Lunch at 12:30pm, talk at 1pm, in 148 Fitzpatrick

Title: Seeking Parallels, Percieiving Structures: On the Detection of Rhetorical Parallelism

Abstract: How do we describe the degree to which a text is “stylized” or rhetorically sophisticated? One potential measure of stylization is the degree to which a text exhibits rhetorical parallelism–that is, the notion of juxtaposing certain linguistic features in an organized manner to produce a greater effect. In this talk, I will define the task of parallelism detection and describe the dataset, derived from the sermons of St. Augustine of Hippo, which we have collected for this task. Then, I will outline our proposed solutions for said task. To start with, I will detail some heuristic and probabilistic baselines involving edit distance. Then, I will consider a neural approach incorporating Siamese networks, elaborating upon augmentations we wish to make in order to better equip the networks for our task of interest. Finally, I will also briefly mention other areas of NLP to which I believe the ability to capture parallelism would be valuable.

Bio: Stephen Bothwell is a third-year Ph.D. student in the NLP Group at the University of Notre Dame and is advised by Dr. David Chiang. Broadly speaking, his research centers on computational linguistics and stylistics. In particular, he frequently contemplates methods based on edit distance, enjoys investigating methods for interpretability in neural networks, and is passionate about the incorporation of linguistic information in NLP. Currently, his work includes a stylistic analysis of the texts of St. Augustine and a study of diachronic historical linguistics in the Italic region.