Research
My research sits at the intersection of natural language processing, human-computer interaction, and psycholinguistics. The central question is how conversational AI systems should adapt their language to the people they talk to, and what happens, in terms of trust, comprehension, persuasion, and cognitive load, when they do or do not. My doctoral work at the University of Twente has focused on one specific mechanism, lexical alignment, i.e., the tendency of dialogue partners to reuse each other's words. I have looked at lexical alignment as a measurement problem, i.e., how it should be quantified in dialogue data, as an implementation problem, i.e., how it can be controlled in LLM-based dialogue generation, and as a user-experience problem, i.e., how it affects the people on the other side of the conversation.
Publications
Your Word Against Mine: Lexical Alignment in Argumentative Dialogues with a Conversational Agent
Sumit Srivastava, Mariët Theune, Alejandro Catala, and Chris Reed
Cognitive Science (2026), Being revised for resubmission
Investigates lexical alignment in argumentative dialogues between humans and conversational agents, examining how word choice adaptation affects persuasion and understanding.
Measuring and Implementing Lexical Alignment: A Systematic Literature Review
Sumit Srivastava, Suzanna Wentzel, Mariët Theune, and Alejandro Catala
Computer Speech & Language (2024)
DOI: 10.1016/j.csl.2024.101731
A systematic literature review on measuring and implementing lexical alignment in dialogue systems, covering psycholinguistic foundations and computational approaches.
Trust in a Human-Computer Collaborative Task with or without Lexical Alignment
Sumit Srivastava, Mariët Theune, Alejandro Catala, and Chris Reed
Adjunct Proceedings of the 32nd ACM Conference on User Modeling, Adaptation and Personalization (UMAP Adjunct '24) (2024)
Examines the effect of lexical alignment by a conversational agent on user trust during a human-computer collaborative task.
The Role of Lexical Alignment in Human Understanding of Explanations by Conversational Agents
Sumit Srivastava, Mariët Theune, and Alejandro Catala
Proceedings of the 28th International Conference on Intelligent User Interfaces (ACM IUI '23) (2023)
Studies how lexical alignment in conversational agent explanations affects human understanding, cognitive load, and trust.
Real Time Mono-Vision Based Customizable Virtual Keyboard Using Finger Tip Speed Analysis
Sumit Srivastava and Ramesh Chandra Tripathi
Human-Computer Interaction: 15th International Conference, HCI International 2013, Springer (2013)
DOI: 10.1007/978-3-642-39330-3_53
Presents a real-time virtual keyboard system using mono-vision and fingertip speed analysis for touchless human-computer interaction.