Complex logic at the start of conversation

Last modified: 26 September 2021

To be more human-like, it may be necessary to implement special behavior at the start of the conversation. Generally, it could be described with the following requirements:

Do not speak until interlocutor starts speaking
But still ping interlocutor with "hello" if silence is too long
Start to speak as soon as interlocutor start speaking
When interlocutor finishes a speech, proceed with the dialogue whether the phrase was recognized or not

Let's start from a relatively simple example, implementing 4 and partially implementing 1 and 3. In this example, we won't use a common dialogue library.

context
{
    input phone: string,
}

start node greeting
{
    do
    {
        #connectSafe($phone);
        // Wait until 500 is left or speech is detected.
        #waitForSpeech(500);
        #sayText("Hello!");
        wait *;
    }
    transitions
    {
        inviteToTalk: goto inviteToTalk on true;
    }
}

node inviteToTalk
{
    do
    {
        // Just proceed with a standard flow from now on.
        #sayText("Do you want to talk with me?");
        exit;
    }
}

Now, let's add common dialogue and fully implement all requirements. This example is pretty monstrous, in particular, the digression control interface is verbose, but we're working on it.

import "commonDialogue.dsl";

context
{
    input phone: string,
    output status:string?,
    output serviceStatus:string?,
    output callBackDetails:string?,
}

start node greeting
{
    do
    {
        if(#getVisitCount("greeting") < 2)
        {
            #connectSafe($phone);
        }
        // Disable all common dialogue digressions except `hello`, `answeringMachine`,
        // `iWillCallBack` and `sorryWontCall` to prevent undesired digressions
        // (such as "dont understand") from triggering.
        // Note, that non-disabled digressions still may trigger.
        digression disable
        {
            can_hear_you,
            dont_understand,
            dont_understand_preprocessor,
            i_am_robot,
            reactive_listening,
            repeat_preprocessor,
            repeat,
            @wait
        };
        // Wait until 500 is left or speech is detected.
        if (!#waitForSpeech(500))
        {
            // If 500 is left but no speech is detected,
            // re-enter this node.
            // Note that here `hello` digression also may trigger.
            wait
            {
                self
            };
        }
        
        #sayText("Hello!");
        // Here, only `inviteToTalk`, `iWillCallBack` and `sorryWontCall` may be triggered.
        wait
        {
            inviteToTalk
        };
    }
    transitions
    {
        inviteToTalk: goto inviteToTalk on true;
        // This transition will compete with `hello` digression.
        // If enough time is left for `hello` digression to trigger,
        // it will trigger because its priority is higher.
        // When `hello` return, this transition will be triggered by the following tick.
        self: goto greeting on true priority -1000 tags: ontick;
    }
    onexit
    {
        inviteToTalk: do
        {
            // Enable all digressions back.
            digression enable
            {
                can_hear_you,
                dont_understand,
                dont_understand_preprocessor,
                i_am_robot,
                reactive_listening,
                repeat_preprocessor,
                repeat,
                @wait
            };
        }
    }
}

node inviteToTalk
{
    do
    {
        // Just proceed with a standard flow from now on.
        #sayText("Do you want to talk with me?");
        exit;
    }
}

Found a mistake? Let us know.

Complex logic at the start of conversation

Enroll in beta