Text-based question answering is about finding answers to arbitrary questions by searching on text documents. This task is more targeted to the user needs than simply returning a list of documents or webpages that may (or may not) contain the answer the user is looking for. Text-based question answering benefits from the use of a number of language technology techniques from areas such as document retrieval and information extraction. A question-answering system typically combines generic tools such as search engines and named entity recognisers with tools developed specifically for the QA task.

In this talk I will present the work currently being done in the AnswerFinder project. I will focus on the components that we are building specifically for the task, and I will pay special attention to the answer extraction component that detects and extracts the specific answer from a sentence. Our answer extraction methodology is based on the use of graph-based rules that are automatically learnt during a training stage. This method combines machine learning techniques with graph manipulation techniques to generate precise rules that are able to cover paraphrases of the answer sentence.