Join us on Tuesday, July 12, at 2 pm (EDT) when All Analytics Radio hosts a discussion on the chan
Conditional Action Programmer (CAP), an experimental project from Microsoft's Technology and Research group, offers a way to automate app interactions using natural language rather than code.
By Microsoft's own admission, Conditional Action Programmer is the "worst name ever." Microsoft Bob, Microsoft Kin, and PlaysForSure, you may step down now.
The app automation service, which can be reduced to the more appealing abbreviation CAP, is similar to glue services IFTTT and Zapier that allow apps to interact with each other. It's also similar to another recently introduced Microsoft offering, Flow, so much so that it's tempting to wonder whether the company's various teams talk to one another.
But conversational comprehension is CAP's reason for being. The experimental project from Microsoft's Technology and Research group offers a way to automate app interactions using natural language rather than code.
CAP supports Flow's menu-driven programming model in which a user directs the service, for example, to send an SMS notification to a mobile device when an email arrives in Outlook. This is a proven method for interaction and is easy enough for almost anyone to manage.
But CAP's aims is to provide a natural language interface, a way to describe app interactions in conversational text. By selecting "Add a new task" and typing "Send me an email when Microsoft stock goes up" in the input form text box, CAP responds by presenting that task in a structured web form that provides a way to set a threshold stock price and email subject and body fields.
CAP can do this because it is able to understand the input and match the parsed text with available services. It appears to have trouble parsing text that's all lower-case, however. Told to "send me an email when i get an email," CAP returned a form to send an email when an RSS feed with a specified term is found. The command input as "Send me an email when I get an email," nonsensical though it may be, returned a form to configure the requested inbox flood.
But Microsoft acknowledges that CAP needs further training. The company is making CAP available so it can gather the data necessary to train its language model better. "We need your help to continue to improve natural language understanding," the company explains on the CAP website. "You can help us and benefit from automatic task completion at the same time by using CAP!"
Data is the fuel that makes machine learning models work well. Without enough data, speech recognition, image recognition, and natural language parsing tend to perform poorly. Microsoft says CAP will request access to the applications and internet services that users link and it may retain such data for up to a year to improve its products.
Beyond its utility as a way to automate processes using conversational text, CAP could prove useful as a way to connect cloud services through spoken commands, using the speech-to-text capabilities of voice-based assistants like Microsoft Cortana. But first Microsoft has to nail the basics.
CAP is available on the web, and apps for Android and iOS are said to be forthcoming.