A New Framework for Unsupervised Semantic Discovery

This paper presents a new framework for the unsupervised discovery of semantic in- formation, using a divide-and-conquer ap- proach to take advantage of contextual regu- larities and to avoid problems of polysemy and sublanguages. Multiple sets of doc- uments are formed and analyzed to create multiple sets of frames. The overall proce- dure is wholly unsupervised and domain in- dependent. The end result will be a collec- tion of seets of semantic frames that will be useful in a wide range of applications, in- cluding question-answering, information ex- traction, summarization and text generation.