Adrian Catalan - Party_like_it_s_1970 - Conversational_interfaces_are_back__into_your_Android_app

Transcript

1. PARTY LIKE IT'S 1970: CONVERSATIONAL INTERFACES ARE BACK (INTO YOUR APP) Adrián Catalán @ykro https://goo.gl/yYoRAs
2. (very) BRIEF HISTORY LESSON #DCBERLIN18
3. • 1952 - Audrey(Bell Labs), recognized strings of digits • 1970 – Hidden Markov Model (HMM), which helped speech recognition move away from literal matching to prediction • 70's - DARPA Projects
4. • 80's - Template based brute-force & HMM become popular • 1984 – Speechworks, Automated Speech Recognition over IVR • 90's - Industry starts
5. • 1997 - Dragon NaturallySpeaking, recognition of continuous speech • 2007 - Siri • 2014 - Amazon Echo • 2016 - Google Home
6. DESIGN PROCESS #DCBERLIN18
7. https://www.pexels.com/photo/fashion-sunglasses-woman-girl-46244/ CREATE A PERSONA
8. https://www.pexels.com/photo/red-vintage-shoes-sport-1994/ • Tone • Style • Word and phrase choices • Technique • Voice • Functional design
9. https://www.pexels.com/photo/wood-explosion-fire-hot-8504/ DESTROY THE BOX!
10. TRADITIONAL THINKING WON’T HELP
11. https://www.flickr.com/photos/81909228@N07/36169378343 WRITE A SCREENPLAY
12. https://www.pexels.com/photo/ball-close-up-colors-dark-593467/ COOPERATION! Turn-taking Context Threading “Reading between the lines”
13. https://www.flickr.com/photos/16105375@N00/1764118705 THERE ARE NO ERRORS
14. REAL LIFE CAN, SOMETIMES, SUCK... http://depts.washington.edu/antiwar/may1970strike.shtml
15. UNDERSTANDING IS KEY http://depts.washington.edu/antiwar/may1970strike.shtml
16. USERS RULE!
17. THEY KNOW HOW TO TALK
18. THEY KNOW WHAT THEY WANT
19. THEY WANT TO HELP
20. CONTEXT MATTERS
21. BACK UP PLAN
22. USER GOOGLE ASSISTANT NUMBER GENIE USER NUMBER GENIE USER NUMBER GENIE Canonical “Happy path” First time experience Tapered experience (Return user) Repair SAMPLE DIALOGS
23. SAMPLE DIALOGS USER GOOGLE ASSISTANT NUMBER GENIE USER NUMBER GENIE USER NUMBER GENIE Canonical “Happy path” First time experience Tapered experience (Return user) Repair
24. SNACKS #DCBERLIN18
25. • Avoid written language and use spoken one • Kick off the conversation
26. • Guide user through the conversation • Keep our TTS (Text to speech) short and clear
27. • Avoid data points unrelated to user query • Follow natural turn taking in the conversation
28. • Use conversational makers
29. ADDING MACHINE LEARNING #DCBERLIN18
30. “ — — @
31. https://dialogflow.com/docs/agents
32. INVOKE Say Hello Start the experience with the agent
33. INTENT
34. TRAINING PHRASES
35. ENTITIES
36. FULLFILLMENT Backend APP
37. RESPONSE
38. CONTEXT move between intents achieve more fluid conversations
39. Integrating with AIService class MainActivity : AppCompatActivity(), AIListener { val config = AIConfiguration( "ACCESS_TOKEN", ai.api.AIConfiguration.SupportedLanguages.English, AIConfiguration.RecognitionEngine.System ) aiService = AIService.getService(this, config) aiService.setListener(this) btnListen.setOnClickListener { aiService.startListening() } }
40. override fun onResult(response: AIResponse?) { var strParameters = "" val result = response?.result result?.parameters?.let { params -> params.forEach { key, value -> strParameters += "($key,$value)" } txtResult.text = "Query: ${result.resolvedQuery}" + "\nAction: ${result.action}" + "\nParameters: $strParameters" } } Integrating with AIService
41. PARTY LIKE IT'S 1970: CONVERSATIONAL INTERFACES ARE BACK (INTO YOUR APP) Adrián Catalán @ykro https://goo.gl/yYoRAs