Annotating meta-discourse in academic lectures from different disciplines

The use of discourse structure was shown to be effective in various applications. Meta-discourse is often used as an expression to signal discourse structure. Previous work focused on using the meta-discourse structure in written texts, or spoken material in very clean conditions. This paper presents a metadiscourse annotated corpus in a more challenging educational context. The corpus comprises of academic lectures from two different disciplines: physics and economics. The schema used focuses on five categories: Introduction, Conclusion, Previewing, Reviewing and Enumerating. The annotation task is described in detail, including instructions and strategies used by expert annotators. Annotation results are reported in terms of inter-annotator agreement, self-reported confidence and number of occurrences. Results show that meta-discourse is frequently used in academic lectures and this is observed in the two selected disciplines. Further analysis of the corpus is conducted showing that some of these categories, namely Introduction and Previewing, are correlated with labelled topic boundaries, which is also consistent in both disciplines. This finding shows the potential for using meta-discourse information in topic segmentation task.