Natural Language Processing Seminar
From NLP to socially and personally contextualized understanding | Modeling empathy in NLP for Support-Oriented Interactions
This event is free and open to the publicAdd to Google Calendar
Title: From natural language processing to socially and personally contextualized understanding
Abstract: People express themselves in different ways and leveraging these differences can be beneficial for NLP applications. In this talk, I explore methods for interpreting the language together with its user-dependent aspects – personal history, beliefs, and social environment – and their effect on NLP classification tasks. I will discuss our recent perspectivism-motivated experiments, where we apply personalization methods to the modeling of annotators and compare their effectiveness for predicting the perception of social norms, assessing where personalization helps the most.
Bio: Lucie Flek is a computer science professor at the University of Bonn, leading the research group on Conversational AI and Social Analytics (CAISA). Her research interests revolve around machine learning applications in Natural Language Processing (NLP), particularly in user modeling and stylistic variation. Lucie has a background in particle physics and extensive industry research experience e.g. from Google and Amazon Alexa teams. Her work also delves into robustness and fairness in NLP, e.g. performance of models on underrepresented groups. Her Ph.D. focused on meaning ambiguity, incorporating expert lexical-semantic resources into DNN classification tasks.
Title: Modeling empathy in NLP for Support-Oriented Interactions
Abstract: Empathy recognition and empathetic response generation tasks have become well-established research directions in natural language processing. In this talk, I discuss work on interpersonal support-oriented conversations and open challenges in modeling empathy indicators and behaviors through language. I will present our recent findings on leveraging empathy as an opposing attribute to “toxic” language for controllable generation with LLMs. Finally, I will discuss my ongoing work on investigating cognitive-empathetic strategies in medical conversations for supporting patient-oriented treatment and therapeutic goals. In particular, this work is exploring an application of a medical educational framework for studying the particular goals of breaking bad news clinical conversations, as well as a linguistic framework for clinical empathy in these conversations.
Bio: Allison Lahnala is a Ph.D. Candidate at the University of Bonn, Germany and currently a Visiting Scholar in SBU CS, HLAB. Her research focuses on computational social science and conversation dynamics in goal-oriented settings and interpersonal interactions involving particular social intents, such as persuasion and offering support. She is working to establish theory-driven empathy research approaches in Natural Language Processing (NLP) that consider the complex affective and cognitive processes and social factors that influence empathetic expression and perception. Currently, she is investigating cognitive-empathetic strategies in medical conversations for supporting patient-oriented treatment and therapeutic goals.
(This seminars includes a 30 min. Q&A.)