Revanth Kumar Gundam supervised by Dr. Radhika Mamidi  received his Master of Science – Dual Degree in Computational Linguistics (CLD). Here’s a  summary of his research work on Bridging the Resource Gap: Evaluation and Dataset Creation for Enhanced Reasoning and Semantic Understanding in Low-Resource Languages

This thesis investigates the capabilities and limitations of multilingual language models in lowresource settings, focusing on Telugu. It introduces TeluguEval, a human-curated benchmark of 949 instances across mathematical, commonsense, scientific, legal, and ethical reasoning. The research identifies a significant performance gap in reasoning when transitioning from English to native or Romanized Telugu. Beyond generative reasoning, the work provides a comparative study of BERT-based transformer architectures for multilingual semantic understanding. It details a methodology for multi-label emotion classification and intensity prediction, demonstrating that fine-tuned models like mBERT and RoBERTa effectively capture affective nuances. Additionally, the thesis addresses semantic fidelity in translation by developing an entity-aware fine-tuning strategy using LoRA to ensure named entity preservation in low-resource machine translation. 

June 2026