NLP is exciting, in this roundup we look at latest happening in NLP domain

Picture is worth a thousand words, but words in those picture is worth even more. Words have power beyond our imagination. It affect our thinking, our emotion and our behaviors. To truly understand power of words spend 3 mins of your time listening to Mohammed Qahtani’s “The Power of Words” his winning speech at 2015 World Public Speaking competition.

In this blog post we will be focusing on surveying the landscape at high level, upcoming posts will have more technical details to it. Never before in era did we as humans had access to vast sources of information but also tools to process it to meaningful insights. Computers now can Summarize Multiple Documents, Win Jeopardy, Carryout Dialog with Humans and Translate 50 languages. In Business what this implies is you can find out why you’re customers are not satisfied. Have instant access to dynamically changing competitive environment and most importantly truly understand customer’s intents.

Here at Fafadia Tech we’re on a constant lookout for tools that will help our customers attain their business objectives. Following are collections of tools and research we believe that will help Businesses in 2016

Automatic Text Summarization

Getting 360 degree picture is important, but the pile of information that one has to sift through is increasing day by day. Text Summarization is active area of research with many solved problem, many yet to be solved. Imagine that you’re doing Credit Analysis for a potential borrower. Financial Institutes are looking beyond Balance Sheets, Income Tax Return statements. But also to online reputation and social score of their founders.

Automatic Text Summarization works by extracting and ranking key facts from multiple document that have a greater information contents than individual bits. The way to think about it is threading pearls of information into a nice necklace. General approach to most automatic summarization is as follows

  • Retrieve relevant documents
  • Extract snippets of textual bits that are of importance
  • Rank these extracted snippets, and group them into some logical order
  • Natural Language Generation - Use these extracted information to generated a single document

Question Answering

Turing test is no longer the only way in which the AI community looks to evaluate and advance research. Recently AI2 a Paul Allen company is looking to reframe how efficient AI systems are by making them take Science and Math test which 7th graders do. Recently they hosted a competition on Kaggle where participants were asked to build model which will allow them to pass 7th grade science test. Participants were allowed to use corpus like and Wikipedia to build their models.

Google’s Knowledge Graph and Microsoft’s Santori are some of repository of human knowledge. Google has started serving answers as opposed to just returning collection of ranked documents. This has been made possible because of Knowledge graphs, which it has been curating semi-automatically. Knowledge graph not only stores factual information for different entities, but also associations among them. QA engineers are building blocks to Chat Bots and Virtual Assistants which we’re covering next.

Chat Bots and Virtual Assistants

Customer support is a tricky business, for commoditized businesses it is important to create value by providing the right support at the right time. Big tech companies have entered into an Artificial Intelligence arms race to create Virtual Assistants. List of things that these Virtual Assistants can do include

  • Find Phone numbers to nearby restaurants
  • Order Pizza
  • Book a Flight ticket
  • Alert you on traffic conditions
  • Perform Simple scheduling tasks

With Mobile AI now has rich context to Location, Time of Day and other sources of Publicly available information. IBM’s Watson, Google’s Now, Facebook’s M, Microsoft’s Cortana, Apple’s Siri and Amazon’s Echo are few to increasing number of players in this domain. These companies are getting sophisticated in sense they can convert your voice commands to actual actions.

Image an scenario where a Car’s Manufacturer App can walk through the fault detection steps, in certain cases based on user’s’ inputs dispatch a mechanic nearest to location of car breakdown. Or Imagine a Health Virtual Assistant that can handle your be in a better shape by recommending you different diet and exercises based on your parameters from fitness band.

Advances in Deep Learning, improvement in GPUs and ASIC for Machine Learning and growing number of open source tools are indicators of some of advances to come. Microsoft has open sourced their Bot framework. Facebook has opensource their GPU hardware and Google has open sourced their Deep Learning software which is optimized for their ASICs.