Human Languages in Source Code: Auto-Translation for Localized Instruction

on YouTube ↗12m

" summary

A tool called CodeInternational automatically translates identifiers and comments while preserving Java keywords by splitting identifiers with lightweight AI, handling transliteration for Arabic and left/right script issues, and supporting Java Python and HTML. A GitHub study of 2.9 million Java repos shows ~88% English identifiers and zero right‑to‑left identifiers, informing heuristics. The system translates a Karel reader into 100 spoken languages, compiles and runs, and sees over 10,000 classroom uses.

AI-written summary · about

" chapters (15)

1. 0:00Intro
2. 0:37Empathy: What does this code say?
3. 1:02Most of the world does not speak English
4. 3:11Some Vocabulary
5. 4:18How did we analyze human language in code?
6. 4:35Script
7. 5:12Language Patterns on GitHub
8. 7:14Translating Identifiers
9. 7:33Watch out for short identifiers
10. 7:47Method and Functions have a verb primer
11. 7:57Transliterate right to left languages
12. 8:20Code International
13. 9:01Translating a textbook
14. 9:36Used by people around the world
15. 10:13What about other languages?