appropriate dialog processing with the end user.

  As described above, the dialogue log DB 30 stores a dialogue log that records a history of dialogue between the dialogue processing unit 10 and the end user. The dialogue log stored in the dialogue log DB 30 is used when an alias generation unit 110 described later automatically generates an alias. Further, the dialog log stored in the dialog log DB 30 can be used to update the dialog knowledge stored in the dialog knowledge DB 20. For example, an event in which the dialogue is not smoothly performed is extracted from the dialogue log, a factor that inhibits the smooth dialogue is specified, and the dialogue knowledge is updated so that the smooth dialogue can be realized. This process can be performed not only by the operation of a system administrator or the like, but also automatically using, for example, a neural network technique.

  As described above, the phrase DB 40 stores various phrases handled by the spoken dialogue system 1. FIG. 2 is a diagram for explaining a configuration example of the phrase DB 40. The phrase DB 40 includes an official name master 41, a type master 42, and an alias table 43, for example, as shown in FIG. 2.

  The official name master 41 is a table for managing the official names registered in the phrase DB 40. For each official name registered in the phrase DB 40, a unique ID (official name ID) assigned to each official name and The notation and reading of the official name are stored in association with each other. Formal names registered in the phrase DB 40 include those registered as general-purpose words before the system introduction, and those additionally registered by, for example, the operation of the system administrator during or after the system introduction. For example, the additional registration of the formal name can be performed by using a dedicated tool for creating a template at the time of creating the template, or can be performed through an operation on a registered word management screen described later. For example, a formal name is additionally registered in the phrase DB 40 by adding a record corresponding to a new formal name to the formal name master 41 in accordance with an operation by a system administrator or the like.

  The type master 42 is a table for managing the types of aliases (including registration candidate aliases) registered in the phrase DB 40. In this embodiment, as an example, an alias to be registered in the phrase DB 40 is input by the system administrator on an alias automatically generated by the alias generation unit 110 described later (an alias for “system proposal”) and on a registered word management screen described later. Categorized into two types (named “user input”), and further categorized “system proposal” into three types: automatic generation by rules, automatic generation by dialog log analysis, and automatic generation by language analysis To do. The type master 42 stores unique IDs (type IDs) assigned to these various types and their types in association with each other.

  The alias table 43 is a table for managing aliases registered in the phrase DB 40 (including registration candidate aliases), and assigned to each alias for each alias recorded in the phrase DB 40 as registration candidate aliases. A unique ID, a formal name ID identifying the formal name corresponding to the alias, a notation and reading of the alias, a type ID identifying the type of the alias, a score representing the importance of the alias, and the A valid flag indicating whether registration of an alias is valid or invalid is stored in association with each other. The score is calculated by an alias generation unit 110 described later for an alias of “system proposal” (an alias whose type ID is 1 to 3). For example, the valid flag is set to “false” indicating that the initial value is invalid for the alias of “system proposal”, and “true” indicating that the initial value is valid for the alias of “user input”. Switching is possible according to the operation by the above system administrator. When a new alias is automatically generated by the alias generation unit 110 described later, or when the system administrator performs an operation of adding a new alias on the registered word management screen described later, the new alias is added to the alias table 43. A record corresponding to the alias is added, and the new alias is recorded in the phrase DB 80 as a registration candidate alias.

  In the spoken dialogue system 1 according to the present embodiment, registered words (formal names and aliases) in the phrase DB 40 are managed by the registered word management apparatus 100. The registered word management apparatus 100 updates the formal name master 41 and the alias table 43 of the phrase DB 40 when the system administrator performs an operation of inputting a formal name or an alias, or when a registration candidate alias is automatically generated. The registered words in the phrase DB 40 are managed. The registered word management apparatus 100 includes the alias generation unit 110 and the UI control unit 120 as described above.

  The alias generation unit 110 automatically generates aliases of registered formal names and records them in the phrase DB 40 as registration candidate aliases. The alias generation unit 110 includes a log analysis generation function 111, a rule generation function 112, a language according to three types of “system proposal” among the types of aliases defined in the type master 42. And a generation function 113 by analysis.

  The log analysis generation function 111 is a function for automatically generating aliases for registration candidates by analyzing a dialog log stored in the dialog log DB 30. The log analysis generation function 111 is, for example, a part where the end user speaks a plurality of times to fill one slot 51 by analyzing a dialogue log, that is, a part where the end user speaks repeatedly while changing the wording. Is identified. Then, a phrase that has not been filled in the slot 51 (a phrase that has failed in the slot fill) is extracted as an alias for the registration candidate and recorded in the phrase DB 40.

  More specifically, the log analysis generation function 111 adds a new record to the alias table 43 of the phrase DB 40, and automatically generates an alias ID and a formal name corresponding to the alias (filled in the slot 51 described above). ), The notation and reading of the alias, and the type ID (= 3) indicating that the type of the alias is automatically generated by dialog log analysis are stored in association with each other. The generation function 111 by log analysis calculates a score representing the importance of the alias extracted from the dialogue log, and stores it together with the alias table 43. The score can be calculated based on the analysis result of the dialog log, for example, the higher the number of times that the registration candidate alias appears in the dialog log (the number of times the slot fill has failed). For example, “false” representing invalidity is set as an initial value for the valid flag, but “true” representing validity may be set as an initial value.

  FIG. 3 is a flowchart showing a processing procedure of the generation function 111 by log analysis of the alias generation unit 110. The generation function 111 based on log analysis, for example, executes a series of processes shown in the flowchart of FIG. 3 at a predetermined timing such as a time zone in which the processing load of the voice interaction system 1 becomes low, and automatically generates registration candidate aliases. .

  That is, the generation function 111 based on log analysis obtains a dialogue log from the dialogue log DB 30 (step S101), and an end user's utterance was made multiple times to fill the same slot 51 in the obtained dialogue log. A place is searched (step S102). If there is no place where a plurality of utterances have been made (step S103: No), the processing is terminated as it is. On the other hand, if there is a place where a plurality of utterances have been made (step S103: Yes), the word / phrase that failed in the slot fill at that place is recorded in the word / phrase DB 40 as an alias for the registration candidate (step S104), and the process is terminated. To do. At this time, when a plurality of locations where a plurality of utterances have been performed are found in the processing of step S102, the generation function 111 by log analysis performs the processing of step S104 at each location and ends the processing.

  The rule generation function 112 is a function for automatically generating an alias for a registration candidate by applying a predetermined alias generation rule to a formal name registered in the phrase DB 40. The alias generation rule is, for example, a phrase conversion rule that deletes a word registered as an official name if it ends with “stock,” or a word registered as an official name is a product name. For example, it is composed of a rule group including a phrase conversion rule such as replacement with a product number. The rule generation function 112 sequentially applies these conversion rules to the formal names that are the targets of alias generation, and when the aliases are generated, records them in the phrase DB 40 as registration candidate aliases.

  Specifically, the rule generation function 112 adds a new record to the alias table 43 of the phrase DB 40, automatically generates an alias ID, a formal name ID corresponding to the alias, and the alias name. The notation and reading are stored in association with the type ID (= 1) indicating that the alias type is automatically generated by the rule. Also, the rule generation function 112 calculates a score representing the importance of the alias generated by applying the alias generation rule, and stores it together with the alias table 43. The score can be determined in advance for each rule applied to generate the alias, for example. For example, “false” representing invalidity is set as an initial value for the valid flag, but “true” representing validity may be set as an initial value.

  FIG. 4 is a flowchart illustrating a processing procedure of the generation function 112 based on the rules of the alias generation unit 110. The rule generation function 112 automatically generates an alias by applying an alias generation rule in the past, such as a formal name newly registered by the operation of the system administrator, among the formal names registered in the phrase DB 40. For example, a series of processes shown in the flowchart of FIG. 4 is executed at a predetermined timing such as a time period when the processing load of the voice interaction system 1 is reduced, and registration candidate aliases are automatically generated.

  In other words, the rule generation function 112 acquires one of the official names registered in the phrase DB 40, which has not been subjected to automatic generation of an alias that has been applied with an alias generation rule in the past (step S201). Then, an alias generation process is performed by applying one of the rule groups constituting the alias generation rule to the official name acquired in step S201 (step S202). When the alias is generated (step S203: Yes), it is recorded in the phrase DB 40 as an alias for the registration candidate (step S204). The rule generation function 112 repeats the process of step S202 until all the rules constituting the alias generation rule are applied to the official name acquired in step S201, and all the rules constituting the alias generation rule are displayed. Once applied (step S205), the process is terminated. When there are a plurality of formal names for which aliases are to be generated, a series of processes shown in the flowchart of FIG. 4 may be repeatedly executed for each formal name.

  The generation function 113 by language analysis is a function for automatically generating an alias for a registration candidate by performing language analysis on the official name registered in the phrase DB 40. The generation function 113 by language analysis, for example, decomposes a formal name that is a target of alias generation into morphemes, removes a morpheme of a specific part of speech and generates an alias, or a formal name that is a target of alias generation. In the case of a phrase option filled in the above-described option type slot 51, an alias can be automatically generated by removing a phrase common to other options and using the remaining phrase as an alias. Then, when the alias is generated, the generation function 113 by language analysis records it in the phrase DB 40 as the registration candidate alias.

  Specifically, the generation function 113 by language analysis adds a new record to the alias table 43 of the phrase DB 40, and automatically generates an alias ID, an official name ID corresponding to the alias, and the alias. Is stored in association with a type ID (= 3) indicating that the alias type is automatically generated by language analysis. Further, the generation function 113 by language analysis calculates a score representing the importance of the alias generated by language analysis, and stores it together with the alias table 43. The score can be determined in advance for each method of language analysis, for example. For example, “false” representing invalidity is set as an initial value for the valid flag, but “true” representing validity may be set as an initial value.

  FIG. 5 is a flowchart showing a processing procedure of the generation function 113 by language analysis of the alias generation unit 110. The generation function 113 by language analysis is an official name that has not been automatically generated by alias analysis in the past, such as an official name newly registered by the operation of the system administrator, among the official names registered in the phrase DB 40. For a name, for example, a series of processing shown in the flowchart of FIG. 5 is executed at a predetermined timing such as a time zone when the processing load of the voice interactive system 1 is reduced, and registration candidate aliases are automatically generated.

  That is, the generation function 113 by language analysis acquires one official name that has not been automatically generated by the language analysis in the past from among the official names registered in the phrase DB 40 (step S301). Then, alias generation processing by language analysis is performed (step S302). If no alias is generated (step S303: No), the process ends. On the other hand, if an alias is generated by language analysis (step S303: Yes), it is recorded in the phrase DB 40 as an alias for the registration candidate (step S304), and the process is terminated. When there are a plurality of formal names for which aliases are to be generated, a series of processes shown in the flowchart of FIG. 5 may be repeatedly executed for each formal name.

  For example, the UI control unit 120 manages formal names and aliases (including aliases of registration candidates) registered in the phrase DB 40 in response to a request from a system administrator who has accessed the voice interactive system 1 according to the present embodiment. A registered word management screen, which is a user interface, is displayed on a display device of a terminal used by the system administrator. Then, the system administrator's operation (user operation) on the registered word management screen is accepted, and the content of the operation is reflected in the phrase DB 40.

  An example of a registered word management screen displayed by the UI control unit 120 is shown in FIG. The registered word management screen 200 is created based on the formal name master 41 and the alias table 43 of the phrase DB 40 and is displayed on a display device of a terminal used by the system administrator. For example, as shown in FIG. 6, the registered word management screen 200 includes a formal name presentation area 210 that presents a formal name registered in the phrase DB 40, an alias presentation area 220 that presents an alias corresponding to the formal name, A type presentation area 230 that presents the type of the alias, a score presentation area 240 that presents the score of the alias, a check box 250 that presents the valid / invalid state of the alias in a switchable manner, and the registered word management screen 200 And an “update” button 260 operated to reflect the above user operation in the phrase DB 40.

  The formal name presentation area 210 can present a plurality of selected formal names among the formal names registered in the phrase DB 40. For example, the formal name to be presented may be selected according to the operation of the system administrator or automatically by the system. When selecting a formal name to be presented according to the operation of the system administrator, for example, there is a method of displaying a list of formal names on another screen and presenting the formal name selected by the system administrator from among them. . When the system automatically selects a formal name, for example, a method of preferentially presenting a formal name in which a new alias name is automatically generated, or options of the option type slot 51 in accordance with the creation of the template described above. There is a method of preferentially presenting the formal name defined as.

  For each formal name to be presented, the formal name presentation area 210 is provided with a text box 211 in which the formal name is written and a text box 212 in which the official name is entered. The system administrator can correct the notation and reading of the official name registered in the phrase DB 40 by rewriting the character strings entered in these text boxes 211 and 212.

  Further, in the formal name presentation area 210, an “add” button 213 (screen part for the user to input a formal name to be registered in the phrase DB 40) operated to additionally register a new formal name in the phrase DB 40 is provided. Is provided. When the system administrator operates this “add” button 213, new text boxes 211 and 212 for writing a new formal name notation and reading are displayed, and a new formal name is entered by entering a character string here. Additional registration in the DB 40 is possible.

  The alias presentation area 220 presents, for each formal name presented in the formal name presentation area 210, an alias registered in the phrase DB 40 in association with the formal name (including registration candidate aliases). In the alias presentation area 220, for each alias to be presented, a text box 221 in which the name of the alias is entered and a text box 222 in which the reading of the alias is entered are provided. By rewriting the character strings entered in these text boxes 221, 222, the system administrator can correct the notation and reading of aliases (including registration candidate aliases) registered in the phrase DB 40.

  In addition, the alias name display area 220 is operated so that, for each formal name presented in the formal name presentation area 210, the system administrator inputs a new alias name corresponding to the formal name and adds it to the phrase DB 40. "Add" button 223 (screen part for the user to input an alias to be registered in the phrase DB 40) is provided. When the system administrator operates this “add” button 223, new text boxes 221 and 222 for writing new alias notations and readings are displayed. By entering a character string here, a new alias is registered as a candidate for registration. It can be added to the phrase DB 40 as an alias.

  The type presentation area 230 presents, for each alias presented in the alias presentation area 220, whether the alias is an alias of “system proposal” or an alias of “user input”. As described above, the alias of “system proposal” is an alias automatically generated by the generation function 111 by log analysis, an alias automatically generated by the generation function 112 by rule, or by language analysis. A more detailed type such as an alias automatically generated by the generation function 113 may be presented. The system administrator can recognize the type of the presented alias by referring to the type presentation area 230 of the registered word management screen 200 and determine whether registration of the alias should be validated or invalidated. it can.

  The score presentation area 240 presents a score representing the importance of the alias for the alias of the “system proposal” type among the aliases presented in the alias presentation area 220. This score is a score calculated when the alias generation unit 110 automatically generates an alias as described above and stored in the alias table 43 of the phrase DB 40. Whether the system administrator recognizes the importance of the alias whose type is “system proposal” by referring to the score presentation area 240 of the registered word management screen 200, and should the registration of the alias be enabled or disabled? Can be judged.

  A check box 250 is provided for each alias presented in the alias presentation area 220, and the alias for which the check box 250 is checked indicates that registration is valid. The system administrator switches the validity / invalidity of registration of the alias by checking or unchecking the check box 250 provided for each alias on the registered word management screen 200. Can do. The alias for which the registration is valid can be used as a phrase that replaces the formal name when the dialog processing unit 10 performs the dialog processing with the end user. The alias whose registration is invalidated is maintained in the phrase DB 40 as an alias for the registration candidate, but is not used for the dialogue processing in the dialogue processing unit 10.

  The UI control unit 120 displays the registered word management screen 200 configured as described above, for example, on a display device of a terminal used by the system administrator, and the operation of the system administrator performed on the registered word management screen 200. Accept. When the “update” button 260 on the registered word management screen 200 is operated by the system administrator, the received operation details by the system administrator are reflected in the phrase DB 40.

  FIG. 7 is a flowchart illustrating a processing procedure of the UI control unit 120. When the system administrator requests display of the registered word management screen 200, the UI control unit 120 executes a series of processes shown in the flowchart of FIG. 7 and updates the phrase DB 40 according to the operation of the system administrator.

  That is, when the display of the registered word management screen 200 is requested by the system administrator, the UI control unit 120 first registers as shown in FIG. 6 based on the formal name master 41 and the alias table 43 of the phrase DB 40, for example. The word management screen 200 is generated and displayed on the display device of the terminal used by the system administrator (step S401). Then, if there is an operation for adding a formal name on the registered word management screen 200 (step S402: Yes), the UI control unit 120 receives the operation and acquires the formal name to be added to the formal name master 41 ( Step S403). If there is an operation for adding an alias on the registered word management screen 200 (step S404: Yes), the UI control unit 120 accepts the operation and acquires the alias to be added to the alias table 43 (step S405). . Further, the UI control unit 120 has an operation on the check box 250 corresponding to any of the aliases on the registered word management screen 200, that is, an operation for switching validity / invalidity of registration of the aliases (step S406: Yes). ), And accepts the operation and acquires whether the alias to be switched and registration of the alias are valid or invalid (step S407).

  While the registered word management screen 200 is displayed on the display device of the terminal used by the system administrator, the UI control unit 120 always monitors the operation on the “update” button 260 on the registered word management screen 200 (step). S408). Then, while the “update” button 260 is not operated (step S408: No), the determination and processing from step S402 to step S407 described above are repeated. When the “update” button 260 is operated, the phrase DB 40 is updated based on the information acquired in at least one of the above-described steps S403, S405, and S406 (step S409), The process ends.

  As described above in detail with specific examples, according to the present embodiment, the alias generation unit 110 automatically generates an alias for the official name registered in the phrase DB 40, and the registration candidate alias. It is recorded in the phrase DB 40 as (an alias for “system proposal”). Also, the UI control unit 120 displays the registered word management screen 200 on the display device of the terminal used by the system administrator, for example, in response to a request from the system administrator, and the alias recorded in the phrase DB 40 is It is possible to identify whether the alias is a “system proposal” alias or a “user input” alias entered by the system administrator, and whether to enable or disable the registration of that alias. If possible, present it to the system administrator. Then, the UI control unit 120 receives an operation of the system administrator for the registered word management screen 200 and reflects the operation content in the phrase DB 40. Therefore, according to the present embodiment, the automatically generated alias and the alias entered by the system administrator can be managed in an integrated manner, including whether or not the registration is valid.

  In addition, according to the present embodiment, since the “system proposal” alias is presented on the registered word management screen 200 as a score indicating its importance, registration of the “system proposal” alias is effective. It is possible to make the system administrator accurately determine whether to make or invalidate.

  In addition, according to the present embodiment, the formal name or alias to be added to the phrase DB 40 can be input on the registered word management screen 200, so that the system administrator can refer to the registered word management screen 200 while referring to the registered word management screen 200. Name and alias name input operations can be performed easily and appropriately.

<Supplementary explanation>
The above-described voice interaction system 1 can be realized by, for example, cooperation between hardware constituting a general computer and a program (software) executed on the computer. For example, when the computer executes a predetermined program, functional units such as the above-described dialog processing unit 10, alias generation unit 110 of the registered word management device 100, and UI control unit 120 can be realized. In addition, the above-described dialogue knowledge DB 20, dialogue log DB 30, word / phrase DB 40, and the like can be realized using a mass storage device provided in the computer.

  FIG. 8 is a block diagram illustrating a hardware configuration example of the above-described voice interaction system 1. For example, as shown in FIG. 8, the above-described spoken dialogue system 1 includes a processor (hardware processor, processor circuit) 2 such as a CPU, a memory 3 such as RAM and ROM, and a storage (mass storage) such as an HDD and SSD. Apparatus) 4 and a communication hardware I / F 5 that communicates with the outside via a network.

  At this time, the above-mentioned program executed by the computer is provided by being recorded on, for example, a magnetic disk, an optical disk, a semiconductor memory, or a similar recording medium. As long as the recording medium for recording the program is a computer-readable recording medium, the storage format may be any form. In addition, the above-described program executed by the computer may be installed in the computer in advance, or the program distributed via the network may be installed in the computer as appropriate.

  The above-described program executed by the computer has a module configuration including functional units such as the above-described dialog processing unit 10, the alias generation unit 110 and the UI control unit 120 of the registered word management device 100, and a processor. When the program 2 reads and executes the program as appropriate, these functional units are generated on a memory 3 such as a RAM. The program for realizing the dialogue processing unit 10 and the program for realizing the alias generation unit 110 and the UI control unit 120 of the registered word management apparatus 100 may be independent programs.

  The above-described voice interaction system 1 has a configuration in which some or all of the above functional units are realized by dedicated hardware such as an application specific integrated circuit (ASIC) or a field-programmable gate array (FPGA). There may be.

  In addition, the above-described voice interaction system 1 may be configured as a network system in which a plurality of computers are communicably connected, and the above-described functional units may be distributed and realized in a plurality of computers. For example, the dialogue processing unit 10 and the alias generation unit 110 and the UI control unit 120 of the registered word management apparatus 100 may be realized on different computers. Further, the above-described voice interaction system 1 may be a virtual machine that operates on a cloud system.

  As mentioned above, although embodiment of this invention was described, this embodiment is shown as an example and is not intending limiting the range of invention. The novel embodiment can be implemented in various other forms, and various omissions, replacements, and changes can be made without departing from the scope of the invention. These embodiments and modifications thereof are included in the scope and gist of the invention, and are included in the invention described in the claims and the equivalents thereof.

1 Spoken Dialogue System 10 Dialogue Processing Unit 20 Dialogue Knowledge DB
30 Dialogue log DB
40 words DB
DESCRIPTION OF SYMBOLS 100 Registration word management apparatus 110 Alias generation part 111 Generation function by log analysis 112 Generation function by rule 113 Generation function by language analysis 120 UI control part 200 Registered word management screen 210 Formal name presentation area 213 Formal name "add" button 220 For alias name Area 223 Alias “Add” button 230 Type presentation area 240 Score presentation area 250 Check box 260 “Update” button

Claims (10)

A registered word management device for managing registered words in a phrase database used in a spoken dialogue system,
An alias generation unit that automatically generates an alias that is a phrase rephrasing a formal name registered in the phrase database, and records the generated alias in the phrase database as a registration candidate;
Whether the alias recorded in the phrase database can be identified whether the alias is automatically generated or entered by the user, and whether registration of the alias is valid or invalid A UI control unit that displays a registered word management screen that is presented to the user in a switchable manner, accepts a user operation on the registered word management screen, and reflects the operation content in the phrase database.   The registered word management apparatus according to claim 1, wherein the registered word management screen further presents a score representing the importance of the alias for the alias generated by the alias generation unit.   The registered word management apparatus according to claim 1, wherein the registered word management screen includes a screen part for a user to input an alias to be registered in the phrase database.   4. The registered word management apparatus according to claim 1, wherein the registered word management screen includes a screen part for a user to input a formal name to be registered in the phrase database. 5.   5. The registered word management apparatus according to claim 1, wherein the alias generation unit automatically generates an alias by analyzing a dialogue log of the voice dialogue system. 6.   6. The registered word management according to claim 1, wherein the alias generation unit automatically generates an alias by applying a predetermined alias generation rule to a formal name registered in a phrase database. 7. apparatus.   The registered name management device according to any one of claims 1 to 6, wherein the alias generation unit parses a formal name registered in a phrase database to automatically generate an alias. The registered word management device according to any one of claims 1 to 7,
The phrase database;
And a dialogue processing unit that performs dialogue processing using the phrase database. A registered word management method for managing registered words in a phrase database used in a spoken dialogue system,
Automatically generating an alias that is a phrase rephrasing a formal name registered in the phrase database, and recording the generated alias as a registration candidate in the phrase database;
Whether the alias recorded in the phrase database can be identified whether the alias is automatically generated or entered by the user, and whether registration of the alias is valid or invalid Displaying a registered word management screen that is presented to the user in a switchable manner;
Receiving a user operation on the registered word management screen and reflecting the operation content in the phrase database. On the computer,
A function of automatically generating an alias that is a phrase obtained by rephrasing the formal name registered in the phrase database, and recording the generated alias as a registration candidate in the phrase database;
Whether the alias recorded in the phrase database can be identified whether the alias is automatically generated or entered by the user, and whether registration of the alias is valid or invalid A function to display a registered word management screen that is presented to the user in a switchable manner,
A program for realizing a function of accepting a user operation on the registered word management screen and reflecting the operation content in the phrase database.

Images