[Paper Review] Anusaaraka: Overcoming the Language Barrier in India

Akshar Bharati, Vineet Chaitanya+3 more|ArXiv.org|Aug 7, 2003

Multilingual Education and Policy20 citations

TL;DR

Anusaaraka proposes a human-assisted machine translation system that enables cross-lingual text access among closely related Indian languages by transforming source text into a phonetically and visually similar target language form. The system reduces translation burden on humans by leveraging linguistic proximity, allowing users to learn the output language in ~2 weeks, with post-editing for grammatical accuracy and style adjustment.

ABSTRACT

The anusaaraka system makes text in one Indian language accessible in another Indian language. In the anusaaraka approach, the load is so divided between man and computer that the language load is taken by the machine, and the interpretation of the text is left to the man. The machine presents an image of the source text in a language close to the target language.In the image, some constructions of the source language (which do not have equivalents) spill over to the output. Some special notation is also devised. The user after some training learns to read and understand the output. Because the Indian languages are close, the learning time of the output language is short, and is expected to be around 2 weeks. The output can also be post-edited by a trained user to make it grammatically correct in the target language. Style can also be changed, if necessary. Thus, in this scenario, it can function as a human assisted translation system. Currently, anusaarakas are being built from Telugu, Kannada, Marathi, Bengali and Punjabi to Hindi. They can be built for all Indian languages in the near future. Everybody must pitch in to build such systems connecting all Indian languages, using the free software model.

Motivation & Objective

• To address the language barrier in multilingual India by enabling text access across Indian languages.
• To reduce the cognitive and linguistic load on human translators by offloading language processing to machines.
• To create a scalable, free software-based framework for building translation systems between Indian languages.
• To leverage linguistic similarities among Indian languages to minimize user training time for understanding machine-generated output.
• To support post-editing for grammatical correctness and stylistic refinement, enabling use as a human-assisted translation system.

Proposed method

• The system generates a visual and phonetic approximation of the source text in the target language, preserving morphological and syntactic structures.
• It uses a notation system to represent source language constructions that lack direct equivalents in the target language.
• The output is designed to be visually and phonetically close to the target language, enabling rapid user comprehension after minimal training.
• Users are trained to interpret the output as a phonetic and orthographic approximation, not a literal translation.
• Post-editing by trained users corrects grammatical errors and improves style, transforming the output into a natural target language text.
• The approach is implemented using a free software model, enabling community-driven development and extension to all Indian languages.

Experimental results

Research questions

• RQ1 Can a machine-generated approximation of text in a target Indian language enable rapid comprehension by users unfamiliar with that language?
• RQ2 To what extent can linguistic similarity between Indian languages reduce the training time required for users to interpret machine-generated output?
• RQ3 How effective is the system in supporting human-assisted translation through post-editing for grammatical correctness and style?
• RQ4 What role does visual and phonetic similarity play in reducing the cognitive load of cross-lingual text access?
• RQ5 Can a free software model sustain the development and scaling of multilingual translation systems across India’s diverse linguistic landscape?

Key findings

• Users can learn to read and understand the machine-generated output in approximately two weeks due to the phonetic and orthographic similarity between source and target Indian languages.
• The system successfully reduces the burden on human translators by handling the majority of language processing, leaving interpretation and refinement to the user.
• Post-editing by trained users results in grammatically correct and stylistically appropriate text in the target language.
• The approach is scalable and extensible, with anusaaraka systems already being developed for Telugu, Kannada, Marathi, Bengali, and Punjabi to Hindi.
• The system functions effectively as a human-assisted translation pipeline, combining machine assistance with human oversight for improved output quality.
• The free software model enables collaborative development, supporting the creation of anusaaraka systems for all Indian languages in the near future.