Using runtime hints to improve recognition of slot values - Amazon Lex

Using runtime hints to improve recognition of slot values

With runtime hints you can give Amazon Lex V2 a set of slot values based on context to get better recognition in audio conversations and improved slot resolutions. You can use runtime hints to provide a list of phrases at runtime that become candidates for the resolution of a slot value.

For example, if a user interacting with a flight reservation bot frequently travels to San Francisco, Jakarta, Seoul, and Moscow you can configure runtime hints with a list of these four cities when eliciting for the destination to improve recognition for frequently travelled cities.

Runtime hints are only available in the English (US) and English (UK) languages. They can be used with the following slot types:

  • Custom slot types

  • AMAZON.City

  • AMAZON.Country

  • AMAZON.FirstName

  • AMAZON.LastName

  • AMAZON.State

  • AMAZON.StreetName

Runtime hints basics

  • Runtime hints are used only when eliciting a slot value from a user.

  • When you use runtime hints, the values of the hints are preferred over similar values. For example, for a food ordering bot, you can set a list of menu items as runtime hints while eliciting for food items in a custom slot to prefer “fillet” over similar sounding “fella”.

  • If the user input is different from the values provided in runtime hints, the original user input will be used for the slot.

  • For custom slot types, values provided as runtime hints will be used for resolution of the slot even if they are not part of the custom slot during bot creation.

  • Runtime hints are supported only for 8 kHz audio input. They are available with contact center integrations supported by Amazon Lex V2. Runtime hints aren't provided for audio input from the test window on the Amazon Lex V2 console because it uses 16 kHz audio input.

Note

Before you can use runtime hints with an existing bot, you must first rebuild the bot. Existing versions of a bot don't support runtime hints. You must create a new version of the bot to use them.

You can send runtime hints to Amazon Lex V2 using the PutSession, RecognizeText,RecognizeUtterance, or StartConversation operation. You can also add runtime hints using a Lambda function.

You can send runtime hints at the beginning of a conversation to configure the hints for each slot used in the bot, or you can send hints as part of the session state during a conversation. The runtimeHints attribute maps a slot to the hints for that slot.

Once you send a runtime hint to Amazon Lex V2, they persist for every turn of the conversation until the session ends. If you send a null runtimeHints structure, the existing hints are used. You can modify the hints by:

  • Sending a new runtimeHints structure to the bot. The contents of the new structure replace the existing ones.

  • Sending an empty runtimeHints structure to the bot. This clears the runtime hints for the bot.

Adding slot values in context

Add context for your bot by providing expected slot values as runtime hints when your application has information about the user's next likely utterance. This enables your bot to use information that your application has to improve the conversation with the user.

For example, in a banking app you can generate a list of account nicknames for a specific user, and then use the list when eliciting the account that the user wants to access.

Send runtime hints at the start of the conversation when you have context to help your bot interpret user input. For example, if you have the user's phone number, you can use this information to look up the user so that you can use the PutSession or StartConversation operation to pass first and last name hints to the bot if you are eliciting for the user's name to validate their credentials.

During a conversation, you might gather information from one slot value that can help with another slot value. For example, in a car care app when you have the user's account number you can do a look up to find the cars that the customer owns and pass them along as hints to another slot.

Enter acronyms, or other words whose letters should be pronounced individually, as single letters separated by a period and a space. Don't use individual letters unless they are part of a phrase, such as "J. P. Morgan" or “A.W.S”. You can use upper- or lower-case letters to define an acronym.

Adding hints to a slot

To add runtime hints to a slot, you use the runtimeHints structure that is part of the sessionState structure. The following is an example of the runtimeHints structure. It provides hints for two slots, "FirstName" and "LastName" for the "MakeAppointment" intent.

{ "sessionState": { "intent": {}, "activeContexts": [], "dialogAction": {}, "originatingRequestId": {}, "sessionAttributes": {}, "runtimeHints": { "slotHints": { "MakeAppointment": { "FirstName": { "runtimeHintValues": [ { "phrase": "John" }, { "phrase": "Mary" } ] }, "LastName": { "runtimeHintValues": [ { "phrase": "Stiles" }, { "phrase": "Major" } ] } } } } } }

You can also use a Lambda function to add runtime hints during a conversation. To add runtime hints, you add the runtimeHints structure to the session state of the response that your Lambda function sends to Amazon Lex V2. For more information, see Response format.

You must specify a valid intentName and slotName in the request, otherwise Amazon Lex V2 returns a runtime error.