Automatic labelling of transitivity functional roles

Abstract

Systemic functional linguistics (SFL) is a functionally oriented linguistic framework that has gained increasing influence in recent years, with important applications in the description and analysis of text/discourse. Despite its popularity, relatively little has been done to automate the parsing of functional structures using this framework. Previous attempts have largely depended on non-statistical, rule-based methods, which have limited their application in more complex scenarios. In this article, we present a data-driven method for the classification and labelling of SFL-based functional roles, trained on a recently developed corpus resource. We describe our efforts to engineer lexical, semantic and contextual features in constructing a system for labelling the process types and participant roles in the transitivity system based on the SFL framework. Initial evaluation shows accuracies of 80.5% and 91.8% for the classification of process types and participant roles, respectively. The system is expected to be an important step in achieving fully automated analysis of functional roles in SFL. In addition to applications requiring analysis of English functional structure, we discuss some of the difficulties and future directions in extending the current system to handle less other languages such as Chinese.