Finding Discourse Relations in Student Essays

摘要

The aim of the work reported here is to provide a tool to help secondary school (high school) age students to reflect on the structure of their essays. Numerous tools are available to help students check their spelling and grammar. Very little, however, has been done to help them with higher level problems in their texts. In order to do this, we need to be able to analyse the discourse relations within their texts. This is particularly problematic for texts of this kind, since they contain few instances of explicit discourse markers such as 'however', 'moreover', 'therefore'. The situation is made worse by the fact that many texts produced by such students contain large numbers of spelling and grammatical errors, thus making linguistic analysis extremely challenging. The current paper reports on a number of experiments in classification of the discourse relations in such essays. The work explores the use of machine learning techniques to identify such relations in unseen essays, using a corpus of manually annotated essays as a training set.