In this article, we propose a new automatic topic and subtopic detection method from a document called paragraph extension. In paragraph extension, a document is considered as a set of paragraphs and a paragraph merging technique is used to merge similar consecutive paragraphs until no similar consecutive paragraphs left. Following this, similar word counts in merged paragraphs are summed up to construct subtopic scores by using a corpus which is designed so that we can find words related to a subtopic. The paragraph vectors are represented by subtopics instead of the words. The subtopic of a paragraph is the most frequent one in the paragraph vector. On the other hand, topic of the document is the most dispersive subtopic in the document. An experimental topic/subtopic corpus is constructed for sport and education topics. We also supported corpus by WordNet to obtain synonyms words. We evaluate the proposed method on a data set contains randomly selected 40 documents from the education and sport topics. The experiment results show that average of topic detection success ratio is about %83 and the subtopic detection is about %68. (C) 2015 The Authors. Published by Elsevier Ltd.