| This thesis deals with the automation of information services, especially with the problem of automated email answering for which we present an architecture based on question answering. This architecture was developed in the context of the investor relations services of a large public company. Our solution to this problem will improve the quality of those services, since it will allow most emails to be processed automatically, thus leaving more time to process the others.;Our architecture is divided into modules, each corresponding to a processing stage in a factual question answering system: preprocessing of reference documents, analysis of emails, selection of answer sources, analysis of candidate documents and identification of answer. The experiments we carried out on some modules validated the relevance of the architecture.;The link between automated email answering and question answering has been established in two ways: by a description of an email corpus and by an analysis of the interrogative focus of the question. This study showed that questions found in emails are harder to process than factual questions answered by other question answering systems.;The development of a system based on a question answering architecture benefits from different approaches to natural language processing. For email processing, we use a symbolic approach based on rules to identify the questions, while for candidate documents analysis, we rely on a statistical approach to label the semantic roles of a predicate within a sentence.;The answers to questions found in emails come from several sources that we coordinate by describing a model of the investor relations domain and referring to the annotation language TimeML to annotate time and events. We then describe a graphical representation of these annotations, which is used in the T ANGO TimeML annotation tool.;We conclude by describing how we can use the components of our architecture to implement an automated email answering system and by suggesting some improvements.;Keywords. question answering, automated email answering, natural language processing, semantic role labeling, temporal representation, information system. |
