CSML Studio allows you to use your own pre-trained Natural Language Processing (NLP) service directly in your CSML chatbots with very little configuration. You can easily setup your favorite NLP provider in AI & NLU > NLU Configuration:

By default, all bots run in Strict Mode, where the input needs to exactly match one of the given rules to trigger a flow.

When a NLP provider is configured, all text events will be sent to the NLP provider and returned either as a payload event if an intent is found, or untouched if no intent is found. When an intent is found, the value of the payload will be intent:nameOfIntent.

No matter what NLP provider you pick, any event that passes through this process will have the following properties:

// if an intent is found:
event.intent = {
  "name": "..." // contains the name of the intent, same
  "confidence": "..." // the confidence score that this is the right intent
}

// or, when no intent is found:
event.intent == null

// if other intents match the request:
event.alternative_intents = [ Intent {}, ... ]

// when entities are found, event.entities containes a map of found entities
event.entities.ENTITY_NAME = [
  {
    "value": "the value",
    "metadata": {
      // ... additional metadata as returned by the NLP provider
    }
  }
]

// when this information is known, the processing language code
event.language = "en"

A few additional properties are also set in the resulting event:

• event._nlp_provider: contains details about the NLP provider integration

• event._nlp_result: contains the raw response from the NLP provider

• event.text: contains the original text input

Using NLU in your flows

You can now use the event by matching a found intent with a flow command, by setting that intent as one of the accepted commands for a flow, using the AI Rules feature.

Alternatively, you can also decide to match buttons or other actions within a flow with the found data. For instance, you can use NLP to detect a YES_INTENT and match it without having to list all the possible ways to say "yes" in the button's accepts array.

start:

  say Question(
    "Do you like chocolate?",
    buttons=[
      Button("Yes", accepts=["intent:YES_INTENT"]) as btny,
      Button("No", accepts=["intent:NO_INTENT"]) as btnn,
    ]
  )
  hold
  
  if (event.match(btny)) say "I knew it!"
  else if (event.match(btnn)) say "I think you are lying..."
  
  goto end

Using SaaS providers

CSML Studio currently supports the following NLP providers:

• Dialogflow (read our blog post)

• SAP Conversational AI (formerly Recast.ai)

• Amazon Lex

• Microsoft LUIS

• Rasa

• Wit.ai

• IBM Watson

• Custom Webhook

Please refer to each provider's documentation for more information on how to generate credentials. CSML Studio supports the most common and secure way of connecting to these providers.

If your favorite provider is not in this list, please reach out to contact@csml.dev.

Using the Custom Webhook integration

You can also add a custom NLP engine by using the custom Webhook integration. This is especially useful if we don't currently support your NLP provider of choice or if you are running your NLP services on premise.

To use the Webhook integration, simply provide a URL that we can send a POST request to, with the following body:

{
  "text": "TEXT_TO_ANALYZE"
}

The URL may contain query parameters, for example with an API Key for authentication purposes.

When called, this endpoint must return the response in the following form:

{
  // intent can also be null if no intent is found
  "intent": { 
    "name": "NAME_OF_INTENT", 
    "confidence": 0.84
  },
  "alternative_intents": [ 
    // ... other matching intents
  ],
  "entities": {
    "someEntity": [
      {
        "value": "somevalue",
        "metadata": { 
          // ... any additional information about entity
        },
      },
      // ... other values for the same entity
    ],
    // ... other entities
  }
  // ... any additional data: sentiment analyzis, language...
}

You can test this endpoint directly on the configuration page.

In the AI Rules section of CSML Studio, you will find a way to configure flow commands for any of your flows. This section makes it easy to see in one single screen all the rules that govern the triggering of any of your flows!

Commands are words or sentences (or intents, when NLU is configured) that will trigger a given flow. For example, in the example below, if a user says "Hi", no matter where they are in the bot, they will be redirected to the Welcome flow:

You can have as many AI Rules as you want on as many flows as you want. The commands are case-insensitive, so for example "Hi", "hi", "HI" and "hI" will all trigger the Welcome flow. Also, if the same rule is used as an input to two different flows, the triggered flow will be randomly chosen among the target flows.

This feature can also be used in conjunction with any Preprocessing or NLP configured in your bot. For instance, if after going through a NLP service, an intent is detected, the intent name can be used as a command for any given flow (with the intent: prefix, i.e intent:myIntentName). For instance, this is how an intent called DemoContext would be handled with AI Rules: