Skip to main content
School of Electronic Engineering and Computer Science

CogSci seminar - 7th October

When: Wednesday, October 7, 2026, 1:00 PM - 2:00 PM
Where: Room GC114, Graduate Centre, Mile End

We are excited to have Ravi Shekhar from the University of Essex giving a talk on his recent work with SpeechLLMs. Ravi has a strong connection to QMUL. Before taking up his current lectureship at Essex, he was a postdoctoral researcher with Matt Purver for many years, so it is our great pleasure to have him back with us. 

Title:
Task Arithmetic for Fairer SpeechLLMs and Language Expansion
Abstract:
Speech-enabled large language models (SpeechLLMs) offer an efficient approach to integrating speech capabilities into LLMs. However, improving fairness and extending these models to new languages while preserving existing performance remain key challenges. This talk presents work from the ELOQUENCE project that explores task arithmetic to address both of these challenges. The first part examines how task arithmetic can reduce performance disparities across demographic groups while maintaining overall performance. The second part explores how it can be adapted to introduce new language capabilities while retaining performance on source languages.
Bio: Dr Ravi Shekhar is a Lecturer in Natural Language Processing at the School of Computer Science and Electronic Engineering, University of Essex. His research lies at the intersection of natural language processing, speech, and multimodal AI, with a particular focus on conversational AI, multilingual and low-resource language technologies, speech-enabled large language models, and responsible AI. His work also addresses online harms, including abusive-language detection and content moderation. He is a Local Co-Investigator on the EU Horizon ELOQUENCE project, Principal Investigator of the Innovate UK Chorus KTP, and Co-Investigator on the Centerprise and Hivedome KTP projects. Before joining Essex, Ravi was a postdoctoral researcher at the Computational Linguistics Lab, Queen Mary University of London, where he was part of the EMBEDDIA and SoDeStream projects. He obtained his PhD from the University of Trento, was a visiting researcher at the University of Amsterdam, and holds an MS by Research from IIIT Hyderabad.
If you aren't able to join in person, you can tune in via MS Teams using the link below:

Back to top