SEMINAR: Socio-demographic considerations in natural language...

Guest: Silviu Oprea, Samsung R&D Institute
Title:  Socio-demographic considerations in natural language processing: case studies from investigating sarcasm
Date/Time: December 13, 2024, 14:40
Location: https://sabanciuniv.zoom.us/j/2193563576?omn=97216429137 Meeting ID: 219 356 3576

Abstract: The presence of sarcasm in online communication has motivated an increasing number of computational investigations of the phenomenon across the community. In this talk, we first overview an analysis into the socio-demographic ecology of sarcastic exchanges between humans, focusing on how the effectiveness of such exchanges can be influenced by the socio-demographic similarity between the humans. Next, we challenge the motivation of a specific endeavour of the community: mainly, that of augmenting dialogue systems with the ability to generate sarcastic responses. We overview a series of social experiments that provide guidelines for dialogue systems concerning the appropriateness of generating sarcastic responses, and the formulation of such responses. To enable the experiments, we introduce a sarcastic response generator grounded in the implicit display theory.

Bio: I am a Principal Engineer at Samsung R&D Institute (UK). My recent work has focused on the safety of AI agents, with a focus on parameter-efficient guardrailing methods for large language models, to enable on-device content moderation. I am also interested in training AI agents, particularly in a way that maximises multilingual performance and avoids memorising harmful biases; and evaluating them, particularly on tasks where the correct output is a function of socio-demographic factors (such a task is commonsense inference). Before Samsung, I was an Applied Scientist at Amazon working on Alexa. Notable previous collaborations include: with Huawei, where I worked on learning transformations between monolingual word embedding spaces, to enable unsupervised translation and transfer learning to low-resource languages; and with Frontier Development, where we worked on flood segmentation in satellite images, collaborating with the European Space Agency and UNICEF. I finished my PhD in 2023 at the University of Edinburgh, where I used computational methods to detect and understand the phenomenon of sarcasm, as it is manifested in online textual communication. I completed previous degrees at the University of Edinburgh, the University of Oxford, and Jacobs University in Bremen, Germany. I am from Romania.