24th – 27th March 2018
Testing an integrated method for the automated analysis of rhetorical intent in academic genres.
Genre analysis has become one of the most prolific frameworks in the study of academic discourse. Focusing on the communicative practices of specific discourse communities and describing the rhetorical composition and linguistic features characteristic of their texts, such research has generated valuable knowledge for Language for Specific Purposes (LSP) pedagogy. Recently, genre analysis has begun to make headway into advanced computational applications for automated rhetorical feedback (Cotos, Huffman, & Link, 2015). The promise of this approach is immense. However, realizing its full potential for technological LSP innovations calls for merging analytic paradigms to develop theoretically-grounded methods for automated analysis of rhetorical intent.
This paper integrates move analysis (Swales, 1990) and concept-matching analysis (Sándor, 2007), which has gained prominence in natural language processing studies for detecting rhetorical strategies. Our goal is to explore the realization of rhetorical intent through functional language from two perspectives: interpreted as moves and steps and represented as linguistically-instantiated patterns of concepts, where the instantiations are at least pairwise syntactically related with each other. In this integrated approach, a rhetorical move is modeled as a pattern of linked constituent concepts. Our data come from a large cross-disciplinary corpus of research articles. First, the corpus was manually annotated with moves and steps; then, each move/step category was extracted to identify concept patterns (e.g., novelty, emphasis, trend, contrast). This led to improving the model of linguistic indicators of rhetorical intent and their automatic detection. Through this work, we show: a) that functional language can be detected using an underlying common representation of moves that involves syntactic relationships among concepts, and b) that the two analytic frameworks are mutually informative, as move analysis can enrich the inventory of concepts and concept-based analysis, in turn, can provide a fine-grained representation of instantiations of functional language used to accomplish communicative goals.