There has been a lot of chatter, of late, about how machine learning and artificial intelligence can help solve a lot of problems in the real world. We are already seeing AI make significant inroads into our lives with intelligent virtual assistants like Siri, Google Assistant and most recently Bixby from Samsung. I recently used Bixby for a couple of weeks and was impressed by how it works at the back and keeps a lot of information ready without you having to ask for it. Go to the Bixby screen and a lot of sites I am likely to open are already there as quick links, even some stories that I am likely to click. Even apps like Facebook and Tripadvisor make a lot of curation, based on what we have done and not with us asking it to do anything. Some might find it obtrusive, but the fact is that a lot of us will not bat an eyelid before consuming that information or making ourselves available for more data gleaning. We are just getting used to being assisted more.
One area where this data powered AI is making a bigger impact without us realising it is in language technologies. Recently, I had the opportunity of spending some quality time understanding how technology is helping children with autism make conversation using an app called Avaz. Ajit Narayanan, the man behind the technology and the resultant series of apps, was inspired by what Indian linguist and grammarian Panini taught us over 2,000 years ago. “It struck me that his ideas can be combined with the ideas in the field today and bundled into an app format for kids with special needs,” he told me, sitting in his office at the IIT Chennai’s research park.
Avaz, his iPad app that helps speech therapists teach autistic kids to communicate, is now used by children with special needs all over the world. The Tamil Nadu government, which has already purchased scores of Apple iPads with a Tamil version of the app pre-burned, will start rolling out the programme in aided schools from the next academic year.
But Narayanan’s bigger idea is also simpler. During his initial work with children with special needs he started investigating the boundaries of language, the difference between them as well as the thread that binds them. What struck him most was the fact that a two-year-old could pick up any language without any trouble. “That seems to be a remarkable property in all languages,” he says, highlighting how he started working on generative linguistics.
This is where Narayanan’s technology, because of its deep understanding of language and its pattern-based structure, offers a larger opportunity. “If this can be used by a child with autism, can it be used by any of us to communicate in any language?” This is also an area where companies like Google are working on and spending a lot of time, resources and money, of course aided by their advances in AI and machine learning. However, Narayanan believes in a simpler way to reach this common goal. Though his team too works on Google’s technologies like Tensor Flow, their approach is very different. “For instance, Avaz was created with a conscious design decision that it would work offline.” That makes it so different from say Google products that need constant cloud support. His approach is to convert the large amount of data to a small set of rules which can capture the patterns in language.
“One of our core inventions is a very concise way of capturing linguistic patterns from this enormity of data and using those patterns to express language,” says Narayanan, proud that his engine “is only about 10,000 lines of code”. Could this new, “elegant,” way of tackling big data and the problems and opportunities it poses be better than the data-driven “brute force” way of tackling such problems? Maybe there is a simpler way to crack big data, the way a child would.