towards a context-based dialog management layer for expert systems victor hung, avelino gonzalez...

20
Towards A Context-Based Dialog Management Layer for Expert Systems Victor Hung, Avelino Gonzalez & Ronald DeMara Intelligent Systems Laboratory University of Central Florida eKNOW Conference February 4, 2008

Upload: tiffany-chandler

Post on 13-Jan-2016

214 views

Category:

Documents


2 download

TRANSCRIPT

Page 1: Towards A Context-Based Dialog Management Layer for Expert Systems Victor Hung, Avelino Gonzalez & Ronald DeMara Intelligent Systems Laboratory University

Towards A Context-Based Dialog Management Layer for Expert Systems

Victor Hung, Avelino Gonzalez & Ronald DeMaraIntelligent Systems Laboratory

University of Central Florida

eKNOW ConferenceFebruary 4, 2008

Page 2: Towards A Context-Based Dialog Management Layer for Expert Systems Victor Hung, Avelino Gonzalez & Ronald DeMara Intelligent Systems Laboratory University

University of Central Florida www.ucf.edu

Agenda

Introduction Background Approach Evaluation Project LifeLike

Page 3: Towards A Context-Based Dialog Management Layer for Expert Systems Victor Hung, Avelino Gonzalez & Ronald DeMara Intelligent Systems Laboratory University

University of Central Florida www.ucf.edu

Expert System

Introduction

Interface KnowledgeBase

User

Page 4: Towards A Context-Based Dialog Management Layer for Expert Systems Victor Hung, Avelino Gonzalez & Ronald DeMara Intelligent Systems Laboratory University

University of Central Florida www.ucf.edu

Introduction

Focus on Interface Component Text-based Conversation-based

Idealistic view of Human Computer Interaction Media and science fiction expectations Open, unconstrained dialog Natural conversational flow

Page 5: Towards A Context-Based Dialog Management Layer for Expert Systems Victor Hung, Avelino Gonzalez & Ronald DeMara Intelligent Systems Laboratory University

University of Central Florida www.ucf.edu

Background Natural Language Processing (NLP)

Relevant techniques Not primary focus

Dialog System Design General issue at hand

Context-Based Reasoning (CxBR) Main architectural method

Page 6: Towards A Context-Based Dialog Management Layer for Expert Systems Victor Hung, Avelino Gonzalez & Ronald DeMara Intelligent Systems Laboratory University

University of Central Florida www.ucf.edu

Natural Language Processing

Major issues in NLP (Wilks, 2005) Linguistic Systems

Resolve ambiguities (ASR, syntactic, semantic) Knowledge Representation

Relationship of language and logic Information Corpora

Ontologies, tree banks, Semantic Web project Statistical and Quantitative Methods

Machine learning techniques on linguistic data

Page 7: Towards A Context-Based Dialog Management Layer for Expert Systems Victor Hung, Avelino Gonzalez & Ronald DeMara Intelligent Systems Laboratory University

University of Central Florida www.ucf.edu

Dialog System Design

Dialog Manager

Dialog History

Interpreter

Generator

Dialog Model System Task Model

Domain KnowledgeManager

Domain Task Model

Knowledge Base 1

Knowledge Base 2

Knowledge Base 3

Flycht-Eriksson and Jönsson, 2000

Page 8: Towards A Context-Based Dialog Management Layer for Expert Systems Victor Hung, Avelino Gonzalez & Ronald DeMara Intelligent Systems Laboratory University

University of Central Florida www.ucf.edu

Chatbot Systems Conversation Agents

General chatting techniques ELIZA (Weizenbaum, 1966) Persona-AIML (Galvão et al, 2004) Genericity (Sansonnet et al, 2006)

Embodied Conversation Agents Sam (Cassell et al, 2000) Laura (Bickmore and Picard, 2004) Mel (Lee et al, 2005) Sergeant Blackwell (Traum, 2006)

Page 9: Towards A Context-Based Dialog Management Layer for Expert Systems Victor Hung, Avelino Gonzalez & Ronald DeMara Intelligent Systems Laboratory University

University of Central Florida www.ucf.edu

Dialog Systems

Open issues Evaluation process is too subjective No tremendous breakthroughs since ELIZA Naturalness still a barrier Scarce work on context-based dialog

management Cognitive model-based systems require

extensive expertise

Page 10: Towards A Context-Based Dialog Management Layer for Expert Systems Victor Hung, Avelino Gonzalez & Ronald DeMara Intelligent Systems Laboratory University

University of Central Florida www.ucf.edu

Context-Based Reasoning

Contexts Context-Transition Logic Missions Agent Interface

Context Topology

Mission

Context C0

Context C4

Context C1

Context C3

Context C2

Agent Interface

EnvironmentData

Agent Action

Inference Engine

Stensrud et al, 2004

Page 11: Towards A Context-Based Dialog Management Layer for Expert Systems Victor Hung, Avelino Gonzalez & Ronald DeMara Intelligent Systems Laboratory University

University of Central Florida www.ucf.edu

Context-Based Methods in NLP

Situational contexts used for general NLP tasks Speech recognition

Fügen et al, 2004 Sarma and Palmer, 2004 Serridge, 1997 Young, 1989

Context-based ML (Lieberman et al, 2005)

Page 12: Towards A Context-Based Dialog Management Layer for Expert Systems Victor Hung, Avelino Gonzalez & Ronald DeMara Intelligent Systems Laboratory University

University of Central Florida www.ucf.edu

Context-Based Methods in NLP

Open issues Context-based methods treated as a

gimmicky technique Still at a state of infancy

Use of context heavily prevalent for NLP ambiguity resolution tasks

Dialog management applications not fully explored

Page 13: Towards A Context-Based Dialog Management Layer for Expert Systems Victor Hung, Avelino Gonzalez & Ronald DeMara Intelligent Systems Laboratory University

University of Central Florida www.ucf.edu

Approach

Goal Management Allow for non-linear conversation flow Support for goals

Asynchronous Multiple

Illusion of open dialog Limited to an expert domain, and the user is

cognizant of the dialog system’s functionality as an expert entity

User’s goals are limited to those related to the chatbot’s expertise

Page 14: Towards A Context-Based Dialog Management Layer for Expert Systems Victor Hung, Avelino Gonzalez & Ronald DeMara Intelligent Systems Laboratory University

University of Central Florida www.ucf.edu

Contexts In Goal Management

Context Internal and external circumstances

Goal Desired end state

CxBR Context-goal relationships

Page 15: Towards A Context-Based Dialog Management Layer for Expert Systems Victor Hung, Avelino Gonzalez & Ronald DeMara Intelligent Systems Laboratory University

University of Central Florida www.ucf.edu

Knowledge In Goal Management

ContextualizedKnowledge Base

User Data Base

General KnowledgeBase

Domain-SpecificKnowledge Base

Contextualized KnowledgeContext

Page 16: Towards A Context-Based Dialog Management Layer for Expert Systems Victor Hung, Avelino Gonzalez & Ronald DeMara Intelligent Systems Laboratory University

University of Central Florida www.ucf.edu

Framework for Goal Management

Goal Recognition NLP-based context

activation Goal Bookkeeping

Discourse Goal Stack Model (Branting et al, 2004)

Context Topology Agent Goals User Goals

Context Topology

Mission Goals

Context C0

Context C4

Context C1

Context C3Context C2

Disambiguated Utterance

Inference Engine

Contextualized Knowledge

Page 17: Towards A Context-Based Dialog Management Layer for Expert Systems Victor Hung, Avelino Gonzalez & Ronald DeMara Intelligent Systems Laboratory University

University of Central Florida www.ucf.edu

Evaluation

Plagued by subjectiveness Measuring Naturalness

PARADISE (Walker et al, 1997) User questionnaire (Semeraro et al, 2003)

(Rzepka et al, 2005) Universal chatbot evaluation system

(Shawar and Atwell, 2007)

Page 18: Towards A Context-Based Dialog Management Layer for Expert Systems Victor Hung, Avelino Gonzalez & Ronald DeMara Intelligent Systems Laboratory University

University of Central Florida www.ucf.edu

Project LifeLike

Page 19: Towards A Context-Based Dialog Management Layer for Expert Systems Victor Hung, Avelino Gonzalez & Ronald DeMara Intelligent Systems Laboratory University

University of Central Florida www.ucf.edu

KnowledgeManager

Speech Disambiguator

Context-based Dialog Manager

LifeLike Dialog System

LifeLikeRecognizer NSF User

Data

GeneralKnowledge

AskAlexOntology

Spell Check

Semantics Check

ContextSpecific

Knowledge

LifeLikeSpeechOutput

Dictation String

Phrase String

ContextC

on

text

Disa

mb

igu

ate

dS

tring

Context

Dataset

Context

Dataset

Response String

Response String

Updated Data

Page 20: Towards A Context-Based Dialog Management Layer for Expert Systems Victor Hung, Avelino Gonzalez & Ronald DeMara Intelligent Systems Laboratory University

University of Central Florida www.ucf.edu

Questions