Adapt - Text to Intent

Adapt is a keyword based parser. Intents are built atomically from keywords that are either "required" or "optional". This is quite different from the way intent templates are entered into Naomi, as complete sentences representing ways in which a user might ask for a service. Adapt, in it's default form, is also not set up to avoid name collisions between intents. For these reasons, we do a pre-parsing step to both ensure that every intent name is unique, even if written by different authors with no knowledge of each other, then parse through the intents in order to identify words that either appear in every template for the intent (required) or that appear in some intents (optional), plus use intent keywords.

For example, a Naomi intent might look like this:

    'CalendarIntent': {
        'locale': {
            'en-US': {
                'keywords': {
                    'WhenKeyword': [
                'templates': [
                    "ADD CALENDAR EVENT",
                    "ADD AN EVENT TO MY CALENDAR",
                    "DO I HAVE ANY CALENDAR EVENTS {WhenKeyword}",
                    "WHAT IS ON MY CALENDAR {WhenKeyword}"
        'action': self.handle

We would see that the word "CALENDAR" appears in every template, so that would become a "required" word. "EVENT" appears in three templates, and "ADD" and "{WhenKeyword}" both appear in two templates, so they would be "optional". Words such as "AN", "TO", "MY", "I", "ANY", "HAVE", "WHAT", "IS" and "ON" are likely to appear in other intents, and thus will have lower weights.

This means the Adapt intent would look something like


When processing a request, if any words match one of the words under {WhenKeyword} then that word is stored as a value for WhenKeyword. So the request "WHAT DO I HAVE ON MY CALENDAR TODAY" comes in, it is converted to "WHAT DO I HAVE ON MY CALENDAR {CalendarEvent01_WhenKeyword}, CalendarEvent01_WhenKeyword: TODAY" before being matched.

Project home:

Installation: sudo pip3 install adapt-parser


tti_engine: Adapt TTI