Suchergebnisse

This thesis is focused on temporal information processing, which is a task framed in the field of natural language processing. The aim of this task is to obtain the temporal location and ordering of the events expressed in text or discourse, which requires the automatic interpretation of temporal expressions, events, and the temporal relations between them. The majority of current approaches are based on morphosyntactic knowledge. However, temporal entities are often ambiguous at that language analysis level. Our hypothesis is that the linguistic expression of time is a semantic phenomenon and therefore, to achieve a better extraction performance, temporal information must be processed using semantics. To prove this hypothesis, we present a semantic approach to temporal information processing: TIPSem. This is an automated system that includes features based on lexical semantics, semantic roles, and temporal semantics, in addition to morphosyntactic features. TIPSem has been empirically evaluated through the participation in the TempEval-2 international evaluation exercise and subsequent experiments based on this test. The results obtained firmly support the presented hypothesis and their analysis demonstrates that semantic features aid in handling morphosyntactic ambiguity and favour generalization capabilities. These conclusions have been reached for different languages (i.e., English, Spanish, Italian, and Chinese), which supports the defended hypothesis at a multilingual level. Finally, as compared with the state of the art, TIPSem obtains a very competitive performance level and introduces a remarkable improvement in event processing. TIPSem has been applied to the problem of graphical representation of temporal information. We have developed a dynamic interface which brings users time-based access to information: Time-Surfer. The results obtained through a user-oriented evaluation of this interface demonstrate that TIPSem's performance is also satisfactory from an extrinsic standpoint. ; This PhD thesis has been supported by the Spanish Government, in projects: TIN-2006-15265-C06-01, TIN-2009-13391-C04-01 and PROMETEO/2009/119, where Hector Llorens is funded under a FPI grant (BES-2007-16256).

Los conjuntos de preguntas utilizados normalmente para evaluar los sistemas de búsqueda de respuestas (BR) están principalmente constituidos por preguntas cuyas respuestas son entidades nombradas (NE), por tanto, la mayoría de estos sistemas usan reconocedores de entidades para extraer las posibles respuestas. Últimamente, el etiquetado de roles semánticos y su contribución a la BR es un tema de especial interés. Sin embargo, los sistemas basados en NEs siempre funcionarán mejor que los basados en roles a la hora de extraer respuestas para preguntas cuya respuesta sea una NE. El objetivo de este artículo es evaluar ambos métodos para preguntas de lugar bajo las mismas condiciones, usando, no sólo preguntas basadas en nombres propios sino también basadas en nombres comunes. Para ello se presentan tres propuestas diferentes de un módulo de extracción de respuestas embebidas en un sistema de BR: una basada en entidades nombradas y dos basadas en roles semánticos. Los resultados obtenidos indican que mientras la propuesta de NE contesta mejor las preguntas basadas en nombres propios (+49,57% MRR), las propuestas de roles obtienen los mejores resultados en las preguntas basadas en nombres comunes (+223,48% MRR) siendo sus resultados de una precisión más alta para ambos tipos de preguntas. ; Question sets normally used to evaluate QA systems are mainly based on questions whose answers are named entities, therefore most of these systems rely on NERs to extract possible answers. Nowadays, semantic role labeling and its contribution to question answering has recently become an interesting issue. Nevertheless, NE-based systems will always work better than SR-based ones extracting answers for questions with NE-based answers. The aim of this paper is to evaluate both of approaches for location questions under the same conditions and using not only NE-based questions but also common noun-based ones. In order to achieve this goal we present three different proposals of an answer extraction module embedded into a QA system: one based on named entities and two based on semantic roles. Results show that while NE-based approach performs better with NE-based questions (MRR +49.57%), SR-based approaches show the best results in common noun-based ones (MRR +223.48%) and obtain a higher precision in both types of questions. ; This paper has been partially supported by the Spanish government, project TIN-2006-15265-C06-01 and project GV06/028, and by the framework of the project QALL-ME, which is a 6th Framework Research Programme of the European Union (EU), contract number: FP6-IST-033860.

This paper presents a multilayered architecture that enhances the capabilities of current QA systems and allows different types of complex questions or queries to be processed. The answers to these questions need to be gathered from factual information scattered throughout different documents. Specifically, we designed a specialized layer to process the different types of temporal questions. Complex temporal questions are first decomposed into simple questions, according to the temporal relations expressed in the original question. In the same way, the answers to the resulting simple questions are recomposed, fulfilling the temporal restrictions of the original complex question. A novel aspect of this approach resides in the decomposition which uses a minimal quantity of resources, with the final aim of obtaining a portable platform that is easily extensible to other languages. In this paper we also present a methodology for evaluation of the decomposition of the questions as well as the ability of the implemented temporal layer to perform at a multilingual level. The temporal layer was first performed for English, then evaluated and compared with: a) a general purpose QA system (F-measure 65.47% for QA plus English temporal layer vs. 38.01% for the general QA system), and b) a well-known QA system. Much better results were obtained for temporal questions with the multilayered system. This system was therefore extended to Spanish and very good results were again obtained in the evaluation (F-measure 40.36% for QA plus Spanish temporal layer vs. 22.94% for the general QA system). ; This paper has been partially supported by the Spanish government, project TIN-2006-15265-C06-01, and by the framework of the project QALL-ME, which is a 6th Framework Research Programme of the European Union (EU), contract number: FP6-IST-033860.