University of Bari

Department of Informatics

Intelligent Interfaces


Home Research People Meetings Projects


Present interests:

Past experiences:



Natural Human-Computer dialogues

Within the large domain of Human-Computer Interaction in Natural Language, we are interested in studying some the factors that contribute to make dialogues 'natural': deception attitudes, personality, emotions, and integration of verbal and nonverbal communication forms in Embodied Animated Agents.

a: ATTITUDES AND AFFECT IN HUMAN-COMPUTER INTERACTION.

The aim of this research is to investigate the role played by personality and emotional factors in Human-Computer Interaction.

We are involved in the European "Human Interaction Network on Emotions" (HUMAINE), in which we contribute to research on affective dialogues, user modeling and new evaluation methods.

In the scope of the MAGICSTER Project, we developed a model and a tool to simulate how mixed emotions are activated and decay with time, according to the Agent's personality.
Dynamic Belief Networks are employed in these models, to represent the dynamics of this phenomenon under uncertainty conditions.

In the MOUTH OF TRUTH Project, we studied how deception may be introduced in argumentation monologs and dialogues. The mental state of the two Interlocutors is represented with belief networks.
We developed a domain-independent simulation tool and tested it in several application domains (e.g., Turing Imitation Game and a toy Detective Story).
As a more application-oriented case study, we considered natural doctor-patient conversations, in which various sources of conflict exist and several forms of deception are employed.

In the scope of the XANTHIPPE Project, we simulated dialogues between two agents, aimed at resolving conflicts about a goal. The two agents were programmed, by a logic language, as individuals with a mental state and a set of reasoning and communication capabilities. Conflicts between the two agents were due to holding opposite goals; they might originate from misunderstandings about some of the beliefs that support these goals or could be 'real'. The dialogue was aimed at discovering the causes of conflict and at trying to solve them through clarification or persuasion. Personality traits influenced the mental state, the form of reasoning employed and the sequence of reasoning steps.

We cooperate, in this research topic, with the Institute of Psychology, National Research Council and the Department of Computer Science, University of Liverpool.

Recent publications:

b: DIALOGUES WITH EMBODIED ANIMATED AGENTS.

In the scope of the EC Network of Excellence HUMAINE, we are studying how to simulate persuasion dialogues which combine rational with a-rational arguments.
In the scope of MAGICSTER, we studied how verbal and nonverbal communication could be integrated to produce 'Believable' behaviors in Embodied Animated Agents. Our interest was focused on the 'simulation of 'believable' dialogues between a User and an Agent: we used TRINDIKIT to model and implement mixed-initiative dialogues in which the Agent 'reacted' to the User's moves by revising its plans.
We also studied how dialogue moves can made independent of the 'Body' that is employed to implement them (3D face, 2D animated character or others). To this aim, we defined a XML-like markup language (APML) in which the 'meanings' to be attached to every component of a sentence are specified. We concluded our prototype implementations with systematic evaluation studies.

This research was made in cooperation with the Department of Education Sciences, University of Roma 3, the Department of Computer and System Sciences, University of Roma 'La Sapienza' and the University of Reading.

Recent publications:


User Adaptation in Mobile Computing

With the evolution of technologies for interacting with services in "smart environments" (mobile phones, wearables, "augmented everyday objects"), access to these services depends on the situation in which users and devices are located: context-awareness is, then, increasing of importance to achieve an effective and personalized interaction with these environments.
This means that request, negotiation and presentation of results has to be customized not only to the user but also to other features related to the context; for instance: users’ location, activity, emotional state and technical characteristics of the device.

In developing the D-Me multiagent architecture, we wish to investigate how the context influences service provision, information presentation and interaction in general. At present, our research focuses on the personalization component of this architecture, that establishes how to present service results according to the "user in context" features.

Recent publications:

  • B.De Carolis: Adapting home behavior to its inhabitants
    User Modeling 2005. Springer LNAI 3538
  • S.Pizzutilo, B. De Carolis,  G.Cozzolongo, F. Ambruoso :“ A Group Adaptive System in Public Environments”, in WSEAS Transaction on Systems, Issue 11, vol.4, Nov. 2005,  ISSN 1109-277, pp. 1883-1809.

  • S.Pizzutilo, B.De Carolis,  G.Cozzolongo, V.Silvestri :“Integrated Service Provision for Student Support” in Proceedings of WSEAS AIC 05, Malta, Sept. 2005. , ISBN:960-8457-35-1, pp.319-323.   

Generation of context-adapted documents and hyperdocuments

The aim of this research is to generate explanation texts which are adapted, at the same time, to the attitudes of the person who provides the explanation (the Speaker) and of the person who receives it (the Hearer). These attitudes are stored in two User Models and can be revised during interaction.
Research in natural language and hypermedia generation is supported by systematic evaluation studies, which are made in cooperation with the Department of Psychology, University of Reading.

a: NATURAL LANGUAGE, ONE-SHOT TEXTS.

Text generation is made up by several modules, which perform distinct tasks. A discourse planner establishes information content and order according to the characteristics of the two users. A meta-level rule-based Negotiator resolves conflicts between Speaker and Hearer in the content planning, when they exist. Local and global plan revision algorithms revise the discourse plan before surface generation, to emphasize subjects of particular relevance and to avoid repetition of unfavourable information. Finally, an ATN-based surface generator transforms the text plan into a natural language message by adding, at the same time, empathy to the text. This method was applied to generate explanations about drug prescriptions in the scope of the European Community Project OPADE.

Recent publications:

b: HYPERMEDIA INSTRUCTION MANUALS AND CLINICAL GUIDELINES

Clinical guidelines are "systematically developed statements to assist practitioner and patient decisions about appropriate health care in specific clinical circumstances" (M Field and K Lohr, 1990). They usually include two main components: (a) a decision-support system that provides appropriate suggestions about diagnosis and treatment and (ii) an explanation component, that illustrates concepts and terms mentioned in the suggestions.
With ARIANNA, we wish to investigate how hypermedia guidelines including the two mentioned components can be generated dynamically from two knowledge sources: a decision tree and an ontology of concepts. Both the decision-support and the explanation component are adapted to the level of experience of the users and to their specialisation.
This Project is made in cooperation with radiol ogists of the Department of Experimental Medicine, University of Roma, La Sapienza.
A previous Project, GENET, was aimed at generating adaptive hypermedia from a discourse plan. The generation method was established according to the hypermedia navigation mode, which was tuned, in its turn, to the user's characteristics. Intentional and rhetorical knowledge in the plan tree was employed to construct the hypermedia elements, to assemble them into pages and to introduce links among pages.
A set of evaluation studies, in cooperation with the Department of Psychology, University of Reading, enabled us to validate the generation method adopted in GENET and to revise it.

Recent publications:


From Formal Models to Software Manuals

a: FORMAL SPECIFICATION AND EVALUATION OF USER-ADAPTED INTERFACES

The aim of research in this area is to investigate whether extentions of Petri Nets can be employed to simulate user-adapted interaction in windowing systems, to make pre-empirical usability evaluation studies and to generate software documentation.
The formalism proposed is an integration of coloured Petri Nets with Card Moran and Newell's theory of KLM operators. This formalism was initially employed in the framework of a European Community Project (OPADE), to design the interface of a decision-support system in the area of drug prescription by close cooperation with partners with different roles and competences.

We prototyped a tool (XDM) which enables designers to build these models and to use them in simulating the interface behaviour in specific contexts. Several usability measures (layout and action consistency and complexity) can be automated or made easier by this tool, before empirical testing.

We are, at present, employing this formalism in a project (XDM-Agent) in which an Animated Character endowed with a 'personality' illustrates the tasks the application enables performing and how each task may be performed.
This research is financed by the National Research Council, in the scope of a Project involving several research centers in Italy (DIMMI BENE)

Recent publications:

  • S. Pizzutilo,, B. De Carolis and F. de Rosis:
    Cooperative Interface Agents
    In K Dautenhahn, A Bond, D Canamero and B Edmonds:
    Socially Intelligent Agents: creating relationships with computers and robots.
    Kluwer Ac Publ, in press.
  • B. De Carolis and S. Pizzutilo:
    Formal Models for User-Adapted Interface Design.
    Encyclopedia of Microcomputers'.2001
  • S. Pizzutilo, B. De Carolis, F. de Rosis:
    Cooperating with Personality-Rich Interface Agents.
    International Workshop on "Socially Intelligent Agents 2000", AAAI Fall Symposium.
  • F. de Rosis, B. De Carolis, S. Pizzutilo:
    Automated generation of agent behavior from formal models of interaction.
    Proceedings of 'Advanced Visual Interfaces 2000'.
  • F. de Rosis, B. De Carolis, S. Pizzutilo:
    Software documentation with Animated Agents .
    5th ERCIM Workshop on 'User Interfaces For All'. Dagstuhl, 1999.
  • F. de Rosis, S. Pizzutilo and B. De Carolis:
    Formal description and evaluation of user-adapted interfaces.
    International Journal of Human-Computer Studies, 49,1998.
  • F. de Rosis, S. Pizzutilo and B De Carolis:
    A tool to support specification and evaluation of context-customized interfaces.
    SIGCHI Bulletin, 28, 3, 1996.

b: PERSONALITY-BASED COOPERATION ATTITUDES.

This research was aimed at investigating how personality traits affect delegation-help attitudes in multiagent activities. We built a testbed (GOLEM) that enabled us to create agents with different combinations of personality traits and attitudes and different mental states, through an ad hoc logic programming language and a graphical interface.

This Project was made in cooperation with the Institute of Psychology, National Research Council (Cristiano Castelfranchi and Rino Falcone) and was based on their theory of socially cooperating intelligent agents.

Recent publications:


Home Research People Meetings Projects