Classifications

• • G—PHYSICS
  • G06—COMPUTING; CALCULATING OR COUNTING
  • G06F—ELECTRIC DIGITAL DATA PROCESSING
  • G06F40/00—Handling natural language data
  • G06F40/20—Natural language analysis
  • G06F40/279—Recognition of textual entities
• • G—PHYSICS
  • G06—COMPUTING; CALCULATING OR COUNTING
  • G06F—ELECTRIC DIGITAL DATA PROCESSING
  • G06F40/00—Handling natural language data
  • G06F40/30—Semantic analysis
  • G06F40/35—Discourse or dialogue representation
• • G—PHYSICS
  • G06—COMPUTING; CALCULATING OR COUNTING
  • G06F—ELECTRIC DIGITAL DATA PROCESSING
  • G06F40/00—Handling natural language data
  • G06F40/40—Processing or translation of natural language
  • G06F40/55—Rule-based translation
  • G06F40/56—Natural language generation
• • G—PHYSICS
  • G06—COMPUTING; CALCULATING OR COUNTING
  • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
  • G06N5/00—Computing arrangements using knowledge-based models
  • G06N5/04—Inference or reasoning models
  • G06N5/043—Distributed expert systems; Blackboards

Definitions

• the invention relates to a model and a method of one implementation of a rational dialog agent as the core of a dialog system or a multi-agent system.
• the invention applies to human-agent interaction systems (human-machine dialogue) but also to agent-agent interaction systems (inter-agent communication and cooperation).
• plan-oriented approaches consider an intervention in a communication situation not only as a collection of signs (for example a sequence of words) but as the observable realization of communicative actions (also called depending on the context, acts of language or dialogue) such as informing, asking, confirming, engaging.
• the depositor has developed a new approach based on rational interaction or rational agent dialoguing.
• the depositor first sought to maximize the user-friendliness of the interactions between users and automatic services.
• Sadek 91a Sadek M.D. Mental attitudes and rational interaction: towards a formal theory of communication. PhD in Computer Science, University of Rennes I, France, 1991.
• Sadek 91b D. Sadek Dialogue acts are rational plans. Proceedings ESCA tutorial and Research Workshop on the Structure of Multimodal Dialogue, Maratea, Italy, 1991.
• Sadek 92 Sadek M.D. A study in the logic of intention. Proceedings of the 3rd Conference on Principles of Knowledge Representation and Reasoning (KR'92), pages 462-473, Cambridge, MA, 1992. Sadek 93: Sadek M.D. Foundations of dialogue: rational interaction. Proceedings of the 4th summer school on Natural Language Treatments, pages 229-255, Lannion, France, 1993.
• Sadek 94a Sadek M.D. Mental attitudes and foundation of cooperative behavior. Pavard, B., editor, Cooperative systems: from modeling to design, Octares Eds. , pages 93-117, 1994.
• Sadek 94c Sadek MD Towards a theory of belief reconstruction: application to communication. In (SPECOM94): 251-263.
• Sadek et al 94 Sadek MD, Ferrieux A., & Cozannet A. Towards an artificial agent as the kernel of a spoken dialogue system: A progress report. Proceedings of the AAI'94 Workshop on Integration of Natural Language and Speech Processing, Seattle, WA, 1994.
• Sadek et al 95 D. Sadek, P. Bretier, V. Cadoret, A. Cozannet, P. Dupont, A. Ferrieux, & F. Panaget: A cooperative spoken dialogue system based on a rational agent model: A first implementation on the AGS application. Proceedings of the ESCA tutorial and Research Workshop on Spoken Dialogue Systems, Hanstholm, Denmark, 1995.
• Sadek et al 96a Sadek M.D., Ferrieux A., Cozannet
• Sadek et al 91 M.D. Sadek, P. Bretier, & F.
• Bretier 95 P. Bretier. Cooperative oral communication: contribution to logical modeling and to the implementation of a rational agent dialoguing.
• the user-friendliness of the interaction manifests itself among other things by the system's ability to negotiate with the user, by his capacity to interpret requests by taking into account the context, by his capacity to determine the implied intentions of the user and to carry out with him a flexible interaction which does not follow a preconceived plan once for all.
• the technology developed by the depositor is based on the basic principle which is that: for an automatic system to be able to carry out intelligent dialogues, this system cannot be simulated by an automaton.
• the object of the present invention is the production of a software agent which by its construction is rational.
• the addition of appropriate principles also makes it communicative and cooperative.
• the technology developed by the applicant allows the implementation of a rational agent dialoguing as well as the core of a dialogue system as well as the agent of a multi-agent system.
• communication between such agents is no longer done using natural language but a formal (logical) language adapted to the interaction capacities of said agents.
• the invention relates more particularly to a model and a method for implementing a rational agent dialoguing as the core of a dialogue system or a multi-agent system.
• the method of implementing a rational agent dialoguing as the core of a dialogue system and / or as an element (agent) of a multi-agent system comprises the following steps: - definition of an architecture concept of a rational agent dialoguing,
• the definition of the implementation mechanisms is carried out so as to obtain a direct correspondence between these mechanisms and said model.
• the formal specification of the different components of formal architecture and their combination includes a level of axioms of rationality, a level of axioms of communication, a level of axioms of cooperation.
• the definition of the software architecture implementing the formal architecture comprises: a rational unit comprising an implementation layer of the level of rationality axioms, a layer of implementation of the level of communication axioms, a layer of implementation of the level of cooperation axioms, corresponding respectively to the axioms of the formal model.
• the definition of the software architecture implementing the formal architecture also includes:
• the rational unit, the generation module and the understanding module implement mechanisms for implementing the formal model.
• the generation module is able to transcribe a logical statement produced by the rational unit in natural language for the use of the system.
• the comprehension module is able to interpret a statement of the user into an understandable logical statement of the rational unit.
• the subject of the invention is also a rational agent dialoguing as the core of a dialogue system and / or as an element (agent) of a multi-agent system, comprising: - a definition of a conceptual architecture, a formal specification of different components of this architecture and their combination to obtain a formal model, mainly characterized in that it comprises: - a definition of a software architecture implementing the formal architecture,
• the data comprise data for the implementation of a formal model comprising: - a layer of implementation of rationality axioms, a layer of 1 implementation of communication axioms, a layer of implementation of axioms of cooperation, corresponding respectively to the axioms of the formal model.
• the agent further comprises: a module for generating a statement in natural language from a logical statement resulting from the rational unit and a module for understanding to supply a statement in logical language to the rational unit from a statement in natural language, these modules thus implementing a communication level layer in natural language.
• the invention also relates to an information server, comprising means for implementing a human-machine dialogue system, the core of which is based on the implementation of a rational dialoging agent as defined above.
• the invention also relates to a multi-agent system comprising communicating agents, each agent comprising means for implementing an interaction, the system comprising at least one agent, the core of which is based on the implementation of a rational agent interacting as described above.
• FIG. 1 represents the software architecture of a rational agent interacting
• FIG. 3 represents in more detail the software architecture of a dialoging agent as the core of a dialogue system (in particular oral)
• - Figure 4 shows an architecture showing a rational agent dialoguing as the core of a multi-agent system.
• the depositor has implemented these principles by means of a rational unit 100 which constitutes the core of each agent and which determines his reactions to external events, whether these are solicitations (requests, responses, confirmations etc) of human users or requests from other software agents (this is the case when an agent is the core of a multi-agent system).
• the rational unit 100 is animated by an inference engine which automates reasoning according to the principles of rational interaction which the agent's programmer can adapt or enrich, in a declarative manner, according to the task to be accomplished. To this end, as will be explained below, these reasonings are guided by predetermined axiom diagrams (listed in appendices) and entered into the unit by the agent's programmer declaratively according to the task. to be completed by said agent.
• FIG. 1 illustrates the diagram of a software architecture of an agent in the case where such an architecture is applied to the constitution of a system of dialogue with users.
• FIG. 1 therefore represents the architecture of an agent interacting with a user, through, as will be seen, an understanding module 150 and a generation module 160.
• This architecture corresponds to a first possible family of applications which is (user-friendly) user-service interaction.
• the rational unit 100 is connected to an outside interface 140.
• This interface therefore includes the comprehension module 150 which receives statements in natural language and interprets these statements in a logical statement which serves as an input to the rational unit 100.
• the interface also includes the generation module 160 which expresses the reaction of the rational unit 100 into a natural language statement for the user.
• the rational unit 100 is the central service entity to provide either information (train times, stock market prices, weather forecasts, etc.), reservations or purchases, or even the search for information from the Internet.
• the principles of cooperation established in the rational unit and the natural language processing modules ensure user-friendly interaction with the user. This interaction can be carried out directly by speech by integrating into the dialogue system thus formed speech recognition and synthesis modules (not shown in this figure).
• the rational unit 100 can alone constitute the core of an autonomous software agent.
• this unit interacts with other software agents by means of a communication language between agents such as "Agent Communication Language" (A. CL. Adopted as standard by the FIPA consortium).
• Agent Communication Language A. CL. Adopted as standard by the FIPA consortium.
• the services which the agent can render are then for example transactions on electronic markets, tasks of administration of networks, diffusion of information.
• the rational unit 100 implements principles from the theory of rational interaction whose objective is to formalize and automate the rational behavior of an agent when interacting with other agents. or service users.
• This theory is based on two main notions: the notion of modal logic on the one hand, the objective of which is to allow the mental attitudes of autonomous agents to be represented and, the notion of speech acts on the other, whose objective is to specify the effects of communication on the mental attitudes of agents.
• the state of an agent at a given moment in a communicating exchange is thus characterized by a set of mental attitudes.
• the mental attitudes that can be represented are for example the belief usually noted by the operator K and the intention noted by the operator I.
• Ks designates the belief operator for the system and Ku this same operator for the user.
• Modeling consists of a logical statement or logical language, for example:
• the computation of these reactions is carried out by the inference engine 101.
• the rational unit 100 therefore comprises a data set 102 which includes the axioms of the formal model of the dialoguing rational agent. These data implement the layers of rationality of communication and cooperation of the agent.
• Environmental requests for example user requests, or those of other software agents are transmitted to the rational unit 100 in the form of a logical statement ACL of the theory of rational interaction.
• the inference engine 101 is capable of calculating the consequences of this statement and in particular the possible responses or requests for clarifications to be provided to the interlocutor (whether it is an agent software or a human user) but also other non-communicative actions.
• the inference engine 101 examines whether it does not have a behavioral principle which can be applied to this statement to deduce the logical consequence or consequences. This procedure is then applied to these new consequences until the possibilities are exhausted.
• the inference engine 101 isolates the communication or other actions which it must perform and which then form the reaction of the rational agent.
• the first step of the inference procedure is to put the processed statements into normal form in order to ensure that each statement is presented only in a single given syntactic form in order to be able to sort and compare the statements.
• the inference procedure then consists, for each statement processed, in verifying whether this statement corresponds to one of the schemes of axioms 102 which code the principles of rational behavior adopted.
• the mechanism of this verification is mainly based on the Prolog language unification operation.
• axiom diagrams can be modified by the programmer of the rational unit 101 who can remove or add axiom diagrams or modify those already existing to refine the behavior of the rational unit. These changes can be made dynamically. In this case, the rational unit changes its behavior progressively.
• the reasoning of the rational unit is based on a set of data which strongly depend on the application sought for the rational agent.
• the semantic network 120 makes it possible to express notions of classes and subclasses, and of instance of each class. It also defines the notion of relationship between classes which applies to the different instances of the classes.
• the semantic network 120 will include at least the classes "person” (whose instances will be the set of people known in the agenda) and "function" (whose instances will be known functions ).
• the semantic network includes the fact Prolog: the -function- of (John, Advertising).
• Access to the semantic network 120 is made at any time during the inference procedure when the consequences of the inference depend on the nature of the data.
• the response of the rational agent will depend on his interrogation of the semantic network 120.
• the semantic network 120 can also have notions of semantic proximity which are partially useful for producing cooperative reactions of the rational agent.
• the advertising function will probably be determined as semantically closer to the marketing engineer function than to the garage owner.
• relaxation or relaxation
• constraint restriction This construction allows two symmetrical operations called relaxation (or relaxation) and constraint restriction.
• relaxation or relaxation
• constraint restriction aims to give close responses to requests close to the initial request when the response to the latter does not exist.
• the inference procedure can trigger a relaxation step to be able to give the contact details of advertisers.
• the restriction seeks to seek how to specify a request that is too broad. If there are 500 advertisers registered in the agenda, a restriction step will give the most discriminating dimension of this too large set (for example the company or the advertiser works) in order to be able to ask a relevant question to identify the user's request.
• FIG. 2 also makes it possible to illustrate that the rational unit 100 of a rational agent comprises a generic part independent of the application and a part dependent on the application.
• the diagram in FIG. 3 illustrates in more detail the software architecture of an agent according to the invention.
• the natural language comprehension module 150 interprets a statement of the user into a logical statement understandable by the rational unit 100.
• the vocabulary covered by this module depends in part on the service that the rational agent must provide. This application-dependent part is mainly present in the semantic network 120 of the rational unit, which explains why the understanding module 150 uses a lot of data coming from the semantic network 120.
• the comprehension module 150 is able to take into account the utterance of the user as a series of small syntactic structures (most often words) which will each activate one (or more in the case of synonyms) notion (s) resulting (s) of the semantic network 120.
• the link between the user's input vocabulary and the semantic network 120 is therefore made by means of a concept activation table 131 which indicates which semantic notion (s) corresponds to the words ( or series of words) of vocabulary.
• the understanding module therefore has a list of activated concepts (or even several in the case of synonyms). It is able to transform them into a logical statement formed by a semantic completion process. This process starts from the assumption of semantic connexity of the statement of the user, that is to say that the concepts which he evoked are in relation to each other.
• the module 150 is able to link together, through relationships present in the semantic network, including by creating new concepts if necessary. The process determines the concepts implied in the user's statement.
• the semantic completion calls for a weighting function 132 which makes it possible to fix a numerical weight for each relation of the semantic network, representing the likelihood of the vocations of this relation.
• the completion process takes into account a notion of likelihood when it has to determine which concepts are implied by the user.
• These weights also allow a cost to be associated with each possible interpretation in the case of a synonym. Thus, only one statement will ultimately be retained by the comprehension module, the one with the lowest cost.
• the understanding module 150 must take into account the context of the user's statement. For this, it has the concepts previously mentioned both by the user and by the agent himself in his responses to the user. Some of these can therefore be used during the completion process.
• the comprehension module 150 does not use a syntactic or grammatical analyzer. This allows him to correctly interpret syntactically incorrect statements, which is particularly important in the context of oral dialogue (and the use of voice recognition), the syntax of spontaneous speech being looser,.
• the semantic data of the network indeed represent universal notions. This point particularly facilitates the transfer of an application from one language to another language.
• the generation module 160 accomplishes the reverse task of the understanding module. It is able to transcribe a sequence of communicative acts produced by the rational unit 100 into a statement of the natural language of the user.
• the generation process operates in two phases.
• the first phase consists in making all the decisions as to the linguistic choice which is offered to verbalize the sequence of communicative acts provided at the input of the module.
• the generator 160 uses, among other things, elements from the context of the dialogue to construct the statement most suited to the current situation.
• the module 160 will have to make a choice between equivalent formulations such as "John's telephone number is” or “John's number is” or “his number is”, “it is the ... "depending on the context of the dialogue.
• the objective of this first phase is to construct an intermediate representation of the utterance using a notion of abstract linguistic resources 133.
• An abstract linguistic resource represents either a lexical resource 135, for example common nouns, verbs, adjectives, or a grammatical resource, ie the syntactic structure.
• the second phase uses this abstract representation to construct the final statement.
• the comprehension modules 150 and generation 160 use written texts as input format and respectively output format.
• the recognition module 170 records a voice signal from the user in a text corresponding to the spoken utterance. This module 170 is for example essential when a rational agent is used as a telephone server: the only possible interaction is then vocal.
• this agent can communicate with other software agents, for example across the network.
• ACL communication primitives defined by the theory of rational interaction constitute a language of communication between agents which allow them to perform unambiguous interactions.
• the agents trained by a rational unit 100 and their semantic network 120 without the interaction components in natural language are particularly well suited to the use of the ACL communication language between software agents to form multi-system agent as shown in Figure 4.
• the invention has been implemented with a SUN Ultral station (provided with a 166 megahertz processor) and on a SUN Ultra2 station (having two 64-bit processors and a frequency of 300 megahertz).
• a random access memory is used, the size of which can be around 32 egabytes minimum.
• the maximum system response time is 2 seconds on the Ultra2 platform and 5 seconds on Ultral.
• Connection can be made to a digital network using an ISDN-Basic Rate Interface service integrated digital network card.
• the three modules that have been described, comprehension 150, generation 160 and rational unit 100 have been implemented in Prolog (Quintus version 3.3 for Solaris 2.5). Communication between the various modules and the speech recognition and synthesis systems is carried out by a program written in C language, a prototype of the invention has been developed under Solaris, but a version not including the recognition and speech synthesis was brought under WINDOWS NT 4.0.
• the mental attitudes considered as semantically primitive, namely belief, uncertainty and choice (or preference) are formalized respectively by the modal operators K, U, and C.
• Formulas such as K ( ⁇ , p), U (i, p), and C (i, p) can be read respectively "i believe (or think) p (is true)", "i is uncertain of (the truth of) p” and "i want p is currently true.
• the logical model adopted for the operator K accounts for properties of interest to a rational agent, such as the consistency of his beliefs or his capacity for introspection, formally characterized by the validity of logical schemes such as
• the logic model also guarantees the validity of desirable properties such as, for example, the fact that an agent cannot be uncertain of his own mental attitudes (
• -. T7 (i, M ( i, ⁇ ), M belonging to (K, sK, C, -sC, U, -> ⁇ etc ⁇ ).
• the logical model for the choice entails properties such as the fact that an agent "assumes" the logical consequences of his choices or that an agent cannot not choose the courses of events in which he thinks he is already there
• attitude of intention which is not semantically primitive, is formalized by the operator I which is defined (in a complex way) from the operators C and K.
• a formula such as I (i, p) can be read "i intend to make p”.
• the schematic variables ⁇ , - ⁇ ⁇ 2 • • •, are used to denote expressions of
• the operators Feasible, Done and Agent (i,) are also introduced, such as the formulas Feasible ( ⁇ p), Done (p) and Agent (i,) respectively mean that (action or expression of action) ⁇ can take place after which p will be true, ⁇ has just taken place before which p was true, and i denotes the sole agent of the events appearing in ⁇ .
• the second principle stipulates that an agent who intends to take a given action necessarily adopts the intention that this action is doable, if he does not think that it is already done: this is expressed formally by the validity of the following scheme:
• avoiding redundancy is a component of cooperative behavior, which can be expressed as follows, in terms of elementary property (which, in reality, is not primitive but derives directly from the very definition of the concept of intention ): if an agent intends to let an agent know about a proposition p, then i must think that j does not already know it. This is formally reflected in the validity of the following scheme:
• a corrective response is generated with the intention of correcting a belief of the interlocutor, judged to be erroneous.
• This belief generally constitutes an inferred presupposition (by implicature (Grice 75;) from the recognized communicative act.
• the intention in question is generated in an agent each time his belief about a proposition which in no one believes not its competent interlocutor, is in contradiction with that of its interlocutor. This translates formally by the validity of the following diagram:
• the first part of this property stipulates that at the end of a phenomenon that an agent perceives and to which either he cannot associate an intelligible event, or any event that he can associate with him is unacceptable in light of his beliefs, the agent will adopt the intention of knowing what has been achieved, typically by generating a request for repetition.
• the second part of this property concerns only the case where the agent cannot, in accordance with his mental state, accept any event achievable by what he has observed; in this case, the agent will adopt the intention of letting the author of the event know his disapproval of what he has "understood", which, in terms of language statements, may be manifested, for example, by the statement of what prohibits the officer from admitting the act in question.
• the two parts of this property are expressed by the validity of the following two schemes, the predicates Observe (i, o) and Réalise (o, e) meaning respectively that the agent i has just observed the observable entity o (such as a statement, for example), and that the observable entity o is a way of realizing the event e:
• T 1 can be:

Landscapes

• Engineering & Computer Science (AREA)
• Theoretical Computer Science (AREA)
• General Engineering & Computer Science (AREA)
• Artificial Intelligence (AREA)
• Computational Linguistics (AREA)
• Physics & Mathematics (AREA)
• General Physics & Mathematics (AREA)
• Health & Medical Sciences (AREA)
• Audiology, Speech & Language Pathology (AREA)
• General Health & Medical Sciences (AREA)
• Computing Systems (AREA)
• Mathematical Physics (AREA)
• Software Systems (AREA)
• Evolutionary Computation (AREA)
• Data Mining & Analysis (AREA)
• Machine Translation (AREA)
• Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
• Stored Programmes (AREA)
• Information Transfer Between Computers (AREA)

Applications Claiming Priority (3)

Publications (1)

Family

ID=9534443

Family Applications (1)

Country Status (11)

Families Citing this family (210)

Family Cites Families (26)

• 1998
  • 1998-12-23 FR FR9816374A patent/FR2787902B1/fr not_active Expired - Fee Related
• 1999
  • 1999-12-21 JP JP2000591502A patent/JP2002533827A/ja not_active Abandoned
  • 1999-12-21 CN CN99816347A patent/CN1335959A/zh active Pending
  • 1999-12-21 CA CA002356384A patent/CA2356384A1/fr not_active Abandoned
  • 1999-12-21 BR BR9916421-3A patent/BR9916421A/pt not_active IP Right Cessation
  • 1999-12-21 KR KR1020017007978A patent/KR100694734B1/ko not_active IP Right Cessation
  • 1999-12-21 EP EP99961137A patent/EP1141823A2/de not_active Withdrawn
  • 1999-12-21 US US09/869,205 patent/US7376632B1/en not_active Expired - Fee Related
  • 1999-12-21 AU AU17851/00A patent/AU773217B2/en not_active Ceased
  • 1999-12-21 WO PCT/FR1999/003242 patent/WO2000039672A2/fr active IP Right Grant
  • 1999-12-21 IL IL14381299A patent/IL143812A0/xx unknown

Non-Patent Citations (1)

Also Published As

Similar Documents

Legal Events