WordTranslation — Offline Word Translation Packs (vocabradar/wordfreq vocabulary) / 离线单词翻译包（vocabradar/wordfreq 词表）

What it is

Offline word-translation packs for 43 languages around English and Chinese: every word of a language's cleaned vocabulary, paired with its translations. The vocabulary comes straight from the vocabradar/wordfreq dataset — that dataset is this one's direct source: the words translated here are its small_<code> lists, in the order they are stored there. 166 gzipped JSONL files, one per language pair (83.5 MiB in total, 0.09–1.07 MiB each); en ⇄ X and zh ⇄ X share the same 41 partner languages (40 + Thai).

What it is for

• An offline lexicon. Unzip one file and the whole pair is in memory — no network, no API key, no per-lookup cost.
• Word hints in your own app — the same packs are what the VocabRadar browser extension shows while you read.
• Frequency-ordered work. Lines follow the rank order of the vocabradar/wordfreq list (most frequent word first), so a pack also works as an MT / dev lexicon for a low-resource pair.
• Gap checking. The hit-rate tables below give, per language, how much of that vocabulary actually has a translation — enough to judge whether a pair suits your use.

How it is built

Words: the small_<code> list of the source language, taken as-is from the vocabradar/wordfreq dataset, its frequency order preserved.

Translations: professional bilingual dictionaries come first wherever a license-compatible edition exists — ECDICT (en→zh, MIT), CC-CEDICT (zh→en, CC BY-SA 4.0), JMdict (ja↔en, CC BY-SA 4.0), HanDeDict (zh↔de, CC BY-SA 2.0), CFDICT (zh↔fr, CC BY-SA 3.0) and Folkets lexikon (sv↔en, CC BY-SA 2.5). One-way dictionaries are inverted gloss→headword for their reverse direction, and only target words that occur in the vocabulary are kept. The rest is base data — human-curated dictionaries only: English / Chinese Wiktionary via wiktextract / kaikki.org (CC BY-SA 4.0) + PanLex (CC0) where available. Remaining vocabulary gaps are filled by the local model Hunyuan-MT2-1.8B Q4_K_M (Apache-2.0, official GGUF served by llama.cpp); languages it does not support are filled by a second local model layer, LMT-60-1.7B Q4_K_M (Apache-2.0, NiuTrans) — also machine-generated. Reverse directions obtained by gloss inversion are less reliable for closed-class words (articles, copulas, particles): German der can surface a content-word gloss, and French est can be read as “east”. HanDeDict/CFDICT entries for grammatical particles are usage descriptions rather than single-word equivalents; both cases are kept as-is and marked here.

Quality control. Every pack passes a content-level QC sweep before release (tools/qc_check.py): runaway model output (repeated-character runs, e.g. a French aïe once glossed as 呜呜呜呜…), placeholder/model-leak strings, mojibake and script violations are removed; stray leading combining marks are repaired; emptied words are re-translated by the local model cascade and re-checked. Rules and the 2026-10 cleanup record live in docs/.

All directions live in one directory: one gzip JSONL file per language pair, data/<src>_<dst>.jsonl.gz (the en ⇄ zh pair is stored once). Every line is one JSON object with a fixed pair of fields — a source word and its translations.

中文版请见下方：中文版说明

Files

├── README.md
├── meta.json                   # per-direction source, cap and language stats
├── files.json                  # per-file size (bytes) and SHA-256
└── data/                       # 166 files: <src>_<dst>.jsonl.gz

Data format

{"word": "house", "translations": ["casa", "hogar"]}

• Field names are constant on every line (word, translations), so the Hub's Dataset Viewer renders each file as a two-column table (word: string, translations: list<string>) instead of one column per source word.
• data/<src>_<dst>.jsonl.gz translates the src word list (the small_<src> list of vocabradar/wordfreq) into dst; lines are ordered by that list's rank — deterministic rebuilds.
• Translations are deduplicated case-insensitively and capped at 10 per word; gap-fill entries are capped at 3 and pass the same per-language Unicode script whitelist as every other layer.
• The gap-fill layers are machine-generated and may contain errors.

Languages & hits

hits is the share of that language's list with at least one translation — not the cumulative coverage column of vocabradar/wordfreq, which measures token coverage of a corpus; a hit here only means "this vocabulary word has a translation in this file". native name is the language's own name for itself — the same autonym the vocabradar/wordfreq card lists. first line is the first line of that row's own file, verbatim — the highest-frequency source word (from the vocabradar/wordfreq list this dataset is keyed to) together with the translations found for it, so one glance checks both ends: the vocabulary (is that word really first?) and the target language (do the translations look like it?). The total row adds up words translated and entries; its hits is the weighted aggregate (Σ translated / Σ vocabulary), since each reverse-direction language has its own denominator.

en → X (EnToX)

English Wiktionary source; vocabulary = small_en of vocabradar/wordfreq — 41 languages

code	language	native name	first line	words translated	hits	entries
ar	Arabic	العربية	{"word":"the","translations":["الـ","ال"]}	28,223	100.0%	111,486
bg	Bulgarian	български	{"word":"the","translations":["-та","-то","-те","по"]}	28,220	100.0%	74,713
bn	Bengali	বাংলা	{"word":"the","translations":["-রা","-এরা","-গুলা","-গুলো","-টা","-টি","গাছা","গাছি","মহাবিশ্ব"]}	28,223	100.0%	67,972
ca	Catalan	català	{"word":"the","translations":["el","la","l' els m pl les","es","sa","ets","sos","ses","els","les"]}	28,223	100.0%	84,343
cs	Czech	čeština	{"word":"the","translations":["ta","ten","to","určitý člen","člen určitý"]}	28,223	100.0%	106,829
da	Danish	dansk	{"word":"the","translations":["de","den","det","en","et","-et","-en","-ne"]}	28,223	100.0%	63,132
de	German	Deutsch	{"word":"the","translations":["der","die","das","de","dr","den","dieser","umso","Platz","der die das die"]}	28,223	100.0%	146,827
el	Greek	Ελληνικά	{"word":"the","translations":["ο","ὁ","στον","στη","στο","άρθρο","η","οι","παράμεσος","στα"]}	28,179	99.8%	90,613
es	Spanish	español	{"word":"the","translations":["el","la","los","las","lo","el la los las","entonces","pero","pues"]}	28,223	100.0%	126,135
fa	Persian	فارسی	{"word":"the","translations":["این","آن","ال","باسلیقون","ثلاثة غساله","دلیجه","ساز","سفیدرود","سنده","قبلة"]}	28,223	100.0%	91,682
fi	Finnish	suomi	{"word":"the","translations":["se","-ni","-si","-nsa","-nsä","-mme","-nne","minun","sinun","hänen"]}	28,223	100.0%	128,415
fil	Filipino	Filipino	{"word":"the","translations":["ang","nang","mga"]}	28,223	100.0%	48,367
fr	French	français	{"word":"the","translations":["le","la","les","l'","comp.","colloquial or regional usage","acquitter","au","aux","ce"]}	28,223	100.0%	132,443
he	Hebrew	עברית	{"word":"the","translations":["ה־","ה"]}	28,222	100.0%	50,356
hi	Hindi	हिन्दी	{"word":"the","translations":["जिस काम के पूरे करने में सभी लगे हों","ईसामसीह","दि","यह","वह","वही"]}	28,223	100.0%	106,614
hu	Hungarian	magyar	{"word":"the","translations":["a","az"]}	28,223	100.0%	104,597
id	Indonesian	Bahasa Indonesia	{"word":"the","translations":["si","sang","itu","al","-nya","kian"]}	28,223	100.0%	93,819
is	Icelandic	íslenska	{"word":"the","translations":["-inn","hinn","hið","-in","-nar","-ið","-nir"]}	28,223	100.0%	57,806
it	Italian	italiano	{"word":"the","translations":["il","lo","la","l'","i","gli","le","li","lì","gl’"]}	28,223	100.0%	120,328
ja	Japanese	日本語	{"word":"the","translations":["その","正に","正しく","ザ","這","あの","其の","定冠詞","当該","死因"]}	28,212	100.0%	145,314
ko	Korean	한국어	{"word":"the","translations":["-을수록","그","그만큼$2$","저","하면 할수록","하면 할수록 그만큼 더"]}	28,217	100.0%	104,823
lt	Lithuanian	lietuvių	{"word":"the","translations":["tai"]}	28,223	100.0%	42,549
lv	Latvian	latviešu	{"word":"the","translations":["sirdi"]}	28,211	100.0%	38,168
mk	Macedonian	македонски	{"word":"the","translations":["-ов","-те","-от","многуте","највеќе","најголем дел","најмногу","онака","постојано","-во"]}	28,184	99.9%	47,766
ms	Malay	Bahasa Melayu	{"word":"the","translations":["sang","-nya","itu","ini","si","semakin..., semakin...","lagi..., lagi..."]}	28,223	100.0%	85,179
nb	Norwegian Bokmål	norsk bokmål	{"word":"the","translations":["det","den","-et","-en","-a","-i","de","desto","jo","-one"]}	28,223	100.0%	68,517
nl	Dutch	Nederlands	{"word":"the","translations":["de","het","aan de","aan het","des te","die","hoe","naar de","naar het","’t"]}	28,223	100.0%	107,151
pl	Polish	polski	{"word":"the","translations":["ten","czym","im","przedimek określony","ta","to","te"]}	28,223	100.0%	95,604
pt	Portuguese	português	{"word":"the","translations":["o","a","os","as","o a os as"]}	28,223	100.0%	107,945
ro	Romanian	română	{"word":"the","translations":["-ul","-a","cel","cea","-i","cele","-le"]}	28,223	100.0%	61,507
ru	Russian	русский	{"word":"the","translations":["на ходу","определенный артикль","определённый артикль","подходящий","такой","тем","тот","чем","чем... тем"]}	28,219	100.0%	148,990
sh	Serbo-Croatian	srpskohrvatski / српскохрватски	{"word":"the","translations":["Alpe","Antarktika","Atlantik","Balkan","Baltik","Drava","Drina","Dunav","Holandija","Jadran"]}	28,209	100.0%	118,146
sk	Slovak	slovenčina	{"word":"the","translations":["ten","to","tá"]}	28,223	100.0%	98,458
sl	Slovenian	slovenščina	{"word":"the","translations":["the"]}	28,223	100.0%	76,078
sv	Swedish	svenska	{"word":"the","translations":["den","dess","desto","-en","de","det","ju","vederbörande","vintergatan","överklassen"]}	28,223	100.0%	100,554
ta	Tamil	தமிழ்	{"word":"the","translations":["அது"]}	28,223	100.0%	72,940
th	Thai	ไทย	{"word":"the","translations":["นั้น"]}	28,188	99.9%	28,312
tr	Turkish	Türkçe	{"word":"the","translations":["bir","harfi tarif","ne kadar","üstlenmek"]}	28,223	100.0%	99,531
uk	Ukrainian	українська	{"word":"the","translations":["те"]}	28,221	100.0%	71,637
ur	Urdu	اردو	{"word":"the","translations":["ال"]}	28,223	100.0%	60,976
vi	Vietnamese	Tiếng Việt	{"word":"the","translations":["con","càng","cái","duy nhất","này","ấy"]}	28,223	100.0%	95,230
total	41 languages	—	—	1,156,972	100.0%	3,681,852

X → en (XToEn)

English Wiktionary source, inverted; vocabulary = small_<code> of vocabradar/wordfreq — 41 languages

code	language	native name	first line	words translated	hits	entries
ar	Arabic	العربية	{"word":"في","translations":["a","in","on","at","by","into","through","during","times","within"]}	55,606	100.0%	101,618
bg	Bulgarian	български	{"word":"на","translations":["of","in","on","at","onto","lo"]}	35,848	100.0%	55,492
bn	Bengali	বাংলা	{"word":"না","translations":["not","no","don't","never","without","please","non","boat","un","na"]}	33,045	99.9%	50,042
ca	Catalan	català	{"word":"de","translations":["of","in","with","from","about","more","off","again","coral","dee"]}	27,028	100.0%	60,056
cs	Czech	čeština	{"word":"a","translations":["and","a","that","who","which","any","both","recent","double","plus"]}	51,744	100.0%	92,063
da	Danish	dansk	{"word":"i","translations":["to","of","a","in","for","you","on","with","at","by"]}	29,321	100.0%	50,479
de	German	Deutsch	{"word":"die","translations":["the","a","that","this","but","an","all","what","who","which"]}	38,989	100.0%	114,757
el	Greek	Ελληνικά	{"word":"και","translations":["and","that","but","which","also","too","both"]}	43,188	100.0%	61,778
es	Spanish	español	{"word":"de","translations":["of","in","for","on","with","be","as","at","by","from"]}	34,740	100.0%	98,038
fa	Persian	فارسی	{"word":"و","translations":["and","both"]}	30,108	100.0%	57,014
fi	Finnish	suomi	{"word":"ja","translations":["and","so","also","too","both","plus","likewise","wedge"]}	59,143	100.0%	107,690
fil	Filipino	Filipino	{"word":"sa","translations":["to","of","a","in","for","on","with","at","by","from"]}	30,098	100.0%	40,648
fr	French	français	{"word":"de","translations":["to","of","a","in","for","on","with","be","as","at"]}	31,203	100.0%	100,218
he	Hebrew	עברית	{"word":"את","translations":["you","with","me","alongside","amongst","thou","shovel","spade"]}	57,376	99.9%	68,773
hi	Hindi	हिन्दी	{"word":"के","translations":["of","in","for","with","as","have","by","from","his","about"]}	24,031	100.0%	56,346
hu	Hungarian	magyar	{"word":"a","translations":["the","a","black","amps"]}	46,525	100.0%	81,700
id	Indonesian	Bahasa Indonesia	{"word":"yang","translations":["that","who","which","second","third","whose","rel"]}	30,967	100.0%	77,623
is	Icelandic	íslenska	{"word":"og","translations":["and","but","which","also","both","plus"]}	42,207	100.0%	56,957
it	Italian	italiano	{"word":"di","translations":["to","of","in","for","that","on","with","be","at","by"]}	35,954	100.0%	96,114
ja	Japanese	日本語	{"word":"の","translations":["indicates possessive","nominalizes verbs and adjectives","substitutes for \"ga\" in subordinate phrases","indicates a confident conclusion","indicates emotional emphasis","indicates question","field","plain","hidden interior part","wild"]}	27,856	100.0%	138,071
ko	Korean	한국어	{"word":"이","translations":["and","that","you","this","be","he","but","she","only","two"]}	26,807	100.0%	48,935
lt	Lithuanian	lietuvių	{"word":"ir","translations":["and","even","afterward"]}	63,973	100.0%	71,616
lv	Latvian	latviešu	{"word":"un","translations":["and"]}	45,464	100.0%	51,239
mk	Macedonian	македонски	{"word":"на","translations":["of","for","on","at","per"]}	32,925	100.0%	42,850
ms	Malay	Bahasa Melayu	{"word":"yang","translations":["that","one","which","change","cash","coins","umbrella"]}	28,571	100.0%	69,966
nb	Norwegian Bokmål	norsk bokmål	{"word":"i","translations":["to","of","a","in","for","you","on","with","at","by"]}	25,994	100.0%	48,694
nl	Dutch	Nederlands	{"word":"de","translations":["the","you","this","one","able","bus","exposure","repaid","earthy"]}	28,715	100.0%	78,061
pl	Polish	polski	{"word":"w","translations":["to","of","a","in","for","on","with","at","by","an"]}	49,664	100.0%	81,502
pt	Portuguese	português	{"word":"de","translations":["of","for","with","by","from","about","something","old","someone","hunger"]}	33,123	100.0%	82,780
ro	Romanian	română	{"word":"de","translations":["of","for","on","by","from","since","against"]}	43,279	100.0%	62,582
ru	Russian	русский	{"word":"в","translations":["to","of","a","in","for","on","with","as","at","by"]}	59,604	100.0%	122,769
sh	Serbo-Croatian	srpskohrvatski / српскохрватски	{"word":"je","translations":["is","are","if","has","her","did","sued"]}	54,698	100.0%	97,376
sk	Slovak	slovenčina	{"word":"a","translations":["and","amps"]}	59,414	100.0%	101,829
sl	Slovenian	slovenščina	{"word":"je","translations":["is"]}	53,882	100.0%	77,313
sv	Swedish	svenska	{"word":"är","translations":["is","are","am","result","arise"]}	30,515	100.0%	66,415
ta	Tamil	தமிழ்	{"word":"ஒரு","translations":["a","an","hour","sole","sheep","mono"]}	64,751	100.0%	83,531
th	Thai	ไทย	{"word":"ที่","translations":["place","that","which","who"]}	39,499	99.5%	39,856
tr	Turkish	Türkçe	{"word":"ve","translations":["and","so","both","species","vee"]}	63,130	100.0%	95,737
uk	Ukrainian	українська	{"word":"в","translations":["to","in","have","at","into","per"]}	46,285	100.0%	62,505
ur	Urdu	اردو	{"word":"کے","translations":["of","at"]}	22,442	100.0%	37,850
vi	Vietnamese	Tiếng Việt	{"word":"là","translations":["is","that","was","be","as","are","were","been","how","some"]}	10,519	100.0%	26,597
zh	Chinese	中文	{"word":"的","translations":["of","s","used after an attribute when it modifies a noun","used at the end of a declarative sentence for emphasis","also pr. [di4] or [di5] in poetry and songs","a taxi","a cab","really and truly","bull's-eye","target"]}	35,723	100.0%	145,872
total	42 languages	—	—	1,683,954	100.0%	3,161,352

zh → X (ZhToX)

Chinese Wiktionary source; vocabulary = small_zh of vocabradar/wordfreq — 41 languages

code	language	native name	first line	words translated	hits	entries
ar	Arabic	العربية	{"word":"的","translations":["ال"]}	35,723	100.0%	37,413
bg	Bulgarian	български	{"word":"的","translations":["ето"]}	35,702	99.9%	36,852
bn	Bengali	বাংলা	{"word":"的","translations":["ের"]}	35,723	100.0%	36,298
ca	Catalan	català	{"word":"的","translations":["de"]}	35,721	100.0%	36,571
cs	Czech	čeština	{"word":"的","translations":["Jeho"]}	35,723	100.0%	36,903
da	Danish	dansk	{"word":"的","translations":["Dan"]}	35,722	100.0%	36,601
de	German	Deutsch	{"word":"的","translations":["wird nach einem Adjektiv oder einer Attributphrase verwendet","wird nach einem Verb verwendet, um das Subjekt, die Zeit","Ziel, das Schwarze","klar"]}	35,721	100.0%	60,922
el	Greek	Ελληνικά	{"word":"的","translations":["του"]}	35,229	98.6%	36,199
es	Spanish	español	{"word":"的","translations":["de"]}	35,723	100.0%	38,381
fa	Persian	فارسی	{"word":"的","translations":["که"]}	35,721	100.0%	37,234
fi	Finnish	suomi	{"word":"的","translations":["Finnish"]}	35,704	99.9%	37,635
fil	Filipino	Filipino	{"word":"的","translations":["ay"]}	35,722	100.0%	36,531
fr	French	français	{"word":"的","translations":["de","utilisé après un adjectif, un nom","centre de la cible","but","objectif","voir","vrai"]}	35,723	100.0%	67,607
he	Hebrew	עברית	{"word":"的","translations":["של"]}	35,714	100.0%	36,359
hi	Hindi	हिन्दी	{"word":"的","translations":["का"]}	35,722	100.0%	37,985
hu	Hungarian	magyar	{"word":"的","translations":["És"]}	35,720	100.0%	37,238
id	Indonesian	Bahasa Indonesia	{"word":"的","translations":["adalah"]}	35,723	100.0%	36,491
is	Icelandic	íslenska	{"word":"的","translations":["Er"]}	35,723	100.0%	36,535
it	Italian	italiano	{"word":"的","translations":["di"]}	35,723	100.0%	37,266
ja	Japanese	日本語	{"word":"的","translations":["の"]}	35,706	100.0%	38,413
ko	Korean	한국어	{"word":"的","translations":["-의"]}	35,706	100.0%	36,859
lt	Lithuanian	lietuvių	{"word":"的","translations":["ž"]}	35,721	100.0%	36,204
lv	Latvian	latviešu	{"word":"的","translations":["Latv"]}	35,663	99.8%	36,253
mk	Macedonian	македонски	{"word":"的","translations":["Македонски"]}	35,587	99.6%	36,528
ms	Malay	Bahasa Melayu	{"word":"的","translations":["nya"]}	35,723	100.0%	36,659
nb	Norwegian Bokmål	norsk bokmål	{"word":"的","translations":["Det"]}	35,721	100.0%	36,475
nl	Dutch	Nederlands	{"word":"的","translations":["van"]}	35,723	100.0%	37,389
pl	Polish	polski	{"word":"的","translations":["tak"]}	35,723	100.0%	37,637
pt	Portuguese	português	{"word":"的","translations":["do"]}	35,719	100.0%	37,492
ro	Romanian	română	{"word":"的","translations":["de”"]}	35,719	100.0%	36,937
ru	Russian	русский	{"word":"的","translations":["дом"]}	35,694	99.9%	63,021
sh	Serbo-Croatian	srpskohrvatski / српскохрватски	{"word":"的","translations":["Srbijanski"]}	35,494	99.4%	37,198
sk	Slovak	slovenčina	{"word":"的","translations":["So"]}	35,723	100.0%	36,532
sl	Slovenian	slovenščina	{"word":"的","translations":["Slovenski"]}	35,721	100.0%	36,300
sv	Swedish	svenska	{"word":"的","translations":["av"]}	35,723	100.0%	37,092
ta	Tamil	தமிழ்	{"word":"的","translations":["அதன்"]}	35,723	100.0%	36,048
th	Thai	ไทย	{"word":"的","translations":["ของ"]}	34,357	96.2%	34,368
tr	Turkish	Türkçe	{"word":"的","translations":["oldu"]}	35,723	100.0%	37,256
uk	Ukrainian	українська	{"word":"的","translations":["Українською"]}	35,699	99.9%	37,623
ur	Urdu	اردو	{"word":"的","translations":["کی"]}	35,723	100.0%	36,683
vi	Vietnamese	Tiếng Việt	{"word":"的","translations":["của"]}	35,722	100.0%	37,470
total	41 languages	—	—	1,462,195	99.8%	1,593,458

X → zh (XToZh)

Chinese Wiktionary source, inverted; vocabulary = small_<code> of vocabradar/wordfreq — 41 languages

code	language	native name	first line	words translated	hits	entries
ar	Arabic	العربية	{"word":"في","translations":["在"]}	55,553	99.9%	55,814
bg	Bulgarian	български	{"word":"на","translations":["在"]}	35,768	99.8%	35,904
bn	Bengali	বাংলা	{"word":"না","translations":["从不"]}	32,791	99.2%	32,874
ca	Catalan	català	{"word":"de","translations":["德"]}	27,006	99.9%	27,148
cs	Czech	čeština	{"word":"a","translations":["a"]}	51,708	99.9%	51,825
da	Danish	dansk	{"word":"i","translations":["i"]}	29,310	100.0%	29,437
de	German	Deutsch	{"word":"die","translations":["是","这些"]}	38,975	100.0%	61,494
el	Greek	Ελληνικά	{"word":"και","translations":["和"]}	42,952	99.4%	43,061
en	English	English	{"word":"the","translations":["那"]}	28,219	100.0%	84,155
es	Spanish	español	{"word":"de","translations":["德"]}	34,731	100.0%	35,055
fa	Persian	فارسی	{"word":"و","translations":["和"]}	30,070	99.9%	30,277
fi	Finnish	suomi	{"word":"ja","translations":["是"]}	59,030	99.8%	59,234
fil	Filipino	Filipino	{"word":"sa","translations":["萨"]}	30,093	100.0%	30,167
fr	French	français	{"word":"de","translations":["的","与","于","从","由","从中","乎","为了"]}	31,196	100.0%	50,536
he	Hebrew	עברית	{"word":"את","translations":["你"]}	57,300	99.8%	57,431
hi	Hindi	हिन्दी	{"word":"के","translations":["科"]}	23,800	99.0%	24,036
hu	Hungarian	magyar	{"word":"a","translations":["a"]}	46,519	100.0%	46,669
id	Indonesian	Bahasa Indonesia	{"word":"yang","translations":["杨"]}	30,957	100.0%	31,060
is	Icelandic	íslenska	{"word":"og","translations":["og"]}	42,189	100.0%	42,280
it	Italian	italiano	{"word":"di","translations":["迪"]}	35,947	100.0%	36,154
ja	Japanese	日本語	{"word":"の","translations":["的"]}	27,822	99.9%	28,038
ko	Korean	한국어	{"word":"이","translations":["牙齿","二"]}	26,758	99.8%	26,847
lt	Lithuanian	lietuvių	{"word":"ir","translations":["ir"]}	63,905	99.9%	63,978
lv	Latvian	latviešu	{"word":"un","translations":["不"]}	45,397	99.9%	45,471
mk	Macedonian	македонски	{"word":"на","translations":["在"]}	32,840	99.7%	32,954
ms	Malay	Bahasa Melayu	{"word":"yang","translations":["杨"]}	28,562	100.0%	28,673
nb	Norwegian Bokmål	norsk bokmål	{"word":"i","translations":["你"]}	25,985	100.0%	26,067
nl	Dutch	Nederlands	{"word":"de","translations":["de"]}	28,708	100.0%	28,879
pl	Polish	polski	{"word":"w","translations":["w"]}	49,651	100.0%	49,832
pt	Portuguese	português	{"word":"de","translations":["de"]}	33,117	100.0%	33,370
ro	Romanian	română	{"word":"de","translations":["德"]}	43,271	100.0%	43,415
ru	Russian	русский	{"word":"в","translations":["就是","同时","如何","其中","影响","目前","参加","具有","实现","是不是"]}	59,388	99.6%	72,489
sh	Serbo-Croatian	srpskohrvatski / српскохрватски	{"word":"je","translations":["你"]}	54,687	100.0%	54,724
sk	Slovak	slovenčina	{"word":"a","translations":["a"]}	59,390	100.0%	59,484
sl	Slovenian	slovenščina	{"word":"je","translations":["je"]}	53,870	100.0%	53,934
sv	Swedish	svenska	{"word":"är","translations":["是"]}	30,507	100.0%	30,680
ta	Tamil	தமிழ்	{"word":"ஒரு","translations":["一个"]}	64,563	99.7%	64,609
th	Thai	ไทย	{"word":"ที่","translations":["地点","地點"]}	34,098	85.9%	34,634
tr	Turkish	Türkçe	{"word":"ve","translations":["ve"]}	63,011	99.8%	63,200
uk	Ukrainian	українська	{"word":"в","translations":["在"]}	46,084	99.6%	46,236
ur	Urdu	اردو	{"word":"کے","translations":["的"]}	22,415	99.9%	22,547
vi	Vietnamese	Tiếng Việt	{"word":"là","translations":["是"]}	10,510	99.9%	10,589
total	42 languages	—	—	1,668,653	99.5%	1,785,261

Regenerating

python tools/build_pro.py                       # rebuild professional-dictionary layer
python tools/wt.py probe --pair en              # vocabulary hit / gap report
python tools/wt.py fill  --pair en --pass p2    # local Hy-MT2 gap fill
python tools/wt.py pack  --pair en              # re-pack data/ + meta/files/README

Toolchain

WordTranslation/
├── README.md / meta.json / files.json   # dataset card + sidecars (published)
├── data/                                # the only published artifact: <src>_<dst>.jsonl.gz
├── build/                               # build inputs (not published)
│   ├── raw/                             #   source dumps (wiktextract en/zh)
│   ├── pro/                             #   professional dictionary layer (<src>_<dst>.jsonl)
│   └── <pair>/{cache,llm,manual,...}    #   per-pair intermediates
├── tools/wt.py                          # probe | fill | pack, single file
├── docs/                                # plans and research
└── work/<pair>/{hymt,lmt}/<pass>/       # model gap-fill layers (pack input)

subcommand	effect
probe --pair en|zh	vocabulary hits and gaps, both directions
fill --pair en --pass p1|p2 [--engine hymt|lmt] [--langs a,b] [--conc 8] [--limit N]	fill the gaps word by word → work/<pair>/<engine>/<pass>/<code>.jsonl (append-only, resumable, Ctrl-C safe)
pack --pair en [--pass both] [--base clean|merged] [--no-pro]	pro + base + hymt/lmt layers → re-pack data/, rebuild meta.json / files.json / README.md

• --base clean (default) = professional dictionaries (pro_) + Wiktionary (pairs_) + PanLex (panlex_), human-curated only; the reverse direction is its inversion. merged = the old four-layer merge (deprecated, kept for comparison).
• The en ⇄ zh pair is produced twice with different content; the flat layout keeps one of each: en_zh from the zh p2 pipeline (XToZh) and zh_en from the en p2 pipeline (XToEn), matching the "definition-language family first" rule of the site consumer VocabRadar/web/scripts/init_web.js. Each of the two p1 passes skips that code (wt.CROSS_SKIP) so packing cannot overwrite the other.
• The model is served by llama.cpp's llama-server; wt.py starts, reuses and shuts it down on its own.
• Environment variables: WT_LLAMA_SERVER, WT_HYMT_GGUF, WT_LMT_GGUF, WT_PORT (default 8791), WT_WF_DATA (the vocabradar/wordfreq vocabulary directory).

License

• Professional dictionaries: ECDICT (MIT), CC-CEDICT (CC BY-SA 4.0), JMdict (CC BY-SA 4.0), HanDeDict (CC BY-SA 2.0), CFDICT (CC BY-SA 3.0), Folkets lexikon (CC BY-SA 2.5) — all compatible with this dataset's CC BY-SA 4.0 licence; attribution is given here and in meta.json.
• wiktextract/kaikki.org extraction: CC BY-SA 4.0 (Wiktionary, dual-licensed GFDL).
• PanLex supplement: CC0 (2024-03-01 dump).
• vocabulary, taken from the vocabradar/wordfreq dataset: MIT.
• Hunyuan-MT2-1.8B / LMT-60-1.7B gap-fill layers: machine-generated, disclosed here and in meta.json; both models Apache-2.0.

About VocabRadar

This dataset is a VocabRadar project — the vocabulary radar. The VocabRadar browser extension marks up the words you do not know yet while you watch videos or browse the web: definitions in place, translation, bilingual subtitles and annotated subtitles. Whatever you looked up can go on to vocabradar.com, which turns it into related exercises and games so the words actually stick.

---

WordTranslation —— 离线单词翻译包（vocabradar/wordfreq 词表）

这是什么

43 种语言的离线单词翻译包，围绕英文与中文两翼：把某语言清洗后的词表里的每个词配上译文。 词表直接来自 vocabradar/wordfreq 数据集 —— 它就是本数据集的直接源头：这里翻译的就是它的 small_<code> 词表，按它在数据集里的 原顺序逐词来。共 166 个 gzip JSONL 文件，每对语言一个（合计 83.5 MiB， 单个 0.09–1.07 MiB）；en ⇄ X 与 zh ⇄ X 共用同一批 41 种伙伴语言（40 + 泰语）。

有什么用

• 离线词典：解压一个文件，整对语言就是内存里的字典 —— 不联网、不要 API key、没有按次计费。
• 给自己的应用做生词提示 —— VocabRadar 浏览器扩展 阅读时显示的释义与字幕就是这份数据。
• 按频序取词与实验：行序即 vocabradar/wordfreq 词表的排名（最高频的词在最前），也可当低资源语言对的翻译 / 词表实验数据。
• 查缺口：下面的命中率表按语言给出「多少词真的有译文」，据此判断某一对够不够用。

怎么构建的

词：源语言的 small_<code> 列表，原样取自 vocabradar/wordfreq 数据集，保留词频顺序。

译文：有许可兼容的专业双语词典时优先用专业词典 —— ECDICT（en→zh，MIT）、 CC-CEDICT（zh→en，CC BY-SA 4.0）、JMdict（ja↔en，CC BY-SA 4.0）、 HanDeDict（zh↔de，CC BY-SA 2.0）、CFDICT（zh↔fr，CC BY-SA 3.0）、 Folkets lexikon（sv↔en，CC BY-SA 2.5）。单向词典的反向按「释义→词条」反转， 且只保留目标语词表里真实存在的词。其余基础数据只取人工编撰的词典：英文 / 中文 Wiktionary（经 wiktextract / kaikki.org 抽取，CC BY-SA 4.0）+ 可得的 PanLex（CC0）。 剩余词表缺口由本地模型 Hunyuan-MT2-1.8B Q4_K_M 补译（Apache-2.0，llama.cpp 跑官方 GGUF）；它不支持的语言再由第二层本地模型 LMT-60-1.7B Q4_K_M 补 （Apache-2.0，NiuTrans）—— 同为机器生成。反查得到的译文对封闭类词 （冠词、系动词、助词）可靠性较低：德语 der 可能落到实词义，法语 est 可能被读成“东”。 HanDeDict/CFDICT 中语法助词的词条是用法说明而非单词对译；两类情况都按原样保留并在此注明。

发布前质检：每个包发布前过一遍内容级质检（tools/qc_check.py）——模型失控复读 （如法中 aïe 曾被译成 呜呜呜呜… 这类单字连跑）、占位语/模型泄漏、乱码与脚本违规 一律剔除；游离附加号开头的词典残留原位修复；被清空的词走本地模型级联补译并复检。 规则明细与 2026-10 清理记录见 docs/。

所有方向共用一个目录：每对语言一个 gzip JSONL 文件 data/<src>_<dst>.jsonl.gz （en ⇄ zh 只存一份）。每行一个 JSON 对象，字段固定 —— 源词 word 及其 translations。

文件结构

├── README.md
├── meta.json                   # 各方向的来源、上限与语言统计
├── files.json                  # 每个文件的大小（字节）与 SHA-256
└── data/                       # 166 个文件：<src>_<dst>.jsonl.gz

数据格式

{"word": "house", "translations": ["casa", "hogar"]}

• 每行字段名固定（word、translations），HF 的 Dataset Viewer 才能把每个文件渲染成 两列表格（word: string、translations: list<string>），而不是每个源词一列。
• data/<src>_<dst>.jsonl.gz 把 src 词表（vocabradar/wordfreq 的 small_<src> 列表）译成 dst；行序按该列表的排名，重建结果确定。
• 译文按大小写不敏感去重，每词上限 10 条；补译条目上限 3 条， 并与其他层一样过每语言的 Unicode script 白名单。
• 补译层是机器生成的，可能有错。

语言与命中率

命中率 是该语言词表里有至少一条译文的词所占比例 —— 不是 vocabradar/wordfreq 那张表的 cumulative coverage（那是语料 token 覆盖率）；这里的「命中」只表示 「该词表词在本文件里有译文」。自称 是该语言自己的叫法（与 vocabradar/wordfreq 卡的 native name 同源）。首行 是该行自己那个文件的第一行，原样照抄 —— 最高频的源词（即本数据集所依据的 vocabradar/wordfreq 词表里的第一个词） 加上给它找到的译文。一眼同时核对两头：词表（第一个词对不对）与译文语种（译文像不像该语言）。 末行 合计 的已译词数与条目数为直接相加；命中率为加权合计 （Σ 已译词数 / Σ 词表分母）—— 反向方向各语言分母不同，算术平均没有意义。

en → X (EnToX)

英文维基词典来源；词表 = vocabradar/wordfreq 的 small_en — 40 种语言

代码	语言	自称	首行	已译词数	命中率	条目数
ar	Arabic	العربية	{"word":"the","translations":["الـ","ال"]}	28,223	100.0%	111,486
bg	Bulgarian	български	{"word":"the","translations":["-та","-то","-те","по"]}	28,220	100.0%	74,713
bn	Bengali	বাংলা	{"word":"the","translations":["-রা","-এরা","-গুলা","-গুলো","-টা","-টি","গাছা","গাছি","মহাবিশ্ব"]}	28,223	100.0%	67,972
ca	Catalan	català	{"word":"the","translations":["el","la","l' els m pl les","es","sa","ets","sos","ses","els","les"]}	28,223	100.0%	84,343
cs	Czech	čeština	{"word":"the","translations":["ta","ten","to","určitý člen","člen určitý"]}	28,223	100.0%	106,829
da	Danish	dansk	{"word":"the","translations":["de","den","det","en","et","-et","-en","-ne"]}	28,223	100.0%	63,132
de	German	Deutsch	{"word":"the","translations":["der","die","das","de","dr","den","dieser","umso","Platz","der die das die"]}	28,223	100.0%	146,827
el	Greek	Ελληνικά	{"word":"the","translations":["ο","ὁ","στον","στη","στο","άρθρο","η","οι","παράμεσος","στα"]}	28,179	99.8%	90,613
es	Spanish	español	{"word":"the","translations":["el","la","los","las","lo","el la los las","entonces","pero","pues"]}	28,223	100.0%	126,135
fa	Persian	فارسی	{"word":"the","translations":["این","آن","ال","باسلیقون","ثلاثة غساله","دلیجه","ساز","سفیدرود","سنده","قبلة"]}	28,223	100.0%	91,682
fi	Finnish	suomi	{"word":"the","translations":["se","-ni","-si","-nsa","-nsä","-mme","-nne","minun","sinun","hänen"]}	28,223	100.0%	128,415
fil	Filipino	Filipino	{"word":"the","translations":["ang","nang","mga"]}	28,223	100.0%	48,367
fr	French	français	{"word":"the","translations":["le","la","les","l'","comp.","colloquial or regional usage","acquitter","au","aux","ce"]}	28,223	100.0%	132,443
he	Hebrew	עברית	{"word":"the","translations":["ה־","ה"]}	28,222	100.0%	50,356
hi	Hindi	हिन्दी	{"word":"the","translations":["जिस काम के पूरे करने में सभी लगे हों","ईसामसीह","दि","यह","वह","वही"]}	28,223	100.0%	106,614
hu	Hungarian	magyar	{"word":"the","translations":["a","az"]}	28,223	100.0%	104,597
id	Indonesian	Bahasa Indonesia	{"word":"the","translations":["si","sang","itu","al","-nya","kian"]}	28,223	100.0%	93,819
is	Icelandic	íslenska	{"word":"the","translations":["-inn","hinn","hið","-in","-nar","-ið","-nir"]}	28,223	100.0%	57,806
it	Italian	italiano	{"word":"the","translations":["il","lo","la","l'","i","gli","le","li","lì","gl’"]}	28,223	100.0%	120,328
ja	Japanese	日本語	{"word":"the","translations":["その","正に","正しく","ザ","這","あの","其の","定冠詞","当該","死因"]}	28,212	100.0%	145,314
ko	Korean	한국어	{"word":"the","translations":["-을수록","그","그만큼$2$","저","하면 할수록","하면 할수록 그만큼 더"]}	28,217	100.0%	104,823
lt	Lithuanian	lietuvių	{"word":"the","translations":["tai"]}	28,223	100.0%	42,549
lv	Latvian	latviešu	{"word":"the","translations":["sirdi"]}	28,211	100.0%	38,168
mk	Macedonian	македонски	{"word":"the","translations":["-ов","-те","-от","многуте","највеќе","најголем дел","најмногу","онака","постојано","-во"]}	28,184	99.9%	47,766
ms	Malay	Bahasa Melayu	{"word":"the","translations":["sang","-nya","itu","ini","si","semakin..., semakin...","lagi..., lagi..."]}	28,223	100.0%	85,179
nb	Norwegian Bokmål	norsk bokmål	{"word":"the","translations":["det","den","-et","-en","-a","-i","de","desto","jo","-one"]}	28,223	100.0%	68,517
nl	Dutch	Nederlands	{"word":"the","translations":["de","het","aan de","aan het","des te","die","hoe","naar de","naar het","’t"]}	28,223	100.0%	107,151
pl	Polish	polski	{"word":"the","translations":["ten","czym","im","przedimek określony","ta","to","te"]}	28,223	100.0%	95,604
pt	Portuguese	português	{"word":"the","translations":["o","a","os","as","o a os as"]}	28,223	100.0%	107,945
ro	Romanian	română	{"word":"the","translations":["-ul","-a","cel","cea","-i","cele","-le"]}	28,223	100.0%	61,507
ru	Russian	русский	{"word":"the","translations":["на ходу","определенный артикль","определённый артикль","подходящий","такой","тем","тот","чем","чем... тем"]}	28,219	100.0%	148,990
sh	Serbo-Croatian	srpskohrvatski / српскохрватски	{"word":"the","translations":["Alpe","Antarktika","Atlantik","Balkan","Baltik","Drava","Drina","Dunav","Holandija","Jadran"]}	28,209	100.0%	118,146
sk	Slovak	slovenčina	{"word":"the","translations":["ten","to","tá"]}	28,223	100.0%	98,458
sl	Slovenian	slovenščina	{"word":"the","translations":["the"]}	28,223	100.0%	76,078
sv	Swedish	svenska	{"word":"the","translations":["den","dess","desto","-en","de","det","ju","vederbörande","vintergatan","överklassen"]}	28,223	100.0%	100,554
ta	Tamil	தமிழ்	{"word":"the","translations":["அது"]}	28,223	100.0%	72,940
th	Thai	ไทย	{"word":"the","translations":["นั้น"]}	28,188	99.9%	28,312
tr	Turkish	Türkçe	{"word":"the","translations":["bir","harfi tarif","ne kadar","üstlenmek"]}	28,223	100.0%	99,531
uk	Ukrainian	українська	{"word":"the","translations":["те"]}	28,221	100.0%	71,637
ur	Urdu	اردو	{"word":"the","translations":["ال"]}	28,223	100.0%	60,976
vi	Vietnamese	Tiếng Việt	{"word":"the","translations":["con","càng","cái","duy nhất","này","ấy"]}	28,223	100.0%	95,230
合计	41 种语言	—	—	1,156,972	100.0%	3,681,852

X → en (XToEn)

英文维基词典来源反转；词表 = vocabradar/wordfreq 的 small_<code> — 41 种语言

代码	语言	自称	首行	已译词数	命中率	条目数
ar	Arabic	العربية	{"word":"في","translations":["a","in","on","at","by","into","through","during","times","within"]}	55,606	100.0%	101,618
bg	Bulgarian	български	{"word":"на","translations":["of","in","on","at","onto","lo"]}	35,848	100.0%	55,492
bn	Bengali	বাংলা	{"word":"না","translations":["not","no","don't","never","without","please","non","boat","un","na"]}	33,045	99.9%	50,042
ca	Catalan	català	{"word":"de","translations":["of","in","with","from","about","more","off","again","coral","dee"]}	27,028	100.0%	60,056
cs	Czech	čeština	{"word":"a","translations":["and","a","that","who","which","any","both","recent","double","plus"]}	51,744	100.0%	92,063
da	Danish	dansk	{"word":"i","translations":["to","of","a","in","for","you","on","with","at","by"]}	29,321	100.0%	50,479
de	German	Deutsch	{"word":"die","translations":["the","a","that","this","but","an","all","what","who","which"]}	38,989	100.0%	114,757
el	Greek	Ελληνικά	{"word":"και","translations":["and","that","but","which","also","too","both"]}	43,188	100.0%	61,778
es	Spanish	español	{"word":"de","translations":["of","in","for","on","with","be","as","at","by","from"]}	34,740	100.0%	98,038
fa	Persian	فارسی	{"word":"و","translations":["and","both"]}	30,108	100.0%	57,014
fi	Finnish	suomi	{"word":"ja","translations":["and","so","also","too","both","plus","likewise","wedge"]}	59,143	100.0%	107,690
fil	Filipino	Filipino	{"word":"sa","translations":["to","of","a","in","for","on","with","at","by","from"]}	30,098	100.0%	40,648
fr	French	français	{"word":"de","translations":["to","of","a","in","for","on","with","be","as","at"]}	31,203	100.0%	100,218
he	Hebrew	עברית	{"word":"את","translations":["you","with","me","alongside","amongst","thou","shovel","spade"]}	57,376	99.9%	68,773
hi	Hindi	हिन्दी	{"word":"के","translations":["of","in","for","with","as","have","by","from","his","about"]}	24,031	100.0%	56,346
hu	Hungarian	magyar	{"word":"a","translations":["the","a","black","amps"]}	46,525	100.0%	81,700
id	Indonesian	Bahasa Indonesia	{"word":"yang","translations":["that","who","which","second","third","whose","rel"]}	30,967	100.0%	77,623
is	Icelandic	íslenska	{"word":"og","translations":["and","but","which","also","both","plus"]}	42,207	100.0%	56,957
it	Italian	italiano	{"word":"di","translations":["to","of","in","for","that","on","with","be","at","by"]}	35,954	100.0%	96,114
ja	Japanese	日本語	{"word":"の","translations":["indicates possessive","nominalizes verbs and adjectives","substitutes for \"ga\" in subordinate phrases","indicates a confident conclusion","indicates emotional emphasis","indicates question","field","plain","hidden interior part","wild"]}	27,856	100.0%	138,071
ko	Korean	한국어	{"word":"이","translations":["and","that","you","this","be","he","but","she","only","two"]}	26,807	100.0%	48,935
lt	Lithuanian	lietuvių	{"word":"ir","translations":["and","even","afterward"]}	63,973	100.0%	71,616
lv	Latvian	latviešu	{"word":"un","translations":["and"]}	45,464	100.0%	51,239
mk	Macedonian	македонски	{"word":"на","translations":["of","for","on","at","per"]}	32,925	100.0%	42,850
ms	Malay	Bahasa Melayu	{"word":"yang","translations":["that","one","which","change","cash","coins","umbrella"]}	28,571	100.0%	69,966
nb	Norwegian Bokmål	norsk bokmål	{"word":"i","translations":["to","of","a","in","for","you","on","with","at","by"]}	25,994	100.0%	48,694
nl	Dutch	Nederlands	{"word":"de","translations":["the","you","this","one","able","bus","exposure","repaid","earthy"]}	28,715	100.0%	78,061
pl	Polish	polski	{"word":"w","translations":["to","of","a","in","for","on","with","at","by","an"]}	49,664	100.0%	81,502
pt	Portuguese	português	{"word":"de","translations":["of","for","with","by","from","about","something","old","someone","hunger"]}	33,123	100.0%	82,780
ro	Romanian	română	{"word":"de","translations":["of","for","on","by","from","since","against"]}	43,279	100.0%	62,582
ru	Russian	русский	{"word":"в","translations":["to","of","a","in","for","on","with","as","at","by"]}	59,604	100.0%	122,769
sh	Serbo-Croatian	srpskohrvatski / српскохрватски	{"word":"je","translations":["is","are","if","has","her","did","sued"]}	54,698	100.0%	97,376
sk	Slovak	slovenčina	{"word":"a","translations":["and","amps"]}	59,414	100.0%	101,829
sl	Slovenian	slovenščina	{"word":"je","translations":["is"]}	53,882	100.0%	77,313
sv	Swedish	svenska	{"word":"är","translations":["is","are","am","result","arise"]}	30,515	100.0%	66,415
ta	Tamil	தமிழ்	{"word":"ஒரு","translations":["a","an","hour","sole","sheep","mono"]}	64,751	100.0%	83,531
th	Thai	ไทย	{"word":"ที่","translations":["place","that","which","who"]}	39,499	99.5%	39,856
tr	Turkish	Türkçe	{"word":"ve","translations":["and","so","both","species","vee"]}	63,130	100.0%	95,737
uk	Ukrainian	українська	{"word":"в","translations":["to","in","have","at","into","per"]}	46,285	100.0%	62,505
ur	Urdu	اردو	{"word":"کے","translations":["of","at"]}	22,442	100.0%	37,850
vi	Vietnamese	Tiếng Việt	{"word":"là","translations":["is","that","was","be","as","are","were","been","how","some"]}	10,519	100.0%	26,597
zh	Chinese	中文	{"word":"的","translations":["of","s","used after an attribute when it modifies a noun","used at the end of a declarative sentence for emphasis","also pr. [di4] or [di5] in poetry and songs","a taxi","a cab","really and truly","bull's-eye","target"]}	35,723	100.0%	145,872
合计	42 种语言	—	—	1,683,954	100.0%	3,161,352

zh → X (ZhToX)

中文维基词典来源；词表 = vocabradar/wordfreq 的 small_zh — 40 种语言

代码	语言	自称	首行	已译词数	命中率	条目数
ar	Arabic	العربية	{"word":"的","translations":["ال"]}	35,723	100.0%	37,413
bg	Bulgarian	български	{"word":"的","translations":["ето"]}	35,702	99.9%	36,852
bn	Bengali	বাংলা	{"word":"的","translations":["ের"]}	35,723	100.0%	36,298
ca	Catalan	català	{"word":"的","translations":["de"]}	35,721	100.0%	36,571
cs	Czech	čeština	{"word":"的","translations":["Jeho"]}	35,723	100.0%	36,903
da	Danish	dansk	{"word":"的","translations":["Dan"]}	35,722	100.0%	36,601
de	German	Deutsch	{"word":"的","translations":["wird nach einem Adjektiv oder einer Attributphrase verwendet","wird nach einem Verb verwendet, um das Subjekt, die Zeit","Ziel, das Schwarze","klar"]}	35,721	100.0%	60,922
el	Greek	Ελληνικά	{"word":"的","translations":["του"]}	35,229	98.6%	36,199
es	Spanish	español	{"word":"的","translations":["de"]}	35,723	100.0%	38,381
fa	Persian	فارسی	{"word":"的","translations":["که"]}	35,721	100.0%	37,234
fi	Finnish	suomi	{"word":"的","translations":["Finnish"]}	35,704	99.9%	37,635
fil	Filipino	Filipino	{"word":"的","translations":["ay"]}	35,722	100.0%	36,531
fr	French	français	{"word":"的","translations":["de","utilisé après un adjectif, un nom","centre de la cible","but","objectif","voir","vrai"]}	35,723	100.0%	67,607
he	Hebrew	עברית	{"word":"的","translations":["של"]}	35,714	100.0%	36,359
hi	Hindi	हिन्दी	{"word":"的","translations":["का"]}	35,722	100.0%	37,985
hu	Hungarian	magyar	{"word":"的","translations":["És"]}	35,720	100.0%	37,238
id	Indonesian	Bahasa Indonesia	{"word":"的","translations":["adalah"]}	35,723	100.0%	36,491
is	Icelandic	íslenska	{"word":"的","translations":["Er"]}	35,723	100.0%	36,535
it	Italian	italiano	{"word":"的","translations":["di"]}	35,723	100.0%	37,266
ja	Japanese	日本語	{"word":"的","translations":["の"]}	35,706	100.0%	38,413
ko	Korean	한국어	{"word":"的","translations":["-의"]}	35,706	100.0%	36,859
lt	Lithuanian	lietuvių	{"word":"的","translations":["ž"]}	35,721	100.0%	36,204
lv	Latvian	latviešu	{"word":"的","translations":["Latv"]}	35,663	99.8%	36,253
mk	Macedonian	македонски	{"word":"的","translations":["Македонски"]}	35,587	99.6%	36,528
ms	Malay	Bahasa Melayu	{"word":"的","translations":["nya"]}	35,723	100.0%	36,659
nb	Norwegian Bokmål	norsk bokmål	{"word":"的","translations":["Det"]}	35,721	100.0%	36,475
nl	Dutch	Nederlands	{"word":"的","translations":["van"]}	35,723	100.0%	37,389
pl	Polish	polski	{"word":"的","translations":["tak"]}	35,723	100.0%	37,637
pt	Portuguese	português	{"word":"的","translations":["do"]}	35,719	100.0%	37,492
ro	Romanian	română	{"word":"的","translations":["de”"]}	35,719	100.0%	36,937
ru	Russian	русский	{"word":"的","translations":["дом"]}	35,694	99.9%	63,021
sh	Serbo-Croatian	srpskohrvatski / српскохрватски	{"word":"的","translations":["Srbijanski"]}	35,494	99.4%	37,198
sk	Slovak	slovenčina	{"word":"的","translations":["So"]}	35,723	100.0%	36,532
sl	Slovenian	slovenščina	{"word":"的","translations":["Slovenski"]}	35,721	100.0%	36,300
sv	Swedish	svenska	{"word":"的","translations":["av"]}	35,723	100.0%	37,092
ta	Tamil	தமிழ்	{"word":"的","translations":["அதன்"]}	35,723	100.0%	36,048
th	Thai	ไทย	{"word":"的","translations":["ของ"]}	34,357	96.2%	34,368
tr	Turkish	Türkçe	{"word":"的","translations":["oldu"]}	35,723	100.0%	37,256
uk	Ukrainian	українська	{"word":"的","translations":["Українською"]}	35,699	99.9%	37,623
ur	Urdu	اردو	{"word":"的","translations":["کی"]}	35,723	100.0%	36,683
vi	Vietnamese	Tiếng Việt	{"word":"的","translations":["của"]}	35,722	100.0%	37,470
合计	41 种语言	—	—	1,462,195	99.8%	1,593,458

X → zh (XToZh)

中文维基词典来源反转；词表 = vocabradar/wordfreq 的 small_<code> — 41 种语言

代码	语言	自称	首行	已译词数	命中率	条目数
ar	Arabic	العربية	{"word":"في","translations":["在"]}	55,553	99.9%	55,814
bg	Bulgarian	български	{"word":"на","translations":["在"]}	35,768	99.8%	35,904
bn	Bengali	বাংলা	{"word":"না","translations":["从不"]}	32,791	99.2%	32,874
ca	Catalan	català	{"word":"de","translations":["德"]}	27,006	99.9%	27,148
cs	Czech	čeština	{"word":"a","translations":["a"]}	51,708	99.9%	51,825
da	Danish	dansk	{"word":"i","translations":["i"]}	29,310	100.0%	29,437
de	German	Deutsch	{"word":"die","translations":["是","这些"]}	38,975	100.0%	61,494
el	Greek	Ελληνικά	{"word":"και","translations":["和"]}	42,952	99.4%	43,061
en	English	English	{"word":"the","translations":["那"]}	28,219	100.0%	84,155
es	Spanish	español	{"word":"de","translations":["德"]}	34,731	100.0%	35,055
fa	Persian	فارسی	{"word":"و","translations":["和"]}	30,070	99.9%	30,277
fi	Finnish	suomi	{"word":"ja","translations":["是"]}	59,030	99.8%	59,234
fil	Filipino	Filipino	{"word":"sa","translations":["萨"]}	30,093	100.0%	30,167
fr	French	français	{"word":"de","translations":["的","与","于","从","由","从中","乎","为了"]}	31,196	100.0%	50,536
he	Hebrew	עברית	{"word":"את","translations":["你"]}	57,300	99.8%	57,431
hi	Hindi	हिन्दी	{"word":"के","translations":["科"]}	23,800	99.0%	24,036
hu	Hungarian	magyar	{"word":"a","translations":["a"]}	46,519	100.0%	46,669
id	Indonesian	Bahasa Indonesia	{"word":"yang","translations":["杨"]}	30,957	100.0%	31,060
is	Icelandic	íslenska	{"word":"og","translations":["og"]}	42,189	100.0%	42,280
it	Italian	italiano	{"word":"di","translations":["迪"]}	35,947	100.0%	36,154
ja	Japanese	日本語	{"word":"の","translations":["的"]}	27,822	99.9%	28,038
ko	Korean	한국어	{"word":"이","translations":["牙齿","二"]}	26,758	99.8%	26,847
lt	Lithuanian	lietuvių	{"word":"ir","translations":["ir"]}	63,905	99.9%	63,978
lv	Latvian	latviešu	{"word":"un","translations":["不"]}	45,397	99.9%	45,471
mk	Macedonian	македонски	{"word":"на","translations":["在"]}	32,840	99.7%	32,954
ms	Malay	Bahasa Melayu	{"word":"yang","translations":["杨"]}	28,562	100.0%	28,673
nb	Norwegian Bokmål	norsk bokmål	{"word":"i","translations":["你"]}	25,985	100.0%	26,067
nl	Dutch	Nederlands	{"word":"de","translations":["de"]}	28,708	100.0%	28,879
pl	Polish	polski	{"word":"w","translations":["w"]}	49,651	100.0%	49,832
pt	Portuguese	português	{"word":"de","translations":["de"]}	33,117	100.0%	33,370
ro	Romanian	română	{"word":"de","translations":["德"]}	43,271	100.0%	43,415
ru	Russian	русский	{"word":"в","translations":["就是","同时","如何","其中","影响","目前","参加","具有","实现","是不是"]}	59,388	99.6%	72,489
sh	Serbo-Croatian	srpskohrvatski / српскохрватски	{"word":"je","translations":["你"]}	54,687	100.0%	54,724
sk	Slovak	slovenčina	{"word":"a","translations":["a"]}	59,390	100.0%	59,484
sl	Slovenian	slovenščina	{"word":"je","translations":["je"]}	53,870	100.0%	53,934
sv	Swedish	svenska	{"word":"är","translations":["是"]}	30,507	100.0%	30,680
ta	Tamil	தமிழ்	{"word":"ஒரு","translations":["一个"]}	64,563	99.7%	64,609
th	Thai	ไทย	{"word":"ที่","translations":["地点","地點"]}	34,098	85.9%	34,634
tr	Turkish	Türkçe	{"word":"ve","translations":["ve"]}	63,011	99.8%	63,200
uk	Ukrainian	українська	{"word":"в","translations":["在"]}	46,084	99.6%	46,236
ur	Urdu	اردو	{"word":"کے","translations":["的"]}	22,415	99.9%	22,547
vi	Vietnamese	Tiếng Việt	{"word":"là","translations":["是"]}	10,510	99.9%	10,589
合计	42 种语言	—	—	1,668,653	99.5%	1,785,261

重新生成

python tools/build_pro.py                       # 重建专业词典层 build/pro
python tools/wt.py probe --pair en              # 词表命中与缺口报告
python tools/wt.py fill  --pair en --pass p2    # 本地 Hy-MT2 补缺口
python tools/wt.py pack  --pair en              # 重打包 data/ + meta/files/README

工具链

WordTranslation/
├── README.md / meta.json / files.json   # 数据集卡 + 边车文件（发布）
├── data/                                # 唯一发布产物：<src>_<dst>.jsonl.gz
├── build/                               # 构建输入（不发布）
│   ├── raw/                             #   源 dump（wiktextract en/zh）
│   ├── pro/                             #   专业词典层（<src>_<dst>.jsonl）
│   └── <pair>/{cache,llm,manual,...}    #   各对的中间产物
├── tools/wt.py                          # probe | fill | pack 单文件工具链
├── docs/                                # 规划与调研
└── work/<pair>/{hymt,lmt}/<pass>/       # 模型补译层（pack 的输入）

子命令	作用
probe --pair en|zh	正/反双方向命中与缺口报告
fill --pair en --pass p1|p2 [--engine hymt|lmt] [--langs a,b] [--conc 8] [--limit N]	逐词补缺 → work/<pair>/<engine>/<pass>/<code>.jsonl（追加幂等，Ctrl-C 断点续跑）
pack --pair en [--pass both] [--base clean|merged] [--no-pro]	pro + base + hymt/lmt 层 → 重打包 data/，重建 meta.json / files.json / README.md

• --base clean（缺省）= 专业词典(pro_) + 维基(pairs_) + PanLex(panlex_) 纯人工来源， 反向为其反转；merged = 旧管线四层合并包（已弃用，留作对照）。
• en ⇄ zh 这一对由两套管线各产一份且内容不同；扁平布局下只保留一份： en_zh 取 zh 对 p2（XToZh 管线），zh_en 取 en 对 p2（XToEn 管线）， 与网站消费端 VocabRadar/web/scripts/init_web.js 的「释义语族优先」口径一致。 对应两对的 p1 自动跳过该 code（wt.CROSS_SKIP），避免 pack 时互相覆盖。
• 模型服务由 llama.cpp 的 llama-server 提供，wt.py 自动拉起/复用/关闭。
• 环境变量：WT_LLAMA_SERVER、WT_HYMT_GGUF、WT_LMT_GGUF、WT_PORT（缺省 8791）、 WT_WF_DATA（vocabradar/wordfreq 词表目录）。

许可证

• 专业词典：ECDICT（MIT）、CC-CEDICT（CC BY-SA 4.0）、JMdict（CC BY-SA 4.0）、 HanDeDict（CC BY-SA 2.0）、CFDICT（CC BY-SA 3.0）、Folkets lexikon（CC BY-SA 2.5）—— 均与本数据集的 CC BY-SA 4.0 兼容；署名见本节与 meta.json。
• wiktextract/kaikki.org 抽取：CC BY-SA 4.0（Wiktionary，另以 GFDL 双许可）。
• PanLex 补充：CC0（2024-03-01 dump）。
• 词表：取自 vocabradar/wordfreq 数据集，MIT。
• Hunyuan-MT2-1.8B / LMT-60-1.7B 补译层：机器生成，已在此与 meta.json 中披露； 两个模型均为 Apache-2.0。

关于 VocabRadar

这项工作是 VocabRadar（词汇雷达）的工作：VocabRadar 浏览器扩展 让你在看视频、刷网页时把还不认识的生词就地标注出来 —— 释义、翻译、双语字幕、注释字幕； 查过的词还能送到 vocabradar.com，自动生成相关习题与游戏，把词真正记住。