Over the last few years, Grape, the enterprise messaging app backed by Betaworks and Zynga founder Mark Pincus, has been toiling away to build a smarter chat box powered by Natural Language Processing (NLP).
The aim was to go beyond reducing the “look-up factor” — the time it takes to link to and reference external information, achieved via the Austrian startup’s IndexAPI that sees it able to integrate with all manner of public, enterprise and bespoke data — to also enable Grape to detect and trigger workflows automatically, such as the ability to recognise dates, questions, assignments and to-dos.
The latter functionality has been quietly rolled out for customers who opt in, and TechCrunch was given an exclusive first-look. On the surface at least, the results are impressive.
Dubbed ‘Intelligence Amplification’, the new feature was developed in cooperation with research partners, the Austrian Institute for Artificial Intelligence (ÖFAI) and the Austrian Research Promotion Agency (FFG).
The goal, says co-founder and COO Leo Fasbender, is to put an end to “slash-commands and bots,” which he says work great for techies but have less adoption with casual, normal users “outside our little tech industry bubble”.
In practice this works by Grape performing ‘linguistic analysis’ on each messaging string. This is based on rules that the startup has defined in combination with Random Forest, a machine learning approach. “Using this, we classify messages into different Speech Acts: utterances that have some performative function in language and communication,” explains Fasbender.
The result is that Grape is able to recognise when a chat message relates to a possible calendar entry, question, to-do, and more. When it determines that a message requires a resulting action, such as scheduling a meeting, an action button appears below the message. Clicking on it adds that entry to a user’s calendar, with all related details pre-populated.
Grape’s add to calendar action
Grape’s add to todo action
Grape’s NLP processing: categorising chat messages based on language, string-type, if it contains a question, an action and more
Grape’s NLP engine is multilingual, supporting English and German
Adds Fasbender: “Our approach is also quite innovative in that the messaging and the NLP system work together in an asynchronous manner. This means the chat system can retain it’s “snappiness” and doesn’t have to wait for the potentially slow analysis to complete. This is important for the sole reason that Grape is a communication solution, primarily, with our language processing functioning as a kind of smart assistant in the background, to automate workflows, schedule meetings and assign tasks.”
Comment