Datasets:

---

legesher

/

language-decoded-community

like 0

Follow

Legesher 6

Brings this dataset card into alignment with the canonical project source-of-truth at legesher/language-decoded-experiments (HF PR #43, merged 2026-05-25).

Key changes

• Cond-4 renamed to "Community-Contributed Native Code" (was "Strictly Native Code"). Design goal clarified: code whose problem-solving logic is itself native — written as if a native speaker were approaching the problem, not English code that was later translated. Pending sufficient direct community contributions to the legesher/legesher-native-code HF Space.
• Cond-3 vs cond-4 distinction surfaced: cond-3 ("Mixed Native Sources") uses code pulled from real-world public-source repositories — incidentally available code that humans wrote in or with the target language. Cond-4 would gather code contributors specifically chose to share as representative of native-language programming. Cond-5's fully-translated data served as Phase 3's practical proxy for cond-4 because gathering native-authored code at scale proved difficult.
• Citation title updated to "Language Decoded: Exploring the Impact of Native Code on Multilingual Models".
• Related-datasets section updated to reflect the canonical project structure (experiments dataset as source-of-truth; cond-5 origin via c4ai-aya-expanse-32b via Cohere API; cond-4 contribution pipeline via the HF Space).
• Limitations added: clarified that non-Python files in this dataset are intentional for cond-3 (unlike cond-1/2/5 which are Python-only), and noted that this corpus is closer to native-authored than transpiled but does not enforce native-thinking authorship — cond-4 will eventually distinguish that more cleanly.
• Title callout added at the top with the proposed paper title.

No JSON, parquet, or schema files touched — README-only change. YAML frontmatter preserved verbatim.