Released in 2014 Over 100k Skills on Alexa 4-5 customers that have Echo devices uses Skills Skills are available in 80 countries, 30 language variants
One-shot interactions Multi-turn interactions (series of questions)
Happy path -single dialog turn -linear multi turn (easy to build)
Not a happy path -single dialog turn -non-linear multi turn, conversations deviate from the script (tracking state, context)
Alexa Conversations - deep learning, tracks the context, less code for state management
Alexa CConversations - opens up new experiences for your customers
Default Dialog Manager Custom Skill Dialog Manager (portion of the dialog)
5 build-time components
Dialogs Utterance Set (dialog acts - Invoke, Inform, Affirm) Slots (singular slot, compound slot) API definitions Response Templates (dialog acts - Request, Confirm, Notify)
Request type API name API arguments
API response slot
Pros State management out of the box Confirmations User corrections (user changes their mind while defining) Proactive offers Interoperability (extend existing Alexa skill with Alexa Conversations)
Dialog Simulator consists of simulated dialog flows which produce dialog with variations Using paraphrasing the natural language variations are generated Last step is to train the dialogs
Goal Oriented dialog (take turns and preform actions, Alexa follow up with questions to get more information)
User more cooperative (gives all information) User changes their mind (not sure) User is less cooperative (gives less information)
NER (named entity recognition) model (recognizes slots in user utterances) AP (action prediction) model (predicts what is the next Alexa model) AF (argument filling) model (fills in the values of API arguments from the dialog context)
Easier to capture and handle user corrections Less time spent for writing slots and confirming new slots Easy to extend existing skill
Reference skills alexa.design/ac-github Technical documentation for Alexa Conversations alexa.design/ac-docs Learn the fundamentals of building with Alexa CConversations alexa.design/ac-tutorial
NLU Evolution Utterance Profiler Intent Chaining
ASR (automatic speech recognition) - converts user speech to text Automatic Speech Recognition (ASR) Evaluation tool
ASR evaluation tools saves time testing your skill
Out of Domain utterances
Sensitivity Tuning - review your customer's utterances in Intent History
Change fallbackIntentSensitivity setting level:
Low - your skill requires large numbers of different words and phrases that might not be in your sample utterances Medium - specific utterances are covered by your skill model High - interaction model supports most utterances
allows richer and more consistent experiences
Shared Slots - share slot with multiple skills Catalog Management -
Resolve common entities (more detail about the utterance, giving a slot link)
More personalization, more engagement
3 different ways of personalization:
Account Linking Customer Profile API (fetch user details) Voice profiles
App-to-App Account Linking (from your app or Alexa App)
link Amazon identity with the identity in your app (removes friction of remembering account credentials, both on iOS and Android)
Enable Universal Links (iOS), App Links (Android) Configure the skill for account linking Handle the authorization request in your app Redirect the user back to the Alexa App Use the access token in the skill
Enable permissions Prompt the user to give consent Get the access token (refresh on each request) Call Personal Profile API
http://alexa.design/codegenerator
ASK-SDK Code Generator for Alexa Skill Language Models
Alexa Skills Kit SDK for Node.js
The Alexa Design Guide
Zero to Hero: A comprehensive course to building an Alexa Skill
