Machine Translation in Bengali বাংলা ভাষার যন্ত্র-অনুবাদ

The session traced the journey from the earliest decisions about representing speech in visual form, through the challenges of encoding Bengali script on a machine, to the foundational problem that precedes all translation: making a language computationally legible. Participants examined what is at stake — linguistically and politically — when a spoken language is formalised for a machine to read.

Prof. Chaudhuri discussed the development of the earliest Bengali language resources — dictionaries, thesauri, and the pioneering experiments in OCR, spell-checking, and text processing that established the infrastructure for all subsequent work. The session raised a central question: when we choose which texts to feed a machine, whose Bengali are we teaching it to speak?

Dr. Bhaskar examined a pivotal transitional moment in the field: systems that could not yet translate fluently, but had learned to retrieve meaning across languages. He discussed how, at this stage, translation was no longer about finding the 'right' equivalent word, but about locating relevant meaning — a reframing that had significant consequences for how Bengali text was processed and ranked.

Participants worked hands-on with IndicTrans2, BanglaBERT, Google Translate, and ChatGPT in a guided comparative exercise. They assessed the same Bengali literary passage, newspaper editorial, and conversational excerpt across multiple MT systems, developing practical evaluative vocabulary to articulate what each system got right — and what it lost.

Pritam Dey discussed how AI agents can be leveraged for more targeted Bengali translation — moving beyond generic output toward context-aware, domain-specific results. He conducted hands-on sessions using Google's Antigravity, walking participants through agent-driven translation workflows and their implications for language practitioners. The session drew to a close with a close-reading exercise centred on a passage from Jhumpa Lahiri's The Lowland and its Bengali translation নাবাল জমি (Nabal Jomi), translated by Poulomi Sengupta. Participants examined the translation's choices with care — its handling of cultural register, untranslatable affect, and the texture of Lahiri's prose in Bengali — and offered their observations directly to Pritam, who then attempted to feed those nuanced, humanistic insights back into the agentic workflow in real time, testing how far the machine could stretch toward what the readers had felt.