Brochette: Toward A Continuous Learning Platform for Knowledge Acquisition and Integration

Question-Answering (QA) systems like IBM Watson are particularly challenging to design and need to cover areas including computational linguistics, information retrieval, knowledge representation and reasoning, and machine learning. ‘Exobrain’ is a Korean Research Program which aims at building such a high-performing QA system for the Korean Language. In this paper, we describe ‘Brochette’, a continuous learning platform that iteratively acquires a large volume of unstructured data and extracts sentences related to entries of an ontology, a formal representation of knowledge used for further queries and reasoning purposes. As various Machine Learning modules must be trained and tuned to suit new content and adapt to new domains, Brochette platform also provides a framework for continuously re-training and evaluating machine learning components for syntactic and semantic analysis. Integrating semantic information from the ontology and results of these machine learners, the system is able to discover new knowledge which is then incorporated again in the Knowledge Base, making it continuously evolving.