Artificial intelligence (AI) technology serves various purposes, one being facilitating access to digital services in diverse native languages. In a country as vast as India, where people speak over 121 languages, ensuring the availability of digital services in local languages is a challenging endeavor. The government is addressing this issue through Bhashini, an AI-driven language translation system creating open-source datasets in regional languages to build AI tools, ultimately enhancing digital service delivery.
Notably, a limited number of these 121 languages are covered by natural language processing (NLP), a branch of AI enabling computers to comprehend text and spoken words. Consequently, a considerable portion of the population is excluded from accessing valuable information. Kalika Bali, principal researcher at Microsoft Research India, emphasized the importance of AI tools catering to those who don’t speak widely used languages like English, French, or Spanish.
To overcome the challenge of data scarcity in less common languages, the Bhashini initiative adopts a crowdsourcing approach. It encourages individuals to contribute sentences in various languages, validate transcriptions, translate texts, and label images. This crowdsourced effort aids in building language datasets crucial for training AI models. The datasets generated by Bhashini are already in use in translation tools for education, tourism, and legal proceedings, showcasing the practical applications of AI in facilitating linguistic diversity.
AI models are traditionally trained on datasets, often consisting of written texts. However, many Indian languages have primarily oral traditions, resulting in limited textual records. Bhashini addresses this gap by actively involving contributors in the creation of datasets. The initiative becomes particularly crucial as AI models like ChatGPT or Llama are layered on top of generative AI models, making them more accessible for languages with fewer textual resources.
Earlier this year, Meta’s CEO Mark Zuckerberg announced an AI-powered speech translation model capable of translating and transcribing speech in up to 100 languages. This model, designed for speech-to-text, text-to-speech, speech-to-speech, text-to-text translation, and speech recognition, serves as a valuable tool for understanding and communicating information in languages without widely used writing systems or sufficient available texts for training AI models.
In conclusion, the integration of AI technology, as exemplified by initiatives like Bhashini, plays a pivotal role in bridging linguistic gaps and ensuring that digital services are accessible to a diverse population speaking numerous languages. The collaborative and innovative use of AI in addressing linguistic diversity challenges contributes to more inclusive and equitable digital experiences for people across India.
