An experiment environment for question answering research

Abstract

Building tests and experiments for Question Answering (QA) tends to be a hardworking task due to the fact that a lot of time is spent in the production of repeated code and tasks that could be automated. For example, there is a great number of datasets available and each of them organizes and structures information differently. Also, there are several QA benchmarks, which makes the task of testing an approach in different databases laborious. This work proposes the creation of a system that facilitates researchers to implement new techniques through the easy availability of different datasets, tasks, and pre-implemented techniques. Furthermore, the system architecture has been developed in such a way that having knowledge of the Python programming language is enough to implement and test new techniques. We have a functional prototype that performs the reading of different datasets and implements different techniques related to Question Processing, Information Retrieval, and Answer Processing phases. Empirical tests have shown that the system facilitates the implementation of techniques for the stages of question processing, infor mation retrieval and answer processing.