Virtual Resource Centre for Language Computing (VRCLC), DUK

Happy to be a part of MoSiCom 2024, BITS, Dubai. A very nice experience. Apart from networking and information gain, we felt conferences are meant for enjoyment too. Thank you to the organizers. 

🌟 Thrilled to Share an Exciting Milestone! 🌟 I am incredibly honored to announce that my research paper has been selected for presentation at IEEE MoSICom 2024, hosted at the prestigious BITS Dubai Campus. This opportunity is a dream come true, allowing me to showcase my work on a global platform and engage with some of the brightest minds in the field. A heartfelt thank you to the BITS organizers, DST, my guide Dr. Elizabeth Sherly & my team members for their constant support and inspiration. Your encouragement has been key to this achievement. Additionally, I’m excited to explore the unique and vibrant city of Dubai—renowned for its architectural marvels, cultural richness, and innovative spirit. Here’s to pushing boundaries and embracing new experiences! 🌍✨ #Research #IEEE #MoSICom2024 #Dubai #Innovation #Gratitude #Teamwork

These researchers from Kerala have found the solution to Indic language accuracy problems. Recently, researchers from the Digital University Kerala (DUK) identified a critical flaw in automatic speech recognition (ASR) models, including OpenAI’s Whisper, when dealing with Indic scripts. The research team, led by Elizabeth Sherly from the Virtual Resource Centre for Language Computing (VRCLC), DUK, found that ASR models struggle to accurately evaluate Indian languages, such as Malayalam, Hindi, and Tamil. In an exclusive interview with AIM, Sherly, who has almost 24 years of experience in the field, revealed that the team was able to achieve 50-55% results for the Malayalam language after conducting multi-step fine-tuning on OpenAI’s Whisper. Sherley’s colleague, Kavya Manohar, a computational linguist at DUK, discussed the use of tools like OpenAI’s Whisper and Meta’s SeamlessM4T, emphasising the need for quality data and proper normalisation. "For example, if the Malayalam script of Digital University യൂണിവേഴ്സിറ്റി—loses these components, it becomes ഡ ജ റ റ ൽ യ ണ വ ഴ സ റ റ, leading to a loss in readability,” Leena G Pillai, a research scientist at DUK, wrote in a LinkedIn post. Read the story in the comments to know more!

WISE Stall at the HuddleGlobal 2024 As the Business Developer of WISE, it was an incredibly proud milestone for me to co-ordinate the event from the Digital University Kerala (DUK) and be part of such a prestigious event organized by the Kerala Startup Mission (KSUM). Rather, it was an absolute honour to interact with many talented women budding entrepreneurs and guide them to enhance their startups through WISE. The event truly was a significant learning experience as well as a wonderful platform for collaborating and networking. The stall representing DUK’s WISE (Women Incubation Startups and Entrepreneurship) project was officially inaugurated on 28th November 2024 at the 'HuddleGlobal 2024' industrial entrepreneurship fair, held at the Leela Raviz in Kovalam, Thiruvananthapuram. The magnificent event, led by KSUM, offered a great platform for startups and budding entrepreneurs to showcase their innovations and ventures. The inauguration of the WISE Stall was done by Dr. A Mujeeb, the Honourable Registrar of DUK. The event held from 28th to 30th November 2024, featured inspiring presentations from emerging entrepreneurs, enriching panel discussions with industry leaders, talks by venture capitalists and government delegates, etc. The sessions were not only valuable learning experiences but also provided excellent networking opportunities for the startups, helping them forge connections and grow. The WISE project headed by Dr. Elizabeth Sherly and the startups under WISE led by Neelima Krishnakumar, Jennifer Joseph, Arunima kiran, Soja Beegum, ZYNECORE GLOBAL Private Limited, Shruthi Thankappan, Pooja Nambiar greatly benefited from this platform; connecting with potential collaborators, investors and partners as the ventures floated by women entrepreneurs have also been generating massive interest at the event. Leading the fray were the startups incubated under Digital University Kerala’s; Women Incubation, Start-ups and Entrepreneurship (WISE) scheme, aimed at promoting a women entrepreneurship ecosystem. The three day event was truly worth rewarding for WISE and its Startups. Kudos!!! to all my colleagues from DUK who attended the event Kavya Manohar, Leena G Pillai, JITHESH V S, surag manikoth, Sabitha Rani B S, Malu G, RAHULRAJ P V and helped making this a grand success. Sincere thanks to BAJIYO BAIJU and his friends MEENAKSHY S KUMAR, Krishna M S ., Abin Nizam for their exceptional dedication in effectively materializing both the onscreen and behind-the-scenes activities. #WISE #WomenEntrepreneurs #WomenledStartups #HuddleGlobal2024 With immense pride and pleasure, I present the glimpses from the event as a testament to the success and potential of the WISE initiative. 😀 😀 😀

🎓 My PhD Journey 🎓 😊 Glad to share that I've officially completed my PhD. After seven years of teaching, I felt a strong desire to further my studies. This led me to the Indian Institute of Information Technology and Management - Kerala, where I was selected for the M.Phil in Computer Science program. Transitioning from a syllabus-oriented teaching environment to the world of research was a significant challenge, but I was fortunate to work under the guidance of Prof. Elizabeth Sherly. Her dedication and professionalism were a constant source of inspiration, and she played a crucial role in giving me the confidence to pursue a PhD. I then joined the Department of Computer Science at the University of Kerala as a Research Scholar in Speech Processing under Dr. D. Muhammad Noorul Mubarak, who became not just my guide, but also a friend, philosopher, and mentor. This academic journey has been both challenging and immensely rewarding, and I am deeply grateful to the many individuals who supported and encouraged me along the way. I am truly blessed to have had the love and encouragement of the entire department—students, research scholars, and both teaching and non-teaching staff. In particular, I am grateful for the valuable guidance from other mentors within the department. My journey would not have been possible without the cooperation of many teachers and children during data collection, whose contribution was invaluable. I owe special thanks to Shijith Sir from the Linguistics Department, Manoj Sir and Manju Madam at NISH for their support. I also want to express my deep gratitude to my colleagues at Nirmala College, IHRD, KVVS College, and Digital University Kerala for their encouragement throughout my journey. Lastly, I am incredibly fortunate to have a wonderful family who has been a constant source of support and joy—they've often seemed happier about my achievements than I am myself! 🥰 Thank to all 🙏. #UOK #DUK #KVVS #Nirmala

The week past was very exciting for us. Our team's work which got published in Empirical Methods in Natural Language Processing (EMNLP 2024) was widely acclaimed. Our work and its impact got good media coverage too. We thank the appreciation and support from our colleagues at Digital University Kerala. Link to Paper: https://lnkd.in/gwFS3Ew6 Mathrubhumi Online: https://lnkd.in/gReknXKf SouthLive: https://lnkd.in/gHbDbfZQ News18: https://lnkd.in/gsmk4tZF The Hindu: https://lnkd.in/geD9tPZj #EMNLP2024 #DUK Leena G Pillai Kavya Manohar Elizabeth Sherly

💡 Exciting Research Discovery of our team VRCLC at Digital University Kerala! I'm excited to share that our research team, Virtual Resource Centre for Language Computing (VRCLC), at Digital University Kerala has uncovered a significant flaw in the evaluation routine of OpenAI's Whisper, a widely used AI system for transcribing speech into text. Our team, led by Dr. Elizabeth Sherly, and including my colleagues Dr. Kavya Manohar and myself (Leena G Pillai), found that the AI system reports higher accuracy for Indian languages—such as Malayalam—than it truly achieves. In our investigation, we discovered that Whisper’s accuracy-checking process does not properly account for critical elements in Indian scripts, specifically vowel signs and modifiers like the chandrakkala (virama). For example, if the Malayalam script of  Digital University - ”ഡിജിറ്റൽ യൂണിവേഴ്സിറ്റി” - loses these components, it becomes “ഡ ജ റ റ ൽ യ ണ വ ഴ സ റ റ,” leading to a loss in readability. These omissions go undetected in accuracy evaluations, creating a misleading picture of the system’s reliability. We observed similar oversights in Meta's evaluations of their AI models. Our findings have been recognized internationally, and we presented our findings at the prestigious EMNLP 2024 conference in Florida, a leading event in the field of Natural Language Processing. We are also honored to share that the Association of Computational Linguistics (ACL) has awarded us a grant to support our presentation of the paper titled "What is Lost in Normalization? Exploring Pitfalls in Multilingual ASR Model Evaluations,". This research highlights the challenges of accurately representing regional languages in AI systems, which require a tailored approach distinct from those used for languages like English. At VRCLC, we are committed to advancing the accuracy and usability of AI for Indian languages, addressing these critical linguistic subtleties. Proud to be part of this journey toward better AI for regional languages! #AImodesls #InclusiveNLP #ResponsibleAI #Multilingualism #EMNLP2024