People

Research activity in 2000

The research activity of year 2000 has involved the following scientific areas, depending on the various projects to which the Natural Language Group partecipated:

1) Syntactic analysis and robust methods for natural language processing.

2) Agent models for dialog.

3) Development of dialog systems.

The work concerning syntactic aspects has focused on the project for the development of a treebank for Italian. This project aims at the definition of a representation format able to account for the specific features of the Italian language. Such a format must be designed by taking into account the general requirements of this type of resource, which aim at making it usable for statistical analyses and for making the collected data reusable by other research centers.

The team has also developed some tools for supporting the annotation activity, an interactive incremental parser which is currently used for the annotation, some procedures for extracting information from the data annotated in the treebank format (as, for instance, the existing permutations of the main components of the verbal argument structure), and for the check of the correctness of the annotation.

During year 2000, we proceeded in the study of BDI agent architectures for dialog modeling. Dialog is interpreted an activity which is strictlyrelated to the intentions of the individuals who are involved in an interaction. In particular, we studied the motivations which lead an agent to adopt a goal of another agent and, possibly, lead him to communicate. Two situations have been investigated: when the interaction among two agents happens in a cooperative context and when the two agents do not have any common goal and their motivations to adopt another agent's goal are due to some norm or obligation.

For this aim, we developed a model of the decision process of a BDI agent when he has to deal with norms. This model, which is similar to the one developed for cooperative interactions, is based on some concepts coming from Decision Theory and Game Theory. Indeed, The BDI agent has both to take a decision among different alternatives and to try to predict which is the influence of his choices on the possibility of action of the other agents.

In particular, when an agent has to deal with norms, he has to foresee which is the reaction of the agent who has the role of watching over the norm and sanctioning the violations.

During the research project held in cooperation with CSELT laboratories, we have dealt with the problem of rapid development of speech dialog systems. Since today vocal systems are increasingly used in many areas to provide information over the telephone, telecom companies have the goal of speed up the development of new applications starting from a dialog system shell.

In the case of CSELT, developing applications starting from the Dialogos shell is a complex task which requires both knowledge about the new application domain and deep acquaintance with the system program.

Our project has developed some tools which, in an automated or semi-automated way, produce the code and the knowledge required for a new speech application. These tools start from an higher level knowledge representation language. For this aim, we chose to use XML for representing - with a suitable Document Type Definition (DTD) - procedural knowledge concerning the generation of natural language phrases.

Starting from this knowledge we produce the code by means of a set of XSLT rules which translate it in C code. Similar tools have been developed for producing other knowledge in the internal format used by Dialogos.