What's New in Alexa Conversations


Since the public beta launch of Alexa Conversations, we've made the following improvements.

March 2022

Clear the context in the Alexa Conversations Description Language (ACDL)

By default, Alexa Conversations carries over all arguments between API calls. If you don't want to carry over arguments, you can now apply a ResetArguments policy:

For complete details, see Clear the Context in the Alexa Conversations Description Language.

February 2022

Dialog evaluation tool for Alexa Conversations

The dialog evaluation tool lets you design high-quality conversational training data with less effort. You can work with the tool interactively to see the predicted action and response at each turn in a dialog. Here's how it works: Type an Alexa conversation, using the same words you'd use if you were talking to a friend or colleague. Then see how your skill responds. If you're satisfied, keep the conversation going. If the responses miss the mark, give the tool some feedback. This process launches a workflow in which you supply corrected output that the tool can learn from. For details, see Improve Your Model with the Dialog Evaluation Tool.

"Light Build" for ASK-CLI template skills

You can use the Alexa Skills Kit Command-Line Interface (ASK CLI) to create an Alexa Conversations skill from a quick-start weather bot template. Now this skill builds in two stages—Light Build and Full Build—for each configured language. Frequent status updates let you monitor the progress of the build, and you can begin testing your skill as it soon as the first light build is complete.

The weather bot reports today's weather for a fixed set of cities. The template includes all of the source code, interaction models, and Alexa Conversations artifacts you need to create a skill in Alexa Conversations Description Language (ACDL). You can use the built-in data or experiment by connecting the weather bot to a public weather API. For instructions on using the template, see Tutorial: Create an Alexa Conversations Skill with the ACDL.

September 30, 2021

July 21, 2021

  • Added locale support (beta) for en-AU, en-CA, en-IN, en-GB, and de-DE to the guided development experience.
  • Added reusable dialog support (beta) to ACDL.

April 2021

  • Updated the functionality of the shouldEndSession flag. For Alexa Conversations skills, the functionality of the shouldEndSession flag is now the same as with other custom skills, as described in Request and Response JSON Reference. Prior to this update, Alexa Conversations skills that didn't set the shouldEndSession value would, by default, have the shouldEndSession value set to false.

March 1, 2021

  • Went General Availability (GA).
  • Added support for Alexa entities for en-US.

February 18, 2021

November 30, 2020

  • Added user interface validations and error messages that notify you of the following:
    • Undefined variables.
    • The Affirm/Deny dialog acts must follow the Confirm API/Confirm Args dialog acts.
    • The Inform Args dialog act must follow the Request Args dialog act.
    • Dialogs must start with a user turn that uses the Invoke APIs dialog act.
    • Dialogs must end with an Alexa turn.
    • Dialogs must alternate between user turns and Alexa turns, except when an Offer Next API dialog act follows an API Success/Failure dialog act.
    • If more than one argument is requested for the Request Args dialog act, the Alternate Responses section contains a line for each required API argument. You can't modify these arguments or delete these lines.
    • User input validations.
  • Prevented you from using the same slot name multiple times in a sample utterance.
  • Made dropdown names visible on hover.
  • Disabled the Evaluate Model button when Alexa Conversations is the default dialog manager.
  • Added comments on the Analytics page to indicate how Alexa Conversations metrics are reflected.
  • Added a pop-up warning message before overwriting changes made from a separate tab.
  • Added Alexa Conversations skill responses in the Alexa app.
  • Automatically updated dialog turns with the Request Args and Confirm APIs dialog acts when changing mandatory API arguments.
  • Added hints on how to create dialogs in the right fly-out.
  • Made the following improvements when you configure responses:
    • Removed the + button. To create a new line, you now use the Add prompt button or press enter.
    • Removed the Create Prompt, Create and Edit APL Audio, and Reset buttons. Instead, you name your APLA document and enter Alexa prompts. The APLA documents are saved when you save your skill. You can still navigate to the APLA editor tool by clicking the APL for Audio editor link in the notification banner.
  • Removed the + button when you configure user utterance sets. To add a sample utterance, you now click the Add Utterance button or press enter.

Previous improvements

  • Resolved data loss issue caused by editing the same skill in multiple tabs.
  • Added documentation to explain how to troubleshoot and resolve the most common validation messages: MissingRequestTemplateV2, MissingDialogAnnotation, MismatchArgumentSize, and MissingSampleUtterance.
  • Reduced friction around multiple save buttons on the response page and added clearer call to actions in the developer console.
  • Enabled developers from outside the US to test their en-US Alexa Conversations skills.
  • Enabled slot types with values, which you can use to represent strings in API definitions and responses, without requiring dummy values.
  • Reduced friction in authoring responses in the dialog annotation flow by automatically filling in default response and prompt names.
  • Resolved the issue where dialogs were not automatically updated when utterance sets, responses, and API definitions were updated from the left navigation.
  • Added new validation messages to fast-fail the model build and advise how to resolve issues such as the following:
    • An utterance set has a null slot type
    • There are slot types (with properties) without defined properties
    • There are inconsistencies between arguments in API definitions and dialogs
    • The argument name of the API must match the arguments of the dialog acts
  • Improved model build error messages to advise how to resolve issues such as the following:
    • The first user utterance in a dialog is missing
    • Annotated dialogs are not yet generated
    • An utterance set has an argument that contains special characters
    • A dialog is missing annotations
    • An API definition is missing
  • Provided minor user experience upgrades such as the following:
    • Improved Firefox support
    • Added support for pressing enter on any dialog turn to create new, blank user and Alexa utterances
    • Added inline warning messages when slots are not set
    • Renamed response condition to API success condition / API failure condition
    • Highlighted the link to known issues on the dialog page
    • Restricted a single list slot per annotated utterance
    • Prevented the right fly-out when annotating slots and clicking outside utterances
    • Fixed the pagination bug
  • Added dialog cloning capability to enable you to create similar dialogs faster.
  • Added autocomplete of response names in the left navigation, utterance sets, and API definitions to reduce the effort required to create new components.
  • Added automatic propagation of changes to annotated variable names and API return variable names across the dialog to reduce the effort required to manually update them.
  • Enabled you to create all required Alexa Conversations components as you annotate dialogs so that you don't have to leave the dialog annotation flow and click separate tabs to create the required components. Specifically, we:
    • Removed the requirement to provide API definitions in utterance sets
    • Removed the requirement to provide API definitions in responses
    • Enabled you to create, edit, and delete API definitions from the right fly-out during dialog annotation

Was this page helpful?

Last updated: Nov 27, 2023