Barbara Plank's abstract:
The talk will present an on-going project aiming at enhancing the OPAC (Online Public Access Catalog) search system of the Library of the Free University of Bozen-Bolzano with multilingual access. The Multilingual search system (MuSiL), we have developed, integrates advanced linguistic technologies in a user friendly interface and bridges the gap between the world of free text search and the world of conceptual librarian search. In this talk we will present the architecture of the system, its interface, functionalities and preliminary evaluations of the precision of the search results.

Pasquale Imbemba's abstract:
In this talk, I will present my BSc thesis project. The aim of the work is to develop a compound splitter for German meant to enhance the MuSiL system (Multilingual Search In Libraries).

Compound building is an important phenomenon in German language and rises implementation issues in Cross Language Information Retrieval (CLIR). In German, word composition is productive, i.e. new words are made of existing building ones. The goal of CLIR is to find the information a user needs even if it's written in a different language. This is achieved by designing a system where a query in one language can be compared with documents in another. Translating a query word by word using a dictionary is a possible solution, but it has serious drawbacks. The dictionary may not contain some of the words needed, and it may not know all the possible meanings of the words it refers to. It is therefore copious and inefficient to maintain a complete lexical resource.

Multilingual search systems, like MuSiL, may gain advantages in information retrieval facets if those systems would be able to reverse the productive compounding aspect, in order to obtain meaningful elements of the composition. In my work, I have adapted the shallow morphological approach proposed by de Rijke and Monz and overcome some of its limitations. In the talk, I will highlight its advantages and drawbacks.

Manuel Kirschner's abstract:
Lately, there has been increasing interest in how to best enrich Question-Answering (QA) applications with dialogue capabilities. While classical QA is concerned with questions posed in isolation, its interactive variant keeps track of the QA process and supports the user in finding the exact solution via natural-language dialogue. The context of each utterance must be considered for handling clarification sub-dialogues and to resolve anaphora, ellipses or fragmentary utterances.

We have started to develop an Interactive QA system for the university library domain. We see it as a fusion of the QA scenario with robust dialogue systems techniques. From the QA point of view, the system's task is to retrieve answers to user questions using a knowledge base, where the user holds the initiative throughout the exchange. Research in discourse structure and dialogue management can provide us with models for properly dealing with co-reference, keeping track of the current topic, and even temporally switching to system initiative when the user is lost. We adopt a bottom-up approach, starting with data collection, and building a simple baseline system.

Andrea Bolognesi's abstract:
This paper describes the construction of a categorial grammar derivations bank for Italian by means of an incremental statistical parser based on proof nets. The main research line behind this work lies on the attempt to merge the advantages of a logical approach to linguistic analysis with the ones of statistical and data-driven methods. In particular, we aim to maintain the basic idea of CG framework and overcome its limitations when turning to real life applications. We first introduce the reader to Categorial Type Logic formalism, a particular extension of CG, then we address the issue of enhancing the syntactic categories composition via statistical information along with the overall procedure of building the derivations bank. Finally, we present our first evaluation of the system.

Camilo Thorne's abstract:
As it is well-known, querying and managing structured data in natural language is a challenging task due to its ambiguity (syntactic and semantic) and its expressiveness. On the other hand, querying, for example, a database is a well-defined and unambigous task, namely, that of evaluating some formal query (e.g. an SQL query) of a limited expressiveness over some finite structure: the instance of the database schema. However these formal query languages may be utterly obscure for the casual user. To bridge this gap, the use of controlled languages has been proposed. Controlled languages are fragments of, say, English, with a limited vocabulary and a very restricted set of grammar rules, but in which ambiguity is minimal if not alltogether unexistent. Moreover, they can be engineered in a way that a meaning representation built out from some logic can be compositionally constructed during parsing. A logic ideally matching the expressive power of a formal query language or being able to be taken as such.

In this talk we will present a controlled language, Lite-English, for which we can compositionally build a meaning representation belonging to DL-Lite, a tractable description logic, for which query answering is FOL-reducible. Its expressive power (w.r.t. say, model-theoretical semantics) coincides exactly with that of DL-lite. We will also argue that, given this, Lite-English is capable of performing the same knowledge representation tasks (declaring a knowledge base, querying it, etc) but in a more user-friendly manner.