Hello! I'm Sarubi, a PhD student in Computational Linguistics in the Department of Language Science and Technology, Saarland Informatics Campus at Saarland University, Germany. In my PhD I work on controllable and personalised text generation and the adaptation of large language models. I'm privileged to be supervised by Prof. Alexander Koller and Prof. Vera Demberg. I'm working in the A8: Adapting Text Generation to Individual Users project of the Collaborative Research Center on Information Density and Linguistic Encoding (SFB 1102).

Recently, I proposed AuthorMix, a modular authorship style-transfer framework built on model merging, where a target-style model is composed from style-specific components and the merging weights are learned layer-wise through reinforcement learning. I also develop in-context methods for controllable text adaptation that rewrite text to target reading levels using fine-grained linguistic features, and I run controlled reading studies to steer generation toward individual readers' cognitive profiles.

In my Masters work with Dr. Surangika Ranathunga and Prof. Sanath Jayasena, I mainly worked on adapting sequence-to-sequence pre-trained models using continual pre-training and multi-stage fine-tuning for Low-Resource Language Neural Machine Translation (LRL-NMT). Also contributed to developing a state-of-the-art (Neural) Machine Translation system for the government sector of Sri Lanka.

I'm interested in controllable and personalised text generation, authorship style transfer, model merging, and parameter-efficient adaptation of Large Language Models, as well as anything in between Neural Machine Translation and Low-resource Languages.