People

In 2001 the group carried out research activity on the following topics, according to the various projects which the Natural Language Group was involved in:

1. Syntactic analysis and robust methods for natural language processing.
2. Agent models.

3. Ontologies and Legal Knowledge

There are two lines of research in syntactic analysis. The first is the development of linguistic resources for Italian (including a treebank). In the past years, we have studied a representation format able to account for the specific features of the Italian language. In 2001, we have continued the annotation of a corpus of Italian sentences, from several sources. The goal is to have a treebank of 2000 sentences by the end of 2002, which is a significant amount for training in a few processing tasks. The second line of research concerns the development of methods for natural language processing in a realistic setting. In particular, we have pursued the use of recursive neural networks for parsing decisions. The basic algorithm is based on the psycholinguistic hypothesis that human language processing is incremental, and we have explored the feasibility of incremental parsing with unrestricted text. The recursive neural network is trained on the Penn treebank of English. The preliminary results we have produced are very promising for both human modelling and parsing performance.

About software prototypes, the team has continued the development of tools for supporting the annotation activity. The basic interactive parser has been augmented with some procedures for producing automatically a partially correct annotation of the sentences. Moreover, we have developed a basic incremental parser informed by a recursive neural network. Finally, a robust Chunk-based parser has been implemented and is currently in use as a support tool for treebank construction.

During year 2001, the research about autonomous agents theory focused on the topic of the interaction among agents. In particular, two situations have been analysed: cooperative settings and obligations among agents. The conceptual instruments at the basis of the model are decision and game theory, which are used in a hierarchical planning paradigm in order to choose the most promising way of action for an agent under the light of what other interactants are predicted to do. Since cooperation and norms can introduce new goals among the agent’s ones while it is acting, the problem of restructuring its plans must be addressed: we device a replanning model for the decision and game theoretic planner which tries to make minimal changes to the plan starting from the critical points. The replanning process proceeds by making the plan more partial and then refining it again without restarting the planning process by first principles.

Finally, the model of norms developed has been applied in a legal setting in order to model the concepts of right and duty according to the proposal of legal relations of Hohfeld.

A new activity, started in 2001, concerns the representation of legal knowledge. This research has been carried on in cooperation with the Department of Legal Sciences of the University of Torino, and its final goal is to devise methods for organizing legal knowledge in a way able to support computational access and reasoning. The goal is twofold: to describe the internal structure of legal texts in terms of a suitable set of tags in a markup language such as XML, and to study the ontological organization of legal knowledge. This second task has firstly been faced on the basis of the group’s experience on agent models (see the previous research area) leading to the design of an agent model which reasons about norms and obligations.

More recently, at the end of the year, the research about ontologies has led to an activity aimed at establishing an inter-university Center for Theoretical and Applied Ontology, in cooperation with various institutions, including Departments of Philosophy and Legal Sciences from many national and international organizations.

C. Bazzanella and C. Bosco. Multimodalità e contesto. In Atti del convegno Multimodalità e Multimedialità nella Comunicazione: XI Giornate di Studio del Gruppo di Fonetica Sperimentale - Università di Padova, 2001.

G. Boella. Social rationality and cooperation. In N. Zhong, J. Liu, S. Ohsuga, and J. Bradshaw, editors, Proc. of 2nd International Conference on Intelligent Agents Technology (IAT-2001), pages 64-68. World Scientific, Singapore, 2001.

G. Boella and L. Lesmo. An approach to anaphora based on mental models. In P. Bouquet, editor, CONTEXT'01: Third International Conference on Modeling and Using Context, pages 413-416. Springer Verlag, Berlin, 2001.

G. Boella and L. Lesmo. Deliberate normative agents. In R. Conte and C. Dellarocas, editors, Social Order in Multi Agent Systems. Springer Verlag, Berlin, 2001.

G. Boella and L. Lesmo. Norms and cooperation: Two sides of social rationality. In Proc. of IJCAI Workshop on Autonomy, Delegation, and Control: Interacting with Autonomous Agents, Seattle (WA), 2001.

G. Boella and L. Lesmo. Socially rational agents and collective intentionality. In Proc. of Seventh International Colloquium on Cognitive Science (ICCS-01), Donostia (ES), 2001.

G. Boella, R. Damiano, M. Danieli, and L. Lesmo. Using XML for representing domain dependent knowledge in dialogos. In Proc. International Conference on Text, Speech and Dialog (TSD) LNAI 2166, pages 381-387, Zelezna Ruda (CZ). Springer Verlag. 2001.

G. Boella, L. Favali, and L. Lesmo. An action-based ontology of legal relations. In Proc. of Eighth International Conference on Artificial Intelligence and Law (ICAIL), pages 227-228, St. Louis (MO), ACM. 2001

G. Boella, L. Lesmo, and L. Favali. The definition of legal relations in a BDI multiagent framework. In LNCS 2175, pages 225-236. Springer Verlag, Berlin, 2001.

C. Bosco. Grammatical relation's system in treebank annotation. In E. Miltsakaki, C. Monz, and A. Ribeiro, editors, Proceedings of Student Research Workshop of Joint ACL/EACL Meeting, pages 1-6, Toulouse, 2001.

C. Bosco and C. Bazzanella. Context and multi-media corpora. In P. Bouquet, editor, CONTEXT'01: Third International Conference on Modeling and Using Context, Berlin, Springer Verlag. 2001.

Fabrizio Costa, Vincenzo Lombardo, Paolo Frasconi, Giovanni Soda, Wide coverage incremental parsing by learning attachment preferences, in LNCS 2175, Springer, 2001, pp. 297-307

Patrick Sturt, Fabrizio Costa, Paolo Frasconi, Vinzenzo Lombardo, A wide-coverage model of first-pass structural preferences in human parsing, Conference on Sentence Processing, Philadelphia (USA) Marzo 2001.