We are excited to welcome Bill Hart-Davidson, Associate Dean of Graduate Education at Michigan State, to give a talk on his research in computational rhetoric to colleagues at Penn State. The lecture will be held November 13, from 4:00-5:30pm, in 124 Sparks. A more detailed description follows.
Locating Topoi, Visualizing Available Means
The field of computational linguistics has long held that machines can be programmed to process language data, and scholars in that field have developed algorithms to code and analyze a wide range of text corpora, from online message boards (Claridge 2007) to corporate annual reports (Rutherford 2005). Rhetoric scholars, however, have been much slower to embrace computational models for studying texts, chiefly because of the prevalent view that the key topics studied in rhetoric — persuasion, argumentation, identification — can require nuanced reading and specialized training. Our research group has been exploring ways that computational methods can assist humans in an effort to examine larger and richer sets of texts. Bill Hart-Davidson will present examples from two projects that demonstrate the potential of computational rhetoric methods.
Over the last year, our research group has completed two different projects resulting in applications that locate rhetorical moves common to a given discourse community and/or genre. Both employ computational methods – unsupervised and supervised machine learning – to locate and classify common rhetorical moves, or topoi, writers commonly make in scientific discourse, in one project, and when facilitating online discussions, in the second project.
Both applications are meant to provide heuristic support to practitioners engaged in these discursive activities. The applications can perform analysis on target texts and visualize the results with a level of speed and accuracy that is beyond what a small group of human raters can do in days or weeks. The applications provide these results in dashboard views, allowing them to guide decision making, assist in training or review of past performances, etc.
In this talk, Bill will demonstrate each application and explain the steps taken to build and train one of the classifiers, starting with the human-coded text corpus. The presentation will then focus on how students and practitioners can use such analytic displays to improve their own practice.