Lemko Machine Translation Moonshot

What I really want - outcome - specific measurable result

• Make $100k in IT in 2018
• Get code out there and working, and an app in the app store, and consolidate and market later.

Why:

• translation rates are in freefall
• have to learn to program machines before AI takes over
• have to make the jump from freelance translator to full-fledged computer programmer

Plan 1

• Get transliteration app out there ASAP

Plan A User stories:

• User pastes in Lemko text and obtains passable Polish (rule-based)
• User pastes in Polish text and obtains passable Lemko (rule-based)
• User pastes in Lemko text and obtains passable English (neural machine translation)
• User pastes in English text and obtains passable Lemko (neural machine translation)

Plan B: Hybrid rule-based plus Google Translate API machine

• User story: User pastes in Lemko text, rule-based machine translation into Polish, Google Translate API translates Polish into English
• User story: User pastes in English text, Google Translate API translates Polish into Lemko, rule-based machine translation into Lemko

Modules

1. Rule-based translation
   • Transcription
     • Cyrillic to Latin
     • Latin to Cyrillic
   • Translation
     • Lemko to Polish
     • Polish to Lemko
2. Neural Machine Translation (Artificial Intelligence)
   • Lemko to English translation
   • English to Lemko translation

User interface

• Web app
• Smartphone app
  • Event app
  • Google Tutorial
  • Lemko conversation youtube livestreams
• On-screen Cyrillic keyboard module
• Refresh button module

Resources

• Lemko-English translation corpus
• Google Developers Machine learning youtubes
• OpenNMT open-source neural machine translation system.
• ACL’16 NMT tutorial produced by researchers at Stanford and NYU
• TensorFlow Neural Machine Translation (seq2seq) Tutorial

Networking sites
• Github
• LinkedIn
• Upwork
• Linguist social networks

Ideas

• Build apps for people
  • localize apps for other people (service)
• TEFL app?