TerraTalk
BP Study 2016/07/27
Twitter: 1664riffs GitHub: ykakihara
・
・
@PyCon JP 2014 Publicity
http://wbawakate.connpass.com/event/
・
Realistic Role-Playing Lessons
Proprietary QLEAR Engine for giving
feedback on users speech
Practice as much as you want
Any time you want
TerraTalkEnglish Tutor As An App
https://www.youtube.com/watch?v=fZs0qjgLlKM
1. Select a role 2. Pick a mission 3. Talk through 4. Get feedback
API
Web I/F
ELB
ELB


EC2 Instances
NLP/ML


EC2 Instances
ELB


EC2 Instances
※ ・Queue
Heroku
• >>>>>
• Python 3
• requirements.txt
• Add-on
• iOS/Android Swift Java
•
•
• iOS + Android
 
• TerraTalk
•
•
•
•
• &
• ・
•
• to
• ・
Finite State Machine
AI
AI AI
•
•
•
•
・
•
•
•
•
:
AI
AI
•
•
• AI
• “Can you speak more slowly
please?” “I’m sorry, could you repeat
what you’ve just said?”
・
API
Let’s go to the library.
Let’s go to try berries.
No
Good night
Good knight
↓
↓
• ・
•
•
• /
Shared Task: Grammatical Error
Correction
ACL 2013, 2014
•
• N-Gram
• ・
•
•
•
•
•
QA
• QA
• QCD
•
• PDCA D
API
Web I/F
ELB
ELB


EC2 Instances
NLP/ML


EC2 Instances
ELB


EC2 Instances
※ ・Queue
• Django app
• Python - 8,000 lines
• models.py - 1,196 lines ・・
• templates (HTML) - 1,190 lines
• JavaScript - 1,400 lines
Data Driven Front-End
• Backbone.js
• Joint JS
• Twitter Bootstrap
• ・
• Backbone.js
• Directed Graph
•
• TerraTalk
•
• CMS+
• ・ Python
開発チームの英語コミュニケーション問題の
事例が知りたいです。
y.kakihara@joyz.co.jp

英会話アプリ TerraTalk の裏側