Tutorials

Integrating Natural Language Learning and Ontology Engineering

Intensive research work on applied semantics is currently driven by the expectations raised by the Semantic Web (SW). The process of modeling and management of ontological knowledge, although increasingly crucial in the design of SW applications, is still a largely empirical enterprise.

From the point of view of natural language processing (NLP), SW ontologies represent a new perspective over older problems along two major lines. First, domain ontologies are theories about a domain, usually expressed in linguistic terms. Most of the work done in knowledge representation and modeling for language understanding has a similar purpose. The research on semantic treatment of Web information is thus naturally connected to results and linguistic resources of the work done in natural language learning. Second, any Web document is (at a large extent) a textual object. In order to map any such instance into a semantically interoperable data object, a structure must be decided. The task of information extraction (IE) can be seen exactly as the process of detecting such structures (i.e. templates or frames), in texts, where they are only implicitly expressed. As, this applies to most of the information currently available on the Web, any realistic Semantic Web application calls for robust and large scale language processing capabilities.

Ontological and lexical knowledge interact critically in the above processes. The ontology determines the domain in which an IE process is immerse. However, it does not capture the "linguistic" ways such properties are realized in texts. Under a NLP perspective, such knowledge is usually encoded in lexicons. Lexicons express linguistic properties somehow connected to relevant concepts in the ontology. However, the ways this connection is realized and represented semantically is an open issue: methods and solutions currently adopted are empirically grounded to the diverse application requirements.

In this tutorial, some recent work in the area of modeling and semi-automatic learning of ontological knowledge will be discussed. After a review of the current approaches, a linguistically principled approach to ontology engineering will be presented. Such an approach conforms to SW knowledge representation standards (i.e. OWL). It aims to support both corpus-driven ontology learning, as a structured natural language learning task, and ontology population, as a specific IE task. In order to discuss the benefits of the proposed methodology, some large scale applications, like IE and Question Answering systems, as developed within International Research projects funded by the European Community, will be presented and extensive empirical investigation will be reported.

Un approccio basato su logica computazionale per la specifica e la verifica dell'interazione fra agenti: il sistema SOCS-SI

Tra gli aspetti fondamentali nella progettazione di un sistema multi-agente, la specifica dell'interazione assume un ruolo fondamentale al fine di garantire un desiderato comportamento da parte del sistema intero. L'interazione fra agenti presenta molteplici aspetti che vanno quindi tenuti in considerazione: sintassi, semantica, verifica di conformita`, dimostrazione di proprieta`. In una societa` aperta, agenti eterogenei possono partecipare senza presentare particolari credenziali e non e' quindi possibile in particolare verificare il loro codice o la base di conoscenza per dimostrare che si atterranno alle regole in vigore. Nel progetto SOCS e` stato proposto un linguaggio, basato su una semantica abduttiva, per definire l'interazione in una societa` aperta di agenti. In tale linguaggio e` possibile definire un protocollo aperto, estendibile, non sovravincolante per il comportamento degli agenti. Al linguaggio corrisponde uno strumento software, SOCS-SI, per verificare in tempo reale che gli agenti si comportino come prescritto dalla societa` e per dimostrare automaticamente che i protocolli godano delle proprieta` espresse nei requisiti. In questo tutorial, mostreremo la teoria e gli strumenti (in particolare, il software SOCS-SI) che possono essere usati per progettare, definire e sperimentare protocolli di interazione fra agenti.

Answer Set Programming and Data Integration Systems

Recent work demonstrates that Datalog and Answer Set Programming can be usefully employed in data integration systems. The Tutorial will introduce motivations and problems of data integration systems, and the use of Datalog for querying these systems. In particular, the "Global as View" (GAV) approach will be illustrated, and a possible implementation in Answer Set Programming will be oulined.