Romanization Ambiguity and the Retroflex Flap in Hindi and Bengali: A Corpus-Based Collision Analysis
Year: 2026
Type: preprint
Abstract: The widespread use of informal and non-standardized Romanization for Indic languages particularly Hindi and Bengali creates significant lexical and orthographic ambiguity in Natural Language Processing (NLP) pipelines. When distinct words written in Devanagari and Bengali scripts map to identical Latin-script representations, orthographic collisions occur, leading to loss of semantic precision and reduced performance in downstream tasks such as transliteration, tokenization, machine translation, and language model training. This paper presents a corpus-based collision analysis investigating systematic Romanization ambiguities across Hindi and Bengali text data. To address these script-mapping conflicts, this study introduces the Reflux Flag a structured orthographic disambiguation mechanism designed to preserve phonetic and script-level distinctions during Romanization and back transliteration. By identifying recurring collision patterns and evaluating deterministic disambiguation strategies, this work provides a practical framework for improving dataset consistency and standardized Romanization in Indic NLP workflows.
Author: Biswas, Subir
Language: English
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W7233585167
Cites:
Cited by:
Related to:
