Please note: This PhD seminar will take place online.
Ankit Vadehra, PhD candidate
David R. Cheriton School of Computer Science
Supervisor: Professor Pascal Poupart
Grammar error correction (GEC) tools are often used as a first pass to improve the grammar and fluency of a piece of text, followed by a final pass done by human editors. Current metrics to evaluate GEC tools (e.g., GLEU, M2, Errant) focus on the accuracy of the edits performed by those tools to obtain revised sentences that are compared to gold reference sentences. However, when a human editor performs the final pass, the main question is: how much time did a GEC tool save the editor?
In this talk, I will present a new metric to evaluate the Post Edit Effort in terms of Time (PEET) of GEC tools. A predictive model is trained to estimate PEET for 2 strong GEC models on the CONNL-14 Test dataset.