Textual data augmentation is a prolific field of study where novel techniques to create artificial data are regularly proposed. For the text classification task, it is accepted that the possible gains depend greatly on the size of the dataset, with larger gains being associated to smaller datasets. In this paper, we challenge those results, showing that data augmentation is simply a way of performing better fine-tuning on the neural networks, and that spending more time fine-tuning before applying data augmentation negates its effect. This is a significant contribution as it answers several questions that were left open in recent years, namely: what DA technique performs best (all of them as long as they generate data close enough to the training set) and why does DA work (facilitate training of network). We furthermore show that zero shot data generation via conversational agents such as ChatGPT or LLama2 can increase the performance, but not as much as collecting and annotating external data.

La présentation sera en français.

Enregistrement: https://umontreal.ca.panopto.com/Panopto/Pages/Viewer.aspx?id=e5d46f43-43b7-4c59-9567-b124013004b5