AI gets hang of Gujarati to translate Chinese, Hindi text

AHMEDABAD: Taking a cue from ChatGPT that has taken the world by storm, demonstrating what a good database coupled with natural language processing (NLP) can do, many other projects are employing similar tech in the field of translation.
The researchers at Gandhinagar-based Dhirubhai Ambani Institute of Information and Communication Technology (DAIICT) have successfully created models for translating works in Hindi and Chinese into Gujarati and vice versa. They are now looking at adding Hebrew to the mix.
As part of the National Translation Mission (NTM) of the ministry of electronics and information technology, the DAIICT had received Rs 2 crore to develop algorithms for Gujarati language.
The project will have implications in the fields of e-governance, health and law.
Prof Prasenjit Majumder, principal investigator at the IRLP Lab at DAIICT, said that the project uses machine learning and artificial intelligence (AI).
"Since Gujarat has business interests in China and Israel, such tools will come in handy to understand, for example, the fine print of contracts or their country's laws," he said.
The laboratory is planning to organize a nationwide machine translation challenge for Indian languages by the end of this year.
Will we see a Gujarati ChatGPT anytime soon? Prof Majumder said that it's a distant possibility. "One must understand that ChatGPT is developed on GPT-3, a neural language model that can generate discourses and requires extensive training data. This remains a challenge for most Indian languages including Gujarati," he said.
The lab is working extensively on the theme of NLP - a recent publication in collaboration with a German university developed a hate speech detector - or more specifically aggressive text on platforms such as Twitter using artificial intelligence.
"Against the general perception, artificial intelligence or meta language (ML) doesn't eliminate human intervention completely.
"It helps sieve out solutions faster and provides better results without bias. Generating natural language data using machines is no longer science fiction and will have implications on how AI looks at languages including Gujarati," said Prof Majumder.