Hi! My name is Pranay. I am a Postdoctoral Researcher at the Language and Translation Technology Team (LT3) and the Text-to-Knowledge group (T2K) at Ghent University. I defended my PhD, Bridging the Gap: Adapting English & Multilingual Language Models for LOTE, in September 2025, supervised by Prof. Dr. Els Lefever and Prof. Dr. Orphée De Clercq.
My current work is on mechanistic interpretability for multilingual models: trying to understand what actually happens inside a multilingual model and how and where knowledge transfers across languages. More broadly, my interests are Cross-Lingual Supervision, Low-Resource NLP, Transfer Learning and Interpretability. In the past I have also worked on multi-modal analysis of memes, aspect-based sentiment analysis and large-scale document layout understanding.
Before Ghent, I worked in the R&D Team of retail AI startup Paralleldots, before which I was a student at BITS Pilani, Goa.
In my free time I enjoy listening to all sorts of music (check Records), reading fantasy and playing games.
The Posts section is mostly a notebook for myself: when I figure something out, I write it down there so I can find it again later. You are very welcome to read along.
Projects
- MIMIC — Mechanistic Interpretability of Multilinguality in LLMs/Chatbots. My current project: locating and characterising the internal components that make multilingual models multilingual.
- CLARIAH-VL — Open Humanities Service Infrastructure, the Flemish contribution to the European CLARIN and DARIAH infrastructures. I contribute NLP tooling, including the LeTTuce multilingual PoS taggers.
- LoCKe — Low-Shot Cross-Lingual Knowledge Transfer (2021–2025). The project my doctoral work was carried out under.
News
- March ‘26: Co-organiser of WASSA 2026, the 15th Workshop on Computational Approaches to Subjectivity, Sentiment & Social Media Analysis, co-located with EACL 2026 in Rabat
- March ‘26: Paper titled “Lost in Activations: A Neuron-level Analysis of Encoders for Cross-Lingual Emotion Detection” presented at EACL 2026
- Sept ‘25: Defended my doctoral thesis, “Bridging the Gap: Adapting English & Multilingual Language Models for LOTE”
- May ‘25: Paper titled “EnerGIZAr: Leveraging GIZA++ for Effective Tokenizer Initialization” accepted to Findings of ACL 2025
- Oct ‘23 - Jan ‘24: Visiting Researcher at the IXA NLP Group at the University of the Basque Country, hosted by Prof. Dr. Eneko Agirre
- Oct ‘23: Paper titled “Misery Loves Complexity: Exploring Linguistic Complexity in the Context of Emotion Detection” accepted to Findings of EMNLP’23
Thanks for reading!
