# Examples 0.3.8

Japanese corpus sentences: Tatoeba contributors, https://tatoeba.org/ . Per-sentence IDs link to attribution. Sentences: CC BY 2.0 FR, https://creativecommons.org/licenses/by/2.0/fr/ . Korean adaptations are marked draft or assistant-edited. Original Japanese is retained where a sentence has been adapted.

Vocabulary linkage: EDRDG JMdict example edition, https://www.edrdg.org/pub/Nihongo/ ; CC BY-SA 4.0, https://creativecommons.org/licenses/by-sa/4.0/ . Input archive SHA256: 18a075c6312692b1beb2cf2a0ac5f377874e7d3df5e7412eb7a83117603344cb.

Project-authored sentences: release/examples038/*.tsv. These are assistant-authored, not independently native-speaker-approved. Original vocabulary IDs and rare readings are preserved.

Build-only models: original OPUS-MT eng-kor Marian, opusTCv20210807-sepvoc_transformer-big_2022-07-28; and facebook/m2m100_1.2B revision 11301d1d63d756517521fc0fd34d81c5bff0d946 (MIT). Actual model and edit provenance are recorded per sentence. Direct JA-KO processing timed out after 4896 records; the complete OPUS corpus supplies the remainder. No model weights or user records are bundled/transmitted. Machine translations may be wrong and remain behind a closed, explicitly labeled disclosure.

Readings are generated with pykakasi 2.3.0 unless otherwise marked; context-sensitive readings may need correction. Full word-ID coverage is NOT proof of semantic correctness.