• Aucun résultat trouvé

Article pp.11-13 du Vol.46 n°3 (2005)

N/A
N/A
Protected

Academic year: 2022

Partager "Article pp.11-13 du Vol.46 n°3 (2005)"

Copied!
4
0
0

Texte intégral

(1)

Preface

Question Answering (QA) has emerged as an innovative perspective in Information Access, proposing advanced modalities with respect to Information Retrieval. As QA systems return an actual answer, rather than a ranked list of documents, in response to a question, new capabilities are needed to address crucial issues in Information Access.

Most research in QA in the last years stemmed from evaluation campaigns firstly launched by the Text Retrieval Conferences (TREC) starting from 1999. The QA task was defined such that the systems were to retrieve small snippets of text that contained an answer to open-domain, closed-class questions (i.e., fact-based, short- answer questions that can be drawn from any domain). Over time the task has incrementally gained complexity (e.g., list questions, event-targeted questions, exact answers have been introduced) while maintaining its original shape.

Following the success at TREC, two initiatives, NTCIR (Evaluation of Information Access Technologies) and CLEF (Cross Language Evaluation Forum) started offering QA tasks for languages other than English, both monolingual and crosslingual, where questions are posed in source language and answers have to be retrieved in a document collection of a different, target language. While NTCIR focuses on Asian languages, CLEF focuses mainly on European languages. Over the years the series of QA competitions at CLEF has registered a steady increment in the number of participants and languages involved: QA@CLEF moved from eight groups and four languages in the first 2003 campaign to almost forty groups and 9 languages in the last 2006 campaign, showing an increasing interest, in particular from Europe, in QA. In addition, several pilot tasks are proposed every year with the purpose of investigating new aspects of the task (e.g., evaluation of answer justification snippets). Besides the annual CLEF workshops, multilingual QA has been the topic of the MLQA workshop at EACL 2006. Among the relevant initiatives on evaluation of QA systems, it is also worth to mention the EQueR Evaluation Campaign, which provided an evaluation framework for Question/Answering systems specifically for the French language.

While evaluation campaigns have been providing a forum for fostering QA technologies and for comparing system performance, several challenging aspects of QA remained out of the scope of evaluation and have been addressed by the scientific community in a number of initiatives. This is the case, for instance, of Interactive Question Answering (workshop at HLT-NAACL 2006), pragmatic issues of QA (workshop at HLT-NAACL 2004), relations between QA and Reasoning (the KRAQ workshop at IJCAI 2005), issues posed by QA for restricted Domains (workshops at ACL 2004 and AAAI 2005) and the interesting relations between

Cet article des Editions Lavoisier est disponible en acces libre et gratuit sur tal.revuesonline.com

(2)

12 TAL. Volume 46 – n° 3/2005

automatic summarization and QA (workshops at ACL 2003 and COLING/ACL 2006).

This Special Issue of the TAL Journal on Question-Answering Systems presents an overview of recent progress in Open Domain Question Answering under several perspectives. Such perspectives are all motivated by the fact that answering a question requires different solving strategies, based on a number of processes.

Question-answering systems classically develop a question analysis module which aims to give the type of the answer, and also to specify pieces of information that will guide the system to retrieve relevant passages and to extract the answer.

Passages are generally one or few sentences long. Whatever they are, solutions have to deal with linguistic variations between the expression of a need (the question) and the expression of the searched information (a passage that contains the answer).

These variations are lexical, syntactic and semantic. Solutions may rely on a rule- based, symbolic approach, for instance for question analysis, named entity tagging of documents or pattern-based answer extraction. It may also rely on numerical methods, often used to estimate the proximity of a question and a passage or even to locate the answer in the passage, which are closer to Information Retrieval methods.

Papers in this special issue address a range of problems occurring in question- answering and cover different approaches, for solving either a specific sub-task or the global task.

In their respective papers, Bouma et al., Gillard et al., Laurent et al. and Rosset et al. present complete systems, each one with a specific point of view. Other papers of the issue are dedicated to a specific problem. We shall therefore introduce the papers according to their specificities.

– Different sources of knowledge: Bouma et al. et Gillard et al. use both a pre- existing knowledge base (i.e., a base of answers) and an on-line resolution of questions by extraction from texts.

– Real time QA: Laurent et al. and Rosset et al. integrate temporal constraints as they both develop their systems for a real world application. To achieve this goal, they both aim at indexing collections with maximum granularity and exhaustivity.

– Interactive QA: Rosset et al. present a QA system with oral input that may entail a dialogue with the user. Van Schooten et al. investigate more specifically follow-up utterances after a question-answer pair.

– Intensive use of linguistic knowledge: this is one of the distinctive features of the systems presented by Bouma et al. and Laurent et al.

– Syntax: Bouma et al. and Ligozat show the importance of syntax for answering questions. In Bouma et al., syntax is used at different stages of the resolution process (passage selection, answer extraction, elaboration of answer base, etc.), while in Ligozat, the syntactic structures of a question and of a candidate sentence constrain the matching process between them.

Cet article des Editions Lavoisier est disponible en acces libre et gratuit sur tal.revuesonline.com

(3)

Preface 13

– Numeric approach: Gillard et al. make use of a density measure for selecting passages and of a compacity measure for extracting the answer.

– Answer generation: Moriceau studies a cooperative formulation of answers in order to generate coherent and justified answers and applies it more specifically to numeric answers.

– Evaluation of the relevance of a QA system: Laurent et al. make a comparison between the performance of their system and the performance of a classical IR system that takes into account the amount of effort left to the user.

Brigitte Grau groupe LIR, LIMSI-CNRS, Orsay & ENSIIE, Evry

Bernardo Magnini ITC-IRST, Trento

Cet article des Editions Lavoisier est disponible en acces libre et gratuit sur tal.revuesonline.com

(4)

14 TAL. Volume 46 – n° 3/2005

SPECIFIC REVIEWERS OF THIS ISSUE Patrice Bellot (LIA, Avignon)

Mohand Boughanem (IRIT, Toulouse) Claude de Loupy (Sinequa, Paris)

Olivier Ferret (CEA-LIST, Fontenay aux roses) Claire Gardent (Loria, Nancy)

Gabriel Illouz (LIMSI, Orsay) Guy Lapalme (RALI, Montreal)

Dominique Laurent (Synapse, Toulouse)

Jimmy Lin (University of Maryland, College Park) Diego Mollá Aliod (Macquarie University, Sydney) Laura Monceaux (LINA, Nantes)

Thierry Poibeau (LIPN, Villetaneuse) Isabelle Robba (LIMSI, Orsay) Sophie Rosset (LIMSI, Orsay)

Horacio Saggion (University of Sheffield)

Richard Sutcliffe (University of Essex, Colchester) Isabelle Tellier (GRAPPA, Lille)

Anne Vilnat (LIMSI, Orsay)

Cet article des Editions Lavoisier est disponible en acces libre et gratuit sur tal.revuesonline.com

Références

Documents relatifs

Today, the emphasis of research and development in object-oriented software engi- neering has moved from classes, methods and other base-level entities to components, frame- works

These issues arise when IT is organized and measured according to the key business processes, using components that participate to the processes through their connection to a

Parallèlement au déroulement des expérimentations dans le cadre du dispositif expérimental Visio-Université Ile-de-France, le Pôle Universitaire Européen de Nancy-Metz,

Cet article des Editions Lavoisier est disponible en acces libre et gratuit sur tal.revuesonline.com.. OP DEN

Les solutions mises en œuvre pour traiter ces phénomènes linguistiques relèvent soit d’une approche symbolique, fondée sur l’écriture de règles, par exemple pour l’analyse

Les concepts présentés sont appuyés par des exemples fondés sur la technologie de la société Lingway (interrogation en langue naturelle sur le site des Pages Jaunes pour

The NTCIR Workshops [1] 1 are a series of evaluation workshops designed to enhance research in information access (IA) technologies including information retrieval

Of the 18 participants of the Ad Hoc IR of Japanese and English documents at the first workshop: 10 groups participated in the equivalent tasks at the second workshop, i.e.,