NASSCOM Community

The Pitch - DheeYantra, regional language conversational AI

Blog Post created by NASSCOM Community on Jul 21, 2017

Language is probably one of the toughest challenges for modern technology and the most critical barrier to it's propagation. A lot of interesting work is being done in the sector and we decided to take a peak into the Indian scenario. We have the Co-Founder & CEO at DheeYantra Research Labs, Vidhu Bennie with us to explain how they are using AI for regional Indian language support. 

 

  1. How important is the role of regional languages in digital transformation?

India is a country with 1.3billion population and, as of now only 12% ( ~150 million) speaks English.  More of these non-English speaking crowd can turn into netizens only if the online use of its regional languages is a reality. As per the study by KPMG & Google, by 2021, there would be 201million Hindi language internet users & another 324 million combining other regional language internet users.  This means, having access to digital information in Indic language, opens up another 500 million digital users by 2021, who otherwise wouldn’t have access to these information’s & technology.

 

DheeYantra is envisioning an India where all Indians, irrespective of the language they speak, can have access to the information they need & use the latest technologies such as IoT products, which are currently driven by the Mobile Apps, which supports only English.

 

We see that, the very user Interface has to be changed. AI is the new UI. One need to have conversational UI, which would be the basis for all communication. In the pre digital era, customers used to walk in to a local shop and ask the shop keeper about the product they are looking at., “ do you have a another shade with the same design ?” etc.  The current user interface for internet is all about, you go in to the website or App & explore all by yourself. With conversational UI, you get back that freedom of asking and getting what you want. With the Indian language conversational UI, DheeYantra brings in an opportunity to non-English speaking crowd to make most of the technology, who otherwise was left with no options.  It’s not far that we can see, non English speaking Housewives & Farmers enjoying the benefits of IoT & Internet.

 

  1. Tell us about DheeYantra, how is it unique?

DheeYantra is focused on regional language conversational AI. With DheeYantra’s patent pending technologies in Deep Learning & Natural Language Processing in regional languages, (a) the product will help end users by giving them an option to communicate, get information & tasks done in their own preferred language, there by bringing more native language users to India’s Digital Grid, being at the core of Digital India initiatives. (b) Our product helps enterprises to improve the productivity of Customer support, Sales & Marketing teams – by automating most of the monotonous and repetitive jobs. This will also result in cost optimization for these enterprises. (c) This product also helps to be the communication brain of IoT devices, where the end users would be able to communicate with these products in their own native language. We see that, this is going to bring in a new paradigm shift with respect to changing the user interfaces we currently see & will start to see our AI taking over as new UI.

 

  1. Conventional translation tools have fallen short in a few areas, for example colloquial terms and slangs. How would AI make a difference in this?

Adapting to the new uses of a language, word or slangs has been the limitations of earlier approaches. The meaning of a word changes over a period of time, depending on the circumstances they are being used & the slangs getting developed. AI with it’s advancements in Natural Language Processing helps to curtail this problem. Dhee.ai  handles these issues. It learns those small changes in meaning over time by examining language as it is used by people.

Coming to conversational translation, there has been a big leap in technology through Deep Learning / Neural Machine translational approaches. While the results are good, it still is not a completely solved problem. This is another ongoing research in DheeYantra Research Labs.

 

  1. There are chatbots on one end and personal assistants on the other. Where does DheeYantra lie in that spectrum and how does it compare to the other two?

Though there are lot of “Chatbots” available in market, most of them are not real chatbots. A real Chatbot should be able to understand the questions asked by user in natural language, either in voice or text format. The bot should be able to understand the meaning of what user asked. Whereas most of the available ones expect us to click on a pre-defined question / product link to take the input, revealing their inability for semantic understanding. Chatbot & AI are an highly abused word in recent times. We do see few good Chatbot platforms, which has good conversational skills.  Most of these Chatbots are tuned towards concierge services / personal assistants supporting only English language.

DheeYantra’s conversational agent “dhee.ai”, is aimed at enterprises to use in their functions such as customer support, Sales & Marketing, so that the end users can communicate with the enterprise on their products and services. Many Banks & Financial Services companies are in talks with us to get this technology piloted. This enables a multichannel communication medium to the end users, through the integration of conversational agent on businesses website, App, and social media sites such as Facebook pages. Since what DheeYantra provides is an intelligent conversational stack with regional language support, this is also getting piloted as IoT communication stack by few IoT companies.

 

  1. Given that AIs can have incremental learning, would it be possible for DheeYantra to be a solution for international languages as well? 

Our current focus is Indian language NLP and Indian market. This is a large enough problem - the solution to which  has great scope for enrichment with multiple approaches and modules. So, international language expansion wouldn't be our immediate focus.

 

  1. Given the multiplicity of platforms and content, is DheeYantra feasible for small businesses?

“dhee.ai”, DheeYantra’s conversational AI agent supports small, medium & large enterprises based on the plans each business are subscribed to. “dhee.ai” comes pre-trained on few domains . Finer learnings happen once dhee.ai has been fed with details pertaining to specifics of products and services & the past data, if any. We have plans which are suitable for start-up’s and local business to large enterprises.

 

  1. With bigger players such as Google, Microsoft etc pursuing the same goals, what is the scope for startups in this niche technologies like this? 

Today, the focus of bigger players are to solve bigger problems such as an AI agent which knows everything under the sky. They are trying for domain agonistic AI. This is a larger problem & solving it in one language “English” itself is a big task. With the kind of data these bigger players have, it’s a good problem for them to solve.

 

DheeYantra is trying to solve issues in specific domains. The challenge that we are taking on is to solve these problems in multiple languages. So, I guess we are not chasing the same goals. Even in the event of bigger players pursing the same goals, India is a huge market. We are aiming at bringing in an untapped market ( non English speaking crowd, who will start using Internet, once we have regional language support). The digital success India has seen so far are only from 12% of Indian population ( ~150 million) who knew English.  There is a huge market waiting out there untapped - 88% of 1.3billion!

 

Want to read about other startups in the series? See the pitch

Outcomes