Alok Kothari from Apple talks NLP

In the initial press conference introducing Siri as an integrated iPhone app back in October 2011, an Apple executive asked the assistant  ‘Who are you?’ to which Siri replied ‘I am a humble personal assistant’. Since then, the Natural Language Understanding of the assistant has evolved from a rule based system to the current use of state-of-the art deep learning techniques.

We’re fortunate enough to have Alok Kothari, Machine Learning Engineer at Apple joining us at the AI Assistant Summit in San Francisco this week to discuss Siri and it’s Natural Language Understanding.

Upon speaking with Alok in advance of the summit, he explained how his work involves ‘keeping up to date with research literature regarding natural language processing and machine learning.’ He works on formulating and experimenting with machine learning models, improving the frameworks that make such research and experiments easier and then helps set up pipelines infrastructure that will deploy those models into production and in the hands of users — as they strive to continually improve Siri’s natural language understanding.I asked how Alok started out in AI, and more specifically in voice assistants. He shared his intrigue in the amount of information, especially news that needs to be processed to understand stock market price movements. e.g. a certain news coming out about a merger, new deals, stepping down of a CEO etc could affect the stock price of a company. This was an interest that came about in college, and Alok realised that it was hard to keep track of all the news coming out about many of the companies one might be interested in, as there were relevant signals that could be extracted from many such news articles. ‘This is how I got started in natural language processing. As a math major what fascinated me even more was that I could develop mathematical models (machine learning) to understand all this unstructured natural language data.’

Alok thought that ‘as the amount of textual information in form of natural language keeps increasing, keeping track of it and attempting to understand it is a very challenging problem. Particularly, as natural language is very versatile, idiosyncratic, context dependent and even colloquially diverse from region to region. However, a natural way to consume information for humans has always been by talking with other humans.

Thus “something” that helps process large amounts of information and is able to “talk” to humans in a way they understand language would be immensely useful and seemed like a very interesting, challenging and natural aspiration.

I have since worked on various aspects of natural language processing and machine learning doing research projects, publishing papers and pursuing formal graduate programs and then have been fortunate to end up working on an actual voice assistant used by millions - Siri.’

How are you using AI for a positive impact?

Siri is useful for a lot of things that add utility to your life, simple ones being navigating directions while driving, playing music, setting calendar appointments or getting answers to trivia. A lot of people are using Siri for a number of things that I could not have easily imagined. e.g. I have found non-native speakers learn english pronunciation by speaking to Siri’s english avatar.

Siri interacts with HomeKit and enables physically challenged people to get around their homes, ordering doors to open, lights to turn on, shades to be closed and even help them dictate things to write. Its voice assistants such as Siri that are hopefully helping add dignity to peoples lives. This makes me really proud, humbled and excited to work towards improving AI and natural language understanding on Siri. Hopefully we can make Siri and voice assistants even more useful for all kinds of people in more and more meaningful ways.

Can't make it to San Francisco this week? We'll be back in the US for the Deep Learning Summit and Machine Intelligence in Healthcare Summit this May 24-25 in Boston, followed by the AI in Industrial Automation Summit and Deep Learning for Robotics Summit in San Francisco on June 28-29. Register now to guarantee your place at our upcoming summits to network with global AI experts.