NLP Research Intern

I worked on two projects during my internship at IIT Guwahati:

• Word Embedding Methods Stability: Explored stability of different word embedding methods such as Word2Vec, fastText, and GloVe and investigated the effects of intrinsic parameters like word frequency, vector dimensions, etc on stability. Additionally, we inspected the effects of stability on downstream tasks like word clustering and fairness evaluation. We found that fastText is the most stable method, and stability directly affects the downstream tasks considered.
• Visual Question Answering (VQA) for Indian Languages: I contributed to research on building VQA models for Indian languages (Bengali, Hindi, Marathi and Telugu), performed a comparison of different translation systems like mBART, M2M, MarianMT, Google Translate etc. using evaluation techniques like BLEU, chrF, bertscore etc. to effectively build the multilingual text dataset for the model.