© 2012 Adobe Systems Incorporated. All Rights Reserved. Adobe Confidential. Moses Tool Set A set of tools based on Adobe technology to simplify your usage of Moses Yu Gong | Software Engineer
Jul 03, 2015
© 2012 Adobe Systems Incorporated. All Rights Reserved. Adobe Confidential.
Moses Tool Set A set of tools based on Adobe technology to simplify your usage of Moses Yu Gong | Software Engineer
© 2012 Adobe Systems Incorporated. All Rights Reserved. Adobe Confidential.
Agenda
§ Addressing Moses Pain Points
§ Advantages of Moses Tool Set
§ Moses Tool Set Architecture
§ Moses Tool Set Features
§ Useful Resources
§ Q&A
© 2012 Adobe Systems Incorporated. All Rights Reserved. Adobe Confidential.
Addressing Moses Pain Points
1. Corpus Cleaning
2. Engine Training
3. Engine Testing
4. Integrating Moses With Linguistic Platform
© 2012 Adobe Systems Incorporated. All Rights Reserved. Adobe Confidential.
Advantages of Moses Tool Set
• User Friendly
• Platform Independent
• Open Source
© 2012 Adobe Systems Incorporated. All Rights Reserved. Adobe Confidential.
Moses Tool Set Architecture
© 2012 Adobe Systems Incorporated. All Rights Reserved. Adobe Confidential.
Moses Tool Set Features – Corpus Cleaning
Moses Func*onality • Tokenizing • Casing • Long Segments
Adobe Func*onality • Placeholder Handling • URL Handling • Number Cleaning • Duplicate Line
Cleaning • Weird Aligned Pairs • Cleaning by regular
expressions
© 2012 Adobe Systems Incorporated. All Rights Reserved. Adobe Confidential.
Moses Tool Set Features – Corpus Splitting & Uploading
Split Corpus by Purpose • Training • Tuning • TesCng
Upload Split Corpus to Moses Server
© 2012 Adobe Systems Incorporated. All Rights Reserved. Adobe Confidential.
Moses Tool Set Features –Training & Tuning
Command Line Pain
Human Unfriendly • Highly Detailed • Error Prone • Difficult To Reproduce
© 2012 Adobe Systems Incorporated. All Rights Reserved. Adobe Confidential.
Moses Tool Set Features –Training & Tuning
UI To Simplify Inputs • Training Run ID • Language Model
Parameters • Corpus ID • Source & Target • Default Alignment • Default Reordering • Remote Server
© 2012 Adobe Systems Incorporated. All Rights Reserved. Adobe Confidential.
Moses Tool Set Features –Testing
• How do you know when an engine is good enough? • How do you know when it is intrinsically flawed? • How do you automate comparing a new engine to old ones?
© 2012 Adobe Systems Incorporated. All Rights Reserved. Adobe Confidential.
Moses Tool Set Features –Testing
• Reliable Scoring • Bleu/Nist/Meteor
• Simplified UI • Dynamic ConnecCon to
exisCng engines • Repeatable
© 2012 Adobe Systems Incorporated. All Rights Reserved. Adobe Confidential.
Moses Tool Set Features –Testing
• Reliable Scoring • Bleu/Nist/Meteor
• Simplified UI • Dynamic ConnecCon to
exisCng engines • Repeatable
© 2012 Adobe Systems Incorporated. All Rights Reserved. Adobe Confidential.
Automation
Corpus Cleaning
Corpus Splitting & Uploading
Training & Tuning
Testing
© 2012 Adobe Systems Incorporated. All Rights Reserved. Adobe Confidential.
Localization Workflow Integration
Moses Tooling Chain
Linguistic Platform
© 2012 Adobe Systems Incorporated. All Rights Reserved. Adobe Confidential.
Resources
Source Code: http://code.google.com/p/m4loc
© 2012 Adobe Systems Incorporated. All Rights Reserved. Adobe Confidential.
Questions
© 2012 Adobe Systems Incorporated. All Rights Reserved. Adobe Confidential.