LEXICALIZING COMPUTATIONAL STYLISTICS For Language Learner Feedback

Computational stylistics refers informally to a collection of tasks within computational linguistics that deal with the style—as opposed to the semantic content—of natural language. The most famous of these tasks is perhaps authorship attribution (Stamatatos et al., 2001), which uses statistical variations in word choice to select the most likely from a fixed set of potential authors. Though applicable to a number of different applications, this framework has been applied specifically to literary analysis, grouping authors by their style (Luyckx et al., 2006). A broader definition of style brings it closer to the definition of register or genre (Biber and Conrad, 2009), which has also received some attention in the context of text classification (Kessler et al., 1997). In educational and literacy contexts, the readability of a text is an important stylistic feature, and the automatic detection of grade level, for instance, has been addressed in recent work (Petersen and Osendorf, 2009). The field of text generation was among the first in computational linguistics to implement sensitivity to style (Hovy, 1990), which has continued into the modern, datadriven era (Paiva and Evans, 2005).