Skip to main content

Interactive Dialogues for Explanation and Learning

Objective

The key topics being researched :
-a theory of explanation which embraces description of different types of explanation, a generic model of the explanation process in terms of action for information seeking and knowledge transfer, and explanatory discourse.
-knowledge representation for task based explanation and retrieval processes to satisfy user requests.
-how user requests for information are analysed in a conversational context of discourse focus, functional analysis of question types, and generation of appropriate answers.
-the development of discourse models for planning contextually appropriate control of human computer interaction in explanation systems.
A unifying framework of dialogue models for explanation and learning was developed. The 3 main research themes were: thetheoretical study of discourse for conversational control in didactic situations; the experimental study of optimal explanatory strategies and media; and the creation of a computational architecture of models and processes to retrieve domain knowledge and control explanatory conversations.

Considerable empirical work has been completed in the analysis of a corpus of explanation dialogues for linguistic features, information provision and explanation strategies. Experimental studies have provided further understanding about the effect of expert style on the quality and effectiveness of explanations. An initial theory of explanation has been developed which considers the ontology, viewpoints and criteria for successful explanation. Planning models describe the process of the explanatory task incorporating strategies for explanation delivery and selection rules for generating appropriate content. The theory of task knowledge structures has been extended to provide a domain knowledge model to support explanation. Experimental analysis of dialogue styles has increased understanding of the effectiveness of active/passive expert roles in promoting learning in explanation. The discourse analysis has produced a novel taxonomic framework for the analysis of speech acts and question types in explanation. The results have led to refined definitions of discourse acts and further improvements to the theoretical models. Models of information seeking and answer generation have been described. Rhetorical structure theory has been used to study the functional aspects of information seeking, conversation control and experts replies in explanation dialogues. Dialogue planning processes have been developed from this analysis using task, topic focus and interaction models. A planning model for explanation generation has been developed as have linguistic formalisms fo r functional analysis of explanation dialogues and planners for discourse control. Computational models have been developed for knowledge representation of task and domain knowledge and their attendant retrieval processes. Prototype implementations of the extended task knowledge structures (TKS) and retrieval processes have been completed using a frame based environment. Final work is in progress to create a unifying example to demonstrate various components of the theory and models.
APPROACH AND METHODS
The research is multidisciplinary involving linguistics, cognitive science and computer science (knowledge engineering and human-computer interaction). Three approaches were followed: theoretical development, experimental studies and computational modelling. The action has three complimentary theoretical/empirical themes:
-Theory of explanation and explanation plannning: Considerations for the theory have included study of the ontology of explanation, necessary criteria for successful explanation, schemas of task based and domain knowledge, and process models of explanati on delivery. This work is integrated with planning models for control of explanatory conversations, covering speech acts, topic focus and strategies for information delivery. Planning models, derived from studies of experts' behaviour and experimental testing cover question analysis, discourse focus maintainance and strategies for explanation generation.
-Linguistically motivated dialogue models involve development of planning models for understanding user needs from language input using formal semantic models of the propositional content of questions. This work has developed specifications of dialogue p rocesses for topic focus maintainance, conversational control and explanation generation which may also form the basis for more generic dialogue management.
-Knowledge representation, retrieval and management has extended the theory of Task Knowledge Structures to include domain knowledge concepts for explanation and developed retrieval processes which search for appropriate domain knowledge according to a t yped definition of the users information need. Knowledge representation has been augmented by fuzzy set theory for expression of imprecise, liguistic information and approximate reasoning has been used to design processes of knowledge retrieval for decision support and explanation.
PROGRESS AND RESULTS
The Action has made steady progress towards understanding the theoretical and practical research issues inherent in the development of computational models for explanation systems. Considerable empirical work has been completed in the analysis of a corpus of explanation dialogues for linguistic features, information provision and explanation strategies. Experimental studies have provided further understanding about the effect of expert style on the quality and effectiveness of explanations. An initial th eory of explanation has been developed which considers the ontology, viewpoints and criteria for successful explanation. Planning models describe the process of the explanatory task incorporating strategies for explanation delivery and selection rules forgenerating appropriate content. The theory of task knowledge structures has been extended to provide a domain knowledge model to support explanation. This has involved extension and formalisation of TKS knowledge representation, and development of theretr eval processes which take question types as input parameters and then generate appropriate retrieval strategies. Experimental analysis of dialogue styles has increased our understanding of the effectiveness of active/passive expert roles in promotinglearning in explanation. The discourse analysis has produced a novel taxonomic framework for the analysis of speech acts and question types in explanation. This work has been validated by empirical analysis of explanation corpus. The results have led to refined definitions of discourse acts and further improvements to the theoretical models. Models of information seeking and answer generation have been described. Rhetorical structure theory has been used to study the functional aspects of information seeking, conversation control and experts replies in explanation dialogues. Dialogue planning processes have been developed from this analysis using task, topic focus and interaction models. A planning model for explanation generation has been develope d which in egrates delivery planning with dialigue control and accounts for task based explanation with demonstration. Linguistic formalisms for functional analysis of explanation dialogues and planners for discourse control have been developed. Computa tional models have been developed for knowledge representation of task and domain knowledge and their attendant retrieval processes. These models are currently being tested and refined. Prototype implementations of the extended TKS knowledge structures and retrieval processes have been successfully completed using a frame-based environment. Final work is in progress to create a unifying example to demonstrate various components of the theory and models.
POTENTIAL
The results will help provide the theoretical basis for the next generation of explanation facilities in knowledge-based systems, cooperative advisory systems and adaptive intelligent tutoring systems. They will also contribute towards better theoretical and practical understanding of the contextual control of human-computer conversations, and representation for task and domain knowledge.

Coordinator

CITY UNIVERSITY
Address
Northampton Square
EC1V0HB London
United Kingdom

Participants (3)

Consiglio Nazionale delle Ricerche (CNR)
Italy
Address
Via Della Faggiola 32
56100 Pisa
UNIVERSITY OF THE AEGEAN
Greece
Address
9 Kanari Street
10671 Athenes
Università degli Studi di Pisa
Italy
Address
Via Risorgimento 9
56126 Pisa