IIT Indore Researchers Develop AI Tool to Ensure Gender-Fair Speech Translation

Researchers at the Indian Institute of Technology (IIT) Indore have developed an artificial intelligence-based tool to identify and correct gender-related errors in speech translation from English into Hindi, Marathi and Gujarati.

Named Hierarchical Gender Arbitration for Speech Translation (HGAST), the tool works as a correction layer over existing AI translation systems instead of replacing them, IIT Indore officials said on Tuesday.

Associate Professor in the Department of Computer Science and Engineering, Dr Chandresh Kumar Maurya, who led the research, said HGAST examines translated output and makes necessary corrections. The tool can be integrated with existing translation systems without costly and time-consuming retraining.

According to the researchers, AI translation systems are often trained on datasets that tend to prioritise masculine verb forms. While English sentences generally do not change according to the speaker's gender, translations into Indian languages can require gender-specific verbs.

For instance, the English sentence “I went to the market” can translate into Hindi differently depending on whether the speaker is male or female. A male speaker would use the masculine form, while a female speaker would use the feminine form.

The researchers said HGAST identifies gender-related cues from the speaker's voice and analyses sentence structure to determine the appropriate verb form. Each correction is subsequently checked to ensure that the original meaning, sentence length and grammar remain unchanged.

The tool was tested on English-to-Hindi, English-to-Marathi and English-to-Gujarati translations using three existing translation systems.

HGAST maintained 99 to 100 per cent accuracy for male speakers, while substantially improving results for female speakers. Female-speaker accuracy rose from 83.15 per cent to 98.88 per cent in Hindi, from 73-76 per cent to 96-97 per cent in Gujarati, and from 90-93 per cent to 100 per cent in Marathi.

IIT Indore Director Prof Suhas Joshi said correctly recognising speakers is important for ensuring fair representation in AI-based translation.