NLP components

Quick example

states:
    welcome:
        component: meya.text
        properties:
            text: "Let's test Dialogflow."
    dialogflow_question:
        component: meya.dialogflow
        properties:
            text: "Try saying 'Do you provide financing?' or 'Do you accept insurance?'"
            project_id: XXXX
            language: en
            error_message: "I don't understand 🤔. Try again."
            min_confidence: 0.8
            require_match: true
        transitions:
            financing: financing_state
            insurance: insurance_state
    financing_state:
        component: meya.text
        properties:
            text: "Yes, we do provide financing."
        return: true
    insurance_state:
        component: meya.text
        properties:
            text: "Yes, we do accept insurance."
        return: true

❗️

Intents are used to transition

In the transitions section of the code snippet above, both financing and insurance must be intents you've created in your NLU agent. If they don't exist, this code snippet will not work.

Similarly, any NLU intent you want to handle in this component must have an entry in the transitions section, otherwise nothing will happen when the user expresses that intent.

Flow scope variables returned by the NLU agent

All of the NLU agents (Dialogflow, api.ai, luis.ai, and wit.ai) set several variables on the flow scope related to the NLU intent matching process. These variables can be accessed from within a flow using Mustache syntax: {{ flow._intent }}, or from a Python component using self.db.flow.get("_intent").

meya.input_cms

Accepts input from the user and matches it against a Bot CMS-trained NLU agent. It also extracts entity data and makes that available on the flow scope.

states:
    input_cms_state:
        component: meya.input_cms
        properties:
            text: "What would you like to update?"
            space: "change"
            key: "phone"
            language: "en"
            error_message: "I don't understand"
            require_match: false
            min_confidence: 0.8
            detect_language: true
            buttons:
                - text: "My phone number"
                  action: phone
                - text: "Something else"
                  action: no_match
        transitions:
            phone: update_phone_number
            no_match: everything_else

    update_phone_number:
        component: meya.text
        properties:
            text: "Okay, let's update your phone number"
        return: true
    
    everything_else:
        component: meya.text
        properties:
            text: "You said something else"
        return: true

Transitions

next: the default transition for an answer (if not present, flow will transition to the next state in sequence)
no_match: the state to transition to if require_match is false and the user inputs something that doesn't match an intent, or matches an intent other than the one specified in this component.

📘

The captured user intent will be available as flow.intent later in the next step(s) of your flow.

meya.dialogflow

Accepts input from the user and matches it against a pre-trained Dialogflow model. It also extracts entity data and makes that available on flow scope. Read the Dialogflow docs

🚧

This component requires a Dialogflow v2 agent and the Dialogflow integration. If you are using a v1 agent, upgrade your agent or use the meya.api_ai component.

Transitions

next: the default transition for an answer (if not present, flow will transition to the next state in sequence)
no_match: the state to transition to if require_match is false and the user inputs something ambiguous

📘

The captured user intent will be available as flow.intent later in the next step(s) of your flow.

dialogflow_state:
    component: meya.dialogflow
    properties:
        text: "How are you doing today?"
        project_id: XXXX
        language: en
        error_message: "I don't understand"
        min_confidence: 0.8
        require_match: false
    transitions:
        bad: bad_state
        good: good_state
        no_match: ambiguous_state
bad_state:
    component: meya.text
    properties:
        text: "That's too bad."
    return: true
good_state:
    component: meya.text
    properties:
        text: "Great!"
    return: true

meya.wit

Accepts input from the user and matches it against a pre-trained wit.ai model. It also extracts entity data and makes that available on flow scope. Read the wit docs

🚧

Categorizing the user intent

If you are going to use the transitions property of this component you should train your model to classify utterances using the intent entity that categorizes the user intent. See example

Transitions

next: the default transition for an answer (if not present, flow will transition to the next state in sequence)
no_match: the state to transition to if require_match is false and the user inputs something ambiguous

📘

The captured user intent will be available as flow.intent later in the next step(s) of your flow.

component: meya.wit
properties:
	text: "How do you feel?"
	token: XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX
	error_message: "I don't understand"
	min_confidence: 0.8
	require_match: false
transitions:
	bad: bad_state
	good: good_state
	no_match: ambiguous_state

meya.api_ai

Accepts input from the user and matches it against a pre-trained api.ai model. It also extracts entity data and makes that available on flow scope. Read the api.ai docs

🚧

This component does not support Dialogflow v2 features, such as multi-language agents. If you wish to make use of those features, use the meya.dialogflow component with your v2 agent.

Transitions

next: the default transition for an answer (if not present, flow will transition to the next state in sequence)
no_match: the state to transition to if require_match is false and the user inputs something ambiguous

📘

The captured user intent will be available as flow.intent later in the next step(s) of your flow.

component: meya.api_ai
properties:
  text: "How do you feel?"
  client_access_token: XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX
  language: en
  error_message: "I don't understand"
  min_confidence: 0.8
  require_match: false
transitions:
  bad: bad_state
  good: good_state
  no_match: ambiguous_state

❗️

Intents are used to transition

In the transitions section of the code snippet above, both bad and good must be intents you've created in your Dialogflow agent. If they don't exist, this code snippet will not work.

Similarly, any Dialogflow intent you want to handle in this component must have an entry in the transitions section, otherwise nothing will happen when the user expresses that intent.

Example - Language Detection

You can use the meya.api_ai component to detect a user's language. Start by creating a Dialogflow agent with both English and French language enabled. Then create a greetings intent in the agent with English and French training phrases, such as hello and bonjour. Finally, create a new flow called language and paste the following code.

states:
    first:
        component: meya.text
        properties:
        		text: "Your language is {{ flow.language }}"

    api_ai_state:
        component: meya.api_ai
        properties:
            text: "Enter some text"
            client_access_token: <YOUR_CLIENT_ACCESS_TOKEN>
        transitions:
            greeting: greeting

    greeting:
        component: meya.text
        properties:
            text: "Your language is {{ flow.language }}"

Test it out in the test chat window. Enter hello and the bot will respond Your language is en. Enter bonjour and the bot will say Your language is fr.

meya.luis

Accepts input from the user and matches it against a pre-trained luis.ai model. It also extracts entity data and makes that available on flow scope. Read the luis docs

Transitions

next: the default transition for an answer (if not present, flow will transition to the next state in sequence)
no_match: the state to transition to if require_match is false and the user inputs something ambiguous

📘

The captured user intent will be available as flow.intent later in the next step(s) of your flow.

component: meya.luis
		properties:
  			text: "How do you feel?"
  			app_id: xxxxxxxxxxxxxxxxxxxxxxxxxx
  			api_key: xxxxxxxxxxxxxxxxxxxxxxxxxx
  			error_message: "I don't understand"
  			min_confidence: 0.8
		transitions:
  			bad: bad_state
  			good: good_state

❗️

Intents are used to transition

In the transitions section of the code snippet above, both financing and insurance must be intents you've created in your Luis agent. If they don't exist, this code snippet will not work.

Similarly, any Luis intent you want to handle in this component must have an entry in the transitions section, otherwise nothing will happen when the user expresses that intent.

meya.nlp_yes_no

Accepts input from the user and matches it against a pre-trained yes/no model (English).

Transitions

next: the default transition for an answer (if not present, flow will transition to the next state in sequence)
no_match: the state to transition to if require_match is false and the user inputs something ambiguous

❗️

yes and no are reserved in YAML so must be escaped with single or double quotes.

component: meya.nlp_yes_no
properties:
	text: Do you like Meya?
transitions:
	"yes": yes_state
 	"no": no_state