Luca Dini

Error Driven Word Sense Disambiguation

In this paper we describe a method for performing word sense disambiguation (WSD). The method relies on unsupervised learning and exploits functional relations among words as produced by a shallow parser. By exploiting an error driven rule learning algorithm (Brill 1997), the system is able to produce rules for WSD, which can be optionally edited by humans in order to increase the performance of the system.

Linking Theory and Lexical Ambiguity: The Case of Italian Motion Verbs

This chapter takes into account the interactions between motion verbs and spatial prepositions in Romance languages, considering Italian as a case study. We will show that, contrary to traditional approaches (Zin-garelli, 1995; Serianni, 1988), there is no need to assume that spatial prepositions in Italian are almost systematically ambiguous between a static and a dynamic interpretation. Such an assumption, which is made in analogy with Germanic languages, hides the generalization that monosyllabic static and dynamic Italian locative prepositions axe not distinct.

Ontology based law discovery

The vast amount of information freely available on the Web constitutes a unparalleled resource for automatic knowledge discovery and learning. In this article we propose a study on Ontology Induction for individual laws based on corpora comparison that exploits a domain corpus automatically generated from the Web; in particular we present a case study on the Italian “Legge Bassanini” (59/1997, 127/1997 - concerning the simplification and decentralization of administrative procedures). We evaluate how the induced ontological characterizations might vary according to different factors, such as the genre (e.g. news vs. social media),the learning algorithm, the text analysis granularity, etc; the main contribution of the paper consists of highlighting the structural difference emerging from the learned predicates, and in showing how the learning mechanism might provide valuable information on how laws are perceived in different layers of the civil society.

Real Time Customer Opinion Monitoring

The paper addresses a crucial topic in current CRM processes, i.e. the one of constant monitoring customer opinions. We use the label “Real Time Customer Opinion Monitoring” to denote the process of retrieving, analyzing and assessing opinions, judgments, criticisms about products and brands, from newsgroups, message boards, consumer associations sites and other public sources on the Internet. We suggest that the use of Language Technologies and — more specifically — of Information Extraction technologies provides a substantial help in Customer Opinion Monitoring, when compared to alternative approaches, including both the “traditional” methodology of employing human operators for reading documents and formalizing relevant opinions/facts to be stored, data mining techniques bases on the non—linguistic structure of the page (web mining) or on statistical rather then linguistic analysis of the text (text mining in its standard meaning). In the light of these considerations, a novel application (ArgoServer) is presented, where different technologies cooperate with the core linguistic information extraction engine in order to achieve the result of constantly updating a database of product or brand-related customer opinions automatically gathered from newsgroups. The paper will emphasize how far the currently implemented shallow parsing techniques can go in understanding the contents of customers and users’ messages, thus extracting database records from relevant textual segments. It will also stress the limits inherently associated to the use of pure shallow techniques for the comprehension of language, and show how a new emerging linguistic technology to be developed in the context of the European project Deep Thought could in principle overcome such limits.

SiSSA: an infrastructure for NLP application development

Recently there has been a growing interest in infrastructures for sharing NLP tools and resources. This paper presents SiSSA, a project that aims at developing an infrastructure for prototyping, editing and validation of NLP application architectures. The system will provide the user with a graphical environment for (1) selecting the NLP activities relevant for the particular NLP task and the associated linguistic processors that execute them; (2) connecting new linguistic processors to SiSSA; (3) checking that the chosen architectural hypothesis corresponds to the functional specifications of the given application.The TRACTOR philosophy is to accept deposits of resources in any format, and to distribute them in the form in which they are received (with small changes if possible such as additional documentation, and putting a browsable version or sample online.) In addition, certain standards are recommended and help is offered to providers who wish to make their resources conformant with the standards. This lack of standardisation is not simply a pragmatic measure in the face of problems of heterogeneity, but is based on a profound scepticism towards current resource standardisation practice. In the future, TRACTOR aims to build up particularly parallel corpora and tools for processing and extracting meaning from such resources.

Two Level Approach to SRL

The Frame Labeling over Italian Texts (FLaIT) task is an SRL evaluation exercise part of the Evalita 2011. In this paper we present CELI’s participation in Evalita 2011 FLaIT task. Based on Markov model reasoning, our system obtained the highest precision in comparison to the other participants. The core of our approach for argument classification is based on a set of general manually encoded rules in two reasoning systems. We have also developed modules for Frame Prediction and Boundary Detection based on lexical parser.

Automatic gazetteer generation from wikipedia

The presence of high quality Named Entity gazetteer within a CLIR system is crucial in order to provide multilingual access to digital resources, particularly in the domain of Digital Libraries. In our paper we investigate an approach for automatically extracting this kind of resources from Wikipedia using an unsupervised approach that leverages the DBpedia classification of the English articles in order to induce the same classification onto encyclopedia pages expressed in other languages. By exploiting the structured information present in Wikipedia we furthermore aim at enriching our standard gazetteer with translations to other languages as well as with the alternative spellings of the entities.