Search Images Maps Play YouTube News Gmail Drive More »
Sign in
Screen reader users: click this link for accessible mode. Accessible mode has the same essential features but works better with your reader.

Patents

  1. Advanced Patent Search
Publication numberCN101656799 A
Publication typeApplication
Application numberCN 200910167056
Publication date24 Feb 2010
Filing date19 Aug 2009
Priority date20 Aug 2008
Also published asCN101656799B, EP2157570A1, EP2157570B1, US8935163, US20100049513
Publication number200910167056.5, CN 101656799 A, CN 101656799A, CN 200910167056, CN-A-101656799, CN101656799 A, CN101656799A, CN200910167056, CN200910167056.5
Inventors胜仓裕, 黄声扬
Applicant阿鲁策株式会社
Export CitationBiBTeX, EndNote, RefMan
External Links: SIPO, Espacenet
Automatic conversation system and conversation scenario editing device
CN 101656799 A
Abstract
The invention provides an automatic conversation system and a conversation scenario editing device, the objective of the invention is that: language models can be generated as well as responsing to users calling even without any special knowledge, which can implement conversation content based on more acurate speech recognition, and the language models are based on implement of conversation scenario of conversation content of pre-prepared conversation procedure. The system includes a conversation device and a conversation server. The conversation device generates an input sentence through speech recognition of an utterance by a user. The conversation server determines the reply sentence based on the conversation scenario when a reply sentence to the input sentence is requested from the conversation device. The editor includes a language model generator for generating a language model to be used for the speech recognition based on the conversation scenario. According to the editor, a non-expert can generate the language model to provide an adequate conversation based on the speech recognition.
Claims(8)  translated from Chinese
1.一种为了自动会话系统而生成会话情节的会话情节编辑装置,所述自动会话系统具有:会话装置,其通过对用户发话进行声音识别来生成输入句子,向会话服务器请求与该输入句子对应的回答句子;会话服务器,其在从所述会话装置请求了回答句子时,根据会话情节决定回答句子,将该回答句子发送到所述会话装置,并将回答句子输出给用户, 该会话情节编辑装置的特征在于, 具有:语言模型生成单元,其根据所述会话情节生成在所述会话装置进行声音识别时使用的语言模型。 Conversation scenario editing device for a system to automatically generate a session conversation scenario, the automatic conversation system comprising: a conversation device, which the user's utterance by speech recognition to generate the input sentence, the session server request corresponds to the input sentence The answer sentence; session server, upon request from the conversation device answer sentence, decided to answer sentence based on the conversation scenario, the reply sentence to the conversation device sending the reply sentence output to the user, the conversation scenario editing apparatus characterized by comprising: language model generating unit that generates the language model in speech recognition apparatus using the session in accordance with the conversation scenario.
2. 根据权利要求1所述的会话情节编辑装置,其特征在于, 所述会话情节由对象和态射组成,该会话情节编辑装置还具有编辑单元,其生成具有作为态射的输入句子、 和作为与该态射对应的对象的回答句子的所述会话情节。 2. The conversation scenario editing device according to claim 1, wherein, wherein the session by an object and morphisms plot composition, the conversation scenario editing device further comprises an editor that generates functions as morphisms input sentence, and In response to the sentence morphism corresponding object of the conversation scenario.
3. 根据权利要求2所述的会话情节编辑装置,其特征在于, 该会话情节编辑装置还具有动态知识生成单元,其生成用于从所述会话情节中4企索与态射对应的对象的动态知识。 3. The conversation scenario editing device according to claim 2, characterized in that the conversation scenario editing device further comprises a dynamic knowledge generator, which generates the plot from the session 4 cable companies and morphisms corresponding object dynamic knowledge.
4. 一种自动会话系统,其特征在于, 具有:会话装置,其通过对用户发话进行声音识别来生成输入句子,向会话服务器请求与该输入句子对应的回答句子;会话服务器,其在从所述会话装置请求了回答句子时,根据会话情节决定回答句子,将该回答句子发送到所述会话装置,并将回答句子输出给用户;以及会话情节编辑装置,其具有语言模型生成单元,该语言模型生成单元根据所述会话情节生成所述会话装置进行声音识别时使用的语言模型。 An automatic conversation system, comprising: a conversation device, which generates input sentence through speech recognition of the user's utterance, the server requests the session to the input sentence and answer sentence; session server, which from the said conversation device requesting the answer sentence, decided to answer sentence based on the conversation scenario, the reply sentence to the conversation device sending the reply sentence output to a user; and a conversation scenario editing device, which has a language model generation unit, the language model generating unit generates the session apparatus when the speech recognition language model used in accordance with the conversation scenario.
5. 根据权利要求4所述的自动会话系统,其特征在于, 所述会话情节由对象和态射组成,该自动会话系统还具有编辑单元,其生成具有作为态射的输入句子、和作为与该态射对应的对象的回答句子的所述会话情节。 The automatic conversation system according to claim 4, wherein the conversation scenario composed by the objects and morphisms, the automatic conversation system also has the editing unit generating functions as morphisms input sentence, and as with The morphisms corresponding object on the conversation scenario reply sentence.
6. 根据权利要求5所述的自动会话系统,其特征在于,该自动会话系统还具有动态知识生成单元,其生成用于从所述会话情节中检索与态射对应的对象的动态知识。 The automatic conversation system according to claim 5, characterized in that the automatic conversation system also has a dynamic knowledge generation unit generates a dynamic knowledge retrieval from the conversation scenario with morphisms corresponding object.
7. 根据权利要求4至6中任意一项所述的自动会话系统,其特征在于, 所述会话服务器,将记载了使所述会话装置进行的动作的动作控制信息发送到所述会话装置,所述会话装置根据所述动作控制信息,在输出回答句子的同时执行动作。 According to the 4-6 automatic conversation system of any one of the preceding claims, wherein the session server will record the action so that the conversation device operation control information to the conversation device, The session control means in accordance with said operation information, the output operation is performed while the reply sentence.
8. 根据权利要求7所述的自动会话系统,其特征在于, 所述会话装置是由所述动作控制信息来控制动作的终端装置。 8. The automatic conversation system according to claim 7, characterized in that the conversation device is a terminal device by the operation control information to control the operation.
Description  translated from Chinese

自动会话系统以及会话情节编辑装置 Automatic conversation system and conversation scenario editing device

技术领域 Technical Field

本发明涉及自动会话系统以及会话情节编辑装置,更详细而言,涉及作为可以自动输出对用户发话进行应答的回答来确立与用户的会话的系统的自动会话系统、以及对在该自动会话系统中使用的会话情节进行生成以及编辑的会话情节编辑装置。 The present invention relates to an automatic conversation system and conversation scenario editing device, and more particularly, relates to a user's utterance can automatically output to respond answer session with the user to establish a system of automatic conversation system, as well as the automatic conversation system It is used in the conversation scenario editing device, and the conversation scenario editing. 背景技术 Background

以往提出了,在取得用户的发话时输出与该发话内容对应的回答的自动会话装置(例如专利文献1)。 Have been proposed, the output obtained when the user's utterance contents of the utterance corresponding to the automatic answering device session (e.g. Patent Document 1). 在现有的自动会话装置中, 一般的方式是使用把用户的发话和与之对应的回答作为一对来存储的数据库,通过检索该数据库来输出与用户的发话对应的回答。 In the conventional automatic conversation device, the general approach is to use the user's utterance and the corresponding answer as a pair to store the database, by searching the database to output to the user's utterance answer.

但是,在以一对一的关系输出与用户的发话对应的回答的现有方式中,通 However, in the conventional manner-one relationship between the output and the user's utterance reply, the pass

过自动会话装置难以实现:关于某个话题,在用户和自动会话装置之间确立自然的会话;或者循序渐进地向用户讲述预先准备的有故事性的内容(例如,制度结构的说明、急救治疗的内容等)。 Too difficult to achieve automatic conversation device: about a topic in the conversation between the user device and the automatic nature of the establishment of the session; or the user step by step to have a story about the pre-prepared content (for example, institutional structure description, first aid treatment content, etc.).

作为这样的用于确立自然流程的会话的技术,提出了在对用户发话进行应答的同时,使用实现遵循预先准备的会话流程的会话内容的情节,但为了根据该情节生成语言模型,只有具有专门知识的KB (knowledge base,知识库)技术人员才能够生成语言才莫型。 Such a process for establishing a natural conversation techniques presented in response to the user's utterance performed simultaneously, followed achieve plot prepared using the session conversation flow content, but in order to generate a language model based on the plot, only in special Knowledge KB (knowledge base, knowledge base) and technical personnel will be able to generate language only Mo type. 而且在多数情况下,由于将KB(知识库)和会话引擎融合,因此即使是KB制作者也无法掌握全体"情节"。 And in most cases, since the KB (Knowledge Base) and the session engine integration, so even KB makers can not grasp all the "plot."

【专利文献1】特开2002-366190号公开公报发明内容 [Patent Document 1] JP 2002-366190 Publication DISCLOSURE OF THE INVENTION

本发明的目的在于,即使不具有专门知识,也可以在应答用户发话的同时生成语言模型,可以实现基于更准确的声音识别的会话内容,所述语言模型以实现遵循预先准备的会话流程的会话内容的情节为依据。 The purpose of the present invention is that even does not have the expertise, can also be generated in response to a user's utterance while the language model, can achieve conversations based on more accurate voice recognition, the language model to follow in order to achieve pre-prepared session flow of conversation the plot is based on the content.

作为用于解决上述问题的手段,本发明具备以下特征。 As a means for solving the above problems, the present invention has the following characteristics. 本发明提出一种会话情节编辑装置。 The present invention provides a conversation scenario editing device. 该会话情节编辑装置为了自动会话系统而生成会话情节,所述自动会话系统具有:会话装置,其通过对用户发话进行声音识别来生成输入句子,向会话服务器请求与该输入句子对应的回答句 The conversation scenario editing device for automatic conversation system generated conversation scenario, the automatic conversation system comprising: a conversation device, which the user's utterance by speech recognition to generate the input sentence, a request to the session server to the input sentence and answer sentence

子;会话服务器,其在从所述会话装置请求了回答句子时,根据会话情节决定 Child; session server, upon request from the conversation device answer sentence, according to the conversation scenario decide

回答句子,将该回答句子发送到所述会话装置,并将回答句子输出给用户,所 Reply sentence, the reply sentence to the conversation device sending the reply sentence output to a user, the

述会话情节编辑装置的特征在于,具有:语言模型生成单元(例如语言模型生成部),其根据所述会话情节,生成在所述会话装置进行声音识别时使用的语言模型。 Wherein said conversation scenario editing device comprising: a language model generating unit (for example, the language model generating unit), which based on the conversation scenario, a language model generating means when the session using voice recognition.

根据该会话情节编辑装置,在应答用户发话的同时生成语言^^莫型,可以实现基于更准确的声音识别的会话内容,所述语言模型以实现遵循预先准备的会话流程的会话内容的会话情节为依据。 The conversation scenario editing device according to the user's utterance in response ^^ Mo while generating language model, based on the content of the session can be achieved more accurate voice recognition, the language model to achieve a conversation scenario follows the process previously prepared conversation session contents as the basis.

所述会话情节由对象和态射(morphism)组成,所述会话情节编辑装置可以进一步具有编辑单元,其生成具有作为态射的输入句子、和作为与该态射对应的对象的回答句子的所述会话情节。 The conversation scenario objects and morphisms (morphism) composed of the conversation scenario editing device further comprises an editor, which has generated as morphisms input sentence, and as the corresponding object morphism reply sentence said the conversation scenario. 在现有的自动会话装置中,仅可以由能够理解根据KB执行会话的会话引擎所特有的知识的KB (知识库)技术人员来生成使会话装置动作的知识,但根据该会话情节编辑装置,一4殳人(最终用户等)也可以自由地构建由"对象和态射的框架,,构成的会话情节,由此可以实现基于更准确的声音识别的会话内容。而且,由此,KB制作者可以掌握全体"会话情节"。 In the conventional automatic conversation device, it can only be understood by the conversation engine according KB performs session peculiar knowledge of KB (Knowledge Base) technical staff to generate knowledge to make conversation device operation, but based on the conversation scenario editing device, a 4 Shu people (end users, etc.) are free to build "objects and morphisms framework constituted by the ,, conversation scenario, which can achieve conversations based on more accurate voice recognition, and, thus, KB Country You can grasp the whole "conversation scenario."

另外,基于本发明,会话情节编辑装置具有以下优点。 In addition, based on the present invention, the conversation scenario editing device has the following advantages.

.可以在将"对象(回答句子)和态射(输入句子)"连接的同时构建会话情节,因此情节的表现易懂。 You can "objects (reply sentence) and a morphism (input sentence)" plot building session of simultaneous connections, so the plot is easy to understand in the performance.

-开发了对象的引用功能,因此关于对象,可以利用引用源的信息(与引用了某对象的对象相关的信息)。 - Develop a reference to the object function, so about an object, you can use the information (with reference to a certain object information related to the object) reference to the source.

•通过对象的迁移目的地和引用源,可以浏览对象的状态迁移关系。 • migration destination and reference source through an object, you can browse the object's state migration relationship.

-通过排列有对象和态射的"回答列",可以表现"对象和态射"的基本性质。 - By arranged objects and morphisms "answer column," can be expressed "objects and morphisms" the basic properties.

-"回答列"正好与"剧本"类似,从一般的制作剧本的观点来看,"回答列是相当易懂的信息"。 - "The answer column" Just similar to "script" from a general point of view of making plays, "replied column is quite understandable information."

5-通过会话情节编辑装置.会话服务器.会话装置的全体,可以实现"使会话装置按照剧本动作"。 5- through conversation scenario editing device. Session server. Plenary Session device can be realized, "the session apparatus according to the script action."

此外,只要会话情节编辑装置是可以处理"由对象和态射构成的会话情节" 得方式,则不限于上述内容。 In addition, as long as the conversation scenario editing device can handle "by the object and the conversation scenario morphisms constituted" get way, is not limited to the above.

上述会话情节编辑装置可以进一步具有动态知识生成单元,其生成用于从所述会话情节中检索与态射对应的对象的动态知识。 The above-mentioned conversation scenario editing device may further comprises a dynamic knowledge generator, which generates a dynamic knowledge retrieval from the conversation scenario with morphisms corresponding object. 根据该会话情节编辑装置,可以高速地检索相当于输入句子的态射以及与该态射对应的对象,并输出作为对象的回答句子。 The conversation scenario editing device according to retrieve the equivalent of high-speed input sentence morphisms and morphisms corresponding to the object, and outputs the reply sentence as an object.

另外,在上述会话情节编辑装置中,会话情节编辑装置可以把预定内容的用户发话以外的全部用户发话作为一个态射来记载。 Further, in the conversation scenario editing device, the conversation scenario editing device can be subscription content to all users outside of the speaker utterance shot to record as a state. 根据该会话情节编辑装置,可以定义能够覆盖无限的用户发话的回答句子。 The conversation scenario editing apparatus, can be defined to cover unlimited user's utterance reply sentence.

本发明的第二形态提出了自动会话系统。 The second aspect of the present invention proposes an automatic conversation system. 该自动会话系统的特征在于,具 The automatic conversation system is characterized by a

有:会话装置,其通过对用户发话进行声音识别来生成输入句子,向会话服务器请求与该输入句子对应的回答句子;会话服务器,其在从所述会话装置请求了回答句子时,根据会话情节决定回答句子,将该回答句子发送到所述会话装置,并将回答句子输出给用户;以及会话情节编辑装置,其具有语言模型生成单元,该语言模型生成单元根据所述会话情节,生成所述会话装置进行声音识别时使用的语言模型。 Are: conversation device, which generates the user's utterance by speech recognition input sentence, a request to the session server to the input sentence and answer sentence; session server, upon request from the conversation device answer sentence, based on the conversation scenario determines the reply sentence, the reply sentence to the conversation device sending the reply sentence output to a user; and a conversation scenario editing device, having a language model generating unit, the language model generating unit based on the conversation scenario, the generated Session device when speech recognition language model used.

根据该自动会话系统,可以在应答用户发话(输入句子)的同时生成语言模型,可以实现基于更准确的声音识别的会话内容,所述语言模型以实现遵循预先准备的会话流程的会话内容的会话情节为依据。 According to the automatic conversation system can be uttered (input sentence) in response user while generating the language model, based on the content of the session can be achieved more accurate voice recognition, the language model to achieve the previously prepared conversation contents to follow the flow of the conversation session plot as the basis.

另外,在上述自动会话系统中,所述会话情节由对象和态射组成,该自动会话系统中还可以具有编辑单元,其生成具有作为态射的输入句子、和作为与该态射对应的对象的回答句子的所述会话情节。 Further, in the automatic conversation system, the conversation scenario composed by the objects and morphisms, the automatic conversation system may also have editing unit generating functions as morphisms input sentence, and as the morphism corresponding object The answer sentence on the conversation scenario.

根据该自动会话系统,在现有的自动会话装置中,仅可以由能够理解根据KB执行会话的会话引擎所特有的知识的KB (知识库)技术人员来生成使会话装置动作的知识,但根据该会话情节编辑装置, 一般人(最终用户等)也可以自由地构建由"对象和态射的框架"构成的会话情节,由此可以实现基于更准确的声音识别的会话内容。 The automatic conversation system according to the conventional automatic conversation device can only be made to understand (KB) technical staff to generate knowledge of the operation in the session according to the session of the Executive Session engine KB KB unique knowledge, but according to The conversation scenario editing device, most people (end-users, etc.) are free to build "objects and morphisms framework" plot constituted by the session, which can be achieved based conversations more accurate voice recognition. 而且,由此,KB制作者可以掌握全体"会话情 And, thus, KB creator can grasp the entire "conversation feeling

6节"。 6. "

另外,在上述自动会话系统中可以进一步具有动态知识生成单元,其生成用于从所述会话情节中检索与态射对应的对象的动态知识。 Further, in the automatic conversation system may further include a dynamic knowledge generating unit that generates a dynamic knowledge retrieval from the conversation scenario with morphisms corresponding object.

另外,在上述会话情节编辑装置中,会话编辑装置可以4巴预定内容的用户发话以外的全部用户发话作为一个态射来记载。 Further, in the conversation scenario editing device, the session editing device 4 users can bar speaker predetermined content to all users outside shot to utterance as a state record. 根据该会话情节编辑装置,可以定义能够覆盖无限用户发话的回答句子。 The conversation scenario editing device, the user can define an unlimited utterance can cover a reply sentence.

另夕卜,在上述自动会话系统中,会话服务器将记载了使会话装置进行的动作的动作控制信息发送到会话装置,会话装置根据动作控制信息,在输出回答 Another evening BU, in the automatic conversation system, the session server will record the session means the operation of the operation control device sends information to the session, the session apparatus according to the operation control information in the output answer

句子的同时执行动作。 Operation is performed while the sentence. 根据该自动会话系统,会话情节制作者按照会话情节不仅可以控制回答句子的内容,也可以控制会话装置可以执行的任何动作。 The automatic conversation system according to the conversation scenario creator can be controlled in accordance with not only the content of the conversation scenario reply sentence can also control sessions means any action that can be performed.

另外,在上述自动会话系统中,会话装置可以是由所述动作控制信息来控制动作的终端装置(例如机器人等)。 Further, in the above-described automatic conversation system, the session may be a terminal apparatus by the operation control means for controlling operation of the information (e.g., a robot, etc.). 根据该发明,会话情节制作者按照会话情节进行终端装置的动作控制,也可以将会话情节作为动作控制程序来利用。 According to the invention, the conversation scenario creator plot operated by session control terminal device, it can also plot the session as an operation control program to use.

根据本发明,可以在应答用户发话(输入句子)的同时生成语言模型,可以实现基于更准确的声音识别的会话内容,所述语言模型以实现遵循预先准备的会话流程的会话内容的会话情节为依据。 According to the present invention, in response user utterance (input sentence) while generating language model, based on the content of the session can be achieved more accurate voice recognition, the language model to achieve the previously prepared conversation contents to follow the flow of the conversation to the conversation scenario in accordance with. 附图说明 Brief Description

图1是表示自动会话系统的结构例的框图。 1 is a block diagram showing an example of automatic conversation system.

图2A是表示会话装置的一个结构例的框图。 2A is a block diagram showing a configuration example of the conversation device. 图2B是表示声音识别部的一个结构例的框图。 2B is a block diagram showing a configuration example of the voice recognition unit. 图3是表示会话服务器的一个结构例的框图。 Figure 3 is a block diagram showing a configuration example of a session server. 图4是表示会话情节编辑装置的一个结构例的框图。 Figure 4 is a block diagram showing a configuration example of the conversation scenario editing device. 图5是表示与谈话的范围相当的会话情节的例子的状态迁移图。 Figure 5 is a conversation with a range equivalent to a conversation scenario example of a state transition diagram. 图6是表示将图5的会话情节作为数据来表现的例子的图。 Figure 6 is a plot showing the conversation Figure 5 as the data to show an example of FIG. 图7是表示包含态射的复合的会话情节的例子的状态迁移图。 Figure 7 shows the composite contains morphisms conversation scenario example of a state transition diagram. 图8是表示将图7的会话情节作为数据来表现的例子的图。 Figure 8 is showing the conversation scenario in Fig. 7 as the data to show an example of FIG. 图9是表示进行基于NULL功能的强制回答的会话情节的例子的状态迁移图。 FIG. 9 is a function-based NULL forced to answer a conversation scenario example of a state transition diagram.

图IO是表示将图9的会话情节作为数据来表现的例子的图。 IO is a diagram showing the conversation scenario in Fig. 9 as an example to show the data in Fig. 图ll是表示通过引用功能,对用户发话进行"固执回答"的会话情节的 Figure ll is a function by reference, the user utterance "stubborn answer" sessions plot

例子的状态迁移图。 State transition diagram of an example.

图12是表示将图11的会话情节作为数据来表现的例子的图。 FIG. 12 is a diagram showing the conversation scenario 11 as the data to show an example of FIG.

图13是表示通过"通过复合构成的单位元"构建了"闭环回答,,的会话 Figure 13 shows the "unit formed by a composite element" to build a "closed loop answer session ,,

情节的例子的状态迁移图。 Plot example of a state transition diagram.

图14是表示将图13的会话情节作为数据来表现的例子的图。 14 is a diagram showing the conversation scenario 13 as data to show an example of FIG.

图15是在态射的复合中结合法则成立的会话情节的例子的状悉迁移图。 Figure 15 is a combination of a conversation scenario holds for morphism composition in the example of state transition diagram noted.

图16是表示将图15的会话情节作为数据来表现的例子的图。 16 is a diagram showing the conversation scenario 15 as data to show an example of FIG.

图17是表示会话情节编辑装置的编辑画面例的图。 Figure 17 shows an example of a conversation scenario editing screen editing device.

图18是表示会话情节保持部的数据结构例的图。 18 is a data structure to keep the conversation scenario example of portion.

图19是表示用于通过会话编辑装置生成会话情节数据的输入画面例的图。 19 is a diagram for an example of the input screen by means for generating the conversation scenario editing session data.

图20接续图19,是表示用于通过会话编辑装置生成会话情节数据的输入画面例的图。 20 connecting 19, is for example of the input screen through the session the conversation scenario editing device data.

图21接续图20,是表示用于通过会话编辑装置生成会话情节数据的输入画面例的图。 Figure 21 connecting 20, shows the data for generating the conversation scenario editing session by means of an example of an input screen in FIG.

图22接续图21,是表示用于通过会话编辑装置生成会话情节数据的输入画面例的图。 22 connecting 21, is for example of the input screen through the session the conversation scenario editing device data.

图23接续图22,是表示用于通过会话编辑装置生成会话情节数据的输入画面例的图。 Figure 23 connecting 22, shows the data for generating the conversation scenario editing session by means of an example of an input screen in FIG.

图24是表示会话情节编辑装置的变形结构例的功能框图。 FIG 24 is a functional block diagram showing a modification example of the structure of the conversation scenario editing device. 图25是回答处理部的功能框图。 FIG. 25 is a functional block diagram of the processing unit of the answer.

图26是表示字符串和从该字符串提取出的形态素的关系的图。 Figure 26 shows strings and extracted from the string morpheme of relationship. 图27是表示"发话句子的类型"、和表示该发话句子的类型的两字符的字母、以及与该发话句子的类型对应的发话句子的例子的图。 Figure 27 shows the "uttered sentence type", and represents the letter of the uttered sentence type two characters, as well as examples of the type of the uttered sentence uttered sentence corresponding to FIG.

图28是表示句子的类型和用于判定该类型的词典的关系的图。 FIG. 28 is a type of sentence and for determining the type of dictionary of FIG relationships. 图29是表示会话数据库存储的数据的数据结构的一例的概念图。 FIG 29 is a diagram showing a data structure of data stored in the session database conceptual diagram of an example. 图30是表示某话题确定信息和其它话题确定信息的关联的图。 FIG. 30 is a topic specification information and other information associated topics identified in FIG. 图31是表示话题标题(也称为"第二形态素信息")的数据结构例的图。 31 is a topic titles (also referred to as "second morpheme information") is an example of the data structure of FIG. 图32是用于说明回答句子的数据结构例的图。 FIG. 32 is a diagram illustrating an example of a data structure answer sentence.

图33是表示与某话题确定信息对应的话题标题、回答句子、下一计划指 33 is a determining information corresponding to a particular topic and topic titles, the reply sentences, the next program means

定信息的具体例的图。 Given a specific example of the information.

图34是用于说明计划空间的概念图。 FIG. 34 is a conceptual diagram program space.

图35是表示计划的例子的图。 FIG. 35 is a diagram showing an example of a program.

图36是表示另一计划的例子的图。 FIG 36 is a plan showing another example of FIG.

图37是表示计划会话处理的具体例的图。 Figure 37 shows a specific example of the plan conversation process.

图38是表示会话控制部的主处理的一例的流程图。 FIG. 38 is a flowchart showing the main processing session control unit an example.

图39是表示计划会话控制处理的一例的流程图。 FIG 39 is a flowchart showing an example of the plan conversation control process.

图40接续图39,是表示计划会话控制处理的一例的流程图。 Figure 40 connection 39, it is a flowchart showing an example of the plan conversation control process.

图41是表示基本控制状态的图。 FIG. 41 is a diagram showing the basic control state.

图42是表示谈话空间会话控制处理的一例的流程图。 FIG 42 is a flowchart showing a discourse space conversation control process of an example.

符号说明 Symbol Description

1自动会话装置、10会话装置、20会话服务器、30会话情节编辑装置、 40会话情节具体实施方式 An automatic conversation device, the conversation device 10, the conversation server 20, the conversation scenario editing device 30, a specific embodiment of the conversation scenario 40

以下,参照附图"^兑明本发明的实施方式。 Referring to the drawings, "the embodiment of the present invention ^ against Ming.

本实施方式提出了,根据预先准备的会话情节对用户的发话等进行应答来输出回答的自动会话系统、以及生成、编辑会话情节的会话情节编辑装置。 The present embodiment proposes, conversation scenario prepared in advance in accordance with the user's utterance reply to the like outputs answered automatic conversation system, and generating, a conversation scenario editing conversation scenario editing device. (1.自动会话系统、会话情节编辑装置的结构例) (Configuration example 1. Automatic conversation system, the conversation scenario editing device)

以下,说明自动会话系统、会话情节编辑装置的结构例。 Hereinafter, a configuration example of automatic conversation system, the conversation scenario editing device. 图l是表示自动会话系统1的结构例的框图。 Figure l is a block diagram showing a configuration example of the automatic conversation system 1. 自动会话系统1由会话装置10、与会话装置10 连接的会话服务器20、以及生成、编辑会话服务器20所使用的会话情节的会话情节编辑装置30构成。 Automatic conversation system 1 conversation device 10, the conversation server 10 connected with the conversation device 20, and the generated conversation scenario editing session server 20 using a conversation scenario editing device 30 is configured.

会话装置10,当用户输入发话时,将其发话内容发送到会话服务器20。 Conversation device 10, when the user enters the caller, the caller sends its content to the conversation server 20. 会话服务器20当取得发话内容时,根据会话情节决定回答和动作控制信息, 并且将回答以及动作控制信息输出到会话装置10,该回答是对发话内容的答复,该动作控制信息是记载了与该回答对应的动作、即由会话装置10执行的动作的信息。 Session server 20 when the speaker made the content, according to the conversation scenario decided to answer and motion control information, and the answer, and the operation control information to the conversation device 10, the answer is that the content of the speaker's reply, the operation control information is recorded with the Answer corresponding action, i.e., information from the operation means 10 performs the session. 会话情节编辑装置30生成、编辑会话情节40,并且输出已生成的、或者 Conversation scenario editing device 30 generates, edited conversation scenario 40, and outputs the generated, or

已编辑好的会话情节。 The edited conversation scenario. 所输出的会话情节40被存储在会话服务器20中。 Output conversation scenario 40 is stored in the session server 20. 另夕卜, 会话情节编辑装置30根据生成的会话情节40生成语言模型50。 Another evening Bu, conversation scenario editing device 30 based on the generated conversation scenario 40 generated language model 50. 为了由会话装置IO进行声音识别而使用语言模型50。 For voice recognition by the conversation device IO and language model 50. 以下,详细描述各个上述装置。 Hereinafter, a detailed description of each of the above-described apparatus. (U.会话装置) (U. conversation device)

会话装置IO具有以下功能:取得用户的发话(用户发话)作为输入,将该输入内容(以下称为输入句子)发送到会话服务器20,接收从会话服务器20答复的回答以及动作控制信息,根据接收内容执行与回答的输出以及动作控制信息对应的动作。 Session IO device has the following functions: to obtain the user's utterance (user's utterance) as an input, the input content (hereinafter referred to as the input sentence) to the conversation server 20, receiving control information from the conversation server 20 replies response and operation, the received Content perform answered output and operation control information corresponding action.

会话装置IO是具备运算处理装置(CPU)、主存储器(RAM)、读出专用存储器(ROM)、输入输出装置(1/0)、在必要时具备硬盘装置等外部存储装置的信息处理装置、或者包含这种信息处理装置的器具、玩具等,例如是计算机、便携电话机、所谓的互联网家电、或者机器人等装置。 IO device is session includes arithmetic processing means (CPU), main memory (RAM), a read-only memory (ROM), input-output means (1/0), when necessary, the information processing apparatus includes a hard disk device and other external storage device, or information processing apparatus comprising such an appliance, toys, for example, a computer, a portable telephone, a so-called Internet appliance, a robot or the like apparatus. 在会话装置10的所述ROM或者硬盘装置等中存储了程序,将该程序加载到主存储器上,由CPU执行该程序,由此实现会话装置10。 In the ROM or hard disk device conversation device 10 stores the program, the program is loaded into the main memory, executed by the CPU, thereby achieving the conversation device 10. 另外,上述程序也不一定存储在信息处理装置内的存储装置中,也可以从外部装置(例如ASP (应用程序服务提供商)的服务器等)提供,将其加载到主存储器上。 Also, the program is not necessarily stored in the storage device of the information processing apparatus, but also from an external device (such as ASP (application service provider) server, etc.), which is loaded into the main memory.

图2A是表示会话装置10的一个结构例的框图。 2A is a block diagram showing a configuration example of the conversation device 10. 会话装置IO具有:输入部11、与输入部11连接的声音识别部15、与声音识别部15连接的会话处理部12、与会话处理部12连接的动作控制部13、与会话处理部12以及动作控制部13连接的输出部14。 Session IO device comprising: an input unit 11, and the voice recognition unit 11 connected to the input unit 15, the conversation process unit 15 connected to the voice recognition unit 12, 13, and the conversation process unit and the operation control unit 12 is connected to the conversation process unit 12, and operation control section 13 connected to the output section 14. 另外,会话处理部12可以与会话服务器20通信。 In addition, the processing unit 12 may be a communication session with the session server 20. 另外,声音识别部15取得由会话情节编辑装置30生成的语言模型50,根据该语言模型将声音信号变换为输入句子。 Further, the voice recognition unit 15 acquires from the conversation scenario editing device 30 generates a language model 50, based on the language model is converted into a sound signal input sentence.

输入部11具有取得用户的发话内容(输入句子),将其变换成电气信号等会话处理部12可以处理的信号来进行传递的功能。 Input section 11 having obtained the content of the user's utterance (input sentence), which is converted into electrical signals, signal processing unit 12 can process a session to be passed function. 输入部11例如是麦克风。 Input unit 11 such as a microphone.

声音识别部15具有根据从输入部11取得的信号推测用户的发话内容,将作为推测结果而得到的用户的发话内容作为输入句子来输出的功能。 According to the voice recognition unit 15 has acquired from the input unit 11 signals the user's utterance contents presumed, as a speculative result of the user's utterance contents obtained as a function of outputting the input sentence. 图2B揭示了表示声音识别部15的结构例的功能框图。 Figure 2B discloses a functional block diagram showing the structure of a voice recognition unit 15 embodiment. 声音识别部15具有:取得声音信号的前处理部110、与前处理部110连接的特征提耳又部120、与特征^是取部 Voice recognition unit 15 has: a front made sound signal processing unit 110, and the characteristics of the pre-processing unit 110 is connected to mention ear 120, and features a measuring section and section ^

120连接的识别部130。 120 Identification unit 130 is connected.

识别部130具有:识别运算部140、与识别运算部140连接的音响模型存储部150、以及语言模型存储部160。 Recognition portion 130 has: recognition calculation section 140, and 150, as well as the language model storage unit 160 acoustic model storage unit 140 connected to the identification operation part. 识别运算部140与特征提取部120相连, 并且与会话控制部20相连。 140 feature extraction unit coupled to the identification operation part 120, and the control unit 20 is connected to the session. 另外,语言模型存储部160的存储内容,根据来自后述的语言模型生成部34的输出被改写。 In addition, the language model storage unit stored content 160, to be described later based on an output from a language model generating unit 34 are rewritten.

前处理部IIO将作为声音信号的模拟信号变换为数字信号,另外进行除去噪声以使特征提取变得容易的噪音除去处理。 The pre-processing unit IIO converted audio signal as an analog signal into a digital signal, for removing additional noise feature extraction becomes easy to make the noise removal process.

特征提取部120取得从前处理部IIO输出的模拟信号,进行从该模拟信号取得判定模式(pattern)属于哪一类(class)的信息的处理。 Feature extraction unit 120 acquire the analog signal pre-processing unit IIO output processing information acquisition checking mode (pattern) The analog signal from which type (class) of. 在此,所谓"模式",是指真实世界的声音、图像等信息,所谓"类",是模式适合的预定的概念。 In this case, the so-called "model" refers to the real world of sound, images and other information, the so-called "class" is the pattern for a predetermined concept. 例如,通过的发音而产生的声音信号是模式,该声音信号成为记号这样的类。 For example, the sound signal is generated pronunciation mode, the sound signal becomes a sign of such a class.

特征提取部120根据作为声音信号的数字信号,计算表示该声音信号的特征的特征量。 120 as a sound signal according to a digital signal, characterized in that the sound signal is calculated using a characteristic feature amount extraction unit. 关于如何计算特征量提出了各种方法,例如存在将数字信号变换为声谱(spectrum)并提取出共振峰(formant),将其作为特征矢量来算出的方法。 About how to calculate the feature quantity proposed various methods, for example, the presence of the acoustic spectra is converted into a digital signal (spectrum) and extracted formants (formant), as a feature vector calculating method thereof. 或者,特征提取部120也可以通过根据数字信号取得MFCC (Mel Frequency Cepstrum Coe伍cient)这样特征量的方法来计算出特征矢量,可以采用任何方法。 Alternatively, the feature extraction unit 120 can also obtain MFCC (Mel Frequency Cepstrum Coe Ng cient) such feature amounts according to the digital signal to calculate the feature vector method, any method may be employed.

识别部130具有判别通过特征提取部120输出的特征矢量属于哪个类的功能。 Unit 130 discriminative feature extraction unit 120 through the output feature vectors which belong to the class of functions. 识别部130预先准备了代表各类的原型矢量(prototype vevtor),计算所输入的特征矢量近似于哪个原型矢量,决定与特征矢量对应的类。 Unit 130 is prepared in advance with representatives of the various types of prototype vector (prototype vevtor), feature vector calculation approximates the input vector which prototype, decided to feature vector corresponding class. ,

当更详细地说明识别部130的结构时,识别部130具有:识别运算部140、 与识別运算部140连接的音响模型存储部150以及语言模型存储部160。 As described structure identifying unit 130 in more detail, the recognition unit 130 has: recognition calculation section 140, and the identification operation part 150 and the language model storage unit 160 acoustic model storing unit 140 is connected.

音响模型存储部150具有存储音响模型的功能。 Audio model storage unit 150 has a function to store audio model. 音响模型是求出表示当对 Audio model is obtained showing when to

单词列W,、W2、…、Wn进行发声时以多大概率观测到特征矢量系列X1、X2、...、 Word string W ,, W2, ..., Wn proceeds at what probability the observed utterance a feature vector series X1, X2, ...,

Xn的带条件的概率的概率模型。 Xn conditional probability model of probability.

语言模型存储部160具有存储语言模型的功能。 Language model storage unit 160 has a function of storing language model. 语言模式是求出单词列 Language mode is determined word list

W,、 W2、…、Wn的发生才既率P ( W" W2、…、Wn)的才既率才莫型。 W ,, W2, ..., Wn incidence rate was either P (W "W2, ..., Wn) the rate was only just Mo type.

11该概率模型,对被称为语料库(corpus)的文例集(例如收集了报纸报道的电子数据、在该实施方式中包含会话情节)中出现的单词Ww的出现次数、 单词列Ww、 Wk的出现次数进行计数,用前者除后者,由此计算单词的连接概率,求出单词列的发生概率。 11 The probability model, example sentences set is called the corpus (corpus) (for example, electronic data collected newspaper reports, including conversation scenario in this embodiment) Ww number of occurrences of words appearing in the word column Ww, Wk of counting the number of occurrences, in addition to the latter by the former, thereby calculating the connection probabilities of words, determining an occurrence probability of a word column.

识别运算部140具有以下功能:使用声音模型,求出对单词列w,、 w2..... Recognition calculation section 140 has the following features: Using voice model, calculated on a word string w ,, w2 .....

Wn进行发声时观测到特征矢量系列XI、 X2..... Xn的概率Pl、和单词列 Wn sound at the time of observation feature vector series XI, X2 ..... Xn probability Pl, and the word column

Wp W2、 ...、 Wn的发生概率P2的积,搜索该积达到最大的单词列W" W2、...、wn。 Wp W2, ..., Wn probability P2 of the product, the product of the maximum search word column W "W2, ..., wn.

识别运算部140将作为搜索结果而得到的单词列作为输入句子而输出,传递到会话控制部处理部12。 The identification operation part 140 as the search result word string obtained as the input sentence is output, is transmitted to the control unit session processing unit 12.

以上,结束了识别部130、声音识別部15的说明。 This concludes the recognition unit 130, a voice recognition unit 15 is described. 返回图2A,继续会话装置10的结构例的说明。 Returning to Figure 2A, illustrates a configuration example of the conversation device 10 continues.

会话处理部12将从声音识别部15取得的输入句子发送到会话服务器20, 请求会话服务器20发送与输入句子对应的回答句子、以及与该回答句子对应的动作控制信息。 The input sentence from the conversation process unit 12 to obtain the voice recognition unit 15 is transmitted to the conversation server 20, and requests the conversation server 20 transmits a reply sentence corresponding to the input sentence, and the reply sentence corresponding to the operation control information. 另外,会话处理部12,当从会话服务器20接收回答句子以及与该回答句子对应的动作控制信息时,将回答句子传递到输出部14来使其输出,同时将动作控制信息传递到动作控制部13。 In addition, the session processing unit 12, when the answer sentence to the reply sentence and operation control information corresponding to the conversation server 20 receives from when the reply sentence is passed to the output unit 14 to its output, while the operation of the operation control information is passed to the control unit 13.

动作控制部13根据从会话处理部12传递来的动作控制信息,执行所指定的动作。 Operation control unit 13 from passing conversation process unit 12 to the operation control information, perform the specified action. 若所指定的动作是执行输出部14的显示(例如,再生所指定的动作), 则使输出部14执行该动作。 If the specified action is to perform the display output unit (e.g., specified regeneration operation) 14, the output unit 14 performs the operation. 另外,当所指定的动作是与从会话服务器20取得的回答句子不同的回答句子的输出(例如,从会话服务器20取得的回答句子是"说什么呢?",不同的回答句子是"请说点什么吧!")时,使输出部14 输出这样的回答句子。 In addition, when the specified action is obtained from the conversation server 20 different answers sentence reply sentence output (for example, from the conversation server 20 to obtain a reply sentence is "talking about?", A different answer sentence is "Please say something What is it! "), the output unit 14 outputs this answer sentence.

输出部14具有以用户可理解的形态输出回答句子的功能。 Output portion 14 has a form understandable to the user the function of the output of the reply sentence. 关于以何种形态输出回答句子,在本发明中没有限制。 About output reply sentence in what form, it is not limited in the present invention. 例如,输出部14可以是向用户提供作为文本信息的回答句子的例如LCD设备的终端设备、向用户提供作为语音(声音)信息的回答句子的人工语音发生器或扬声器。 For example, the output unit 14 may provide the reply sentence as text information of a terminal device such as an LCD device to a user, the user provides a speech (voice) information of the reply sentence artificial voice generator or a speaker. 注意,输出部14可以是如下机器人:能够在设置在机器人上的显示器上显示图像/文本、能够使所机器人上设置的光源发光、能够通过机器人上设置的扬声器输出语音/声音、或者能够移动自己的手臂等。 Note that the output section 14 may be of the robot: the ability to display an image / text on the monitor settings on the robot, enabling the light source provided on the robot, the speaker output can be set by the robot voice / audio, or be able to move their arms and so on. 动作控制部13根据动作控制信息来控制这样的动作。 Operation control unit 13 based on the operation control information to control this kind of action.

(1.2.会话服务器) 会话服务器20是具有根据会话情节决定回答和动作控制信息,并将回答以及动作控制信息输出到会话装置IO的功能的装置,所述回答是针对发话内容的答复,所述动作控制信息是记载了与该回答对应的动作、也就是使会话装置IO执行的动作的信息。 (1.2 Session server) session server 20 is a plot based on the session decided to answer and motion control information and answer and the operation control information to the conversation device IO device function, the answer is for the speaker content reply, the operation control information is recorded with the answer corresponding action, which is to make information sessions IO device to perform the operation.

会话服务器20是具备运算处理装置(CPU)、主存储器(RAM)、读出专用存储器(ROM)、输入输出装置(1/0)、在必要时具备硬盘装置等外部存储装置的信息处理装置等,例如是计算机、工作站、服务器装置等。 The conversation server 20 includes a processing unit (CPU), main memory (RAM), a read-only memory (ROM), input-output means (1/0), when necessary, the information processing apparatus includes a hard disk device and other external storage device , for example, a computer, a workstation, a server device or the like. 在会话服务器20的所述ROM或者硬盘装置等中存储了程序,通过将该程序加载到主存储器上、由CPU对其进行执行,来实现会话服务器。 In the session server or hard disk device 20 of the ROM and the like are stored in the program is loaded into the main memory through the program, it is executed by the CPU, to achieve the session server. 另外,上述程序也不一定存储在信息处理装置内的存储装置中,也可以是从外部装置(例如ASP (应用程序服务提供商的服务器等))提供,将其加载到主存储器中的结构。 In addition, the storage device of the program is not necessarily stored in the information processing apparatus, or may be from an external device (such as a server ASP (application service provider)) are provided, which was loaded into the main memory structures.

会话装置10和会话服务器20可以通过有线或无线相连,另外也可以经由LAN、无线LAN、因特网等通信网(也可以组合多个通信网)相连。 Conversation device 10 and the conversation server 20 may be connected via wired or wireless, it is also possible via a LAN, wireless LAN, a communication network such as the Internet (a plurality of communication networks may be combined) are connected. 另外, 会话装置10和会话服务器20可以不一定是个别独立的装置,即使是通过同一装置实现会话装置10和会话服务器20的结构,本发明也成立。 In addition, the 10 and the conversation server sessions means 20 may not necessarily be separate and independent of the device, even if the structure is to achieve and the conversation server 20 sessions means 10 by the same means, the present invention is also true.

图3是表示会话服务器20的一个结构例的框图。 Figure 3 is a block diagram showing a configuration example of the conversation server 20. 会话服务器20具有:可以和会话装置10通信的回答处理部21;与回答处理部21相连的意思解释词典部23以及会话情节存储部22。 Session server 20 has: You can answer and conversation device processing unit 10 of the 21 communication; and means connected to the processing unit 21 to answer interpretation dictionary unit 23, and the conversation scenario memory 22.

回答处理部21从会话装置10取得输入句子,根据在会话情节存储部22 中存储的会话情节选择或决定与该输入句子对应的回答句子,将所决定的回答句子和与该回答句子对应的动作控制信息发送到会话装置10。 Reply process unit 21 acquires the input sentence from the conversation device 10, based on the conversation scenario stored in the conversation scenario 22 stored in the corresponding selected or decided to reply sentence to the input sentence, a reply sentence corresponding to the determined reply sentence and the operation transmitting control information to the conversation device 10. 另外,回答处理部21参照意思解释词典部23存储的意思解释词典,取得输入句子的同意词或同意句,基于该同意词或同意句选择或决定回答句子。 In addition, reference meaning answered 21 storing dictionary explain the meaning of section 23 interpretation dictionary processing unit, agreed to acquire the input sentence word or sentence agreed to answer based on the consent of a word or sentence agreed sentence choices or decisions.

意思解释词典部23具有存储意思解释词典的功能,该意思解释词典用于进行与输入句子对应的回答句子的改写(基于同意词的扩展等)。 Meaning interpretation dictionary unit 23 has a memory function dictionary interpret the meaning, the meaning and interpretation dictionary is used for the input sentence reply sentence rewriting (based on the agreed terms of expansion, etc.). 意思解释词典相当于具有辞典(thesaurus)那样的功能的数据库。 Meaning interpretation dictionary is equivalent to a database with a dictionary (thesaurus) as a function. 会话情节存储部22具有存储通过会话情节编辑装置30生成或编辑的会话情节40的功能。 Conversation scenario storage unit 22 has a function of storing by the conversation scenario editing device 30 generates or edited conversation scenario 40. 在后面描述会话情节40的说明。 Described later conversation scenario described 40. (1.3.会话情节编辑装置) (1.3 conversation scenario editing device)

会话情节编辑装置30具有以下功能:新生成由所述会话服务器20使用的会话情节,或者变更已生成的会话情节,生成追加内容或者进行内容的一部分的删除而修正后的会话情节。 Conversation scenario editing device 30 has the following functions: the newly generated conversation scenario by the conversation server 20 to use, or change the generated conversation scenario generates additional content or to delete a portion of the contents of the conversation and the revised plot.

会话情节编辑装置30是具备运算处理装置(CPU)、主存储器(RAM)、读出专用存储器(ROM)、输入输出装置(1/0)、在必要时具备硬盘装置等外部存储装置的信息处理装置等,例如是计算机、工作站等。 Conversation scenario editing device 30 includes a processing unit (CPU), main memory (RAM), a read-only memory (ROM), input-output means (1/0), a hard disk device or the like with the information processing external storage device, if necessary devices, such as computers, workstations. 在会话情节编辑装置30的所述ROM或者硬盘装置等中存储了程序,通过将该程序加载到主存储器上、由CPU对其进行执行,来实现会话情节编辑装置30。 In the ROM or a hard disk device or the like in the conversation scenario editing device 30 stores the program, the program is loaded into the main memory by, be executed by the CPU, to achieve the conversation scenario editing device 30. 另外,上述程序也可以不一定存储在信息处理装置内的存储装置中,也可以是从外部装置(例如ASP (应用程序服务提供商的服务器等))提供,将其加载到主存储器上的结构。 Also, the program may be not necessarily stored in the information processing apparatus memory device may also be from an external device (such as a server ASP (application service provider)) are provided, which was loaded into the main memory of the structure .

图4是表示会话情节编辑装置30的一个结构例的框图。 Figure 4 is a block diagram showing a configuration example of the conversation scenario editing device 30. 会话情节编辑装置30具有:输入部31、与输入部31相连的编辑部32、与编辑部32相连的输出部34以及会话情节保持部33、与会话情节保持部33相连的语言模型生成部34。 Conversation scenario editing device 30 includes: editing unit 31, the input unit 31 is connected to an input unit 32, the editing unit 32 is connected to the output section 34 and the conversation scenario retention unit 33, the holding portion 33 is connected to the language model generating unit 34 and the conversation scenario .

输入部31具有以下功能:取得用户的输入,将其变换为电气信号等编辑部32可以处理的信号来传递。 Input section 31 has the following features: Get the user's input, the signal is converted into an electrical signal editing unit 32, etc. can be processed to pass it. 输入部31例如是键盘、指点设备、触摸面板、麦克风的某一个或它们的组合。 Input unit 31 such as a keyboard, pointing device, a touch panel, one or a combination of the microphone.

输出部34具有以会话情节编辑装置30的使用者(操作员)可识别的形态输出编辑中或编辑完成后的会话情节的内容的功能。 Output section 34 has a user (operator) in the conversation scenario editing device 30 may be identifiable format output editor or edit the content of a conversation scenario after the completion of the function. 输出部34例如是液晶显示器装置等。 Output unit 34 such as a liquid crystal display device or the like.

编辑部32具有根据从输入部31输入的内容,进行作为会话情节的数据的生成、以及其编辑(追加、变更、删除)的功能。 Edit section 32 has an input unit 31 based on input from the contents of a conversation scenario were to generate data, and its editing (adding, changing, deleting) function. 此外,通过使输出部34显示编辑中的会话情节的内容,操作员可以实时地掌握会话情节的内容。 In addition, the output unit 34 displays the contents of a conversation scenario editing, real-time operator can grasp the contents of a conversation scenario. 另外, In addition,

编辑部32将编辑完成后的会话情节的数据输出到会话情节保持部33。 Editorial Department 32 After editing a conversation scenario data output to the conversation scenario retention unit 33.

另外,编辑部32也可以具有以下功能:检查在生成的会话情节中是否遵 In addition, the editing unit 32 may have the following features: Check whether the generated conversation scenario Compliance

14守了适当的状态迁移关系,在发生违反的情况下,生成向操作员通知发生了违反、违反发生的输入句子或回答句子的消息等,使输出部进行显示等。 14 keep the proper state transition relations, in the event of a violation generates a violation notice, a violation occurred in the input sentence or a reply sentence message to the operator, etc., the output unit for display and so on.

另夕卜,编辑部32可以进一步具有相当于会话服务器20的意思解释词典部 Another evening BU, editing unit 32 may further have the equivalent of the conversation server 20 to explain the meaning of section dictionary

23的意思解释词典部,编辑部32可以具有以下功能:使用该意思解释词典部,当会话情节中存在重复的意思内容的输入句子或回答句子时,对它们进行整理或者综合,或者促使操作员进行整理、综合。 Meaning interpretation dictionary unit 23, the editorial department 32 may have the following features: Use the dictionary to explain the meaning of section, when there are duplicate content conversation meant plot input sentence or a reply sentence, or a combination of them organize, or cause the operator finishing comprehensive.

会话情节保持部33具有将从编辑部32取得的会话情节40,以随后可以读取的形态进行存储或保持的功能。 Conversation conversation scenario retention unit 33 has acquired 32 episodes from the newsroom 40 can then be read in the form of storing or keeping function. 会话情节保持部33中存储的会话情节40的数据,根据需要或者在预定的时刻等被送到会话服务器20的会话情节存储部22。 The conversation scenario data stored in the conversation scenario retention unit 33 40, or as needed to be sent to the conversation server 20, the conversation scenario storage unit 22 at a predetermined timing and the like. 从会话情节保持部33到会话情节存储部22的会话情节40的转发,可以以经由存储介质的形态来进行,也可以经由通信网、通信电缆来进行。 From the conversation scenario retention unit 33 to the conversation scenario memory 22 of the conversation scenario 40 is forwarded, via a form of storage medium may be, or may be performed via a communication network, the communication cable.

语言模型生成部34具有以下功能:根据在会话情节保持部33中存储的会话情节40,根据在该会话情节40中包含的输入句子、或者输入句子以及回答句子,计算出在它们中包含的单词列的发生概率,将计算出的结果作为语言模型来存储。 Language model generating unit 34 has the following functions: According to the conversation scenario 33 stored in the holding portion of the conversation scenario 40, the input sentence based on the conversation scenario 40 included, as well as the input sentence or a reply sentence, they calculate the word contained in the the probability of a column, the result is calculated as the language model to store.

所存储的语言模型50,根据需要或者在预定的时刻等被转发到会话装置10的声音识别部15。 The stored language model 50, as necessary or at a predetermined timing, etc. are forwarded to the voice recognition unit 10 of the conversation device 15. 从语言模型生成部34到声音识别部15的语言模型50的转发,可以以经由存储介质的形态来进行,也可以经由通信网、通信电缆来进行。 Language model 34 from the language model generating unit 15 forwards the voice recognition unit 50, via a form of storage medium may be, or may be performed via a communication network, the communication cable.

(1.3.1.关于会话情节) (1.3.1 With regard to the conversation scenario)

在此说明会话情节40。 In this description of the conversation scenario 40. 本发明中的会话情节具有以下特征。 The present invention conversation scenario has the following characteristics. (1 )以回答句子为"对象",以用户发话(输入句子)为"态射"(morphism )。 (1) in order to answer the sentence as "objects" to the user's utterance (input sentence) is "morphism" (morphism). 通过具有该特征,可以通过"状态迁移图"表现会话情节决定的会话流程。 By having this feature, you can "state transition diagram" expression session flow conversation scenario decision. 本发明的会话情节,通过使用后述的"其它,,功能,可以输出与全部输入句子(用户发话)对应的回答句子。另外,通过后述的"计时器"发话,可以应对用户的"无言"(无输入)(可以将无言作为态射来处理)。 Conversation scenario of the present invention, through the use of said "other ,, function, can output all input sentence (user's utterance) corresponding to the reply sentence. Further, the later-described" timer "utterance, the user may respond" no words "(no input) (can be silent as morphisms to handle).

图5是表示会话情节的例子的状态迁移图。 Figure 5 shows an example of a conversation scenario state transition diagram. 图中,椭圆框X1、 X2、 X3、X4分别是回答句子,它们相当于"对象"。 Drawing, oval frames X1, X2, X3, X4 are reply sentences, respectively, which correspond to the "object." 图中在箭头附近显示的句子是输入句子,它们相当于"态射"。 Sentence shown near an arrow shows the input sentence, which is equivalent to "state radio." 图中<其它>表示来自XI的态射"喜欢"以及"不喜欢"以外的输入句子。 Figure <Others> from the XI represent morphisms "like" and "dislike" input sentence outside. 图中〈timer〉表示在用户无言的状态下经过了预定期间的状态。 Fig. <Timer> represents the state in the user silent after a predetermined period of state. 另夕卜,"〈其它〉l〈timer〉"这样的标记表示"其它"或"timer"。 Another Xi Bu, "<other> l <timer>" this flag indicates "other" or "timer".

在图5所示的例子中,使得"想吃点什么,,这样的"态射,,迁移到作为回答句子X1、即"你喜欢拉面吗"的"对象"。 In the example shown in Figure 5, the "want something to eat ,, such" state radio ,, migrate to a reply sentence X1, that is, "do you like ramen" "objects." 在回答句子X1的输出后,在第l态射"不喜欢"发生的情况下,迁移到回答句子X4 "遗憾!换个话题吧"。 In response to the output X1 sentence after the first l morphism "dislike" the event, migration to reply sentence X4 "sorry! Change the subject." 另一方面,在回答句子X1的输出后,在第2态射"喜欢,,发生的情况下,迁移到回答句子X3 "那么,向你介绍美味的店"。另一方面,在回答句子X1的输出后,在第1以及第2态射以外的态射发生的情况下、或者在用户无言的状态下经过了一定期间的情况下,迁移到回答句子X2"拉面,喜欢? On the other hand, the output of the reply sentence X1, the state at the second shot "like ,, in the event of transits to a reply sentence X3" So, to introduce you to nice restaurant. "On the other hand, in the reply sentence X1 After the output, other than the first and second Morphisms morphisms the event, or in the user's status after silent under certain circumstances during the transits to a reply sentence X2 "Ramen, like? 不喜欢? I do not like? "。 . "

当将上述图5的会话情节作为数据来表现时,作为一例而成为图6那样的内容。 When the above-mentioned conversation scenario in Fig. 5 as data show, as an example and become the kind of content Figure 6. 在此"XI (发话A) X2"是回答列,记载了XI的回答状态通过发话A迁移到X2的回答状态。 In this "XI (utterance A) X2" is the answer column, describes the state of XI answer migrated to answer state X2 by speaker A.

(2)在态射中可以定义复合 (2) can be defined in the composite morphism

通过该特征,可以接受从主情节分支的发话,并且即使分支也可以返回开始(主情节)。 With this feature, you can accept the speaker from the main branch of the plot, and even branch to be returned to the start (main story). 因此,会话情节的生成者可以构建自己构思的会话流程"故事",使会话系统进行遵照该故事的会话。 Therefore, the session creator of the plot can build your own session flow concept "story", the session in accordance with the system of the story session.

图7是表示包含态射的复合的会话情节的例子的状态迁移图。 Figure 7 shows the composite contains morphisms conversation scenario example of a state transition diagram. 图中的记号以及标记遵照图5。 Figure mark and the mark in accordance with FIG. 在此例的会话情节中,在回答句子X1 "你喜欢拉面吗?"的输出后,在第1态射"不喜欢,,发生的情况下,迁移到回答句子X3 "是么? In this example of a conversation scenario, after the reply sentence X1 "Do you like ramen you?" Output, in the first morphism "dislike ,, the event, migration to reply sentence X3" is not it? 拉面很美味啊"。另一方面,在第1态射以外的态射发生的情况下、或者在用户无言的状态下经过了一定期间的情况下,迁移到回答句子X2 "向你介绍真正美味的店"。 Tasty ramen ah. "On the other hand, other than the first morphism morphisms the event, or in the user's status after silent under certain circumstances during the transits to a reply sentence X2" I introduce you to the real nice restaurant. "

在上述回答句子X3 "是么?拉面很美味啊"的输出后,仅规定了一个态射〈其它〉卜timer、通过任意输入句子(用户发话)或者经过一定期间而迁移到回答句子X2 "向你介绍真正美味的店"。 After the above-mentioned reply sentence X3 "is not it? Ramen was delicious, ah," the output, only specifies a morphism <other> BU timer, by any input sentence (user's utterance) or after a certain period while migrating to reply sentence X2 "I introduce you to a really nice restaurant. "

因为可以使用包含这样的态射的复合的会话情节的例子,所以在本发明中可以在尊重对方的发话的同时,引入想要坚持的自己的发话。 Because you can use the examples of such circumstances include the session morphisms complex, so in the present invention can respect each other's utterance while introducing want to stick to their own speaker.

此外,当将上述图7的会话情节作为回答列来表现时,成为图8那样的内容。 In addition, when the above-mentioned conversation scenario in Fig. 7 as an answer columns performance, as in Fig. 8 as content. 在此,^是X2的引用。 Here, ^ is a reference to X2. 被引用的1的引用源是X2,在形式上相当于在对象X1和X2之间决定了态射"(不喜欢)X3 (〈其它叫〈timer〉)"。 Cited reference source 1 is X2, equivalent to between objects in the form of X1 and X2 decided morphism "(not like) X3 (<other call <timer>)". 该态射是态射"不喜欢"和态射V其它〉Ktimer,的复合。 The morphisms are morphisms "dislike" and morphisms V Other> Ktimer, composite. (3 )可以定义单位元 (3) may be defined unit yuan

在本发明的会话情节中可以定义单位元。 In the conversation scenario of the present invention can be defined identity element. 所谓"单位元",是指不使对象迁移的态射。 The so-called "identity element" refers not to the object morphisms migration. 由于可以定义单位元,以下情况成为可能。 Because you can define the identity element, the following become possible.

(A) 可以对用户发话进行"强制回答"。 (A) of the user's utterance can be "forced to answer."

图9是表示进行强制回答的会话情节的例子的状态迁移图。 FIG. 9 is forced to answer a conversation scenario example of a state transition diagram. 在此例中,在回答句子X1 "我喜欢拉面。拉面才是美食的本质"的输出时,规定了附加了NULL的第1态射<其它>,不管是什么样的输入句子(用户发话)都忽枧输入句子地进行"我喜欢拉面。拉面才是美食的本质"的强制输出。 In this case, in reply sentence X1 "I like ramen. Ramen is the essence of food," the output, provides an additional NULL the first morphism of <Others>, no matter what kind of input sentence (user's utterance) are suddenly soap input sentence be carried out. "I like ramen. Ramen is the essence of food," the forced output. 另一方面,在回答句子X1 "我喜欢拉面。拉面才是美食的本质"的输出后,通过第2态射〈timeP而迁移到回答句子X2 "向你介绍真正美味的店"。 On the other hand, in the reply sentence X1 "I love nature Ramen Ramen is food," the output through the second Morphism <timeP migrate to reply sentence X2 "I introduce you to a really nice restaurant."

在此例中,将忽视对方的发话的情况标记为"NULL"。 In this case, the other party will ignore utterance labeled "NULL". 在图9所示的例子中,由于忽视全部发话,因此对<其它>附加了NULL,但也可以仅忽视"不喜欢"。 In the example shown in FIG. 9, due to neglect all of the speaker, so the <Others> attached NULL, but can be ignored only "do not like."

此外,在将上述图9的会话情节作为回答列来表现时,成为图10那样的内容。 In addition, when the above-mentioned conversation scenario in Fig. 9 as answer columns performance, as 10 such contents Fig. 回答列"XI (<其它>)乂1"中,"(<其它>)"是从X1到X1的态射,是单位元。 Answer column "XI (<other>) qe 1", "(<other>)" is the morphism from X1 to X1, is the identity element.

(B) 可以对用户发话进行"固执回答"。 (B) of the user's utterance can be "stubborn answer."

图ll是表示对用户发话进行"固执回答"的会话情节的例子的状态迁移图。 Figure ll shows the user utterance "stubborn answer" conversation scenario example of a state transition diagram. 在图ll的例子中,在回答句子X1 "拉面,喜欢?不喜欢?"的输出后,发生了第1态射"不喜欢"的情况下,迁移到回答句子X3 "是么?拉面很美味啊"。 In the example of Figure ll, after the reply sentence X1 "ramen, love? Do not like it?", This happened under the first morphism "do not like", the transits to a reply sentence X3 "is not it? Ramen was delicious ah. " 另一方面,在回答句子X1的输出后,发生了第2态射"喜欢"的情况下,迁移到回答句子X2 "向你介绍真正美味的店"。 On the other hand, the output of the reply sentence X1, the state took place under the second shot "like", the transits to a reply sentence X2 "I introduce you to a really nice restaurant." 另一方面,在回答句子XI的输出后,发生了第1以及第2态射以外的态射的情况下、或者在用户无言的状态下经过了一定期间的情况下,再次返回回答句子X1 "拉面喜欢?不喜欢?"。 On the other hand, the output of the reply sentence XI after the occurrence of a first down and the second case morphisms morphisms other than, or in the user silent state has gone on for a certain period, returning again to the reply sentence X1 "Ramen like? Do not like it?." 于是,可以强制性地使用户从"喜欢"或"不喜欢,,两项中选择一项。 Thus, users can be forced to make from "like" or "dislike ,, two select one.

此外,当将上述图11的会话情节作为回答列来进行表现时,成为图12那 In addition, when the conversation scenario described above 11 in reply to the performance of the column, it became 12

17样的内容。 17 kind of content. 在此,XI是X1的引用。 Here, XI is X1 references. 被引用的XL和引用源XI具有相同的迁 XL and references cited sources with the same move XI

移目的地。 Move destination. 这意味着X1和X1同形,这种情况下的态射"(〈其它〉卜timer〉)"也相当于从X1到XI的态射,因此称为单位元。 This means that the X1 and X1 are the same shape, morphisms in this case "(<Others> BU timer>)" also corresponds to the XI from X1 morphisms, so called identity element.

(C)通过"通过复合而构成的单位元,,可以构建"闭环回答"。 (C) by "unit element constituted by a composite can be constructed ,," closed loop answered. "

通过具备该特征,可以在闭合的循环(loop)中催促对方的发话。 By providing this feature, it can urge the other side of the speaker in a closed loop (loop) in. 图13是表示通过"通过复合而构成的单位元"构建了"闭环回答"的会话情节的例子的状态迁移图。 Figure 13 shows the "unit constituted by the composite element" to build a "closed loop reply" conversation scenario example of a state transition diagram. 在此例中,通过回答句子X1、 X2、 X3、 X4构建了闭环,通过该闭环可以控制会话的流程。 In this case, by reply sentences X1, X2, X3, X4 construct closed loop, the closed loop can be controlled by the flow of the session. 当将上述图13的会话情节作为回答列来表现时,成为图14那样的内容。 When the above-mentioned conversation scenario in Fig. 13 as the answer columns performance, as the contents of FIG. 14. 在此情况下也相当于从X1到XI的态射。 In this case, the equivalent of morphisms from X1 to XI of.

将(〈其它叫〈timer〉 ) X2 (〈其它〉卜timer〉) The (<Others called <timer>) X2 (<other> BU timer>)

X3 (〈其它〉l〈timer^) X4 (〈其它〉卜timer〉)称为单位元。 X3 (<other> l <timer ^) X4 (<other> BU timer>) called Unit yuan. 这种情况下的单位元构成了"闭环"。 Units in this case element constitutes a "closed loop."

以上,结束了"项目(3)可以定义单位元"的说明。 This concludes the "item (3) can be defined identity element," the instructions.

(4) 在态射的复合中,结合法则成立 (4) In the morphism composition, the binding rules of the establishment

通过该特征,对于与某态射对应的回答列S,可以构建沿不同的两条路径的回答列S1和S2,而且可以将它们作为等同的回答列来处理。 With this feature, the state radio and a corresponding answer column S, can be built along two different paths answer columns S1 and S2, and can answer them as identical columns to handle. 此时,当设S是与某问题相关的回答列时,Sl和S2是提供与S对应的不同解释的回答列,提供了与问题解决相关的信息。 At this time, when the set S is the answer to a question related to the columns, Sl and S2 are provided with different interpretations S corresponding answer column provides a problem-solving related information. 由于具有该特征,因此在本发明的会话情节中,可以应对有逻辑的用户发话 Because of this feature, a conversation scenario in the present invention, therefore, the logic can respond to the user's utterance

图15中表示在态射的复合中结合法则成立的会话情节的例子的状态迁移图。 Figure 15 shows binding conversation scenario holds for morphism composition in the example of the state transition diagram. 此外,当将上述图15的会话情节作为回答列来表现时,成为图16那样的内容。 In addition, when the above-mentioned conversation scenario in Fig. 15 as the answer columns performance, as the contents of FIG. 16. 在此,X1是X2的引用。 Here, X1 is X2 references. 在形式上,下式成立。 Formally, the following equation holds.

(提示)2£1 (是xx ) (〈其它〉l〈timer〉) (Tip) 2 £ 1 (a xx) (<other> l <timer>)

=(是xx ) X4 (〈其它〉l〈timer^) = (Y xx) X4 (<other> l <timer ^)

=(提示)X3 (〈其它〉l〈timer〉) = (Prompt) X3 (<other> l <timer>)

(5) 描绘交换图 (5) depicted in Fig exchange

通过该特征,可以定义用于到达任意对象的态射。 With this feature, you can define morphisms reach any object. 因此,可以在情节中设定目标,并且可以进行情节整体的掌握。 Therefore, you can set the target in the plot, and the whole episode can grasp.

(6) 其它本发明,在"可以以输入句子作为对象、以回答句子作为态射来处理的谈话的范围"内,4全索的结构完全不同,因此无法与"可以以输入句子作为态射、 以回答句子作为对象来处理的谈话的范围,,进行同样的处理。在本发明中不处理前者那样的谈话的范围。 (6) Other present invention, "can enter a sentence as an object, in order to answer the sentence as a morphism conversation to deal with the range", the 4 whole cable structure is completely different, and therefore can not and "can enter a sentence as a morphism to answer the sentence range as an object of conversation ,, subjected to the same treatment. In the present invention does not deal with the scope of the former kind of conversation.

(1.5.会话情节编辑装置的定位) 在此,关于本发明的会话情节编辑装置30的定位进行汇总。 (1.5. Conversation scenario editing device positioning) Here, on the present invention conversation scenario editing device 30 is positioned to summarize. (1)关于具有对象和态射的会话情节,可以列举出以下特征。 (1) With regard to the objects and morphisms having a conversation scenario can include the following features. •以回答句子为对象、以输入句子为态射(状态迁移) • In answer sentence for the object to the input sentence as morphisms (state transition)

-在尊重输入句子的同时,引入想要坚持的回答句子(维持文理:复合) •与输入句子无关地说定回答句子(强制回答:单位元) -反复催促对方,直到说出需要的发话为止(固执回答:单位元) •在闭合的循环中催促输入句子(闭环:单位元) •进行与解决问题相关的会话(解决问题:结合法则) •进行面向目标的会话(有目标的会话:交换图) - At the same time respect the input sentence, the sentence you want to adhere to the introduction of the answer (to maintain ARTS: Composite) • irrespective of the input sentence settle reply sentences (mandatory answer: Unit Yuan) - repeatedly urged each other to tell the speaker until the required date (stubborn Answer: Unit yuan) • urged in a closed loop input sentence (closed loop: Unit yuan) • conduct and resolve issues related to the session (to solve the problem: combining law) • be goal-oriented sessions (targeted sessions: exchange Figure)

此外,通过回答列也可以整理上述特征。 In addition, the column can also sort by answering the above characteristics. 会话情节编辑装置30具有通过回答列来表现上述会话情节的特征的功能。 Scenario editing session by answering a column 30 having the above-mentioned conversation scenario to show the characteristics of functional devices.

通过利用上述会话情节,会话服务器20进行简单的检索即可。 By using the above conversation scenario, the session server 20 simple search can be. 即,会话服务器将当前的状态作为会话情节的对象(回答句子)来掌握,.在发生了利用者发话的情况下,会话服务器20在进行意思解析的同时检索最佳的态射(输入句子),接下来的状态成为与检索到的态射(,输入句子)对应的对象(回答句子)。 That is, the current status of the session server as a session plot objects (reply sentence) to master, in the event of a case of the use by the next speaker, the conversation server 20 during the same time to retrieve the best means to resolve the morphism (input sentence) , the next state becomes the retrieved morphism (input sentence) corresponds to an object (reply sentence).

此外,上述的会话情节不仅作为状态迁移图或以此为基础的数据(图6、 8、 IO等)来表现,也可以使用图17所示那样的大纲编辑器(outline editor) 那样的GUI来生成、编辑。 In addition, the above-mentioned conversation scenario not only as a state transition diagram or as a basis for the data (Fig. 6, 8, IO, etc.) to show what they can be used as shown in Figure 17 outline editor (outline editor) as a GUI generate, edit.

(2.会话情节生成装置的动作例) 接着,对上述会话情节编辑装置30的动作例进行说明。 (2 conversation scenario generating apparatus operation example) Next, operation of the above-described embodiment the conversation scenario editing device 30 will be described. 本实施方式的会话情节编辑装置30,关于多个不同主题(会话的主题), 可以确立与用户的会话。 This embodiment conversation scenario editing device 30, on a number of different topics (themes session), it can be established with the user's session. 图18是表示会话情节保持部33 及会话情节存储部22 (以下简称为会话情节保持部33)存储的会话情节的数据结构例的图。 18 is a conversation scenario retention unit 33 and the conversation scenario memory 22 (hereinafter referred to as conversation scenario retention unit 33) storing a data structure example of a conversation scenario in Fig. 会话情节保持部33可以针对每个与谈话的范围或主题(会话主题)201 对应的域200保持独立的会话情节数据。 Conversation scenario retention unit 33 can remain independent sessions plot data for each domain and range or topics of conversation (conversation topics) 201 corresponding to 200. 例如,可以具有与"天气,,域和"咖啡豆"域分别相关的会话情节数据,当用户进行与天气相关的发话时,会话服务器20、更详细来说是回答处理部21,将作为"天气,,域的会话情节数据优先,搜索与输入句子(也称用户发话)对应的回答句子(也称系统发话),输出对用户发话进行应答的系统发话。 For example, you can have the "Weather ,, domain and 'coffee beans' domains are related to the conversation scenario data, when weather-related user utterance, the conversation server 20, and more is answered processing unit 21, will serve as" Weather ,, conversation scenario domain data priority, search and input sentence (also known as the user utterance) reply sentence (also known as system utterance), the output of the user utterance response system speaker. 另一方面,当用于进行与"咖啡豆"相关的发话时,回答处理部21将作为"咖啡豆"域的会话情节数据优先,搜索与用户发话对应的系统发话,输出对用户发话进行应答的系统发话。 On the other hand, when used for the "coffee beans" related to the speaker, the reply process unit 21 as the conversation scenario "beans" field data prioritization, search and user utterance of the speaker system, the output of the user utterance reply The system speaker.

各域200具有:用户发话句子(相当于输入句子,以下相同)210和、作为与用户发话句子对应的自动会话系统的回答而准备的系统发话句子220。 The fields 200: a user's utterance sentences (equivalent to the input sentence, the same applies hereinafter) 210 and, as with the user's utterance sentence corresponding automatic conversation system answers prepared system's utterance sentence 220. 在图18所示的例子中,记录了用户发话句子210-1和与其关联的系统发话句子220-1,同时记录了在设想用户应答该系统发话220-1而发话时的用户发话句子210-2,记录了作为与该用户发话句子210-2对应的自动会话系统的回答而准备的系统发话句子220-2。 In the example shown in Figure 18, the system records the user's utterance sentence 210-1 and 220-1 associated uttered sentence, while recording the response of the system envisaged the user utterance and utterance 220-1 when the user utterance sentence 210- 2, recording the system as the user's utterance sentence 210-2 corresponding automatic conversation system answers prepared utterance sentence 220-2.

例如,上述会话情节成为以下这样的用户和系统的会话。 For example, the above-mentioned conversation scenario has become less so user and system sessions.

用户发话句子210-1:"好天气啊" User utterance sentence 210-1: "Good weather ah"

系统发话句子220-1:"喜欢好天气吗?" System utterance sentence 220-1: "Like good weather it?"

用户发话句子210-2:"是的,喜欢啊" User utterance sentence 210-2: "Yes, like ah."

系统发话句子220-2:"不喜欢雨天吗?" System utterance sentence 220-2: "Do not you like the rain?"

图18所示的会话情节表示了最简单的形态的会话情节。 Conversation scenario shown in Figure 18 shows a plot of the easiest conversation form. 在本自动会话系统可以处理的会话情节中,为了可以应对用户对同一系统发话做出不同的反应而返回用户发话的情况,也可以对一个系统发话句子准备多个用户发话句子。 In the conversation scenario automatic conversation system can handle, in order to cope with the user on the same system can react differently to the speaker and return the user utterance to be a system for preparing a plurality of user's utterance sentence uttered sentences.

会话情节编辑装置30具有以下功能:生成由在会话情节保持部33中存储的新的域200、和该域200用户发话句子210、系统发话句子220构成的会话情节数据,使其存储在会话情节保持部33中。 Conversation scenario editing device 30 has the following functions: generation by a new domain in conversation scenario retention unit stores 33 200, and 200 users of the domain utterance sentence 210, the system utterance sentence 220. The conversation scenario data to make it in the conversation scenario memory holding section 33. (3.会话情节的输入例) (Example 3. Enter a conversation scenario)

接着,说明会话情节的输入例。 Next, enter a conversation scenario example. 图19到图23是表示关于某域200输入了会话情节的情况下的输入画面的迁移的一例的图。 19 to FIG. 23 is entered on a field 200 showing an example of the input screen when the session under the circumstances of migration.

图19表示通过会话情节编辑装置30生成的输入界面画面的一例。 Figure 19 shows an example of an input interface screen 30 generated by the conversation scenario editing device. 在此,以域200是关于"咖啡豆',的域来进行说明。 In this case, the domain 200 is about "coffee beans' domains will be described.

会话情节编辑装置30、更详细来说是编辑部32,生成成为输入界面的窗口300,使输出部34对其进行显示。 Conversation scenario editing device 30, is more detailed editing unit 32 generates an input interface window 300, the output unit 34 to display it. 在窗口300中设置了显示区域301,通过由用户操作输入部31,在此输入用户发话句子以及系统发话句子。 Setting the display area 301 in window 300, operated by the user through the input unit 31, enter the user's utterance sentence and a system's utterance sentence. 在图19的例子中显示了域名302,是等待接受该域200中存储的会话情节的输入的状态。 Figure 19 shows an example of the domain 302, is waiting to receive state 200 in the field of conversational episodes of inputs.

图20是输入了作为该域200中存储的会话情节的开始的用户发话句子401 的状态的画面例。 Figure 20 is a user begins to enter the domain 200 stores as uttered sentence conversation scenario example screen 401 of the state.

在实际执行自动会话的情况下,会话服务器20的回答处理部21,当用户发话与在此记载的用户发话句子401 "关于咖啡豆" 一致、或者是可以视为与其相同的发话内容时,作为提取对用户发话进行应答的系统发话句子的域200,从会话情节存储部22中选择将域名303设为"咖啡豆,,的域200,并使该域200优先来选择系统发话句子。 In the case of actual implementation of the automatic session, the reply process unit 21 of the conversation server 20, when the user's utterance described herein with the user utterance sentence 401 "About coffee beans" consistent, or can be regarded as the same utterance with its content, as respond to the user's utterance extraction system utterance sentence domain 200, the conversation scenario storage unit 22 is selected from the domain name 303 is set to "beans ,, domain 200, and the priority field 200 to select the system's utterance sentence.

作为会话情节的输入者的用户,进行与上述用户发话句子401对应的回答、即系统发话句子的输入。 As the conversation scenario input by the user, those performed by the user's utterance sentence 401 corresponding answer, the system uttered sentence input. 图21表示通过用户输入了针对用户发话句子401 "关于咖啡豆"的系统发话句子501的状态的窗口300的显示例。 Figure 21 shows the input by the user against a user's utterance sentence 401 "on the beans," the system's utterance sentence 501 status window shows 300 cases. 在此例中假定记载了以下会话情节:针对"关于咖啡豆,,这样的用户发话句子401,自动会话系统发出作为"关于味道的特征进行回答""摩卡"、"蓝山"、"乞力马扎罗,,中,想知道哪个? In this example it is assumed that the conversation scenario described in the following: for the "coffee beans ,, on this user's utterance sentence 401, the system sends automatic conversation as" on the taste characteristics answer "," mocha "," Blue Mountain "," kilima Mazzarò, Medium, want to know what? "这样的疑问的情节回答句子501。 "This question of plot reply sentence 501.

接着,作为会话情节的输入者的用户,针对上述情节回答句子501输入预想的用户发话句子。 Then, as the session entered the user who plot against the above circumstances the reply sentence 501 input expected user's utterance sentence. 图22表示针对所述情节回答句子501输入了预想的用户发话句子601的状态的窗口300的显示例。 Figure 22 shows the plot for the reply sentence 501 input expected user's utterance sentence 601 status window shows 300 cases. 在此例中,假定针对"关于味道的特征进行回答。"摩卡"、"蓝山"、"乞力马扎罗"中,想知道哪个?"这样的系统发话句子501,预想用户进行了"蓝山"这样的回答,由用户输入了用户发话句子601 "蓝山"。 In this example, it is assumed for the "answer on taste characteristics." Mocha "," Blue Mountain "," Kilimanjaro ", the want to know what?" Such a system's utterance sentence 501, expected users of the "blue Mountain "this answer, the user enters the user utterance sentence 601" Blue Mountains. "

接着,作为会话情节的输入者的用户,输入与上述用户发话句子601对应的系统发话句子。 Then, as the conversation scenario importer user input and said user's utterance sentence 601 system corresponding uttered sentence. 图23表示输入了与所述用户发话句子601对应的系统发话句子701的状态的窗口300的显示例。 Figure 23 shows the input of the user's utterance sentence uttered sentence 601 701 corresponding system status window shows 300 cases. 会话情节的输入者输入系统发话701 来作为用户发话句子601的回答。 Conversation scenario utterance input to input system 701 to 601 as a user utterance sentence answer.

通过这样的会话情节,自动会话系统在用户想知道咖啡豆蓝山时,可以返 Through this conversation scenario, automatic conversation system when the user wants to know Blue Mountain coffee beans, you can return

21回其回答。 21 back to their answer. 此外,从此以后,会话情节的输入者可以继续输入用户发话句子、 系统发话句子,以使用户与自动会话系统的会话持续。 Moreover, since then, a conversation scenario can continue to enter the user input utterance sentence, the system's utterance sentence, so that the user's session with the automatic conversation system continued.

如上述那样输入的会话情节(用户发话句子和系统发话句子的集合),通过编辑部32被写入会话情节保持部33并存储。 As the conversation scenario described above as inputs (user's utterance sentences and uttered sentence collection system), it is written by the editorial department 32 conversation scenario retention unit 33 and stored. 该会话情节被转移到会话服务 The conversation scenario is transferred to the session service

器20的会话情节存储部22。 20 conversation scenario storage unit 22. 此外,在被转移到会话情节存储部22的情况下, 也可以进行会话情节的变换、移植,以便成为与会话服务器20适应的会话情节。 In addition, in the conversation scenario is transferred to the storage unit case 22 to be carried out the conversation scenario transformation, migration, and so as to be adapted to the conversation server 20. The conversation scenario.

会话服务器20的回答处理部21,也参照会话情节存储部22中存储的新的会话情节,以便可以输出与用户发话对应的情节回答。 Reply process unit 21 of the conversation server 20, also referring to the conversation scenario storage unit 22 new conversation scenario stored so that it can output a user utterance plot answer. (3.变形例) (3. Modification)

本实施方式即使如下这样变形也成立。 The present embodiment is also true even if modified as follows. (1)会话情节编辑装置的变形例 (1) a modification of the conversation scenario editing device

图24是变形例涉及的会话情节编辑装置30X的功能框图。 FIG 24 is a modified example of the conversation scenario editing device 30X functional block diagram. 会话情节编辑装置30X基本上具有与前面所述的会话情节编辑装置30相同的结构,不同点是具有与会话情节保持部33相连的动态知识生成部36。 Conversation scenario editing device 30X has substantially the same structure as the front 30 and the conversation scenario editing device, it differs holding portion 33 having a dynamic knowledge generating unit connected with the conversation scenario 36. 此外,关于相同的结构要素赋予相同的参照符号,省略它们的说明。 In addition, with regard to the same components are given the same reference numerals, and their explanation is omitted.

动态知识生成部35具有根据在会话情节保持部33中存储的会话情节40 生成动态知识40X的功能。 Dynamic knowledge generation section 35 has based on the conversation scenario in the storage unit 33 functions to keep the conversation scenario 40 to generate dynamic knowledge to 40X. 动态知识40X是为了使会话服务器20可以更高速且高效率地检索作为态射的输入句子以及作为其对象的回答句子,而根据作为回答列的会话情节40再构成的数据。 Dynamic knowledge 40X is to make the conversation server 20 may be faster and efficiently retrieving Morphisms input sentence and answer sentence as its object, according to the data session as circumstances answer column 40 reconstituted.

根据该变形例,可以降低会话服务器20的处理负荷,可以进行高速的回答句子的答复。 According to this modification, it can reduce the processing load of the conversation server 20 may be made of the response speed of the reply sentence.

(4.会话服务器的结构的另一个例子) (Another example 4. The session server structure)

本发明的会话服务器20、回答处理部21即使采用下述的结构,也可以实现本发明。 Session server 20 of the invention, even if the reply process unit 21 uses the following structure, the present invention can also be implemented. 以下描述会话服务器20、更详细而言是回答处理部21的结构例。 The following description of the session server 20, and more are answered in terms of a configuration example processor 21. 图25是回答处理部21的扩大框图,是表示会话控制部300以及句子解析部400的具体结构例的框图。 FIG 25 is a block diagram illustrating the processing unit 21 to expand the answer is a block diagram showing a specific configuration example of the conversation control unit 300 and a sentence analyzing unit 400. 回答处理部21具有会话控制部300、句子解析部400和会话数据库500。 Reply process unit 21 has a conversation control unit 300, the sentence analyzing unit 400 and the conversation database 500. 会话数据库500具有存储会话情节40或动态知识40X 的功能。 Conversation database 500 has stored the conversation scenario 40 or dynamic knowledge 40X features. (4丄4.句子解析部) (4 Shang The sentence analyzing unit)

接着,参照图25说明句子解析部400的结构例。 Next, with reference to FIG. 25 illustrates a configuration example of the sentence analyzing unit 400.

句子解析部400解析通过输入部100或声音识别部200确定的字符串。 Parsing the sentence analyzing unit 400 determines through the input unit 100 or the voice recognition unit 200 a string. 该句子解析部400在本实施方式中如图25所示,具有字符串确定部410、形态素才是取部420、形态素数据库430、输入种类判定部440和发话种类数据库450。 The sentence analyzing unit 400 in the present embodiment, as shown in Figure 25, with the string determining portion 410, is taken morpheme unit 420, a morpheme database 430, an input type determining unit 440 and the utterance type database 450. 字符串确定部410按照每一个短语来划分通过输入部100以及声音识别部200 确定的一连串的字符串。 String determining portion 410 according to each of a series of phrases divided string input portion 100 and a voice recognition unit 200 determines through. 所谓一个短语,表示以不名皮坏文法的意思的程度尽可能细分字符串而得到的一个划分句子。 Called a phrase that represents a division sentence to the extent possible broken leather string does not name the meaning of bad grammar obtained. 具体而言,字符串确定部410,在一连串字符串中有某一定以上的时间间隔时,在该部分划分字符串。 Specifically, the character string specifying unit 410, when there is more than a certain time interval, which in part is divided in a series string string. 字符串确定部410将该划分而得的各字符串输出到形态素提取部420以及输入种类判定部440。 String specifying unit 410 obtained by dividing the output of each string to the morpheme extracting unit 420 and the input type determining unit 440. 此外,以下说明的"字符串,,表示每一个短语的字符串。 In addition, each phrase string "String described below ,, representation.

(4丄4丄形态素提取部) (4 Shang Shang morpheme extracting unit 4)

形态素提取部420,根据通过字符串确定部410划分而得的一个短语的字符串,从该一个短语的字符串中提取出构成字符串的最小单位的各形态素,作为第一形态信息。 Morpheme extracting unit 420, based on a string by string determining portion 410 obtained by dividing a phrase extracted from a string that a phrase in the smallest unit of morphemes constituting the string, as the first form of information. 在此,所谓形态素,在本实施方式中表示字符串中出现的词结构的最小单位。 Here, the morpheme, represents the smallest unit of a string of words that appear in the structure of the present embodiment. 作为该词结构的最小单位,列举出例如名询、形容词、动词等词类。 As the smallest unit of word structure, such as the name of inquiry include, adjectives, verbs and other parts of speech.

各形态素如图26所示,在本实施方式中可以表现为ml、 m2、 m3、...。 Morphemes shown in Figure 26, in this embodiment can be expressed as ml, m2, m3, .... 图26是表示字符串和从该字符串中提取出的形态素的关系的图。 Figure 26 shows strings and extracted from the morpheme string of relationships. 如图26所示, 从字符串确定部410输入了字符串的形态素提取部420,将该输入的字符串.、 和在形态素数据库430中预先存储的形态素群(该形态素群,关于属于各个词类的各形态素,准备了记载有该形态素的词条•读法*词类•活用形等的形态素词典)进行对照。 As shown in Figure 26, enter the form of strings from a string specifying unit 410 pixel extraction unit 420, the input string., And in the morpheme database 430 previously stored morpheme groups (the morpheme group, About morpheme belonging to each individual part of speech, we have prepared a record entry of the morpheme * Parts of Speech • Reading • conjugation morpheme dictionary, etc.) were compared. 进行了该对照的形态素提取部420,从该字符串中提取出与预先存储的形态素群的某个一致的各形态素(ml、 m2、...)。 The controls carried out morpheme extracting unit 420, extracted morphemes consistent with a previously stored morpheme groups from the string (ml, m2, ...). 除了该提取出的各形态素的要素(nl、 n2、 n3、...),列举出例如助动词等。 In addition to the extracted morphemes elements (nl, n2, n3, ...), and examples thereof include auxiliary verbs like.

该形态素提取部420将提取出的各形态素作为第一形态素信息而输出到话题确定信息检索部320。 The morpheme extracting unit 420 will be extracted morphemes as the first morpheme information output to the topic specification information retrieval unit 320. 此外,第一形态素信素不需要被构造化。 In addition, the first morpheme letter need not be constructed of elements. 在此所谓"构造化",是指根据词类等对字符串中包含的形态素分类排列,例如是指将作为发话句子的字符串变换为像"主语+宾语+谓语"等那样按照预定的顺序 In this so-called "structure" and refers to the parts of speech and other factors classification are ranked according to the form contained in a string, for example, it refers to as the uttered sentence is converted to a string such as "subject + object + predicate" and that in a predetermined order

23排列形态素而形成的数据。 23 are arranged to form a morpheme data. 当然,即使使用构造化后的第一形态素信息,也不会妨碍实现本实施方式。 Of course, even with the construction of the first morpheme information after, it does not prevent the implementation of the present embodiment.

(4丄4.2.输入种类判定部) (4 Shang 4.2 input type determining unit)

输入种类判定部440,根据通过字符串确定部410确定的字符串,判定发话内容的种类(发话种类)。 Input type determining unit 440, based on a string by the character string determining portion 410 determines determines uttered contents type (utterance type). 该发话种类是确定发话内容的种类的信息,在本实施方式中例如表示图27所示的"发话句子的类型"。 The utterance type is the type of information to determine the content of the utterance, in the present embodiment, for example, as shown in FIG. 27 represents "uttered sentence type." 图27是表示"发话句子的类型"、表示该发话句子的类型的两字符的字母、以及与该发话句子的类型对应的发话句子的例子的图。 27 is a "uttered sentence type" means the two-character alpha-uttered sentence type, and the type of the uttered sentence uttered sentence examples corresponding to Fig.

在此,"发话句子的类型"在本实施方式中如图27所示,由陈述句(D; Declaration)、时间句(T; Time )、场所句(L; Location )、否定句(N; Negation) 等构成。 Here, the "uttered sentence types" as shown in this embodiment 27, the declarative sentences (D; Declaration), the time period (T; Time), place the sentence (L; Location), negative sentence (N; Negation ) and the like. 该由各类型构成的句子由肯定句或疑问句构成。 The configuration of each type of sentence consists affirmative or interrogative sentences. 所谓"陈述句",表示表达利用者的意见或想法的句子。 The so-called "declarative", said the use of the expression of the views or ideas of a sentence. 该陈述句,在本实施方式中如图27所示, 举出例如"我喜欢佐藤"等句子。 The declarative sentence, as in the present embodiment 27, include, for example, "I like Sato" and other sentences. 所谓"场所句",表示伴随场所的概念的句子。 The so-called "place sentence" means a place associated with the concept of the sentence. 所谓"时间句",表示伴随时间的概念的句子。 The so-called "time sentence" means a sentence accompanied by the concept of time. 所谓"否定句",表示对陈述句进行否定时的句子。 The so-called "negative sentence" means a sentence to be negative when declarative sentence. 关于"发话句子的类型"的例句,如图27所示那样。 About the "uttered sentence type" of sentences, as shown in Figure 27.

为了由输入种类判定部440判定"发话句子的类型",输入种类判定部440 在本实施方式中如图28所示,使用用于判定是陈述句的定义表现辞典、用于判定是否定句的否定表现辞典等。 For 440 determines the "uttered sentence type" by the input type determining unit, input type determining unit 440 in the present embodiment shown in Figure 28, using a declarative definitions for determining performance dictionaries for determining the Negative Sentences dictionaries and other performance. 具体来说,从字符串确定部410输入了字符串的输入种类判定部440,根据所输入的字符串,将该字符串与发话种类数据库450中存储的各辞典进行对照。 Specifically, from the character string specifying unit 410 inputs the character string input type determination unit 440, based on the input character string, the string with the utterance type database 450 stored in each dictionary were compared. 进行了该对照的输入种类判定部440,从该字符串中提取出与各辞典有关的要素。 Carried out the control input type determining unit 440, the extracted elements associated with each dictionary from the string.

该输入种类判定部440根据所提取出的要素,判定"发话句子的类型"。 The input type determining unit 440 according to the extracted feature mentioned, it is determined "uttered sentence type." 例如,输入种类判定部440,在字符串中包含关于某现象进行了陈述的要素时, 将包含该要素的字符串判定为陈述句。 For example, the input type determining unit 440, when included in the string elements about a phenomena statement, the string will contain the elements determined to be declarative sentence. 输入种类判定部440将判定出的"发话句子的类型"输出到回答取得部380。 "Uttered sentence type" input type determining unit 440 determines that the output 380 to the reply retrieval unit.

(4丄5.会话数据库) (4 Shang 5. Session database)

接着,参照图29说明会话数据库500存储的数据的数据结构例。 Next, with reference to FIG. 29 illustrates the data session configuration example of data stored in the database 500. 图29是表示会话数据库500存储的数据的结构例的概念图。 FIG 29 is a conceptual diagram showing the structure of an example of the session database 500 to store the data.

会话数据库500如图29所示,预先存储了多个用于确定话的话题确定信息810。 Session database 500 shown in FIG. 29, stored in advance for determining a plurality of words of the topic specification information 810. 另外,各个话题确定信息810可以与其它话题确定信息810相关联, 例如在图29所示的例子中,当确定话题确定信息C (810)时,确定与该话题确定信息C( 810)相关联的其它话题确定信息A( 810 )、话题确定信息B( 810 )、 话题确定信息D (810)来存储。 Furthermore, each topic specification information 810 is associated with other information 810 may determine the subject, e.g., in the example shown in FIG. 29, when it is determined topic specification information C (810), it is determined with the topic specification information C (810) associated The other topic specification information A (810), the topic specification information B (810), the topic specification information D (810) to store.

具体来说,话题确定信息810,在本实施方式中表示与从利用者输入时预想的输入内容、或者向利用者的回答句子具有关联性的"关键字"。 Specifically, the topic specification information 810 indicates input from the user and the expected type in the present embodiment, or to take advantage of those who answered the sentence has relevance "keyword."

与话题确定信息810对应地存储了一个或多个话题标题820。 With the topic specification information 810 stored in correspondence to a topic title 820 or more. 话题标题820 通过由一个字符、多个字符串或它们的组合组成的形态素构成。 Topic title 820 form by a single character, multiple strings or combinations thereof constitute prime. 与各话题标题820对应地存储了向利用者的回答句子830。 Correspondence with the topic title 820 stores the user's answer to sentence 830. 另外,表示回答句子830的种类的多个回答种类与回答句子830相对应。 In addition, it indicates the type of sentence answer multiple answer 830 species and 830 corresponding to the reply sentence.

接着,说明某话题确定信息810和其它话题确定信息810的关联。 Next, a topic specification information 810 and other topics related information to determine 810. 图30 是表示某话题确定信息810A和其它话题确定信息810B、 810Q~810C4、 810D广810D3…的关联的图。 FIG. 30 is a topic specification information 810A and other topic specification information 810B, 810Q ~ 810C4, 810D wide association 810D3 ... Fig. 此外,在以下的说明中,所谓"相关联地存储" 是指当读取某信息X时可以读取与该信息X关联的信息Y。 In the following description, the term "stored in association" is one that can read information associated with the information X when a message is read X Y. 例如,把在信息X的数据中存储了用于读出信息Y的信息(例如表示信息Y的存储目的地地址的指针、信息Y的存储目的地物理存储器地址、逻辑地址等)的状态,设为"信息Y与信息X "相关联地存储""。 For example, the stored information is used to read out the information in the information X, Y data (for example, pointer information indicating a storage destination address Y, storage destination information Y physical memory address, the logical address, etc.) state, set "Information and information Y X" is stored in association "."

在图30所示的例子中,话题确定信息,在与其它话题确定信息之间,可以将上位概念、下位概念、同义词、反义词(本图的例子中省略)相关联地存储。 In the example shown in Figure 30, the topic specification information (not the example of the figure) is determined between the information can be superordinate concepts, subordinate concept, synonyms, antonyms and other topics related to storage. 在本图所示的例子中,作为与话题确定信息810A (="电影")对应的上位概念的话题确定信息,与话题确定信息810A相关联地存储了话题确定信息810B (="娱乐"),例如相对于话题确定信息("电影")被存储在上一层次中。 In the example shown in this figure, as with the topic specification information 810A (= "movie") corresponding to the upper concept topic specification information 810A and the topic specification information stored in association with the topic specification information 810B (= "entertainment") , e.g., with respect to the topic specifying information ("movie") is stored in a hierarchy.

另外,与话题确定信息810A相关联地存储了与话题确定信息810A (= "电影")对应的下位概念的话题确定信息810C,(="导演")、话题确定信息810C2 (="主演")、话题确定信息810C3 (="发行公司")、话题确定信息810C4 (="上映时间")以及话题确定信息810D,(="七武士")、话题确定信息81002(="乱,,)、话题确定信息810D3 (="保镖")、…。 Further, with the topic specification information 810A stored in association with the topic specification information 810A (= "movie") corresponding to the lower-level concepts of the topic specification information 810C, (= "director"), the topic specification information 810C2 (= "starring") , topic specification information 810C3 (= "issuing company"), the topic specification information 810C4 (= "release time") and the topic specification information 810D, (= "Seven Samurai"), topic specification information 81002 (= "chaos ,,), topic specification information 810D3 (= "bodyguard"), ....

另外,与话题确定信息810A关联了同义词900。 In addition, with the topic specification information 810A related synonymous 900. 在此例中,表示作为话题确定信息810A即关键字"电影"的同义词,而存储了"作品"、"内容"、"影 In this case, it represents a topic specification information 810A that is the keyword "movie" synonymous, and storage of the "work", "content", "Shadow

院"的情况。通过确定这样的同义词,在发话中未包含关键字"电影",但在发话句子等中包含"作品"、"内容"、"影院,,的情况下,可以认为在发话句子 Hospital, "the situation by determining such a synonym, in the speaker is not included in the keyword" movie ", but included in the uttered sentence like" work "," content "," Cinema ,, circumstances, it is considered that utterance sentence

等中包含话题确定信息81 OA来进行处理。 And other topic specification information included 81 OA for processing.

回答处理部21通过参照会话数据库500的存储内容,当确定某个话题确定信息810时,可以高速地检索.提取与该话题确定信息810相关联地存储的其它话题确定信息810以及该话题确定信息810的话题标题820、回答句子830 等。 Reply process unit 21 by referring to the session database 500 to store the contents, when it is determined a certain topic specification information 810, can be retrieved at high speed. Extracts specification information 810 stored in association with the topic specification information 810 and the other topic specification information to the topic 810 topic title 820, 830 and so the answer sentence.

接着,参照图31说明话题标题820 (也称为"第二形态素信息")的数据结构例。 Next, with reference to FIG. 31 described topic titles 820 (also referred to as "second morpheme information") data structure embodiment. 图31是表示话题标题820的数据结构例的图。 FIG 31 is a diagram showing an example of the data structure of topic title 820.

话题确定信息810D,、 810D2、 810D3、...分别具有多个不同的话题标题 Topic specification information 810D ,, 810D2, 810D3, ... each having a plurality of different topic title

820,、 8202.....话题标题8203 、 8204、...、话题标题8205、 8206、…。 820 ,, 8202 ..... topic headings 8203, 8204, ..., subject headings 8205, 8206, .... 在本 In this

实施方式中,如图31所示,各个话题标题820是由第一确定信息1001、第二确定信息1002和第三确定信息1003构成的信息。 Embodiment, shown in Figure 31, each of the topic titles 820 is the information from the first specification information 1001, second specification information 1002 and third specification information 1003 composed. 在此,第一确定信息IOOI, 在本实施方式中,表示构成话题的主要的形态素。 Here, the first specification information IOOI, in the present embodiment, showing the configuration of the main subject morpheme. 作为第一确定信息1001的例子举出例如构成句子的主语。 As a first example of determining information 1001 include, for example subject constitute sentence. 另外,第二确定信息1002,在本实施方式中表示与第一确定信息1001具有密切的关联性的形态素。 In addition, the second specification information 1002, in the present embodiment, the first specification information 1001 shows a close correlation with the morpheme. 该第二确定信息1002 举出例如宾语。 The second specification information 1002 include e.g. object. 而且,第三确定信息1003,在本实施方式中,表示表示关于某对象的动作的形态素、或修饰名词等的形态素。 Also, the third specification information 1003, in this embodiment, expressed her action on an object morpheme, or modifying a noun like morpheme. 该第三确定信息1003,举出例如动词、副词或形容词。 The third specification information 1003 includes, for example verbs, adverbs or adjectives. 此外,第一确定信息IOOI、第二确定信息1002、 第三确定信息1003各自的意思没必要限定于上述内容,即使对第一确定信息1001、第二确定信息1002、第三确定信息1003赋予别的意思(别的词类), 只要根据它们可以掌握句子的内容,本实施方式就成立。 Further, the first specifying information IOOI, the second specification information 1002 and third specification information 1003 is not necessary to define their meaning to the above, even if the first specification information 1001, second specification information 1002 and third specification information 1003 do not impart It means (other parts of speech), so long as they can grasp the contents according to the sentence, on the establishment of the present embodiment.

例如,在主语为"七武士"、形容词为"有趣"的情况下,如图31所示, 话题标题(第二形态素信息)8202由作为第一确定信息1001的形态素"七武士,,和作为第三确定信息1003的形态素"有趣"构成。此外,在该话题标题8202中不包含与第二确定信息1002对应的形态素,作为第二确定信息1002而存储了用于表示没有相应的形态素的记号"*,,。 For example, in the case of the subject of "The Seven Samurai", the adjective is "interesting", shown in Figure 31, the topic title (second morpheme information) 8202 is determined by the information 1001 as the first morpheme "Seven Samurai ,, and as a third specification information 1003 morpheme "interesting" constituted. In addition, the topic title 8202 is not included in the second specification information 1002 and corresponding morpheme, as the second specification information 1002 for indicating no stored corresponding morpheme the symbol "* ,,.

此外,该话题标题8202 (七武士; *;有趣),有"七武士有趣"的意思。 In addition, the topic title 8202 (Seven Samurai; *; interesting), the "Seven Samurai interesting" means. 在构成该话题标题820的括号内,以下成为从左起第一确定信息1001、第二确定信息1002、第三确定信息1003的顺序。 In parenthesis constituting the topic title 820, the following information is determined to become the first from the left 1001, the second specification information 1002 and third specification information 1003 in order. 另外,在话题标题820中没有从第一到第三确定信息中包含的形态素时,关于该部分示为"*,,。 Further, when not included in the topic title 820 from the first to third specification information of morpheme, on the portion shown as "* ,,.

此外,构成上述话题标题820的确定信息不像上述第一到第三确定信息那样限定于三个,例如也可以进一步具有其它确定信息(第四确定信息、以及第四以上)。 In addition, specification information constituting the topic title 820 unlike the above-described first to third specification information as limited to three, for example, may further have other specification information (fourth specification information, and a fourth above).

接着,参照图32说明回答句子830。 Next, with reference to FIG. 32 illustrates reply sentence 830. 回答句子830如图32所示,在本实 Reply sentence 830 as shown in FIG. 32, in the present

述(D; Declaration )、时间(T; Time)、场所(L; Location),否定(N; Negation) 等类型(回答种类),针对每个类型准备回答句子。 Said (D; Declaration), time (T; Time), places (L; Location), negative (N; Negation) and other types (answer types), be prepared to answer for each type of sentence. 另外,设肯定句为"A", 设疑问句为"Q"。 In addition, the set affirmative sentence is "A", set questions for the "Q".

参照图33说明话题确定信息810的数据结构例。 Referring to FIG. 33 illustrates the data structure of the topic specification information 810 of the embodiment. 图33表示与某个话题确定信息810 "佐藤"对应的话题标题820、回答句子830的具体例。 Figure 33 shows a topic determining information 810 "Sato" corresponds to the topic titles 820, 830 sentence answer specific examples.

与话题确定信息810 "佐藤"对应了多个话题标题(820) 1-1、 1-2、...。 The topic specification information 810 "Sato" corresponds to a number of topic titles (820) 1-1, 1-2, .... 与各个话题标题(820) 1-1、 1-2、…对应地存储了回答句子(830) 1-1、 1-2、…。 And each topic title (820) 1-1, 1-2, ... stored in correspondence with the reply sentences (830) 1-1, 1-2, .... 针对每个回答种类840而准备了回答句子830。 For each answer 840 species and 830 ready to answer sentence.

当话题标题(820) 1-1为(佐藤;*;喜欢){这是提取出"喜欢佐藤"中包含的形态素而得的话题标题}时,与该话题标题(820) ll对应的回答句子(830) 1-1列举出(DA;陈述肯定句"我也喜欢佐藤")、(TA;时间肯定句"我喜欢击球时的佐藤,,)等。后述的回答取得部380, —边参照输入种类判定部440的输出, 一边取得与该话题标题820对应的一个回答句子830。 When the topic title (820) 1-1 (Sato; *; like) {This is the time to extract the form "like Sato" included elements derived from the topic title}, and the topic title (820) corresponding to the answer ll sentences (830) 1-1 include (DA; statements affirmative sentence "I like Sato"), (TA; time affirmative sentence "I like to hit a ball Sato ,,). and answered later acquisition unit 380, - edge reference input type determining output unit 440, while 820 made the topic title 830 corresponding to a reply sentence.

在各回答句子中,与该回答句子对应地确定了指定针对用户发话优先输出的回答句子(称为"下一回答句子")的信息、即下一计划指定信息840。 Each answer sentence, with the determined reply sentence corresponding to a user utterance precedence specified for the output of a reply sentence (referred to as "the next reply sentence") information, i.e., the next-plan designation information 840. 下一计划指定信息840,若是可以确定下一回答句子的信息,则可以是任何信息, 例如是可以从会话数据库500中存储的全部回答句子中确定至少一个回答句子的回答句子ID等。 Next-plan designation information 840, if the next may be determined reply sentence information may be any information, for example, from the session database 500 can store all of at least one of the answer sentences answering sentence determined reply sentence ID and the like.

此外,在本实施方式中,下一计划指定信息840,作为以回答句子单位确定下一回答句子的信息(例如回答句子ID)而进行说明,但下一计划指定信息840也可以是以话题标题820、话题确定信息810单位确定下一回答句子(在这种情况下,作为下一回答句子而指定了多个回答句子,因此称为下一回答句子群。但是,实际上作为回答句子被输出的,成为在该回答句子群中包含的某个回答句子)的信息。 Further, in the present embodiment, the next-plan designation information 840, as a reply sentence to the reply sentence unit determines the next information (e.g., reply sentence ID) which will be described, but the next-plan designation information 840 may be based topic title 820, topic specification information 810 to determine the next unit to answer sentence (in this case, as the next reply sentence and a plurality of answer sentences specified, it is called the next answer sentence group. However, in fact, is output as the reply sentence It becomes the reply sentence group contains a reply sentence) information. 例如,即使作为下一计划指定信息而使用话题标题ID、 话题确定信息ID,本实施方式也成立。 For example, even if a next-plan designation information using the topic title ID, the topic specification information ID, the present embodiment is also established.

(4丄6.会话控制部) 在此,返回图25,说明会话控制部300的结构例。 (4 Shang 6. The session control unit) Here, back to FIG. 25, a configuration example of the conversation control unit 300.

会话控制部300具有以下功能:控制回答处理部21内的各结构要素(声音识别部200、句子解析部400、会话数据库500、输出部600、声音识别词典存储部700)间的数据的收发、以及自身与会话装置的数据的收发,并且进行对用户发话应答的回答句子的决定、输出。 Conversation control unit 300 has the following functions: controlling the respective constituent elements within the response processing unit 21 (the speech recognition unit 200, the sentence analyzing unit 400, the session database 500, an output unit 600, speech recognition dictionary storing unit 700) send and receive data between, as well as the device itself and the session data transceiver, and decided to carry out the user's utterance reply reply sentence output.

会话控制部300,在本实施方式中如图25所示,具有管理部310、计划会话处理部320、谈话空间会话控制处理部330和CA会话处理部340。 Conversation control unit 300, as shown in the present embodiment 25, having a management unit 310, the plan conversation process unit 320, a discourse space conversation control process unit 330 and the CA conversation process unit 340. 以下, 对这些结构要素进行说明。 Hereinafter, explanation of these structural elements. (4.1.6.1.管理部) (4.1.6.1. Management)

管理部310具有存储谈话履历,并且根据需要进行更新的功能。 Management unit 310 has a storage conversation history, and update functions depending on the needs. 管理部310具有根据来自话题确定信息检索部350、省略句补充部360、话题检索部370、回答取得部380的请求,将所存储的谈话履历的全部或一部分传递到所述各部的功能。 In accordance with the management unit 310 from the topic specification information retrieval unit 350, the elliptical sentence complementation unit 360, the topic retrieval unit 370, the reply retrieval unit 380 requests the transfer of all or part of the conversation history stored to the function of each part.

(4.1.6.2.计划会话处理部) (4.1.6.2. The plan conversation process unit)

计划会话处理部320具有执行计划,使得在与用户之间确立遵从计划的会话的功能。 The plan conversation process unit 320 has executed plan so that users establish functional compliance between planned sessions. 所谓"计划"是指按照预定的顺序向用户提供预定的回答。 The so-called "plan" is in accordance with a predetermined order to provide a predetermined answer. 以下, 说明计划会话处理部320。 Hereinafter, the plan conversation process unit 320.

计划会话处理部320具有根据用户发话按照预定的顺序输出预定的回答的功能。 The plan conversation process unit 320 has the user utterance in a predetermined order to output a predetermined answer function.

图34是用于说明计划的概念图。 FIG. 34 is a conceptual diagram for explaining the plan. 如图34所示,在计划空间1401中预先准备了多个计划l、计划2、计划3、计划4等各种计划1402。 As shown, the space program in 1401 prepared a number of plans 34 l, Program 2, Program 3, 4 and other planned 1402 program. 所谓计划空间1401,是指在会话数据库500中存储的多个计划1402的集合。 The so-called plan space 1401, is a collection of more than one program at a session stored in the database 500 1402. 回答处理部21 在装置启动时或会话开始时,选择预先决定用于开始的计划,或者根据各用户发话的内容,从计划空间1401中选择适当的某个计划1402,使用所选择的计划1402进行与用户发话对应的回答句子的输出。 When the reply process unit 21 when the unit starts or start a conversation, select pre-determined for the program began, or according to each user's utterance contents of a program to select the appropriate program space from 1401 1402, using the selected program 1402 corresponding to the user utterance output reply sentence.

图35是表示计划1402的结构例的图。 35 is a plan configuration example of 1402. 计划1402具有回答句子1501、与之关联的下一计划指定信息1502。 Program 1402 having reply sentence 1501 associated with next-plan designation information 1502. 下一计划指定信息1502,是确定在该计划1402中包含的回答句子1501之后,向用户输出的包含预定的回答句子(称为下一候补回答句子)的计划1402的信息。 Next-plan designation information 1502 is determined reply sentence 1501 included in the program after 1402, to the user that contains a predetermined output reply sentence (referred to as the next-reply candidate sentence) of the program information 1402. 在此例中,计划1具有在计划l执行时由回答处理部21输出的回答句子A ( 1501 )、与该回答句子A ( 1501 )关联的下一计划指定信息1502。 In this example, when the program having a program executed by the answer l processing unit 21 outputs the reply sentence A (1501), the next-plan designation information associated with the reply sentence A (1501) 1502. 下一计划指定信息1502,是确定具有作为针对回答句子A ( 1501 )的下一候补回答句子的回答句子B ( 1501)的计划1402 的信息"ID: 002"。 Next-plan designation information 1502 is determined to have as the next candidate for a reply sentence A (1501) of the reply sentence reply sentence B (1501) program information 1402 "ID: 002". 同样地,针对回答句子B ( 1501 )也决定了下一计划指定信息1502,在输出了回答句子B (1501)的情况下,指定包含下一候补回答句子的计划(1402)。 Similarly, for the reply sentence B (1501) also determines the next-plan designation information 1502, in a case where the output reply sentence B (1501), the next-reply candidate sentence plan (1402) contains specified. 于是,计划1402,通过下一计划指定信息1502连锁地连接,实现向用户输出一连串的连续内容这样的计划会话。 Thus, the planned 1402 1502 designation information attached by chain to the next plan, implement the series of continuous output to the user content such plans session. 即,将想要传达给用户的内容(说明句子、指导句子、问巻(questionnaire)等)分割成多个回答句子,并且预先决定各回答句子的顺序,作为计划而准备好,由此可以根据用户的发话,按顺序向用户提供这些回答句子。 That is, you want to convey to the user's content (illustrative sentence, guidance sentence, and asked Volume (questionnaire), etc.) is divided into a plurality of reply sentences and pre-determined order of the answer sentence as planned and ready, it can be based on user utterance, in order to provide these answers sentence. 此外,如果通过下一计划指定信息1502指定的计划1402中包含的回答句子1501,是对之前的回答句子的输出进行应答的用户发话,则不一定需要立即输出,在用户和回答处理部21 之间,该计划也可以间隔关于别的话题的会话后,输出通过下一计划指定信息1502指定的计划1402中包含的回答句子1501。 Further, if the designation information 1502 designated by next-plan plan reply sentence 1402 included in 1501, before the output of the reply sentence response to the user's utterance is not necessarily output immediately, in answer to the user and the processing unit 21 After the room, the plan may also spaced about other topics of conversation, reply sentence output specification information 1502 1501 1402 specified plan contained by the next plan.

此外,图35所示的回答句子1501对应于图33所示的回答句子830中的某一个回答句子字符串,另外,图35所示的下一计划指定信息1502对应于图33所示的下一计划指定信息840。 In addition, the reply sentence 1501 shown in FIG. 35 corresponds to FIG 33 of the reply sentence 830 in reply sentence string of one another, as shown in the next plan designation information 1502 in FIG. 35 corresponds to the case shown in FIG. 33 a plan designation information 840.

此外,计划1402的连接不限于图35所示的一维排列。 Furthermore, plans 1402 is connected is not limited to one-dimensional array 35 as shown in FIG. 图36是表示具有与图35不同的连接方式的计划1402的例子的图。 36 shows have 35 different connections plan view showing an example 1402 of FIG. 在图36所示的例子中,计划1 ( 1402),为了可以指定成为下一候补回答句子的两个回答句子1501、即1402,而具有两个下一计划指定信息1502。 In the example shown in FIG. 36, Plan 1 (1402), can be specified to be the next-reply candidate sentence two reply sentence 1501, i.e. 1402, having two next-plan designation information 1502. 作为具有输出了某个回答句子A (1501 )的情况下的下一候补回答句子的计划1402,为了确定具有回答句子B (1501)的计划2 ( 1402)、以及具有回答句子C (1501)的计划3 ( 1402)这两个计划1402,设置了两个次计划指定信息1502。 As the next candidate of the case having an output reply sentence A (1501) under the plan reply sentence 1402, in order to determine with the reply sentence B (1501) of the program 2 (1402), and having a reply sentence C (1501) in Program 3 (1402), the two plan 1402, set up two sub-plan designation information 1502. 此外,回答句子B、回答句子C是选择性的、择一的,在输出一方的情况下不输出另一方,该计划1 In addition, the reply sentence B, C answer sentence is selective, alternative, and in the case of the other party does not output the output of one of the Program 1

(1402)结束。 (1402) ends. 于是,计划1402的连接不限于一维排列的形态,即使是树形图的连接、网状的连接也可以。 Thus, the program is not limited to one-dimensional array is connected form, even if the connection is a tree diagram, the mesh 1402 may also be connected.

此外,不限定各计划具有几个下一候补回答句子。 In addition, each program is not limited to having a number of next-reply candidate sentence. 另外,针对成为话的结尾的计划1402,也可以不存在下一计划指定信息1502。 Furthermore, for the end of the program, then becomes 1402, may not exist next-plan designation information 1502.

图37表示某一连串的计划1402的具体例。 Figure 37 shows a specific example of a series of program 1402. 该一连串的计划1402「 14024 对应于用于向用户通知与危机管理相关的信息的4个回答句子1501, ~ 15014。 4个回答句子1501, ~ 15014全体构成一段连贯的话(说明文章)。各计划1402, ~ 14024分别具有"1000-01" "1000-2" "1000-3" " 1000-4"这样的ID数据1702】~ 17024。此外,ID数据中的连字符以后的号码是表示输出顺序的信息。另外, 各计划1402, ~ 14024分别具有下一计划指定信息1502, ~ 15024。下一计划指定信息15024的内容是"1000-0F"这样的数据,但该连字符以后的号码"OF" 是表示接下来输出的预定计划不存在,该回答句子是一连串的话(说明文章) 的结尾的信息。 The 1402 series of program "corresponds to 14024 for notifying the user and crisis management related information 4 reply sentences 1501 - 15014.4 reply sentences 1501, ~ 15014 all constitute some coherent words (help article). Each plan 1402, ~ 14024 respectively "1000-01" "1000-2" "1000-3" "1000-4" This ID data 1702] ~ 17024. In addition, ID data after the hyphen represents the output sequence number is information. In addition, the programs 1402, ~ 14,024 respectively next-plan designation information 1502, - the content of the designation information 15024 15024. The next plan is "1000-0F" such data, but the number after the hyphen "OF "The next scheduled output is not present, the answer sentence is a series of words (help article) the end of the message.

在此例中,当用户发话为"请教大地震发生时的危机管理"时,计划会话处理部320开始执行该一连串的计划。 In this example, when a user utterance as "crisis management advice earthquake occurred", the plan conversation process unit 320 starts executing the succession plan. 即,当计划会话处理部320接受用户发话"请教大地震发生时的危机管理"时,计划会话处理部320检索计划空间1401,调查是否存在具有与用户发话"请教大地震发生时的危机管理"对应的回答句子1501,的计划1402。 That is, when the plan conversation process unit 320 accepts a user utterance "crisis management advice earthquake occurred", the plan conversation process unit 320 retrieves program space 1401 to investigate whether there have user utterance "crisis management advice earthquake occurred." corresponding to the reply sentence 1501, the program 1402. 在此例中,假定与"请教大地震发生时的危机管理"对应的用户发话字符串1701,对应于计划1402,。 In this example, assume that the corresponding user "crisis management advice earthquake occurred" uttered a string 1701, corresponding to the planned 1402 ,.

计划会话处理部320,当发现计划1402,时,取得该计划14(^中包含的回答句子1501,,将该回答句子1501,作为对用户发话的回答来输出,同时通过下一计划指定信息1502,确定下一候补回答句子。 The plan conversation process unit 320, when the discovery schedule 1402, when the program get reply sentences (^ 14 contained the reply sentence 1501 ,, 1501, as a response to the user's utterance output, by the next plan designation information 1502 , determining the next candidate reply sentences.

然后,在回答句子1501,的输出后,经由输入部11或声音识别部200等接受用户发话时,计划会话处理部320进行计划14022的执行。 Then, after the reply sentence 1501, the output via the input unit 11 or the voice recognition unit 200, etc., accept the user's utterance, the plan conversation process unit 320 for execution of plan 14022. 即,计划会话处理部320判定是否执行通过下一计划指定信息1502,指定的计划14022,即是否输出第2个回答句子15012。 That is, the plan conversation process unit 320 determines whether to execute the next plan designation information 1502 by the specified program 14022, i.e., whether the output of the first two reply sentences 15012. 具体来说,计划会话处理部320,将与该回答句子15012对应的用户发话字符串(也称为例句)17012或者话题标题820 (在图37中省略图示)与接受的用户发话进行比较,判定它们是否一致。 Specifically, the plan conversation process unit 320, the speaker string (also called sentences), and the reply sentence corresponding user 17012 15012 or comparative topic title 820 (not shown in FIG. 37) and receiving a user utterance, determining whether they are consistent. 在一致的情况下,输出第2个回答句子15012。 In the same case, the output of the first two reply sentences 15012. 另夕卜,在包含第2个回答句子15012 的计划14022中记载了下一计划指定信息15022,因此确定下一候补回答句子。 Another evening BU, including the first two in reply sentences 15012 14022 describes the plan next plan designation information 15022, and therefore determine the next candidate reply sentences.

同样地,根据此后继续进行的用户发话,计划会话处理部320可以依次转移到计划14023、计划14024来进行第3个回答句子15013、第4个回答句子15014的输出。 Also, according to the user utterance thereafter proceed, the plan conversation process unit 320 may be sequentially transferred to the Plan 14023, Plan 14024 to 15013 the first three sentences to answer the first four answered output 15014 of sentence. 此外,第4个回答句子15014是最终回答句子,当第4个回答句子15014的输出完成时,计划会话处理部320结束计划执行。 In addition, the first four reply sentence 15014 is the ultimate answer sentence, when the output of the first four reply sentences 15014 is completed, the plan conversation process unit 320 ends program execution.

于是,通过依次执行计划1402广14024,可以按照既定的顺序向用户提供预先准备的会话内容。 Thus, by 1402 followed by the implementation plan Canton 14024, can provide a pre-prepared session content to the user in accordance with the established order.

(4.1.6.3.谈话空间会话控制处理部) (4.1.6.3. Discourse space conversation control processing unit)

返回图25,继续会话控制部300的结构例的说明。 Returning to Figure 25, to continue the session control unit 300 of the embodiment described configuration.

谈话空间会话控制处理部330具有话题确定信息;险索部350、省略句补充部360、谈话检索部370和回答取得部380。 Discourse space conversation control process unit 330 having a topic specification information; risks cable portion 350, the elliptical sentence complementation unit 360, the conversation retrieval unit 370 and the reply retrieval unit 380. 所述管理部310控制会话控制部300的全体。 The management unit 310 controls the entire conversation control unit 300.

所谓"谈话履历",是确定用户和回答处理部21间的会话的话题或主题的信息,谈话履历是包含后述的"关注话题确定信息""关注话题标题""利用者输入句子话题确定信息""回答句子话题确定信息"中的至少某一个的信息。 The so-called "conversation history", is to determine the 21 session of the user and the reply process unit topic or topics, the conversation history is included after the reference to the "focused topic specification information" "focused topic title" "user input sentence topic specification information at least one message "" reply sentence topic specification information "in. 另夕卜,在谈话履历中包含的"关注话题确定信息,,"关注话题标题""回答句子话题确定信息"不限于通过之前的会话确定的信息,也可以是在过去的预定期间中成为"关注话题确定信息""关注话题标题""回答句子话题确定信息"的信息,或者是它们的累积的记录。 Another evening BU, is included in the conversation history "focused topic specification information ,," focused topic title "" reply sentence topic specification information "is not limited to information sessions determined by the previous, it can be become a predetermined period in the past." focused topic specification information "" focused topic title "" reply sentence topic specification information "message, or their cumulative record.

以下,对构成谈话空间会话控制处理部330的所述各部进行说明。 Hereinafter, the respective portions constituting the discourse space conversation control process unit 330 will be described.

(4丄6.3丄话题确定信息检索部) (4 6.3 Shang Shang topic specification information retrieval unit)

话题确定信息检索部350将通过形态素提取部420提取出的第一形态素信息与各话题确定信息进行对照,从各话题确定信息中检索与构成第一形态素信息的形态素一致的话题确定信息。 Topic specification information retrieval unit 350 through the morpheme extracting unit 420 extracted first morpheme information with the topic specification information were compared to determine the information retrieved from each subject morpheme constituting the first shape information determined prime consistent topic information. 具体来说,话题确定信息检索部350,当从形态素提取部420输入的第一形态素信息由"佐藤,,以及"喜欢"这两个形态素构成时,将输入的第一形态素信息与话题确定信息群进行对照。 Specifically, the topic specification information retrieval unit 350, when the morpheme extracting unit 420 from the input from the first morpheme information "Sato ,, as well as" the first morpheme like "these two morpheme constituting the input of information Groups were compared to determine the information and topics.

进行了该对照的话题确定信息检索部350,当关注话题标题820focus (为了与到前一次为止检索出的话题标题、其它话题标题区别,而标记为820focus )中包含构成第一形态素信息的形态素(例如"佐藤")时,向回答取得部380 We carried out the control topic specification information retrieval unit 350, when the focused topic title 820focus (up to and to retrieve previous topic title, topic titles other distinction, and marked 820focus) contains the form constituting the first morpheme information hormone (for example, "Sato") is made to answer section 380

输出该关注话题标题820focus。 The output of the focused topic title 820focus. 另一方面,当关注话题标题820focus中不包含构成第一形态素信息的形态素时,话题确定信息检索部350根据第一形态素信息,决定利用者输入句子话题确定信息,将输入的第一形态素信息以及利用者输入句子话题确定信息输出到省略句补充部360。 On the other hand, when the focused topic title 820focus does not contain information constituting the first morpheme morpheme, the topic specification information retrieval unit 350 according to the first embodiment of pixel information to determine the user input sentence topic specification information, the input of the first morpheme information and the user input sentence topic specification information to the elliptical sentence complementation unit 360. 此外,"利用者输入句子话题确定信息",是指相当于第一形态素信息中包含的形态素内、与利用者作为话题的内容相对应的形态素的话题确定信息,或者是相当于第一形态素信息中包含的形态素内、有可能与利用者作为话题的内容相对应的形态素的话题确定信息。 In addition, the "user input sentence topic specification information" refers to the area corresponding to the first morpheme morpheme information contained in the user content as topic morpheme corresponding topic specification information, or the equivalent of the first morpheme within a morpheme information contained in it is possible to determine information about the user as a content subject morpheme corresponding topic.

(4.1.6.3.2.省略句补充部) (4.1.6.3.2. Elliptical sentence complementation unit)

省略句补充部360,利用到前一次为止检索到的话题确定信息810 (以下称为"关注话题确定信息")以及在前一次的回答句子中包含的话题确定信息810(以下称为"回答句子话题确定信息"),对所述第一形态素信息进行补充, 由此生成多个种类的补充后的第一形态素信息。 Elliptical sentence complementation unit 360, the use of the previous date of the retrieved topic specification information 810 (hereinafter referred to as "focused topic specification information") and its previous reply sentence topic specification information 810 included in (hereinafter referred to as "reply sentence topic specification information "), on the first morpheme information is added, thereby producing a plurality of types of first morpheme information after the supplement. 例如,在用户发话为"喜欢" 这样的句子时,省略句补充部360将关注话题确定信息"佐藤"包含在第一形态素信息"喜欢"中,生成补充后的第一形态素信息"佐藤、喜欢"。 For example, the user's utterance is "like" such a sentence, the sentence complementation unit 360 will be omitted focused topic specification information "Sato" included in the first morpheme information "like", the generation supplemented first morpheme information "Sato like. "

即,当将第一形态素信息设为"W"、将关注话题确定信息或回答句子话题确定信息的集合设为"D"时,省略句补充部360在第一形态素信息"W" 中包含集合"D"的要素来生成补充后的第一形态素信息。 That is, when the first morpheme information is set to "W", will focus on the topic specification information or answer sentence topic specification information is set to "D", omit the sentence complementation unit 360 in the first morpheme information "W" in It contains a collection of "D" factor to generate the supplemented first morpheme information.

由此,使用第一形态素信息构成的句子为省略句,在作为曰语无法理解等情况下,省略句补充部360可以使用集合"D",将该集合"D"的要素(例如"佐藤")包含在第一形态素信息"W"中。 Thus, sentences using the first morpheme information is omitted sentence constituted, as said in the incomprehensible language, etc., the elliptical sentence complementation unit 360 can use aggregate "D", the feature collection "D" (for example, "Sato ") contained in the first morpheme information" W "in. 结果,省略句补充部360可以作出对第一形态素信息"喜欢"进行补充后的第一形态信息"佐藤、喜欢"。 As a result, the elliptical sentence complementation unit 360 can be made to the first aspect information first morpheme information "like" supplement after the "Sato, like." 此外,补充后的第一形态信息"佐籐、喜欢"对应于"喜欢佐藤"这样的用户发话。 In addition, after the first supplementary form information "Sato, like" corresponds to "like Sato" This user utterance.

即,省略句补充部360,即使在利用者的发话内容是省略句等情况下,也可以使用集合"D"来补充省略句。 That is, the elliptical sentence complementation unit 360, even when the user utterance content is lower elliptical sentence, etc. You can also use a collection of "D" to add the elliptical sentence. 结果,省略句补充部360,即使由第一形态素信息构成的句子是省略句,也可以使该句子成为适当的曰语。 As a result, the elliptical sentence complementation unit 360, even if the sentence constituted by the first morpheme information is omitted sentence, the sentence can also be designed such that said language.

另外,省略句补充部360根据所述集合"D" 4企索与补充后的第一形态素信息一致的话题标题820。 In addition, the sentence complementation unit 360 is omitted in accordance with the set of the first morpheme information "D" 4 cable companies and supplemented the topic title 820. 当发现与补充后的第一形态素信息一致的话题标题 When the first morpheme discovery and consistent information supplemented topic title

820时,省略句补充部360将该话题标题820输出到回答取得部380。 820, the elliptical sentence complementation unit 360 outputs the topic title 820 to the reply retrieval unit 380. 回答取得部380可以根据通过省略句补充部360检索到的适当的话题标题820,输出最适合于利用者的发话内容的回答句子830。 Reply retrieval unit 380 can retrieve 360 the sentence complementation unit by appropriate topic title 820 is omitted, the output of the most suitable content to the user's utterance reply sentence 830.

此外,省略句补充部360不仅限于在第一形态素信息中包含集合"D"的要素。 In addition, the elliptical sentence complementation unit 360 is not limited to the elements contained in the first morpheme information collection of "D". 该省略句补充部360也可以根据关注话题标题,在提取出的第一形态素信息中包含在构成该话题标题的第一确定信息、第二确定信息或第三确定信息的某一个中包含的形态素。 The elliptical sentence complementation unit 360 can also focus on the topic title in the extracted first morpheme information contained in the first specification information constituting the topic title in the second specification information or one third specification information contained in morpheme. (4.1.6.3.3.话题检索部) (4.1.6.3.3. Topic retrieval unit)

话题检索部370,当未通过省略句补充部360确定话题标题820时,将第一形态素信息、和与利用者输入句子话题确定信息对应的各话题标题820进行对照,从各话题标题820中检索最适合于第一形态素信息的话题标题820。 370, while 360 have not determined when the topic titles 820, the first morpheme information, and a user input sentence topic retrieval unit through supplementary section omitted sentence topic specification information corresponding to each topic title 820 controls, 820 from each topic title retrieving the most suitable for the first morpheme information, the topic title 820.

具体来说,从省略句补充部360输入了检索命令信号的话题检索部370, Specifically, from the elliptical sentence complementation unit 360 inputs a search command signal topic retrieval unit 370,

一形态素信息,从与该利用者输入句子话题确定信息对应的各话题标题中检索最适合于该第一形态素信息的话题标题820。 A morpheme information from the input sentence topic specification information corresponding to the user for each topic title retrieved most suitable for the first morpheme information, the topic title 820. 话题^r索部370将该检索到的话题标题820作为检索结果信号,输出到回答取得部380。 ^ R topic search unit 370 retrieves the topic title 820 as the retrieval result signal to the reply retrieval unit 380.

先前披露的图33表示与某话题确定信息810 ( - "佐藤")对应的话题标题820、回答句子830的具体例。 Figure 33 shows the determined previously disclosed information 810 and a topic (- "Sato") corresponding to the topic titles 820, 830 sentence answer specific examples. 如图33所示,例如话题检索部370,由于在所输入的第一形态素信息"佐藤、喜欢"中包含话题确定信息810(="佐藤"), 因此确定该话题确定信息810 (="佐籐"),然后,将与该话题确定信息810 (="佐藤")对应的各话题标题(820) 1-1、 1-2、…和所输入的第一形态素信息"佐藤、喜欢"进行对照。 33, for example, the topic retrieval unit 370, since the first morpheme information inputted "Sato, like" includes topic specification information 810 (= "Sato"), thus determining the topic specification information 810 (= " Sato "), then, will determine the topic information 810 (=" Sato ") corresponding to each topic title (820) 1-1, 1-2, ... and the input of the first morpheme information" Sato, like "were compared.

话题纟企索部370根据其对照结果,从各话题标题(820) 1-1 ~ 1-2中确定与所输入的第一形态素信息"佐藤、喜欢" 一致的话题标题(820) 1-1 (佐藤: *;喜欢)。 Si topic search unit 370 companies under its control results from each of the topic titles (820) 1-1 and 1-2 to determine the first morpheme information inputted "Sato, like" the same topic titles (820) of 1- 1 (Sato: *; like). 话题检索部370将检索到的话题标题(820) 1-1 (佐藤;*;喜欢) 作为检索结果信号,输出到回答取得部380。 Topic retrieval unit 370 retrieves the topic title (820) 1-1 (Sato; *; like) as retrieval result signal to the reply retrieval unit 380. (4丄6.3.4.回答取得部) (4 Shang 6.3.4. Reply retrieval unit)

回答取得部380才艮据通过省略句补充部360、或者话题4企索部370检索出的话题标题820,取得与该话题标题820对应的回答句子830。 Burgundy was the reply retrieval unit 380 according to the elliptical sentence complementation unit 360 by, or subject 4 Enterprise search unit 370 retrieves the topic title 820, made the topic title 820 reply sentence 830. 另外,回答取得部380根据通过话题检索部370检索出的话题标题820,将与该话题标题820 对应的各回答种类、和通过输入种类判定部440判定出的发话种类进行对照。 In addition, the reply retrieval unit 380 820, each answer types with the topic title 820 corresponding to, and be controlled by the input type determining unit 440 determines that the utterance type by topic retrieval unit 370 according to the retrieved topic title. 进行了该对照的回答取得部380从各回答种类中检索与判定出的发话种类一致的回答种类。 Carried out the controlled reply retrieval unit 380 retrieves the answer from each category consistent with the type of utterance determines the type of answer.

在图33所示的例子中,回答取得部380,当通过话题^f企索部370检索出的话题标题是话题标题1-1 (佐藤;*;喜欢)时,从与该话题标题1-1对应的回答句子1-1 (DA、 TA等)中,确定与通过输入种类判定部440判定出的"发话句子的类型"(例如DA) —致的回答种类(DA)。 In the example shown in FIG. 33, the reply retrieval unit 380, when prices by topic ^ f search unit 370 retrieves the topic title topic title 1-1 (Sato; *; like) when the topic title from 1- 1 reply sentence 1-1 (DA, TA, etc.), it is determined by the input type determining unit with 440 determines the "uttered sentence type" (such as DA) - induced answers species (DA). 确定了该回答种类(DA) 的回答取得部380,根据所确定的回答种类(DA),取得与该回答种类(DA) 对应的回答句子ll ("我也喜欢佐藤。")。 The answer is determined species (DA) of the reply retrieval unit 380, based on the determined kind of answer (DA), made the answer species (DA) corresponding to the answer sentence ll ("I like Sato.").

在此,上述"DA"、 "TA"等中的"A,,表示肯定形式。因此,在发话种类以及回答种类中包含"A"时,表示关于某事情进行肯定。另外,在发话种类以及回答种类中也可以包含"DQ"、 "TQ,,等种类。 Here, the "DA", "TA" and the like "A ,, affirmed form. Therefore, in the utterance type and kind reply with" time A ", represents about a thing be sure. In addition, the speaker type and Answer species may also contain "DQ", "TQ ,, and other species. 该"DQ"、 "TQ"等中的"Q"表示关于某事情的疑问。 The "DQ", "TQ" and the like "Q" indicates something about a question.

当回答种类由上述疑问形式(Q)构成时,与该回答种类对应的回答句子由肯定形式(A)构成。 When the answer to these questions form the species (Q) constituted the answer corresponding to the type of sentence consists affirmative reply form (A). 作为以该肯定形式(A)生成的回答句子,列举出对疑问事项进行回答的句子等。 As with the affirmative form (A) generated reply sentence, include the answer to the question of sentence and other matters. 例如,当发话句子是"你对投币游戏机进行过操作吗?"时,关于该发话句子的发话种类成为疑问形式(Q)。 For example, when uttered sentence is "you have been a slot machine operation?", The utterance type on the uttered sentence into question form (Q). 与该疑问形式(Q)对应的回答句子,列举出例如"我对投币游戏机进行过操作"(肯定形式(A))。 The interrogative form (Q) corresponding to the answer sentence, includes, for example, "I have been operating the slot machine" (affirmative form (A)).

另一方面,当发话种类由肯定形式(A)构成时,与该回答种类对应的回答句子由疑问形式(Q)构成。 On the other hand, when the utterance type by the affirmative form (A) constitute a time, and the answer corresponding to the type of questions to answer in the form of a sentence consists of (Q). 作为以该疑问形式(Q)生成的回答句子,列举出对发话内容反问的疑问句、或者询问特定事情的疑问句等。 As to the question form (Q) generated reply sentence, include the contents of the speaker to ask the questions, or ask specific questions and other things. 例如,当发话句子是"我爱好玩投币游戏机"时,关于该发话句子的发话种类成为肯定形式 For example, when uttered sentence is "I like to play slot machines," the speaker on the uttered sentence types become affirmative form

(A)。 (A). 与该肯定形式(A)对应的回答句子,列举出例如"不爱好玩弹子机吗?" With the affirmative form (A) corresponding to the answer sentence, includes, for example, "I do not like to play pinball machine?"

(询问特定事情的疑问句(Q))。 (Ask specific questions matter (Q)).

部310。 310. 从回答取得部380输入了回答句子信号的管理部310,将所输入的回答句子信号输出到输出部600。 Obtained from the input unit 380 to answer the answer sentence signal managing section 310, outputs the reply sentence signal input to the output portion 600. (4.1.6.4. CA会话处理部) CA会话处理部340具有以下功能:在针对用户发话,在计划会话处理部320以及谈话空间会话控制处理部330的任意一个中都没有决定回答句子的情况下,根据用户发话的内容输出可以继续与用户的会话的回答句子。 (. 4.1.6.4 CA conversation process unit) CA conversation process unit 340 has the following features: In for the user utterance, in any plan conversation process unit 320, and the discourse space conversation control processing section 330 are not decided on a case to answer sentence According to the content of the user utterance output can continue with the user's session reply sentence. 以上,结束了回答处理部21的结构例的il明。 This concludes the il clear answer processor configuration example 21. (4.2.会话控制方法)具有上述结构的回答处理部21通过以下这样进行动作来实现会话控制方法。 Answer processing unit (4.2. Session Control Method) 21 having the above structure operates in the following to achieve such a session control method. 说明本实施方式的回答处理部21、更详细而言是会话控制部300的动作。 Description reply process unit 21 of the present embodiment, more specifically the conversation control unit 300 of the operation. 图38是表示会话控制部300的主处理的一例的流程图。 FIG. 38 is a flowchart showing the main processing session control unit 300 an example. 该主处理是每当会话控制部300接受用户发话时执行的处理,通过进行该主处理,进行与用户发话对应的回答句子的输出。 The main process is a process whenever the conversation control unit 300 accepts a user utterance performed by performing the main processing, the user performs utterance corresponding to the reply sentence output. 会话装置10和会话服务器20 (回答处理部21 ) 间的会话(对话)确立。 The conversation device 10 and the conversation server 20 (answer 21 processing unit) between sessions (dialogue) established. 进入主处理时,会话控制部300、更详细而言是计划会话处理部320,首先执行计划会话控制处理(S1801)。 Into the main process, the session control unit 300, more specifically, is the plan conversation process unit 320, the first implementation plan conversation control process (S1801). 计划会话控制处理是执行计划的处理。 The plan conversation control process is a process executed plan. 图39、图40是表示计划会话控制处理的一例的流程图。 39, FIG. 40 is a flowchart showing an example of the plan conversation control process. 以下,参照图39、 图40说明计划会话控制处理的例子。 Hereinafter, with reference to FIG. 39, FIG. 40 illustrates an example of the plan conversation control process. 当开始计划会话控制处理时,计划会话处理部320首先进行基本控制状态信息检查(S1901)。 When starting the plan conversation control process, the plan conversation process unit 320 first checks the basic control state information (S1901). 基本控制状态信息,将计划1402的执行是否结束作为基本控制状态信息,存储在预定的存储区域中。 The basic control state information, the implementation of the plan has ended 1402 as the basic control state information in a predetermined storage area. 基本控制状态信息具有记载计划的基本控制状态的作用。 The role of the basic control information has a documented plan of the basic control. 图41是表示关于被称为情节的类型的计划而产生的4个基本控制状态的图。 41 is a plot about to be called the type of plan and produce four basic control Fig. 以下说明各个状态。 The following descriptions of each state. (1) 结束该基本控制状态是用户发话与执行中的计划1402 —致、更详细而言是与计划1402所对应的话题标题820或例句一致的情况。 (1) The basic control state is the end user's utterance and implementation plan 1402 - To, more specifically with the planned 1402 corresponding topic title 820 or identical sentence case. 在这种情况下,计划会话处理部320结束该计划1402,转移到与通过下一计划指定信息1502指定的回答句子1501对应的计划1402。 In this case, the plan conversation process unit 320 terminates the plan 1402, the information is transferred to the specified reply sentence 1502 designated by next-plan 1501 corresponding program 1402. (2) 放弃35该基本控制状态,是当判断出用户发话内容是请求了计划1402的结束的情况下,或者判定为用户的关心转移到了执行中的计划以外的事项的情况下设定的基本控制状态。 (2) give up 35 of the basic control state is a case where the content is determined that the user utterance plan requested end of 1402, or to determine a case is transferred to the user's interest in the plan other than the implementation of the set of basic items control. 在基本控制状态信息表示放弃的情况下,计划会话处理部320检索在成为放弃对象的计划1402以外是否存在与用'户发话对应的计划1402,在存在的情况下开始该计划1402的执行,在不存在的情况下结束计划的执行。 In the basic control state information indicates the case of abandonment, the plan conversation process unit 320 retrieves whether there is use 'user utterance plan 1402, in the presence of the beginning of the implementation of the plan in 1402 than 1402 to become the object of abandonment plan, End the absence of implementation of the plan. (3) 维持该基本控制状态,是在用户发话不对应于与执行中的计划1402对应的话题标题820 (参照图33)或例句1701 (参照图37),并且判断出用户发话不对应于基本控制状态"放弃"的情况下,在基本控制状态信息中记载的基本控制状态。 (3) maintaining the basic control case, is in the user's utterance does not correspond to the execution plan 1402 corresponding to the topic title 820 (see FIG. 33) or the example sentence 1701 (see FIG. 37), and determines that the user utterance does not correspond to the basic Under the control of the state "give up", the basic control state information is described in the basic control. 当是该基本控制状态的情况下,计划会话处理部320在接受用户发话时, 首先研究是否重新开始正保留•中止的计划1402,当用户会话不适合于重新开始计划1402时,例如用户发话不对应于与计划1402对应的话题标题802 或例句1702时,开始其它计划1402的执行、或者进行后述的谈话空间会话控制处理(S1802)等。 When the case is that the basic control state, the plan conversation process unit 320 accepts the user's utterance, the first study of whether to re-start the program being retained • Abort 1402, when a user session is not suitable for the re-start program 1402, for example, the user's utterance is not corresponding to the planned 1402 corresponding topic title 802 or 1702 sentence, the discourse space conversation 1402 after the start of the implementation of other programs, or conduct described control processing (S1802) and the like. 当用户发话适合于重新开始计划1402时,根据所存储的下一计划指定信息1502进行回答句子1501的输出。 When the user's utterance is adapted to restart plan 1402, specify the information to answer 1502 1501 output sentence according to the next scheduled stored. 当基本控制状态为"维持"时,计划会话处理部320检索其它计划1402, 以便可以输出与该计划1402对应的回答句子1501以外的回答,或者进行后述的谈话空间会话控制处理等,但在用户发话再次成为与计划1402相关的发话时,重新开始该计划1402的执行。 When the basic control state is "maintenance", the plan conversation process unit 320 retrieves another plan 1402 to 1402 can output a reply sentence 1501 and the plan other than to answer, or to talk space conversation control process will be described later, etc., but in When the user's utterance became associated with the planned 1402 utterance again to resume the implementation of the plan 1402. (4) 继续该状态,是在用户发话不对应于执行中的计划1402中包含的回答句子1501、并且判断出用户发话内容不对应于基本控制状态"放弃"、并且从用户发话解释出的用户的意图不明确的情况下所设定的基本控制状态。 (4) continuing the state, it is in the user's utterance reply sentence 1501 does not correspond to the execution plan 1402 included, and determines that the user utterance does not correspond to the content of the basic control state "abandoned", and the interpretation of the utterance of the user from the user The intent is unclear under the circumstances set basic control. 当基本控制状态为"继续"时,计划会话处理部320在接受用户发话时, 首先研究是否重新开始正保留•中止的计划1402,当用户发话不适合于重新开始计划1402时,进行后述的CA会话控制处理等,以便可以输出用于从用户引出其它发话的回答句子。 When the basic control state is "continue", the plan conversation process unit 320 accepts the user's utterance, the first study of whether to re-start the program being retained • Abort 1402, when the user utterance is not suitable for re-start program 1402, were described later CA conversation control process, in order to be able to output to the user utterance answer leads to another sentence. 返回图39,继续说明计划会话控制处理。 Returning to Figure 39, continued to explain the plan conversation control process. 参照了基本控制状态信息的计划会话处理部320,判定基本控制状态信息表示的基本控制状态是否是"结束"(S1902)。 Referring to the basic control state information plan conversation process unit 320 determines whether the basic control state indicated by the basic control state information is the "end" (S1902). 当判定出基本控制状态是"结束"时(S1902,是),计划会话处理部320在基本控制状态信息表示的执行中的计划1402中判定回答句子1501是否是最终回答句子(S1903 )。 When it is determined that the basic control program is executed in "end" when (S1902, Yes), the plan conversation process unit 320 in the basic control state information indicates the 1402 judgment whether the reply sentence 1501 is the final reply sentence (S1903). 当判定为已输出最终回答句子1501时(S1903,是),计划会话处理部320, 由于在该计划1402中已经全部传达完应该向用户回答的内容,因此为了判定是否开始新的别的计划1402,检索在计划空间中是否存在与用户发话对应的计划1402 (S1904)。 When it is determined as the final output reply sentence 1501 (S1903, Yes), the plan conversation process unit 320, since 1402 the program has been fully completed should be communicated to users answered content, so in order to determine whether other plans to start a new 1402 retrieve whether the user utterance in the plan space programs exist in 1402 (S1904). 当该检索的结果是未发现与用户发话对应的计划1402时(S1905,否),由于不存在应该向用户提供的计划1402,因此计划会话处理部320直接结束计划会话控制处理。 As a result of this search is not found in the user's utterance program 1402 (S1905, No), since the plan should be provided to the user not exist 1402, thus ending the plan conversation directly plan conversation process unit 320 controls the process. 另一方面,当该检索的结果是发现了与用户发话对应的计划1402时(S1905,是),计划会话处理部320转移到该计划1402 (S1906)。 On the other hand, when the result of the search is found in the user's utterance program 1402 (S1905, Yes), the plan conversation process unit 320 is transferred to the program 1402 (S1906). 这是由于, 存在应该向用户提供的计划1402,因此开始该计划1402的执行(输出计划1402 中包含的回答句子1501 )。 This is because the present schedule should be provided to the user 1402, so 1402 is the beginning of the program implementation (planned 1402 output reply sentence 1501 included). 然后,计划会话处理部320输出该计划1402的回答句子1501 (S1908)。 Then, the plan conversation process unit 320 outputs the reply sentence 1402 Program 1501 (S1908). 所输出的回答句子1501成为对用户发话的回答,计划会话处理部320提供想要向用户传达的信息。 The output of the reply sentence 1501 to become speaker of the user answers the plan conversation process unit 320 provides the user wants to convey the message. 回答句子输出处理(S1908)后,计划会话处理部320结束计划会话控制处理。 Reply sentence output process (S1908), the plan conversation process unit 320 terminates the plan conversation control process. 另一方面,在先前输出的回答句子1501是否是最终的回答句子1501的判定(S1903)中,当先前输出的回答句子1501不是最终的回答句子1501时(S1903,否),计划会话处理部320转移到与先前输出的回答句子1501相连的回答句子1501、即通过下一计划指定信息1502确定的回答句子1501所对应的计划1402 (Sl卯7)。 On the other hand, the output of the previous sentence answer whether 1501 is the final reply sentence 1501 decision (S1903), when the previously output reply sentence 1501 is not the final reply sentence 1501 (S1903, No), the plan conversation process unit 320 Transfer to the previously output reply sentence 1501 is connected to the reply sentence 1501, i.e. 1502 designated by next-plan information determined corresponding to the reply sentence 1501 Program 1402 (Sl d 7). 此后,计划会话处理部320输出在对应的计划1402中包含的回答句子1501,进行对用户发话的回答(S1908)。 Since then, the reply sentence 1501 output plan conversation process unit 320 included in the corresponding 1402 program, carried out the answer to the user's utterance (S1908). 所输出的回答句子1501成为对用户发话的回答,计划会话处理部320提供想传达给用户的信息。 The output of the reply sentence 1501 to become speaker of the user answers the plan conversation process unit 320 provides information to convey to the user. 在回答句子输出处理(S1908)后,计划会话处理部320结束计划会话控制处理。 In reply sentence output process (S1908), and finishes the planned session plan conversation process unit 320 controls the process. 在S1902的判定处理中,当基本控制状态信息不是"结束"时(S1902, 否),计划会话处理部320判定基本控制状态信息表示的基本控制状态是否是"放弃"(S1909)。 In S1902 the determination process, when the basic control state information is not "the end" when (S1902, No), the plan conversation process unit 320 determines whether the basic control basic control state information indicates the "abandonment" (S1909). 当判定为基本控制状态是"放弃"时(S1909,是),由于不存在应该继续的计划1402,因此计划会话处理部320为了判定是否存在应该开始的新的别的计划1402,在计划空间1401内斥全索是否存在与用户发话对应的计划1402 (S1904)。 When it is judged as the basic control state is "abandon" (S1909, Yes), because there is no plan 1402 should continue, so the plan conversation process unit 320 in order to determine whether there is another new program should start 1402, in 1401 planned space denounced the existence of a full search within the user's utterance program 1402 (S1904). 此后,与在先描述的S1903 (是)中的处理同样地, 计划会话处理部320执行从Sl卯5到S1908的处理。 Thereafter, S1903 previously described (is) in the process in the same manner, the plan conversation process unit 320 executes the process from Sl d 5 S1908's. 另一方面,在基本控制状态信息表示的基本控制状态是否是"放弃"的判定(S1909)中,当判定为基本控制状态不是"放弃"时(S1909,否),计划会话处理部320进一步进行基本控制状态信息表示的基本控制状态是否是"维持"的判定(S1910)。 On the other hand, the basic control state indicated by the basic control state information is "abandon" the judgment (S1909), when determined that the basic control state is not "give up" (S1909, No), the plan conversation process unit 320 further It is the basic control basic control information indicates "maintenance" of judgment (S1910). 当基本控制状态信息表示的基本控制状态是"维持"时(S1910,是),计划会话处理部320,关于正保留•停止的计划1402调查用户是否再次表示出关心,当表示出关心时进行动作,重新开始临时保留•停止的计划1402地进行动作。 When the basic control state indicated by the basic control state information is "maintain" (S1910, Yes), the plan conversation process unit 320, on positive retention plans • Stop 1402 survey expressed concern whether the user again operates when expressed concern start again stopped temporarily reserved • 1402 to carry out the action plan. 即,计划会话处理部320检查保留停止中的计划1402(图40; S2001 ), 判定用户发话是否对应于保留.停止中的该计划1402 (S2002)。 Specifically, the plan conversation process unit 320 checks to stop the plan to keep 1402 (Figure 40; S2001), it determines that the user utterance corresponds to a retention stop in the program 1402 (S2002).. 当判定为用户发话对应于该计划1402时(S2002,是),计划会话处理部320转移到与该用户发话对应的计划1402 (S2003 )。 When it is determined that corresponds to the user's utterance program 1402 (S2002, Yes), the plan conversation process unit 320 is transferred to the user's utterance program 1402 (S2003). 此后,执行回答句子输出处理(图39; S1908)以便输出在该计划1402中包含的回答句子1501。 Thereafter, the reply sentence output process (Fig. 39; S1908) to output the reply sentence 1402 included in the 1501 program. 通过如此进行动作,计划会话处理部320根据用户发话可以重新开始已保留'中断的计划1402,可以将在预先准备的计划1402中包含的内容全部传递给用户。 By thus operated, the plan conversation process unit 320 can resume the user utterance reserved 'interrupted program 1402, the content can be prepared in advance in the program 1402 included in all delivered to the user. 另一方面,在前面的S2002 (参照图40)中判定出保留.停止中的计划1402不对应于用户发话时(S2002,否),计划会话处理部320为了判定是否存在应该开始的新的别的计划1402,在计划空间1401内检索是否存在与用户发话对应的计划1402 (图39; S1904)。 On the other hand, it is determined that remain in front of the S2002 (refer to FIG. 40). 1402 to stop the program when the user's utterance does not correspond to (S2002, No), the plan conversation process unit 320 determines whether there is a new order to do should begin plan 1402, the plan space 1401 in the presence or absence of the user utterance to retrieve the corresponding program 1402 (FIG. 39; S1904). 此后,与在先描述的S1903 (是)中的处理同样地,计划会话处理部320执行从S1905到S1908的处理。 Thereafter, in the same manner, the plan conversation process unit 320 executes the processing from S1905 to S1908 and S1903 described previously (YES) the processing. 在S1910的判定中,当基本控制状态信息表示的基本控制状态不是"维持" 时(S1910,否),意味着基本控制状态信息表示的基本控制状态是"继续"。 In determining S1910, when the basic control basic control state information indicates not "maintain" when (S1910, No), the basic control means that the basic control state information indicates "continue." 在这种情况下,计划会话处理部320不进行回答句子的输出而结束计划会话控制处理。 In this case, the plan conversation process unit 320 does not output the reply sentence and terminates the plan conversation control process.

以上,结束了计划会话控制处理的说明。 This concludes the description of the plan conversation control process. 返回图38,继续"i兌明主处理。 Returning to Figure 38, continued "i against the next primary process.

当结束计划会话控制处理(S1801 )时,会话控制部300开始谈话空间会话控制处理(S1802)。 When the plan conversation control process (S1801), the session control unit 300 starts discourse space conversation control process (S1802). 但是,在计划会话控制处理(S1801)中进行了回答句子输出的情况下,会话控制部300不进行谈话空间会话控制处理(S1802)以及在后面说明的CA会话控制处理(S1803 )的任何一个,进行基本控制信息更新处理(S1904)来结束主处理。 However, the plan conversation control process (S1801) is carried out under the reply sentence output, session control unit 300 does not perform the discourse space conversation control process (S1802) as well as one of the processing (S1803) is described later in CA conversation control, basic control information update process (S1904) to end the main process.

图42是表示本实施方式的谈话空间会话控制处理的一例的流程图。 FIG 42 is a flowchart showing an example of the process of the present embodiment discourse space conversation control. 首先,输入部11进行取得来自利用者的发话内容的步骤(步骤S2201 )。 First, the input unit 11 acquisition step (step S2201) from the user's utterance contents. 具体而言,输入部ll :f又得构成利用者的发话内容的声音。 Specifically, the input section ll: f they have constituted a sound uttered contents of the user's. 输入部11将所取得的声音作为声音信号,输出到声音识别部200。 The voice input unit 11 as the acquired sound signal is output to the voice recognition unit 200. 此外,输入部11也可以不取得来自利用者的声音,而取得从利用者输入的字符串(例如以文本形式输入的字符数据)。 Further, the input unit 11 may not be made from the user's voice, the string obtained from the user input (e.g., input as text character data). 在这种情况下,输入部ll不是麦克风,而成为键盘或触摸面板等字符输入装置。 In this case, instead of a microphone input portion ll, and a keyboard or a touch panel or the like to become a character input apparatus.

接着,声音识别部200进行根据通过输入部11取得的发话内容确定与发话内容对应的字符串的步骤(步骤S2202)。 Next, the voice recognition unit 200 according to the determination in step utterance string corresponding to the content (Step S2202) via the input unit 11 to obtain the contents of the utterance. 具体而言,从输入部ll输入了声音信号的声音识别部200,根据所输入的声音信号确定与该声音信号对应的单词假设(候补)。 Specifically, the input from the input audio signal portion ll voice recognition unit 200, hypothesis (candidate) corresponding to the sound signal input word is determined according to the sound signal. 声音识别部200取得与所确定的单词假设(候补)对应的字符串,将所取得的字符串作为字符串信号输出.到会话控制部300,更详细而言输出到谈话空间会话控制处理部330。 Voice recognition unit 200 to obtain the determined word hypothesis (candidate) corresponding to the character string, the character string acquired as a character string signal output to the conversation control unit 300, more specifically the discourse space conversation control output to the processing unit 330 .

然后,字符串确定部410进行将通过声音识别部200确定的一连串字符串按照每一短语来划分的步骤(步骤S2203 )。 Then, the character string specifying unit 410 through a series of steps to string the speech recognition unit 200 determines each phrase according to the division (step S2203). 具体而言,从管理部310输入了字符串信号(或者形态素信号)的字符串确定部410,在该输入的一连串字符串中存在某一定以上的时间间隔时,在该部分划分字符串。 Specifically, from the management unit 310 in the string signal (or morpheme signals) string specifying unit 410, the presence of more than a certain time interval of the input string in the series, the division of the string section. 字符串确定部410 将该划分而得的各字符串输出到形态素提取部420以及输入种类判定部440。 String specifying unit 410 obtained by dividing the output of each string to the morpheme extracting unit 420 and the input type determining unit 440. 此外,字符串确定部410,当所输入的字符串是AM泉盘输入的字符串时,优选在标点或空格等所在的部分划分字符串。 In addition, a string specifying unit 410, when the input string is a string AM spring pad to enter the time, preferably partially divided string punctuation or spaces, etc. are located.

此后,形态素提取部420进行根据通过字符串确定部410确定的字符串., Since then, the morpheme extracting unit 420 for string by string specifying unit 410 determined in accordance.,

39提取构成字符串的最小单位的各形态素作为第一形态素信息的步骤(步骤 39 different forms of extraction constitutes the smallest unit pixel string as first morpheme information (step

S2204)。 S2204). 具体而言,从字符串确定部410输入了字符串的形态素提取部420, 将所输入的字符串、和在形态素数据库430中预先存储的形态素群进行对照。 Specifically, section 410 is determined from a string input string prime shape extraction unit 420, the input string, and the morpheme database 430 previously stored morpheme groups were compared. 此外,该形态素群,在本实施方式中,关于属于各个词类的各形态素,准备了记载有该形态素的词条•读法•词类.活用形等的形态素辞典。 In addition, the morpheme group, in this embodiment, with respect to each morpheme belonging to various parts of speech, have prepared a record entry of the morpheme • Reading • parts of speech. Conjugation, etc. morpheme dictionary.

进行了该对照的形态素提取部420,从所输入的字符串中提取出与预先存储的形态素群中包含的各形态素一致的各形态素(ml、 m2、...)。 The controls carried out morpheme extracting unit 420 extracts the various forms of pre-stored morpheme group included the same elements from each morpheme in the input string (ml, m2, ...). 形态素提取部420将提取出的各形态素作为第一形态素信息,输出到话题确定信息检索部350。 Morpheme extracting unit 420 will be extracted morphemes as the first morpheme information output to the topic specification information retrieval unit 350.

接着,输入种类判定部440进行根据构成通过字符串确定部410确定的一句的各形态素,判定"发话句子的类型"的步骤(步骤S2205)。 Subsequently, the input type determining unit 440 according to 410 constituted by the character string determining portion determines each morpheme sentence determined "uttered sentence type" step (step S2205). 具体而言, 从字符串确定部410输入了字符串的输入种类判定部440,根据所输入的字符串,将该字符串和在发话种类数据库450中存储的各辞典进行对照,从该字符串中提取出与各辞典相关的要素。 Specifically, from the character string specifying unit 410 inputs the character string input type determination unit 440, based on the input character string, and the string were compared in the utterance type database 450 stored in the respective dictionaries, from the string extracted elements associated with the dictionary. 提取出该要素的输入种类判定部440,根据提取出的要素,判定该要素属于哪个"发话句子的类型"。 Input type of the extracted feature determination unit 440, based on the extracted feature, which is determined "uttered sentence type" the element belongs. 输入种类判定部440 将判定出的"发话句子的类型"(发话种类)输出到回答取得部380。 "Uttered sentence type" (utterance type) to the input type determining unit 440 determines that the output 380 to the reply retrieval unit.

然后,话题确定信息检索部350进行将通过形态素提取部420提取出的第一形态素信息、和关注话题标题820focus进行比较的步骤(步骤S2206 )。 Then, the topic specification information retrieval unit 350 performs morphological morpheme extracting unit 420 extracted first morpheme information, the comparison (step S2206) and focused topic title 820focus be.

当构成第一形态素信息的形态素和关注话题标题820focus —致时,话题确定信息检索部350将该话题标题820输出到回答取得部380。 When constituting the first morpheme information morpheme and focused topic title 820focus - When delivered, the topic specification information retrieval unit 350 outputs the topic title 820 to the reply retrieval unit 380. 另一方面,当构成第一形态素信息的形态素和话题标题820不一致时,话题确定信息检索部350将所输入的第一形态素信息以及利用者输入句子话题确定信息作为检索命令信号,输出到省略句补充部360。 On the other hand, when the information constituting the first morpheme morpheme and topic titles 820 is inconsistent, the topic specification information retrieval unit 350 of the first morpheme information and the user input sentence topic specification information input as the search command signal output the sentence complementation unit 360 is omitted.

此后,省略句补充部360进行根据从话题确定信息检索部350输入的第一形态素信息,将关注话题确定信息以及回答句子话题确定信息包含在所输入的笫一形态素信息中的步骤(步骤S2207)。 Since then, the sentence complementation unit 360 according to omit determining the first morpheme information information retrieval unit 350 input from the topic, will focus on the topic specification information and the reply sentence topic specification information contained in Zi step morpheme information inputted (step S2207). 具体而言,当将第一形态素信息设为"W",将关注话题确定信息以及回答句子话题确定信息的集合设为"D,, 时,省略句补充部360在第一形态素信息"W"中包含话题确定信息"D"的要素,生成补充后的第一形态素信息,并将该补充后的第一形态素信息、和与集合"D"关联的全部话题标题820进行对照,检索是否存在与补充后的第一形态素信息一致的话题标题820。当存在与补充后的第一形态素信息一致的话题标题820时,省略句补充部360将该话题标题820输出到回答取得部380。 另一方面,当未发现与补充后的第一形态素信息一致的话题标题820时,省略句补充部360将第一形态素信息和利用者输入句子话题确定信息传递给话题检索部370。 Specifically, when the first morpheme information is set to "W", will focus on the topic specification information and the reply sentence topic specification information is set to "D ,, when the sentence complementation unit 360 is omitted in the first morpheme information" All topic title W "contains topic specification information" D "elements, generating supplemented first morpheme information, and will complement the first morpheme information after, and a collection of" related D "820 were compared, Retrieving consistent with the existence of supplemented first morpheme information topic title 820. When there is agreement with the supplemented first morpheme information topic title 820, omitted sentence complementation unit 360 outputs the topic title 820 to the reply retrieval section 380. On the other hand, when the information was not found consistent with the first morpheme supplemented the topic title 820, the elliptical sentence complementation unit 360 of the first morpheme information and the user input sentence topic specification information to the topic retrieval unit 370.

接着,话题检索部370进行将第一形态素信息和利用者输入句子话题确定信息进行对照,从各话题标题820中^f企索适合于第一形态素信息的话题标题820的步骤(步骤S2208)。 Then, the topic retrieval unit 370 of the first morpheme information and the user input sentence topic specification information collated from the various topic headings 820 ^ f suitable topic title cable companies first morpheme information of step 820 (step S2208 ). 具体而言,从省略句补充部360输入了^r索命令信号的话题检索部370,根据在所输入的检索命令信号中包含的利用者输入句子话题确定信息以及第一形态素信息,从与该利用者输入句子话题确定信息对应的各话题标题820中检索适合于该第一形态素信息的话题标题820。 Specifically, from the elliptical sentence complementation unit 360 inputs the search topic retrieval unit ^ r command signal 370, the retrieval based on user input command signal contained in the input sentence topic specification information and the first morpheme information, from the The user input sentence topic specification information corresponding to each topic title 820 retrieved topic title suitable for the first morpheme information 820. 话题检索部370将作为该检索结果而得到的话题标题820作为检索结果信号,输出到回答取得部380。 Topic retrieval unit 370 as the search result obtained by topic title 820 as the retrieval result signal to the reply retrieval unit 380.

接着,回答取得部380根据通过话题确定信息4全索部350、省略句补充部360或者话题检索部370检索出的话题标题820,将通过句子解析部400判定出的利用者的发话种类、和与话题标题820对应的各回答种类进行对照,进行回答句子830的选择(步骤S2209 )。 Then, the reply retrieval unit 380 determines information search unit 4 100 350 according by topic, sentence complementation unit 360 is omitted or topic retrieval unit 370 retrieves the topic title 820, through the sentence analyzing unit 400 determines the type of user utterance, and and topic titles corresponding to 820 species were compared for each answer, to answer sentence 830 selected (step S2209).

具体而言,以下这样来进行回答句子830的选4奪。 Specifically, the following sentence answer to this 830 4 election wins. 即,从话题检索部370 输入了检索结果信号、并从输入种类判定部440输入了"发话句子的类型"的回答取得部380,根据与所输入的检索结果信号对应的"话题标题"和所输入的"发话句子的类型",从与该"话题类型,,对应的回答种类群中确定与"发话句子的类型"(DA等) 一致的回答种类。 I.e., the topic retrieval unit 370 from the retrieval result signal input, and an input unit 440 of the "uttered sentence type" from the reply retrieval unit 380 determines the input type, the search results according to the input signal corresponding to the "topic title" and the Enter the "uttered sentence type", from the same kind of answer to the "Topic type ,, corresponding answer in determining the type of group," uttered sentence type "(DA, etc.).

接着,回答取得部380经由管理部310,将在步骤S2209中取得的回答句子830输出到输出部600 (步骤S2210)。 Then, the reply retrieval unit 380 via the management unit 310, the output acquired in step S2209 reply sentence 830 to the output unit 600 (step S2210). 从管理部310取得了回答句子的输出部600输出所输入的回答句子830。 From the management unit 310 has made a reply sentence output section 600 outputs the inputted answer sentence 830.

以上,结束了谈话空间会话控制处理的说明,返回图38重新开始主处理的说明。 This concludes the description of the discourse space conversation control process returns illustrating 38 main process restarted.

会话控制部300,当结束谈话空间会话控制处理时,执行CA会话控制处理(S1803 )。 Session control unit 300, when the end of the discourse space conversation control process is executed CA conversation control process (S1803). 但是,在计划会话控制处理(S1801)以及谈话空间会话控制处理(S1802)中进行了回答句子输出的情况下,会话控制部300不进行CA会话控制处理(S1803 ),进行基本控制信息更新处理(S1804)来结束主处理。 However, under the plan conversation control process (S1801), and the discourse space conversation control process (S1802) performed a reply sentence output, the conversation control unit 300 does not perform CA conversation control process (S1803), the basic control information update process ( S1804) to end the main process.

CA会话控制处理(S1803 )是判定用户发话是"正在说明某事"、"正在确认某事"、"正在进行非难或攻击"、还是"这些以外,,,输出与用户发话的内容以及判定结果对应的回答句子的处理。通过进行该CA会话控制处理,即使在计划会话控制处理以及谈话空间会话控制处理的任何一个中都无法输出适合于用户发话的回答句子,也可以输出不中断与用户的会话的流程地继续的、 比如"相连"的回答句子。 CA conversation control process (S1803) is to determine the user's utterance is "being described something", "something is being confirmed," "ongoing disapproval or attack", or "these outside ,,, output and user utterance content and determine the results corresponding to the reply sentence processing performed by the CA conversation control process, even when the plan conversation control process and the discourse space conversation control process of any one of the output can not be adapted to the user's utterance reply sentence to be output without interrupting the user's Session process continue, such as "connected" to answer sentence.

然后,会话控制部300进行基本控制信息更新处理(S1804 )。 Then, the conversation control unit 300 performs basic control information update process (S1804). 在该处理中, 会话控制部300、更详细而言是管理部310,在计划会话处理部320进行了回答句子输出的情况下,将基本控制信息设定为"结束",在计划会话处理部320 停止了回答句子输出的情况下,将基本控制信息设定为"放弃",在谈话空间会话控制处理部330进行了回答句子输出的情况下,将基本控制信息设定为"维持,,,在CA会话处理部340进行了回答句子输出的情况下,将基本控制信息设定为"继续"。 In this process, the conversation control unit 300, more specifically, the management unit 310, the plan conversation process unit 320 outputs a reply sentence in the case, the basic control information is set to "end" in the plan conversation process unit 320 stops output reply sentence in the case, the basic control information is set to "give up" in the discourse space conversation control unit 330 outputs a reply sentence under the circumstances, the basic control information is set to "maintain ,,, In the CA conversation process unit 340 outputs a reply sentence in the case, the basic control information is set to "continue."

在该基本控制信息变更处理中设定的基本控制信息,在所述计划会话控制处理(S1810)中被参照,在计划的继续或重新开始中被利用。 In the basic control information changing set of basic control information in the plan conversation control process (S1810) is referring to is the use of continuing or re-start the program.

以上,通过每当接受用户发话时执行主处理,回答处理部21可以才艮据用户发话执行预先准备的计划,并且针对计划中未包含的话题也可以进行适宜的应答。 Above, the main processing executed by the user's utterance whenever acceptable answer processing unit 21 according to the user's utterance can only Burgundy implementation plans prepared in advance, and plan for a topic that is not included can also be an appropriate response.

Referenced by
Citing PatentFiling datePublication dateApplicantTitle
CN103177519A *22 Feb 201326 Jun 2013秦方Remote monitoring system and remote monitoring method with door alarm and real-time voice response functions
CN103177519B *22 Feb 201325 Mar 2015秦方Remote monitoring system and remote monitoring method with door alarm and real-time voice response functions
CN103297503A *8 May 201311 Sep 2013南京邮电大学Mobile terminal swarm intelligent perception structure based on layered information extraction server
CN103297503B *8 May 201317 Aug 2016南京邮电大学基于分层次信息提取服务器的移动终端群智感知系统
CN103578464A *18 Oct 201312 Feb 2014威盛电子股份有限公司Language model establishing method, speech recognition method and electronic device
CN104508629A *25 Jul 20138 Apr 2015托伊托克有限公司Artificial intelligence script tool
CN104601538A *30 Oct 20146 May 2015夏普株式会社Server, speaking control method, speaking device, and speaking system
Classifications
International ClassificationH04M3/42
Cooperative ClassificationG10L15/22
European ClassificationG10L15/22
Legal Events
DateCodeEventDescription
24 Feb 2010C06Publication
28 Apr 2010C10Request of examination as to substance
31 Jul 2013CORBibliographic change or correction in the description
Free format text: CORRECT: APPLICANT; FROM: ARUZE CORP. TO: GLOBAL ENTERTAINMENT CORPORATION
31 Jul 2013C53Change or modification
14 Aug 2013C14Granted