1 / 36

Motivation

Towards a Theoretical Framework for the Integration of Dialogue Models into Human-Agent Interaction. Motivation. How should an intelligent agent incorporate communication? How does communication and behavior integrate within an agent model?

minty
Download Presentation

Motivation

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. Towards a Theoretical Framework for the Integration of Dialogue Models into Human-Agent Interaction

  2. Motivation • How should an intelligent agent incorporate communication? • How does communication and behavior integrate within an agent model? • How can ideas from many different dialogue models and conversation examples by incorporated? • How can one validate the correctness of an agent conversational model?

  3. Motivation Informal Argumentation Persuasion Belief Grounding Integrated Model Cooperative Planning Negotiation Formal Argumentation

  4. Percepts Actions DIALOGUE CAPABLE AGENT Conversation Dialogue Manager Internal API or Language Behavioral Model Dialogue Agent Paradigm • Embedded dialogue manager • Perception processing • Embedded behavior model

  5. Goal • A unified conversational architecture for intelligent agents and assistants • Representation of communication • Incorporation of communication within behavior • Incorporating a variety of models and ideas into a single integrated model • Validation of the conversational model • Independent of dialogue interpreter or agent

  6. Bring behavior to communication as much as possible Focus Agent Implementation Communication Behavior

  7. Focus • Separate dialogue interpreter from agent • Parallel development of each • Interchangeable components Agent Implementation Communication Behavior

  8. Human Interpreter

  9. The Practical CommunicationLanguage (PCL) Hypothesis There exists a language between that of a human conversational participant and that of an intelligent agent. This language is capable of abstracting away the complexity of human language while yet maintaining the practical information of the conversation. Adding to The Practical Dialogue Hypothesis and The Domain-independence Hypothesis stated in Allen 2000.

  10. Current Utterance-Based Languages • Application Programmer Interfaces (API) • Task Management Interface • Specialized Languages • Artificial Discourse Language • Universal Communication Language (Interlingua) • Parameterized Action Representation • Discourse and Speech Act Tags • Agent Communication Languages • Foundation for Intelligent Physical Agents (FIPA-ACL) • Knowledge Query Manipulation Language (KQML)

  11. Searching for the language… True PCL is ideal and volatile Ever expanding definition of ‘practical’ PCL should be abstracted* of • Region and dialect aspects of language. • Informal, Colloquial, Slang and Idiomatic expressions. • Modality (Spoken, Written, Gestural, GUI) *Translated or Incorporated not discarded.

  12. Approach • Task Communication Language (TCL) • Messages to/from Dialogue Interpreter TCL Message • Set of integrated models Task Model Task Communication Model Interaction Model

  13. TCL Messages • Header • Generator: Generated utterance or gesture • Addressee: Intended Receivers of message • Receiver: Participants who saw or heard • Uncertainty in all above fields • Interpretation Stack • Information obtained at al levels of translation • Used by feedback mechanism for improving interpreters • Content • Meaning-Action Concept

  14. Conversational Paradigms

  15. Conversational Paradigms

  16. Conversational Paradigms • Not just endpoint to endpoint • Multiple segmentations • A conversation between people listening in on another conversation

  17. Meaning-Action Concept • Meaning of utterance or gesture • Possible association with action. • “propose( action: )” • “propose( goal: )” • “reject( goal: )” • “counter-propose( action: )” • “query( justification( action: ) )”

  18. Meaning-Action Concepts (MAC) • Defined in ontological format • Allows for rollback to known concepts • Manageable growth of concept space Proposal( ) Commit( ) Counter-Proposal( ) Commit( confidence:30 )

  19. Task Communication Expression • First-order logic expression of MAC. • Conjunction: Multiple Meanings • Disjunction: Ambiguity • Expressiveness and complexity

  20. Focus Communication Behavior

  21. Task Model • Task Concepts:

  22. Task Model • Task Operations:

  23. Task Model Task Model Communication Behavior

  24. Task-Communication Model • Integrate the task concepts and operators • Communication with a dialgue interpreter • Task manipulation of an intelligent agent • Modeling can be language independent • CFSM, CPetriNet, Inference-Based, BDI...

  25. Task-Communication Model • Nested task operators • Lower layers: • Persuasion, inquiry, deliberation, formal argumentation, informal argumentation, clarification, explanation… • Higher layers: • Negotiation, cooperative planning, learning through orders, command and control…

  26. Task-Communication Model Task Model Task-Communication Model Communication Behavior

  27. Trivial Example: Communicative acts TCL Messages Behavioral acts Agent integration Task-Communication Model [IN]: Propose( Action A ) Evaluate( Action A ) [OUT]: Accept( Action A ) [OUT]: Reject( Action A )

  28. Integration Model • Generated automatically through tracing the Task-Communication Model • Represents incoming and outgoing message sequences and possibilities

  29. Task-Communication Model [IN]: Propose( Action A ) Evaluate( Action A ) [OUT]: Accept( Action A ) [OUT]: Reject( Action A )

  30. Interaction Model • Extraction of Input-Output Sequences [IN]: Propose( Action A ) [OUT]: Accept( Action A ) [OUT]: Reject( Action A ) [OUT]: Counter( Action A ) [OUT]: Refine( Action A ) [OUT]: Clarify( Goal G )

  31. Interaction Model Task Model Task-Communication Model Communication Behavior Interaction Model

  32. Interaction Model • Validated with known TCL sequences • If sequence is covered, path validated • If sequence is missing, update model • Assists in integration of models • Prove various properties • Turn Taking • Liveness • Sanity checks

  33. Mixed-Initiative Control • No longer in hands of dialogue interpreter • Also managed by ‘task communication model’ • Task-communication model must • Initiate dialogue sequences • Manage • turn-taking • context tracking • autonomy

  34. Stratagus • Open source real-time strategy engine • Multiple data sets for varying games • Dynamically changing environment • Real time resource management

  35. Discussion

More Related