vocabradar/VocabFreq

Word Frequency Lists (110 languages)

English | 中文说明

Small word-frequency lists for 110 languages, self-built for use as offline frequency dictionaries: one gzip-compressed MessagePack file per language, about 13 MB in total. These lists are also the direct vocabulary source of vocabradar/WordTranslation, which translates them word by word.

The lists come from a single pyicu BreakIterator segmentation pipeline over FineWeb2 HQ+, UltraFineWeb L3, and Wikipedia (each source capped at 1 GiB of text per language). Word selection uses a size-aware frequency gate: fu_w > 1e-6 AND fu_w / word_bytes > 2e-7, packed into 700 centibel bands.

Selection rule: a language is included when its total source text volume exceeds 100 MB (104,857,600 bytes, summed over sources; text_bytes in meta.json). Each source is capped at 1 GiB during corpus build.

Current revision: v7 (2026-10-26). See Version history.

Files

├── README.md
├── meta.json                      # dataset metadata (see fields below)
└── data/
    └── small_<code>.msgpack.gz    # one file per language,
                                   # e.g. small_en.msgpack.gz, small_th.msgpack.gz

meta.json records per language: kept (words in the list), coverage (real corpus coverage of the list), sources (per-source text_bytes / tokens), and derived text_bytes (total source volume) and tokens_total.

All languages use 700 centibel bands with the same cBpack encoding (header + list of bands). See Data format.

Cleaning rules

Single upstream pipeline (pyicu BreakIterator + all-script Unicode rule set), identical for every language:

• Normalization: NFC for Latin/Greek/Cyrillic scripts, NFKC otherwise; abjad vowel-mark removal (ar, he, fa, ur, ps, sd, ug, ...); casefold with dotless-i for tr/az/kk; cedilla→comma fix for ro.
• Rejection rules: no_letter, numeric, has_digit, domain_tld, diacritic_single, foreign (per-language script whitelist).
• Frequency gate: fu_w > 1e-6 AND fu_w / word_bytes > 2e-7, where word_bytes is the word's UTF-8 length (+1 for languages written with spaces).

For implementation details see the source pipeline: WordFrequency/scripts/build_wordfreq.py (in the VocabRadar data repo).

Vocabulary lists

Alphabetical by language code.

code	language	native name	first word	cb	float	words	cumulative coverage
af	Afrikaans	Afrikaans	die	116	6.92e-02	27,289	87.9%
am	Amharic	አማርኛ	እና	176	1.74e-02	30,174	75.6%
ar	Arabic	العربية	في	153	2.95e-02	34,950	78.1%
arz	Egyptian Arabic	مصرى	فى	144	3.63e-02	12,783	79.8%
as	Assamese	অসমীয়া	আৰু	162	2.40e-02	18,601	77.9%
ast	Asturian	asturianu	de	119	6.46e-02	36,867	84.2%
az	Azerbaijani	azərbaycan	və	153	2.95e-02	38,016	80.2%
azb	South Azerbaijani	تۆرکجه	و	168	2.09e-02	10,407	52.7%
ba	Bashkir	башҡортса	һәм	171	1.95e-02	22,247	73.8%
be	Belarusian	беларуская	і	156	2.75e-02	26,836	71.4%
bew	Betawi	Basa Betawi	yang	169	2.04e-02	32,004	91.7%
bg	Bulgarian	български	на	127	5.37e-02	23,184	77.4%
bn	Bengali	বাংলা	এবং	187	1.35e-02	18,147	77.6%
bo	Tibetan	བོད་ཡིག	པ	147	3.39e-02	4,812	97.4%
bs	Bosnian	bosanski	i	144	3.63e-02	43,867	82.5%
ca	Catalan	català	de	122	6.03e-02	30,554	87.0%
ce	Chechen	нохчийн	а	149	3.24e-02	4,772	74.1%
ceb	Cebuano	Cebuano	sa	124	5.75e-02	12,376	87.5%
ckb	Central Kurdish	کوردی	لە	128	5.25e-02	23,055	77.6%
cs	Czech	čeština	a	153	2.95e-02	43,516	78.7%
cy	Welsh	Cymraeg	yn	133	4.68e-02	27,914	84.9%
da	Danish	dansk	i	150	3.16e-02	31,312	83.9%
de	German	Deutsch	der	149	3.24e-02	32,074	80.8%
dv	Dhivehi	ދިވެހި	އެވެ	183	1.48e-02	22,674	81.7%
el	Greek	Ελληνικά	και	151	3.09e-02	20,531	77.2%
en	English	English	the	121	6.17e-02	25,475	90.9%
eo	Esperanto	Esperanto	la	113	7.41e-02	38,703	80.3%
es	Spanish	español	de	114	7.24e-02	29,488	89.4%
et	Estonian	eesti	ja	151	3.09e-02	43,564	76.0%
eu	Basque	euskara	eta	142	3.80e-02	37,926	81.0%
fa	Persian	فارسی	و	140	3.98e-02	18,812	86.3%
fi	Finnish	suomi	ja	148	3.31e-02	43,242	69.4%
fil	Tagalog	Tagalog	sa	120	6.31e-02	25,283	91.6%
fr	French	français	de	123	5.89e-02	30,283	88.9%
fy	West Frisian	Frysk	de	130	5.01e-02	32,813	83.5%
ga	Irish	Gaeilge	an	137	4.27e-02	27,106	88.2%
gl	Galician	galego	de	127	5.37e-02	31,191	86.7%
gmh	Middle High German	Mittelhochdeutsch	und	197	1.07e-02	35,007	86.3%
gu	Gujarati	ગુજરાતી	છે	133	4.68e-02	19,297	80.5%
ha	Hausa	Harshen Hausa	da	119	6.46e-02	23,058	89.0%
he	Hebrew	עברית	של	162	2.40e-02	40,467	80.5%
hi	Hindi	हिन्दी	के	138	4.17e-02	14,413	86.0%
hr	Croatian	hrvatski	i	147	3.39e-02	45,220	81.4%
hu	Hungarian	magyar	a	110	7.94e-02	39,941	74.3%
hy	Armenian	հայերեն	է	140	3.98e-02	22,345	71.0%
id	Indonesian	Bahasa Indonesia	yang	155	2.82e-02	29,234	86.3%
ig	Igbo	Asụsụ Igbo	na	104	9.12e-02	23,765	87.1%
is	Icelandic	íslenska	og	142	3.80e-02	33,521	84.2%
it	Italian	italiano	di	139	4.07e-02	33,439	86.9%
ja	Japanese	日本語	の	129	5.13e-02	27,259	88.8%
ka	Georgian	ქართული	და	148	3.31e-02	19,548	65.9%
kk	Kazakh	қазақша	және	185	1.41e-02	25,922	74.3%
km	Khmer	ខ្មែរ	បាន	173	1.86e-02	12,302	89.9%
kn	Kannada	ಕನ್ನಡ	ಮತ್ತು	172	1.91e-02	21,850	65.1%
ko	Korean	한국어	년	187	1.35e-02	42,580	59.6%
ku	Kurdish	kurdî	li	163	2.34e-02	30,109	82.8%
ky	Kyrgyz	кыргызча	жана	175	1.78e-02	26,184	77.3%
la	Latin	Latina	et	152	3.02e-02	41,416	81.4%
lb	Luxembourgish	Lëtzebuergesch	an	162	2.40e-02	33,684	82.2%
lo	Lao	ພາສາລາວ	ຂອງ	158	2.63e-02	8,165	90.6%
lt	Lithuanian	lietuvių	ir	152	3.02e-02	44,180	79.3%
lv	Latvian	latviešu	un	151	3.09e-02	41,008	82.8%
mg	Malagasy	Malagasy	ny	111	7.76e-02	23,818	80.0%
mk	Macedonian	македонски	на	125	5.62e-02	22,246	80.1%
ml	Malayalam	മലയാളം	ഒരു	191	1.23e-02	19,303	52.5%
mn	Mongolian	монгол	нь	162	2.40e-02	23,142	83.1%
mr	Marathi	मराठी	आणि	178	1.66e-02	20,174	75.3%
ms	Malay	Bahasa Melayu	yang	154	2.88e-02	26,033	89.9%
mt	Maltese	Malti	l	142	3.80e-02	27,749	90.3%
my	Myanmar	မြန်မာ	ကို	163	2.34e-02	8,704	92.2%
nb	Norwegian Bokmål	norsk bokmål	og	149	3.24e-02	28,715	87.4%
ne	Nepali	नेपाली	र	164	2.29e-02	19,634	80.1%
new	Newari	नेपाल भाषा	thumb	294	1.15e-03	2,894	2.0%
nl	Dutch	Nederlands	de	122	6.03e-02	29,041	84.6%
nn	Norwegian Nynorsk	norsk nynorsk	i	140	3.98e-02	31,299	82.1%
no	Norwegian	norsk	i	143	3.72e-02	36,870	76.3%
nrm	Norman	Normaund	de	144	3.63e-02	37,834	65.2%
oc	Occitan	occitan	de	115	7.08e-02	36,968	84.0%
or	Odia	ଓଡ଼ିଆ	ଓ	192	1.20e-02	18,890	81.3%
pa	Punjabi	ਪੰਜਾਬੀ	ਹੈ	156	2.75e-02	15,494	87.4%
pl	Polish	polski	w	141	3.89e-02	44,785	78.6%
pnb	Western Punjabi	پنجابی	دے	145	3.55e-02	19,856	83.2%
ps	Pashto	پښتو	د	105	8.91e-02	19,349	89.3%
pt	Portuguese	português	de	127	5.37e-02	30,823	88.7%
ro	Romanian	română	de	133	4.68e-02	36,662	85.3%
ru	Russian	русский	в	142	3.80e-02	28,737	73.3%
sd	Sindhi	سنڌي	جي	134	4.57e-02	19,326	88.4%
sh	Serbo-Croatian	srpskohrvatski	u	146	3.47e-02	43,563	79.3%
si	Sinhala	සිංහල	සහ	214	7.24e-03	20,592	78.5%
sk	Slovak	slovenčina	a	152	3.02e-02	44,353	79.9%
sl	Slovenian	slovenščina	je	149	3.24e-02	43,561	82.4%
so	Somali	Soomaaliga	oo	154	2.88e-02	33,472	88.8%
sq	Albanian	shqip	të	119	6.46e-02	34,083	88.1%
sr	Serbian	српски	је	155	2.82e-02	22,537	59.8%
sv	Swedish	svenska	och	151	3.09e-02	32,738	81.7%
sw	Swahili	Kiswahili	ya	122	6.03e-02	24,555	88.9%
ta	Tamil	தமிழ்	ஒரு	207	8.51e-03	19,145	61.7%
te	Telugu	తెలుగు	నుండి	205	8.91e-03	19,393	67.3%
tg	Tajik	тоҷикӣ	дар	147	3.39e-02	21,633	77.1%
th	Thai	ไทย	ที่	162	2.40e-02	10,648	88.1%
tr	Turkish	Türkçe	ve	156	2.75e-02	43,447	74.3%
tt	Tatar	татарча	һәм	193	1.17e-02	12,504	68.5%
ug	Uyghur	ئۇيغۇرچە	ۋە	183	1.48e-02	25,287	76.0%
uk	Ukrainian	українська	в	172	1.91e-02	27,817	73.8%
ur	Urdu	اردو	کے	142	3.80e-02	16,966	89.3%
uz	Uzbek	oʻzbekcha	va	173	1.86e-02	32,259	54.6%
vi	Vietnamese	Tiếng Việt	là	186	1.38e-02	12,420	91.9%
war	Waray	Winaray	nga	123	5.89e-02	23,655	84.0%
yi	Yiddish	ייִדיש	די	142	3.80e-02	21,665	85.9%
zh	Chinese	中文	的	137	4.27e-02	35,707	90.8%
Total	110 languages	—	—	—	—	2,997,108	—

(first word = highest-frequency headword, sanity check. cb = centibel band of top word, float = token share. words = number of headwords in the list. cumulative coverage = Σ n_b × 10^(-b/100).)

Data format

Each file is a gzipped MessagePack array of [header, band_0..band_699]. Each band holds words whose frequency was rounded to that centibel (cB), so band b ≈ 10^(-b/100) of running tokens.

import gzip, msgpack

with gzip.open('data/small_en.msgpack.gz', 'rb') as f:
    obj = msgpack.unpackb(f.read(), raw=False)

header, *bands = obj
# header == {'format': 'cB', 'version': 1, 'lang': 'en',
#            'sources': ['ultrafineweb_l3', 'wikipedia']}

words = [(w, 10 ** (-b / 100)) for b, band in enumerate(bands) for w in band]
# words[0] ≈ ('the', ...) — rank 1 word of English

Version history

• v7 (2026-10-26): full rebuild from the self-built pyicu pipeline (single source, no more wordfreq merge). Uniform 700 bands, word gate fu_w > 1e-6 and fu_w / word_bytes > 2e-7. Selection: total source text_bytes > 100 MB → 110 languages (Simple English simple excluded: a wiki edition code, not a language code). Tagalog/Filipino is published under the ISO 639-3 code fil (not the legacy tl), one code for Tagalog/Filipino. Script whitelists fixed for ce, ba, tt, azb, ug, pnb, sd, arz, skr (their non-Latin words were being rejected as foreign, leaving only Latin debris). Cross-source frequency merge uses a global volume denominator (a source the word is absent from counts as zero frequency); an earlier per-word renormalization inflated words unique to small sources (e.g. bew list coverage estimated at 110%).
• v6 (merged 2026-10-26): merged wordfreq v4 (42 langs) into WordFrequency v5. Overlapping 42 languages use wf_v4 files+meta. 28 v5-only languages preserved. th file explicitly kept from v5 (user's own, higher-coverage pyicu result). 3 wf-v4-only languages restored (fil, is, sh).
• v5 (2026-10-26): WordFrequency self-built pyicu BreakIterator corpus (wiki + ultra + fw2). 76 languages (10^-7 threshold, 700 bands).
• v4 (2026-10-12): wordfreq v4 on HF — added digit_mix, possessive, mark_only, diacritic_single, stroke_single, jamo_single. 43 languages, cumulative counts across in-place revisions. Released as https://huggingface.co/datasets/vocabradar/wordfreq.
• v3 (2026-10-10): wordfreq v3 — numeric rule (digit-led codes, CJK numerals, numeric phrases). meta.json reorganized as languages[lang] = {total, removed, kept}.
• v1-v2: wordfreq 1-2 — initial small lists from wordfreq 3.0.2 + Unicode-script whitelist (foreign, pure_num, no_letter).

License

MIT.

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. What you have read can also be on vocabradar.com, which turns it into related exercises and games so the words actually stick.

---

词频表（110 语）

English | 中文说明

（卡片为单文件双语，中文部分完整对应上方英文说明。）

110 种语言的小型词频表，自建管线产出，可直接当离线词频字典使用：每种语言一个 gzip 压缩的 MessagePack 文件，合计约 13 MB。本词表同时是 vocabradar/WordTranslation 的直接词源（逐词翻译成各语种）。

数据来自统一的 pyicu BreakIterator 分词管线，语料为 FineWeb2 HQ+、 UltraFineWeb L3、 Wikipedia（每语种每来源截断至 1 GiB 文本）。收词采用体积感知的频率双门槛： fu_w > 1e-6 且 fu_w / word_bytes > 2e-7，打包成 700 个 centibel 频段。

收录规则： 合计源头语料体积 大于 100 MB（104,857,600 字节，1024 进制；各来源 text_bytes 相加，即 meta.json 中的 text_bytes）的语种入选。构建期每来源封顶 1 GiB。

当前修订版：v7（2026-10-26）。 见版本历史。

文件结构

├── README.md
├── meta.json                      # 数据集元信息（字段见下）
└── data/
    └── small_<code>.msgpack.gz    # 每语种一个文件，
                                   # 如 small_en.msgpack.gz、small_th.msgpack.gz

meta.json 按语种记录：kept（词表词数）、coverage（词表在语料上的真实覆盖率）、 sources（各来源的 text_bytes / tokens），以及派生字段 text_bytes（合计源头 体积）与 tokens_total。

频段数：全部语种统一 700 段，共享 cBpack 编码（header + band 列表）。详见数据格式。

清洗规则

统一的上游管线（pyicu BreakIterator + 全脚本 Unicode 规则集），各语种一致：

• 规范化：拉丁/希腊/西里尔文种 NFC，其余 NFKC；abjad 文种去元音点（ar、he、fa、 ur、ps、sd、ug 等）；casefold（tr/az/kk 用 dotless-i 特殊处理）； ro 的 cedilla→comma 修复。
• 剔除规则：no_letter、numeric、has_digit、domain_tld、diacritic_single、 foreign（按语种的 script 白名单）。
• 频率门槛：fu_w > 1e-6 且 fu_w / word_bytes > 2e-7；word_bytes 为词的 UTF-8 字节数（有空格分词的语言 +1 计入空格代价）。

实现细节见源管线：WordFrequency/scripts/build_wordfreq.py（VocabRadar 数据仓库内）。

词表一览

按语言代码字母序排列。

code	language	native name	first word	cb	float	words	cumulative coverage
af	南非荷兰语	Afrikaans	die	116	6.92e-02	27,289	87.9%
am	阿姆哈拉语	አማርኛ	እና	176	1.74e-02	30,174	75.6%
ar	阿拉伯语	العربية	في	153	2.95e-02	34,950	78.1%
arz	埃及阿拉伯语	مصرى	فى	144	3.63e-02	12,783	79.8%
as	阿萨姆语	অসমীয়া	আৰু	162	2.40e-02	18,601	77.9%
ast	阿斯图里亚斯语	asturianu	de	119	6.46e-02	36,867	84.2%
az	阿塞拜疆语	azərbaycan	və	153	2.95e-02	38,016	80.2%
azb	南阿塞拜疆语	تۆرکجه	و	168	2.09e-02	10,407	52.7%
ba	巴什基尔语	башҡортса	һәм	171	1.95e-02	22,247	73.8%
be	白俄罗斯语	беларуская	і	156	2.75e-02	26,836	71.4%
bew	贝塔维语	Basa Betawi	yang	169	2.04e-02	32,004	91.7%
bg	保加利亚语	български	на	127	5.37e-02	23,184	77.4%
bn	孟加拉语	বাংলা	এবং	187	1.35e-02	18,147	77.6%
bo	藏语	བོད་ཡིག	པ	147	3.39e-02	4,812	97.4%
bs	波斯尼亚语	bosanski	i	144	3.63e-02	43,867	82.5%
ca	加泰罗尼亚语	català	de	122	6.03e-02	30,554	87.0%
ce	车臣语	нохчийн	а	149	3.24e-02	4,772	74.1%
ceb	宿务语	Cebuano	sa	124	5.75e-02	12,376	87.5%
ckb	中库尔德语	کوردی	لە	128	5.25e-02	23,055	77.6%
cs	捷克语	čeština	a	153	2.95e-02	43,516	78.7%
cy	威尔士语	Cymraeg	yn	133	4.68e-02	27,914	84.9%
da	丹麦语	dansk	i	150	3.16e-02	31,312	83.9%
de	德语	Deutsch	der	149	3.24e-02	32,074	80.8%
dv	迪维希语	ދިވެހި	އެވެ	183	1.48e-02	22,674	81.7%
el	希腊语	Ελληνικά	και	151	3.09e-02	20,531	77.2%
en	英语	English	the	121	6.17e-02	25,475	90.9%
eo	世界语	Esperanto	la	113	7.41e-02	38,703	80.3%
es	西班牙语	español	de	114	7.24e-02	29,488	89.4%
et	爱沙尼亚语	eesti	ja	151	3.09e-02	43,564	76.0%
eu	巴斯克语	euskara	eta	142	3.80e-02	37,926	81.0%
fa	波斯语	فارسی	و	140	3.98e-02	18,812	86.3%
fi	芬兰语	suomi	ja	148	3.31e-02	43,242	69.4%
fil	他加禄语	Tagalog	sa	120	6.31e-02	25,283	91.6%
fr	法语	français	de	123	5.89e-02	30,283	88.9%
fy	西弗里西亚语	Frysk	de	130	5.01e-02	32,813	83.5%
ga	爱尔兰语	Gaeilge	an	137	4.27e-02	27,106	88.2%
gl	加利西亚语	galego	de	127	5.37e-02	31,191	86.7%
gmh	中古高地德语	Mittelhochdeutsch	und	197	1.07e-02	35,007	86.3%
gu	古吉拉特语	ગુજરાતી	છે	133	4.68e-02	19,297	80.5%
ha	豪萨语	Harshen Hausa	da	119	6.46e-02	23,058	89.0%
he	希伯来语	עברית	של	162	2.40e-02	40,467	80.5%
hi	印地语	हिन्दी	के	138	4.17e-02	14,413	86.0%
hr	克罗地亚语	hrvatski	i	147	3.39e-02	45,220	81.4%
hu	匈牙利语	magyar	a	110	7.94e-02	39,941	74.3%
hy	亚美尼亚语	հայերեն	է	140	3.98e-02	22,345	71.0%
id	印度尼西亚语	Bahasa Indonesia	yang	155	2.82e-02	29,234	86.3%
ig	伊博语	Asụsụ Igbo	na	104	9.12e-02	23,765	87.1%
is	冰岛语	íslenska	og	142	3.80e-02	33,521	84.2%
it	意大利语	italiano	di	139	4.07e-02	33,439	86.9%
ja	日语	日本語	の	129	5.13e-02	27,259	88.8%
ka	格鲁吉亚语	ქართული	და	148	3.31e-02	19,548	65.9%
kk	哈萨克语	қазақша	және	185	1.41e-02	25,922	74.3%
km	高棉语	ខ្មែរ	បាន	173	1.86e-02	12,302	89.9%
kn	卡纳达语	ಕನ್ನಡ	ಮತ್ತು	172	1.91e-02	21,850	65.1%
ko	韩语	한국어	년	187	1.35e-02	42,580	59.6%
ku	库尔德语	kurdî	li	163	2.34e-02	30,109	82.8%
ky	吉尔吉斯语	кыргызча	жана	175	1.78e-02	26,184	77.3%
la	拉丁语	Latina	et	152	3.02e-02	41,416	81.4%
lb	卢森堡语	Lëtzebuergesch	an	162	2.40e-02	33,684	82.2%
lo	老挝语	ພາສາລາວ	ຂອງ	158	2.63e-02	8,165	90.6%
lt	立陶宛语	lietuvių	ir	152	3.02e-02	44,180	79.3%
lv	拉脱维亚语	latviešu	un	151	3.09e-02	41,008	82.8%
mg	马达加斯加语	Malagasy	ny	111	7.76e-02	23,818	80.0%
mk	马其顿语	македонски	на	125	5.62e-02	22,246	80.1%
ml	马拉雅拉姆语	മലയാളം	ഒരു	191	1.23e-02	19,303	52.5%
mn	蒙古语	монгол	нь	162	2.40e-02	23,142	83.1%
mr	马拉地语	मराठी	आणि	178	1.66e-02	20,174	75.3%
ms	马来语	Bahasa Melayu	yang	154	2.88e-02	26,033	89.9%
mt	马耳他语	Malti	l	142	3.80e-02	27,749	90.3%
my	缅甸语	မြန်မာ	ကို	163	2.34e-02	8,704	92.2%
nb	书面挪威语	norsk bokmål	og	149	3.24e-02	28,715	87.4%
ne	尼泊尔语	नेपाली	र	164	2.29e-02	19,634	80.1%
new	尼瓦尔语	नेपाल भाषा	thumb	294	1.15e-03	2,894	2.0%
nl	荷兰语	Nederlands	de	122	6.03e-02	29,041	84.6%
nn	新挪威语	norsk nynorsk	i	140	3.98e-02	31,299	82.1%
no	挪威语	norsk	i	143	3.72e-02	36,870	76.3%
nrm	诺曼语	Normaund	de	144	3.63e-02	37,834	65.2%
oc	奥克语	occitan	de	115	7.08e-02	36,968	84.0%
or	奥里亚语	ଓଡ଼ିଆ	ଓ	192	1.20e-02	18,890	81.3%
pa	旁遮普语	ਪੰਜਾਬੀ	ਹੈ	156	2.75e-02	15,494	87.4%
pl	波兰语	polski	w	141	3.89e-02	44,785	78.6%
pnb	西旁遮普语	پنجابی	دے	145	3.55e-02	19,856	83.2%
ps	普什图语	پښتو	د	105	8.91e-02	19,349	89.3%
pt	葡萄牙语	português	de	127	5.37e-02	30,823	88.7%
ro	罗马尼亚语	română	de	133	4.68e-02	36,662	85.3%
ru	俄语	русский	в	142	3.80e-02	28,737	73.3%
sd	信德语	سنڌي	جي	134	4.57e-02	19,326	88.4%
sh	塞尔维亚-克罗地亚语	srpskohrvatski	u	146	3.47e-02	43,563	79.3%
si	僧伽罗语	සිංහල	සහ	214	7.24e-03	20,592	78.5%
sk	斯洛伐克语	slovenčina	a	152	3.02e-02	44,353	79.9%
sl	斯洛文尼亚语	slovenščina	je	149	3.24e-02	43,561	82.4%
so	索马里语	Soomaaliga	oo	154	2.88e-02	33,472	88.8%
sq	阿尔巴尼亚语	shqip	të	119	6.46e-02	34,083	88.1%
sr	塞尔维亚语	српски	је	155	2.82e-02	22,537	59.8%
sv	瑞典语	svenska	och	151	3.09e-02	32,738	81.7%
sw	斯瓦希里语	Kiswahili	ya	122	6.03e-02	24,555	88.9%
ta	泰米尔语	தமிழ்	ஒரு	207	8.51e-03	19,145	61.7%
te	泰卢固语	తెలుగు	నుండి	205	8.91e-03	19,393	67.3%
tg	塔吉克语	тоҷикӣ	дар	147	3.39e-02	21,633	77.1%
th	泰语	ไทย	ที่	162	2.40e-02	10,648	88.1%
tr	土耳其语	Türkçe	ve	156	2.75e-02	43,447	74.3%
tt	塔塔尔语	татарча	һәм	193	1.17e-02	12,504	68.5%
ug	维吾尔语	ئۇيغۇرچە	ۋە	183	1.48e-02	25,287	76.0%
uk	乌克兰语	українська	в	172	1.91e-02	27,817	73.8%
ur	乌尔都语	اردو	کے	142	3.80e-02	16,966	89.3%
uz	乌兹别克语	oʻzbekcha	va	173	1.86e-02	32,259	54.6%
vi	越南语	Tiếng Việt	là	186	1.38e-02	12,420	91.9%
war	瓦瑞语	Winaray	nga	123	5.89e-02	23,655	84.0%
yi	意第绪语	ייִדיש	די	142	3.80e-02	21,665	85.9%
zh	中文	中文	的	137	4.27e-02	35,707	90.8%
合计	110 种语言	—	—	—	—	2,997,108	—

（榜首词 = 最高频词条，快速确认语种正确。cb = 榜首词 centibel 频段，float = 榜首词 token 占比。词数 = 词表收录词条数。累计覆盖率 = Σ n_b × 10^(-b/100)。）

数据格式

每个文件是 gzip 压缩的 MessagePack 数组，结构为 [header, band_0..band_699]。每个 band 装有词频四舍五入到该 centibel 的词（约 10^(-b/100) token 概率）。

import gzip, msgpack

with gzip.open('data/small_en.msgpack.gz', 'rb') as f:
    obj = msgpack.unpackb(f.read(), raw=False)

header, *bands = obj
# header == {'format': 'cB', 'version': 1, 'lang': 'en',
#            'sources': ['ultrafineweb_l3', 'wikipedia']}

words = [(w, 10 ** (-b / 100)) for b, band in enumerate(bands) for w in band]
# words[0] ≈ ('the', ...) — 英语榜首词

版本历史

• v7（2026-10-26）：自建 pyicu 管线整体重建（单一来源，不再合并 wordfreq）。统一 700 频段，收词门槛 fu_w > 1e-6 且 fu_w / word_bytes > 2e-7。收录规则：合计源头 text_bytes > 100 MB，共 110 语 （剔除 simple 简明英语：维基站点代码，非语言码）。 语言代码采用 ISO 639-3 的 fil（他加禄语/菲律宾语合并为一个代码），不用旧代码 tl。修复 ce、ba、tt、azb、 ug、pnb、sd、arz、skr 的书写系统白名单（此前其非拉丁词被 foreign 全拒， 只剩拉丁残渣）。 跨来源词频合并改用全局体积分母（缺席来源按 0 频计）；此前的逐词归一化使小来源 独有词频率虚高（如 bew 词表覆盖率估算达 110%）。
• v6（合并于 2026-10-26）：将 wordfreq v4（42 语）合并入 WordFrequency v5。重叠语种使用 wf_v4 文件 + 元数据，28 个 v5 独有语种保留，th 强制保留 v5 文件，3 个 wf_v4 独有语种恢复（fil、is、sh）。
• v5（2026-10-26）：WordFrequency 自建 pyicu BreakIterator 语料（wiki + ultra + fw2），76 语（阈值 10^-7、700 频段）。
• v4（2026-10-12）：wordfreq v4 (HF) —— 新增 digit_mix、possessive、mark_only、diacritic_single、stroke_single、jamo_single，43 语，累计计数。发布于 https://huggingface.co/datasets/vocabradar/wordfreq。
• v3（2026-10-10）：wordfreq v3 —— 新增 numeric 规则（数字起头代码、中文数字与数字短语）。meta.json 重组为 languages[lang] = {total, removed, kept}。
• v1-v2：wordfreq 1-2 —— wordfreq 3.0.2 初始 small 列表 + Unicode 文字白名单清洗（foreign、pure_num、no_letter）。

许可证

MIT。

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； 看过的内容还能在 vocabradar.com，自动生成相关习题与游戏，把词真正记住。