Abstract
For effective and efficient natural language processing (NLP) systems, it is of great significance that a discourse unit must be made a unit of processing. The major obstacle to achieving that goal is the identification of discourse boundaries (DBs). This paper presents an algorithm about the identification of DBs in English text using anaphorically annotated text. The proposed algorithm is based on the structure of anaphoric annotations carried out in Phrase Detectives Corpus 2.1.4 (PD2). It has been tested on anaphorically annotated documents selected at random from PD2 and showed an accuracy of 97.66%.