Dialog Acts for Virtual Agent

  • Release version: Yokohama
  • Updated January 30, 2025
  • 2 minutes to read
    • Summarize
    Summarized using AI
    This content was generated using new OpenAI-powered functionality. Results are provided on an as is basis and are not guaranteed to be accurate or complete.

    Summary of Dialog Acts for Virtual Agent

    The Dialog Acts feature in Virtual Agent, introduced in the Vancouver release and updated in Yokohama, enhances natural language communication by allowing users to interact more fluidly with Virtual Agent. Instead of relying solely on predefined prompts, users can express changes or confirmations in their own words during a conversation, enabling a more dynamic and efficient dialogue.

    Show full answer Show less

    How Dialog Acts Works

    Dialog Acts interprets user inputs using Natural Language Understanding (NLU) to modify conversation variables on the fly. For example, if a user requests a laptop and later wants to change the RAM specification, they can simply say, “Actually, I need 8 GB of RAM,” and the system updates the request immediately without restarting the dialogue flow. If the input doesn’t match the NLU model, it is logged as ignorable. User utterances, feedback, and language data are tracked in specific Open NLU tables.

    Activating Dialog Acts

    Customers enable Dialog Acts by toggling the Enable Dialog Acts switch within the topic creation or properties settings. Once enabled, an additional Confirm modified values toggle appears to prompt users for confirmation when variables are changed through their utterances.

    Implementation Considerations

    • Dialog Acts requires active NLU for the instance and topic.
    • At least one preconfigured response must be active for Dialog Acts options to appear.
    • Dialog Acts can function without a model intent but cannot select entities unless the topic is linked to the model or intent.
    • Decision nodes can be scripted or conditioned to bypass reprompting, but modification toggles do not support topics with multiple date-time nodes.

    Available Dialog Acts Responses

    The feature supports three core response types stored in the Open NLU Dialog Acts table:

    • Modify: Alters a previously recorded variable.
    • Affirm: Confirms a variable.
    • Negate: Declines confirmation of a variable.

    This capability allows Virtual Agent conversations to be more natural, efficient, and user-friendly, reducing the need for repetitive navigation through prompts and improving the overall user experience.

    As part of the Vancouver release, the Dialog Acts feature expands natural language communication with Virtual Agent, leading to more fluid conversations between user and agent.

    How Dialog Acts works

    Dialog Acts interprets and applies user inputs during a Virtual Agent conversation. With this feature active, you engage with Virtual Agent in your own natural language, instead of selecting replies based on prompts in the topic.

    For example, if you request a laptop in a Virtual Agent conversation, you use the topic prompts to select a manufacturer, screen size, and amount of RAM. Without Dialog Acts, you change specifications only by following the prompts in the topic, potentially taking more time to loop through a Virtual Agent conversation as you reset your options. With Dialog Acts, if you change your mind about any of the laptop's specifications before finishing the request, you can input an utterance such as, "Actually, I need 8 GB of RAM." If your input matches the Natural Language Understanding (NLU) model closely enough, the specifications are changed immediately, even if you're on a different point in the topic. If the input doesn't match, it's added to a list of ignorable utterances.

    User utterances are stored in the Open NLU Predict State Tracking [open_nlu_predict_state_tracking] table, while feedback is tracked by the Open NLU Predict Dialog Act Feedback [open_nlu_predict_dialog_act_feedback] table. Languages are tracked in the Open NLU Driver Languages [open_nlu_driver_language] table. In the current release, English is the sole language support for Dialog Acts.

    Activating Dialog Acts

    Activate Dialog Acts by toggling the Enable Dialog Acts switch in the Create a topic form or the Topic Properties tab. When activated, a second Confirm modified values toggle appears. Use this switch to provide a prompt confirming that a variable is to be changed based on your utterances.

    Considerations for implementing Dialog Acts

    The Dialog Acts feature is available as long as NLU is active for the instance and the topic. However, at least one of the preconfigured responses must be Active for the Dialog Acts toggles in the Topic Properties tab to be available. Dialog Acts can work without a model intent, but it can't select an entity if the topic is not associated with that model or intent.

    Decision nodes can be modified by a condition or script to skip reprompting by Dialog Acts. However, the Modify Dialog Acts toggle doesn't work if there are multiple date-time nodes on a topic.

    Available Dialog Acts responses

    The current release includes the following types of responses, stored in the Open NLU Dialog Acts [open_nlu_dialog_act] table:
    Table 1. Dialog Acts response types
    Type Description
    Modify Alter a variable recorded previously.
    Affirm Confirm a variable.
    Negate Decline confirming a variable.