Introduction and Bio
Heya! I am Alok Debnath. Nice to meet you.
I work on the intersection of natural language processing, computational empathy, human computer interaction, and machine learning. My current research goals are to develop a framework to understand and evaluate empathy within small-talk dialogue.
I want to continue to do research in the intersection of language, computation, behaviour, and society, in the hopes of contributing to a seamlessly integrated and ethically consistent digital society. A society where we can trust the intelligence we create to study abstractions such as “the mind” and “consciousness” in a way that we can not yet access.
You can find my CV here and I would love to have a conversation with you if you are interested in anything that I am working on. Through my projects, I have experience with:
- language resources: designing annotation guidelines, data scraping and processing, enhancing corpora, corpus quality evaluation
- computational semantics: analysis of events, states, emotions, associations, as well as lexical and sentence-level semantics, including representation of latent features
- deep learning for NLP: LLMs for a host of detection, tagging, classification, and regression tasks, as well as evaluation.
I am currenly involved in the following projects:
- Evaluating Empathetic Engagement in Conversational Agents (NLP, HCI, Affective Computing, Cognitive Science)
- Context Representation in User Modeling (User Modeling, Personalization, Adaptive Computing)
Evaluating Empathetic Engagement in Conversational Agents
Empathy is a complex idea in cognitive and behavioural psychology, which has been notoriously hard to define for humans. However, if we assume that emotion is a latent characterisitc of conversation (which is quite a big “if” to begin with), we can then identify conversations which display a certain emotion and attempt to replicate what happens when humans encounter said emotion in conversation; or at least that is a summary of the state-of-the-art; ergo the number of “empathetic chatbots” in research.
The goal of my research is to create a system wherein I can evaluate the empathy displayed by these agents, and identify whether they, in fact, communicate empathetically the way humans do, and if not, build a framework to understand what components are missing. The broader goal of this study is to identify if and how large language models tune their responses to latent characteristics in text (because empathy is not just emotion, it is an attuned response to identifying and contextualizing emotion).
Why? Well, first, to study the process of digitalization of the human experience in the form of both language as a means of communication and empathy as a sociopsychological phenomena as it represents itself in interactions, and second, because it provides insights into one of the subcharacteristics of a natural human conversation.
Context Representation in User Modeling
Context is a broad term, often referring to information of the circumstances in which an event, attribute, or action can be fully understood. Within the User Modeling, Adaptive Systems, and Personalization (UMAP) research discipline, context can be seen as the additional information that an agent (i.e. computer) learns about the user (including their environment, history, actions, tone, location, etc.) that provide information relevant to performing or completing a task.
Incorrectly represented contextual information can result in an agent underperforming or malperforming on a given task, and learning information about a user that is irrelevant, temporary, or just wrong. Therefore, not only is the ability to capture context information crucial to intelligent personalized systems, rather how that captured context is represented within the model and changes the agent behaviour is an interesting concept to focus on and capture.
Alongside some of my colleagues at the ADAPT Centre (Jovan Jeromela, Dipto Barman, Hassan Zaal, and Awais Akbar), with the guidance of Prof. Owen Conlan and Prof. Judy Kay, I am co-organizing this workshop!
Please check out CRUM 2023!
In the past, I was involved with the following projects with the wonderful collaborators and advisors mentioned in parantheses. I am very grateful to their contributions in my academic interests.
- Event Analysis in Hindi and Kannada (Jaipal Singh Goud, Pranav Goel, Suhan Prabhu, Priyank Modi, Ujjwal Narayan, Dr. Manish Shrivastava)
- Computational Algebraic Representation of Hindi Syntax (Dr. Manish Shrivastava)
- Non-Euclidian Spaces for Word Representation (Siddharth Bhat, Dr. Manish Shrivastava)
- Sentence Similarity for Sentences with Emojis (Prof. Isabelle Augenstein, Nikhil Pinnaparaju)
- Vagueness in Instructional Texts (Dr. Michael Roth)