Talks at STTP, Gujarat Engineering College, Rajkot | Saṁbhāṣaṇa Research Group

Delivered two talks (online) for Short-Term Training Program (STTP) at Gujarat Engineering College, Rajkot.

Keynote: Indian Language Processing: Challenges and Opportunities (Feb 2)

India’s remarkable linguistic diversity — over 20 constitutionally recognized languages spanning multiple scripts and hundreds of dialects — makes it one of the most challenging yet rewarding settings for NLP research. This talk surveyed the core challenges: severe data scarcity for most Indian languages, morphological richness, script heterogeneity, and the prevalence of code-switching in everyday text and speech. It also covered recent progress in multilingual and cross-lingual models, and highlighted concrete avenues through which undergraduate students can contribute meaningfully to this research landscape.

Lecture: LLMs through the Lens of Computational Linguistics and Structured Knowledge (Feb 5)

Large language models have demonstrated impressive fluency, but their understanding of facts, structure, and meaning remains brittle. This talk examined LLMs through the lens of computational linguistics — asking how concepts from syntax, semantics, and morphology help explain model behavior and failure modes. It then discussed the role of structured knowledge (knowledge graphs, ontologies) in addressing hallucination and factual inconsistency, with a focus on hybrid retrieval-augmented and graph-augmented generation approaches. The talk concluded with a discussion of ethical considerations — bias, fairness, and accountability — especially in multilingual and low-resource settings.