Semantic Segmentation of Text Using Deep Learning

Given a text, can we segment it into semantically coherent sections in an automatic way? Can detect the semantic boundaries, if know how many they are? determine distinct are text? These questions address this paper. To respond, use Bidirectional Encoder Representation from Transformer (BERT) to analyze text and evaluate function that call local incoherence, which expect show maxima at points where boundary is detected. Our results, although preliminary, encouraging suggest our approach be successfully applied. However, quite sensitive with respect quality, as happens case derived audio stream via Automatic Speech Recognition techniques.