Star Japanese — localized course data The English, Korean and Vietnamese glosses and example translations are adaptations of the Japanese and Chinese course materials. They combine dictionary matches, linked parallel examples, offline machine translation and editorial corrections. The translations have not all been individually reviewed. The vocabulary dataset, including these localized editions, is available under Creative Commons Attribution-ShareAlike 4.0 International. https://creativecommons.org/licenses/by-sa/4.0/ Dictionary sources * JMdict, Electronic Dictionary Research and Development Group (EDRDG), CC BY-SA 4.0. English glosses matched to the Japanese headword and reading. Individual entry numbers appear in course-gloss-sources.tsv. https://www.edrdg.org/jmdict/j_jmdict.html https://www.edrdg.org/edrdg/licence.html * CC-CEDICT, Paul Andrew Denisowski and CC-CEDICT contributors, distributed by MDBG, CC BY-SA 4.0. Selected entries supply English equivalents of Chinese glosses. https://www.mdbg.net/chinese/dictionary?page=cc-cedict * Korean and Vietnamese Wiktionary, their respective contributors, CC BY-SA 4.0. Selected entries supply target-language equivalents, with the relevant article pages recorded in course-gloss-sources.tsv. Consult the history of each linked page for its authors. https://ko.wiktionary.org/ https://vi.wiktionary.org/ Data extracted with Wiktextract (Tatu Ylönen, 2022) via Kaikki. https://kaikki.org/kowiktionary/rawdata.html https://kaikki.org/viwiktionary/rawdata.html Examples * Where the Japanese sentence exactly matched an entry with a direct language link, translations were selected from Tatoeba. course-example-translations.tsv identifies the original Japanese sentence, the linked translation and its contributor for each selected pair. Individual Tatoeba texts retain their original licenses, CC BY 2.0 FR or CC0 as applicable. https://tatoeba.org/en/downloads https://tatoeba.org/en/terms_of_use https://creativecommons.org/licenses/by/2.0/fr/ Offline processing * Argos Translate (Argos Open Technologies) and its published language packages. The Chinese–English and English–Vietnamese models derive from OPUS-MT, Jörg Tiedemann and Santhosh Thottingal, 2020, CC BY 4.0. The English–Korean package credits OPUS, Wiktionary/Wiktextract, CCAligned, WikiMatrix and OpenSubtitles2016. https://www.argosopentech.com/argospm/index/ https://aclanthology.org/2020.eamt-1.61/ https://creativecommons.org/licenses/by/4.0/ * Selected terms were also checked using the on-device Google Chrome Translator API. Translation models were used in preparing these editions and are not included in the website. https://developer.chrome.com/docs/ai/translator-api Additional attributions associated with the source vocabulary are in openjlpt-notice.txt and on the course's Sources page. * Some low-confidence preliminary outputs were also compared with Facebook AI's M2M-100 418M multilingual translation model (Fan et al., 2020; MIT). This model is not distributed with the site. https://huggingface.co/facebook/m2m100_418M