by Rahul Kumar
22 October 2025
My research traces the intermedial entanglements between Bombay cinema and the Hindi popular print culture from the 1960s to the 1990s, an archive that survives today in fragile pulp magazines and cheaply printed film periodicals. These materials are often digitized as poor scans, riddled with irregular fonts and faded ink. Until recently, I painstakingly transcribed Hindi passages manually, a slow practice that was as exhausting as it was intimate. AI-powered text extraction now allows me to copy Hindi text directly from scanned pages and feed it into translation models. This saves enormous time, but it also raises questions about what kinds of knowledge are gained and what kinds are lost when archival labor is outsourced to machines.
AI transforms the temporality of research. The tedium of transcription is replaced by rapid extraction, yet this convenience can be misleading. AI text extraction systems routinely stumble over older Devanagari fonts, conjunct characters, and blurred scans, introducing silent distortions. Entire words may be omitted, or characters mangled into gibberish. Ironically, the very qualities that make the archive historically meaningful, such as the cheapness of the paper and the unevenness of typesetting, are those most likely to be erased in the conversion.
The problem compounds with translation. AI translation models, while useful, are trained to domesticate language into the fluency of English. Idioms, slang, or rhetorical flourishes specific to Hindi periodicals often get smoothed out, turned into generic English phrasing. It flattens the very cultural textures my project seeks to recover: the improvisatory humor of gossip columns, the melodramatic excess of serialized fiction, the playful borrowing of cinematic tropes, etc. While doing manual transcription, I was forced to dwell in the grain of the text, but the machinic speed tempts me to bypass that slow encounter.
These technical issues reveal deeper theoretical stakes. My project insists that Bombay cinema cannot be understood apart from its print interlocutors. But AI introduces yet another mediating layer, one that complicates the substitution of human engagement for machinic legibility. The danger lies not simply in error but in the illusion of transparency, the idea that the archive is frictionless data waiting to be extracted. In reality, the messiness, the smudges, and the gaps and illegibilities, are part of the historical record.
Nevertheless, AI cannot be dismissed entirely. It does redistribute labor in enabling ways, freeing time for interpretive work. But it also imposes new responsibilities like checking OCR against originals, annotating mistranslations, and foregrounding the politics of machinic mediation. In this sense, AI is less a neutral assistant than a collaborator whose interventions shape the archive itself. Its mistakes may illuminate typographic conventions of low-cost presses or highlight how Hindi idioms refuse domestication into English. Paying attention to these failures re-centers the stubborn materiality of print culture. AI, then, neither solves nor erases the challenges of researching Hindi print-cinema nexus. Its promise lies in what it makes visible as much as in what it cannot render.