Artificial Intelligence meets Language Translation

International Institute of Information Technology (IIIT) Hyderabad is all

geared to release what it calls the Indian Language Machine Translator. The

project headed by Prof Rajeev Sangal, Director of the institute, is being

carried out within three in-house labs, with each of them working on different

aspects of Natural Language Processing (NLP).

The research at IIIT dealt with several aspects of text and voice of NLP

distinctly and differently. The application is in finishing stages for ten

Indian languages: Hindi, Punjabi, Marathi, Bengali, Urdu, Malayalam, Tamil,

Telugu and Kannada. This product is expected to be ready for commercial use

within a year and is targeted at two distinct areas-Pilgrimage & Tourism and

Health. Prof Sangal explains, “We decided on tailoring the application for

Pilgrimage and Tourism based on usage trends, the Health application comes with

a potential social impact. An ideal example would be a man from Punjab wanting

to take his family for a holiday in Kerala. He should be able to access an

online forum, post his query in Hindi or Punjabi, and a Keralite at the other

end will view this query in Malayalam, reply in Malayalam and our man would see

the answer in Punjabi.”

Though language translation has been tried out globally for research,

development and commercial deployment, it has failed to overcome the challenges

of dialects, grammar changes and colloquialism. The research at IIIT has

understood that Indian scripts are sophisticated but not complex and is

incorporating Artificial Intelligence to language translation.

Rajeev Sangal Professor, IIIT Hyderabad

Prof Sangal, who is also an AI expert embarked on a study of the Panini

Vyakyaran and correlated it to modern Indian literature before adding elements

of AI to enable the computer to learn, condition itself and understand the

requirement of the user. The AI aspect is divided into two components:

Rule-based processing and machine learning. In rule-based processing, an

electronic catalogue of words and phrases is fed to the computer, enabling it to

understand typical usage of grammar elements. As more and more words and phrases

are added to this catalogue, the computer forms more rigid understanding of

rules according to which a particular language operates. Machine learning on the

other hand focuses on providing statistical data based on which the computer

will learn to use the examples. While usual translation software translates

phrases from one language to another, IIIT's research focuses on dependency of

each word on the neighboring one to create a pattern of usage for each language.

This takes care of dialects, localized modifications in language and slangs. In

broad terms, this real-time language transliteration application works on a

three step methodology-analyze, transfer and generate.