Answering questions based on gradually learned knowledge from the web using lightweight semantics

In this paper, we describe an approach to answering natural language questions that is based on knowledge acquired from the Web by a current search engine. Our system gradually builds a lightweight semantic knowledge base that is used to select the answer. Explicit feedback from the interested fellow helps readjust weights of the triplets in a learning-like process. Questions are not restricted to be of specific kinds. Already in an early stage of experimenting, our system was able to bring notable improvement in comparison to its underlying search engine.