Hugging Face's logo

Datasets:
vocabradar
/
wordfreq

Languages:
Arabic
Bulgarian
Bengali
Size:
1M<n<10M
License:
Dataset card Files
xet
 Community
Dataset Viewer
Duplicate
The dataset viewer is not available for this split.
Cannot load the dataset split (in streaming mode) to extract the first rows.
Error code:   StreamingRowsError
Exception:    CastError
Message:      Couldn't cast
version: int64
generatedAt: timestamp[s]
languages: struct<ar: struct<total: int64, removed: struct<digit_mix: int64>, kept: int64>, bg: struct<total: i (... 4407 chars omitted)
  child 0, ar: struct<total: int64, removed: struct<digit_mix: int64>, kept: int64>
      child 0, total: int64
      child 1, removed: struct<digit_mix: int64>
          child 0, digit_mix: int64
      child 2, kept: int64
  child 1, bg: struct<total: int64, removed: struct<digit_mix: int64>, kept: int64>
      child 0, total: int64
      child 1, removed: struct<digit_mix: int64>
          child 0, digit_mix: int64
      child 2, kept: int64
  child 2, bn: struct<total: int64, removed: struct<digit_mix: int64>, kept: int64>
      child 0, total: int64
      child 1, removed: struct<digit_mix: int64>
          child 0, digit_mix: int64
      child 2, kept: int64
  child 3, ca: struct<total: int64, removed: struct<numeric: int64, digit_mix: int64, possessive: int64>, kept: int (... 3 chars omitted)
      child 0, total: int64
      child 1, removed: struct<numeric: int64, digit_mix: int64, possessive: int64>
          child 0, numeric: int64
          child 1, digit_mix: int64
          child 2, possessive: int64
      child 2, kept: int64
  child 4, cs: struct<total: int64, removed: struct<numeric: int64, diacritic_single: int64, digit_mix: int64, poss (... 28 chars omitted)
      child 0, total: int64
      child 1, removed: struct<numeric: int64, diacritic_single: int64, digit_mix: int64, possessiv
...
critic_single: int64
          child 2, digit_mix: int64
          child 3, possessive: int64
      child 2, kept: int64
  child 38, uk: struct<total: int64, removed: struct<digit_mix: int64>, kept: int64>
      child 0, total: int64
      child 1, removed: struct<digit_mix: int64>
          child 0, digit_mix: int64
      child 2, kept: int64
  child 39, ur: struct<total: int64, removed: struct<digit_mix: int64>, kept: int64>
      child 0, total: int64
      child 1, removed: struct<digit_mix: int64>
          child 0, digit_mix: int64
      child 2, kept: int64
  child 40, vi: struct<total: int64, removed: struct<numeric: int64, digit_mix: int64, possessive: int64>, kept: int (... 3 chars omitted)
      child 0, total: int64
      child 1, removed: struct<numeric: int64, digit_mix: int64, possessive: int64>
          child 0, numeric: int64
          child 1, digit_mix: int64
          child 2, possessive: int64
      child 2, kept: int64
  child 41, zh: struct<total: int64, removed: struct<numeric: int64, stroke_single: int64>, kept: int64>
      child 0, total: int64
      child 1, removed: struct<numeric: int64, stroke_single: int64>
          child 0, numeric: int64
          child 1, stroke_single: int64
      child 2, kept: int64
algorithm: string
files: list<item: struct<file: string, bytes: int64, sha256: string>>
  child 0, item: struct<file: string, bytes: int64, sha256: string>
      child 0, file: string
      child 1, bytes: int64
      child 2, sha256: string
to
{'generatedAt': Value('string'), 'algorithm': Value('string'), 'files': List({'file': Value('string'), 'bytes': Value('int64'), 'sha256': Value('string')})}
because column names don't match
Traceback:    Traceback (most recent call last):
                File "/src/services/worker/src/worker/utils.py", line 149, in get_rows_or_raise
                  return get_rows(
                      dataset=dataset,
                  ...<4 lines>...
                      column_names=column_names,
                  )
                File "/src/libs/libcommon/src/libcommon/utils.py", line 272, in decorator
                  return func(*args, **kwargs)
                File "/src/services/worker/src/worker/utils.py", line 129, in get_rows
                  rows_plus_one = list(itertools.islice(safe_iter(ds, dataset=dataset), rows_max_number + 1))
                File "/src/services/worker/src/worker/utils.py", line 544, in safe_iter
                  yield from ds.decode(False) if ds.features else ds
                File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 3052, in __iter__
                  for key, example in ex_iterable:
                                      ^^^^^^^^^^^
                File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 2559, in __iter__
                  for key, pa_table in self._iter_arrow():
                                       ~~~~~~~~~~~~~~~~^^
                File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 2584, in _iter_arrow
                  for key, pa_table in self.ex_iterable._iter_arrow():
                                       ~~~~~~~~~~~~~~~~~~~~~~~~~~~~^^
                File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 558, in _iter_arrow
                  for key, pa_table in iterator:
                                       ^^^^^^^^
                File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 440, in _iter_arrow
                  for key, pa_table in self.generate_tables_fn(**gen_kwags):
                                       ~~~~~~~~~~~~~~~~~~~~~~~^^^^^^^^^^^^^
                File "/usr/local/lib/python3.14/site-packages/datasets/packaged_modules/json/json.py", line 361, in _generate_tables
                  self._cast_table(pa_table, json_field_paths=json_field_paths),
                  ~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
                File "/usr/local/lib/python3.14/site-packages/datasets/packaged_modules/json/json.py", line 132, in _cast_table
                  pa_table = table_cast(pa_table, self.info.features.arrow_schema)
                File "/usr/local/lib/python3.14/site-packages/datasets/table.py", line 2381, in table_cast
                  return cast_table_to_schema(table, schema)
                File "/usr/local/lib/python3.14/site-packages/datasets/table.py", line 2309, in cast_table_to_schema
                  raise CastError(
                  ...<3 lines>...
                  )
              datasets.table.CastError: Couldn't cast
              version: int64
              generatedAt: timestamp[s]
              languages: struct<ar: struct<total: int64, removed: struct<digit_mix: int64>, kept: int64>, bg: struct<total: i (... 4407 chars omitted)
                child 0, ar: struct<total: int64, removed: struct<digit_mix: int64>, kept: int64>
                    child 0, total: int64
                    child 1, removed: struct<digit_mix: int64>
                        child 0, digit_mix: int64
                    child 2, kept: int64
                child 1, bg: struct<total: int64, removed: struct<digit_mix: int64>, kept: int64>
                    child 0, total: int64
                    child 1, removed: struct<digit_mix: int64>
                        child 0, digit_mix: int64
                    child 2, kept: int64
                child 2, bn: struct<total: int64, removed: struct<digit_mix: int64>, kept: int64>
                    child 0, total: int64
                    child 1, removed: struct<digit_mix: int64>
                        child 0, digit_mix: int64
                    child 2, kept: int64
                child 3, ca: struct<total: int64, removed: struct<numeric: int64, digit_mix: int64, possessive: int64>, kept: int (... 3 chars omitted)
                    child 0, total: int64
                    child 1, removed: struct<numeric: int64, digit_mix: int64, possessive: int64>
                        child 0, numeric: int64
                        child 1, digit_mix: int64
                        child 2, possessive: int64
                    child 2, kept: int64
                child 4, cs: struct<total: int64, removed: struct<numeric: int64, diacritic_single: int64, digit_mix: int64, poss (... 28 chars omitted)
                    child 0, total: int64
                    child 1, removed: struct<numeric: int64, diacritic_single: int64, digit_mix: int64, possessiv
              ...
              critic_single: int64
                        child 2, digit_mix: int64
                        child 3, possessive: int64
                    child 2, kept: int64
                child 38, uk: struct<total: int64, removed: struct<digit_mix: int64>, kept: int64>
                    child 0, total: int64
                    child 1, removed: struct<digit_mix: int64>
                        child 0, digit_mix: int64
                    child 2, kept: int64
                child 39, ur: struct<total: int64, removed: struct<digit_mix: int64>, kept: int64>
                    child 0, total: int64
                    child 1, removed: struct<digit_mix: int64>
                        child 0, digit_mix: int64
                    child 2, kept: int64
                child 40, vi: struct<total: int64, removed: struct<numeric: int64, digit_mix: int64, possessive: int64>, kept: int (... 3 chars omitted)
                    child 0, total: int64
                    child 1, removed: struct<numeric: int64, digit_mix: int64, possessive: int64>
                        child 0, numeric: int64
                        child 1, digit_mix: int64
                        child 2, possessive: int64
                    child 2, kept: int64
                child 41, zh: struct<total: int64, removed: struct<numeric: int64, stroke_single: int64>, kept: int64>
                    child 0, total: int64
                    child 1, removed: struct<numeric: int64, stroke_single: int64>
                        child 0, numeric: int64
                        child 1, stroke_single: int64
                    child 2, kept: int64
              algorithm: string
              files: list<item: struct<file: string, bytes: int64, sha256: string>>
                child 0, item: struct<file: string, bytes: int64, sha256: string>
                    child 0, file: string
                    child 1, bytes: int64
                    child 2, sha256: string
              to
              {'generatedAt': Value('string'), 'algorithm': Value('string'), 'files': List({'file': Value('string'), 'bytes': Value('int64'), 'sha256': Value('string')})}
              because column names don't match

Need help to make the dataset viewer work? Make sure to review how to configure the dataset viewer, and open a discussion for direct support.

Cleaned Word Frequency Lists (43 languages)

Small word-frequency lists for 43 languages, derived from wordfreq 3.0.2 and cleaned for use as offline frequency dictionaries: one gzip-compressed MessagePack file per language, roughly 8 MB in total. These lists are also the direct vocabulary source of vocabradar/WordTranslation, which translates them word by word.

Each list is the upstream wordfreq "small" list for that language — every word that occurs at least once per million tokens (frequency ≥ 10⁻⁶) — with noise removed up front: words written in a foreign script, pure numbers and template-digit tokens, possessive forms, and lone character tokens that are not words.

Current revision: v4 (2026-10-12). Cumulative across all revisions: 3,995 entries removed (1,870 numeric tokens from v3 + 2,125 non-word tokens from the v4 rules). See Version history.

中文版请见下方：中文版说明

Files 

├── README.md
├── clean.js                       # preprocessing script that regenerates data/ from raw wordfreq
├── gen_language_table.js          # regenerates the generated sections in this README
├── meta.json                      # dataset metadata (fields below)
├── files.json                     # per-file size (bytes) and SHA-256
└── data/
    └── small_<code>.msgpack.gz    # one file per language,
                                   # e.g. small_en.msgpack.gz, small_zh.msgpack.gz

meta.json records, per language: total (cumulative raw entries ever processed), kept (words in the current list), and removed (entries dropped, grouped by cleaning rule); the top level carries version and generatedAt. The identity total = kept + Σremoved holds for every language. Counts are cumulative across in-place revisions.

Cleaning rules

  • numeric: numeric words — pure digits and CJK numerals (五百, 二零一), digit-led codes (00k, 1990s, 3rd), and numeric phrases (百分之百, 八成, 第十, 五点)
  • digit_mix: letters mixed with a multi-digit run or a decimal point — news-template noise (a00, 0.00pm, 00x00, ال00); single-digit product codes like mp3, f1, b2b are kept
  • possessive: lowercase-led 's genitives (women's, cnn's) — contractions like it's/who's and capitalized genitives like McDonald's are kept
  • mark_only: tokens made only of Unicode modifier letters (ゝ, ヽ, ー)
  • diacritic_single: a lone non-ASCII Latin letter in a language where it isn't a word (â, ó in en/de/…) — languages where such singles ARE words (fr es it pt ca da nb is sv vi), ASCII singles (a, i), and Hungarian ő are kept (see KEEP_DIACRITIC_SINGLE in clean.js)
  • stroke_single: lone CJK strokes (丶, 丨, 丿)
  • jamo_single: lone Hangul jamo components (never standalone syllables)
  • foreign, pure_num, no_letter: fallback rules — letters outside the language's Unicode-script whitelist, digits without letters, and tokens with neither letters nor digits (symbols, emoji)

In-place revisions: counts in meta.json are cumulative. When re-running on an already-cleaned list (input words ≤ previous total), the previous removed breakdown is merged in; when the input is fresh upstream data (input words > previous total), the counters restart from the new baseline instead of double-counting. Rule changes per revision are documented in Version history and in clean.js.

Regenerating 

# raw input: wordfreq v3.0.2 small lists
# https://github.com/rspeer/wordfreq/tree/v3.0.2/wordfreq/data
npm install msgpack-lite
node clean.js --input <raw-wordfreq-data-dir> --output data
node gen_language_table.js

clean.js reads raw small_<code>.msgpack.gz files from the input directory, applies the cleaning rules, and writes cleaned small_<code>.msgpack.gz files plus meta.json (--output defaults to data). Pointing --input at an already-cleaned directory performs an in-place revision (previous counts are carried over).

Vocabulary lists

Sorted by coverage (fraction of the language's word tokens found in this list). Coverage is computed on the raw (uncleaned) upstream corpus, so the stop-token exclusion — which removes the most frequent function words — accounts for most of the gap from 100%.

code language native name first word cb float words cumulative coverage
ar Arabic العربية في 146 3.47e-2 55611 90.2%
bg Bulgarian български на 127 5.37e-2 35848 91.0%
bn Bengali বাংলা না 199 1.02e-2 33062 93.3%
ca Catalan català de 122 6.03e-2 27028 95.1%
cs Czech čeština a 149 3.24e-2 51744 89.9%
da Danish dansk i 151 3.09e-2 29321 93.2%
de German Deutsch die 152 3.02e-2 38989 90.4%
el Greek Ελληνικά και 149 3.24e-2 43191 89.7%
en English English the 127 5.37e-2 28223 93.5%
es Spanish español de 119 6.46e-2 34740 93.6%
fa Persian فارسی و 140 3.98e-2 30108 94.2%
fi Finnish suomi ja 144 3.63e-2 59143 88.7%
fil Filipino Filipino sa 113 7.41e-2 30098 95.2%
fr French français de 132 4.79e-2 31203 93.9%
he Hebrew עברית את 169 2.04e-2 57405 90.5%
hi Hindi हिन्दी के 142 3.80e-2 24031 93.2%
hu Hungarian magyar a 103 9.33e-2 46525 87.7%
id Indonesian Bahasa Indonesia yang 153 2.95e-2 30967 93.8%
is Icelandic íslenska og 132 4.79e-2 42207 91.6%
it Italian italiano di 141 3.89e-2 35954 93.6%
ja Japanese 日本語 の 128 5.25e-2 27856 91.4%
ko Korean 한국어 이 150 3.16e-2 26807 90.2%
lt Lithuanian lietuvių ir 148 3.31e-2 63973 87.4%
lv Latvian latviešu un 148 3.31e-2 45464 92.1%
mk Macedonian македонски на 123 5.89e-2 32925 93.9%
ms Malay Bahasa Melayu yang 150 3.16e-2 28571 95.9%
nb Norwegian Bokmål norsk bokmål i 152 3.02e-2 25994 94.0%
nl Dutch Nederlands de 127 5.37e-2 28715 93.5%
pl Polish polski w 145 3.55e-2 49664 91.6%
pt Portuguese português de 132 4.79e-2 33123 94.3%
ro Romanian română de 128 5.25e-2 43279 91.8%
ru Russian русский в 137 4.27e-2 59604 86.9%
sh Serbo-Croatian srpskohrvatski / српскохрватски je 136 4.37e-2 54698 88.7%
sk Slovak slovenčina a 147 3.39e-2 59414 88.9%
sl Slovenian slovenščina je 133 4.68e-2 53882 90.6%
sv Swedish svenska är 148 3.31e-2 30515 94.0%
ta Tamil தமிழ் ஒரு 199 1.02e-2 64756 86.2%
th Thai ไทย ที่ 170 2.00e-2 39680 71.0%
tr Turkish Türkçe ve 163 2.34e-2 63130 86.9%
uk Ukrainian українська в 160 2.51e-2 46286 90.0%
ur Urdu اردو کے 137 4.27e-2 22442 96.0%
vi Vietnamese Tiếng Việt là 174 1.82e-2 10519 95.4%
zh Chinese 中文 的 121 6.17e-2 35723 90.7%
total 43 languages — — — — 1,712,418 —

first word is the most frequent word in the list — a quick sanity check that the file is in the right language. cb is the centibel band of that word (upstream stores each word's frequency rounded to the nearest centibel, so band b ≈ 10−b/100 of tokens; float is that value for the top word). words is the list size after cleaning. cumulative coverage estimates the share of running words the list covers, summing 10−b/100 over each band's word count — the shortfall from 100% is mostly the 10⁻⁶ entry cutoff. The total row sums the words column across all lists; the other columns are per-language facts and are not additive.

Language codes present in this dataset:

  • ar
  • bg
  • bn
  • ca
  • cs
  • da
  • de
  • el
  • en
  • es
  • fa
  • fi
  • fil
  • fr
  • he
  • hi
  • hu
  • id
  • is
  • it
  • ja
  • ko
  • lt
  • lv
  • mk
  • ms
  • nb
  • nl
  • pl
  • pt
  • ro
  • ru
  • sh
  • sk
  • sl
  • sv
  • ta
  • th
  • tr
  • uk
  • ur
  • vi
  • zh

Data format

Each file is a gzipped MessagePack array of [word, rank, frequency] triples, where rank is the 1-based position in the frequency-sorted list:

import gzip, msgpack

with gzip.open('data/small_en.msgpack.gz', 'rb') as f:
    header, *bands = msgpack.unpackb(f.read())

# header == {'format': 'cB', 'version': 1}
# band b holds words whose frequency was rounded to centibel b (~10**(-b/100));
# bands concatenated in order yield words sorted by frequency rank.
words = [(w, 10 ** (-b / 100)) for b, band in enumerate(bands) for w in band]
# words[0] ≈ (b'the', 5.37e-2) — rank 1, ~5.4% of tokens

This is the upstream wordfreq "cBpack" encoding: 600 frequency bands per language, with each word's frequency rounded to the nearest centibel. For most applications you only need the rank (band order); the band-midpoint frequency is a tie-breaker and thresholding aid. Each file's size and SHA-256 are listed in files.json.

Version history

  • v4 (2026-10-12): added cleaning rules digit_mix, possessive, mark_only, diacritic_single, stroke_single, jamo_single; meta.json counters became cumulative across in-place revisions (total = kept + Σremoved).
  • v3 (2026-10-10): added the numeric rule (digit-led codes, CJK numerals and numeric phrases are not headwords); meta.json reorganized as languages[lang] = {total, removed, kept}.
  • v1–v2: initial releases of the wordfreq 3.0.2 small lists with Unicode-script whitelist cleaning (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. Whatever you watched can go on to vocabradar.com, which turns it into related exercises and games so the words actually stick.

清洗版词频表（43 语）

43 种语言的小型词频表，源自 wordfreq 3.0.2 并做了清洗，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 8 MB。这些词表也是 vocabradar/WordTranslation 的直接词表来源，后者逐词为它们配上译文。

每个文件都是该语种的 wordfreq "small" 列表 —— 即所有每百万词至少出现一次的词（词频 ≥ 10⁻⁶）—— 并在源头剔除了噪声：外文文字的词、纯数字与数字模板 token、属格形式、以及不成词的单字符 token。

当前修订版：v4（2026-10-12）。 跨全部修订累计剔除 3,995 条（v3 的 numeric 数字词 1,870 条 + v4 六类非词条 token 2,125 条）。见版本历史。

文件结构 

├── README.md
├── clean.js                       # 预处理脚本：从原始 wordfreq 重新生成 data/
├── gen_language_table.js          # 重新生成本 README 中的生成段落
├── meta.json                      # 数据集元信息（字段见下）
├── files.json                     # 每个文件的大小（字节）与 SHA-256
└── data/
    └── small_<code>.msgpack.gz    # 每语种一个文件，
                                   # 如 small_en.msgpack.gz、small_zh.msgpack.gz

meta.json 按语种记录：total（累计处理的原始词条数）、kept（当前列表词数）、removed（按清洗规则分组的剔除数）；顶层含 version 与 generatedAt。所有语种满足恒等式 total = kept + Σremoved。计数为跨就地修订的累计值。

清洗规则

  • numeric：数字词 —— 纯数字与中文数字（五百、二零一）、数字起头短代码（00k、1990s、3rd）、数字短语（百分之百、八成、第十、五点）
  • digit_mix：字母与多位数字串或小数点混合的新闻模板噪声（a00、0.00pm、00x00、ال00）；mp3、f1、b2b 等单数字产品码保留
  • possessive：小写起始的 's 属格（women's、cnn's）—— it's/who's 等缩合与 McDonald's 等大写属格保留
  • mark_only：全部由 Unicode modifier letter（Lm）构成的 token（ゝ、ヽ、ー）
  • diacritic_single：非豁免语种中单个注音拉丁字母（en/de 等语种的 â、ó）—— 单字成词语种（fr es it pt ca da nb is sv vi）、ASCII 单字母（a、i）与匈牙利语 ő 豁免（见 clean.js 的 KEEP_DIACRITIC_SINGLE）
  • stroke_single：单个 CJK 笔画（丶、丨、丿）
  • jamo_single：韩文单个字母部件（不成音节）
  • foreign、pure_num、no_letter：兜底规则 —— 含语种文字白名单外字母、有数字无字母、以及既无字母也无数字（符号、emoji）的 token

就地修订：meta.json 的计数是累计值。对已清洗列表重跑（输入词数 ≤ 上版 total）时，上版 removed 明细会合并累计；输入为新上游原始数据（输入词数 > 上版 total）时，计数从新基线重新起算，避免重复计数。每版规则变化见下方版本历史与 clean.js。

重新生成 

# raw input: wordfreq v3.0.2 small lists
# https://github.com/rspeer/wordfreq/tree/v3.0.2/wordfreq/data
npm install msgpack-lite
node clean.js --input <raw-wordfreq-data-dir> --output data
node gen_language_table.js

clean.js 读取输入目录中的原始 small_<code>.msgpack.gz，应用清洗规则，输出清洗后的 small_<code>.msgpack.gz 与 meta.json（--output 默认为 data）。把 --input 指向已清洗目录即为就地修订（自动携带上版统计）。

词表一览

按覆盖率排序（该语种词 token 中出现在本列表内的比例）。覆盖率在原始（未清洗）上游语料上计算，因此停用虚词排除 —— 它移除的正是最高频的功能词 —— 是距 100% 差距的主要来源。

code language native name first word cb float words cumulative coverage
ar Arabic العربية في 146 3.47e-2 55611 90.2%
bg Bulgarian български на 127 5.37e-2 35848 91.0%
bn Bengali বাংলা না 199 1.02e-2 33062 93.3%
ca Catalan català de 122 6.03e-2 27028 95.1%
cs Czech čeština a 149 3.24e-2 51744 89.9%
da Danish dansk i 151 3.09e-2 29321 93.2%
de German Deutsch die 152 3.02e-2 38989 90.4%
el Greek Ελληνικά και 149 3.24e-2 43191 89.7%
en English English the 127 5.37e-2 28223 93.5%
es Spanish español de 119 6.46e-2 34740 93.6%
fa Persian فارسی و 140 3.98e-2 30108 94.2%
fi Finnish suomi ja 144 3.63e-2 59143 88.7%
fil Filipino Filipino sa 113 7.41e-2 30098 95.2%
fr French français de 132 4.79e-2 31203 93.9%
he Hebrew עברית את 169 2.04e-2 57405 90.5%
hi Hindi हिन्दी के 142 3.80e-2 24031 93.2%
hu Hungarian magyar a 103 9.33e-2 46525 87.7%
id Indonesian Bahasa Indonesia yang 153 2.95e-2 30967 93.8%
is Icelandic íslenska og 132 4.79e-2 42207 91.6%
it Italian italiano di 141 3.89e-2 35954 93.6%
ja Japanese 日本語 の 128 5.25e-2 27856 91.4%
ko Korean 한국어 이 150 3.16e-2 26807 90.2%
lt Lithuanian lietuvių ir 148 3.31e-2 63973 87.4%
lv Latvian latviešu un 148 3.31e-2 45464 92.1%
mk Macedonian македонски на 123 5.89e-2 32925 93.9%
ms Malay Bahasa Melayu yang 150 3.16e-2 28571 95.9%
nb Norwegian Bokmål norsk bokmål i 152 3.02e-2 25994 94.0%
nl Dutch Nederlands de 127 5.37e-2 28715 93.5%
pl Polish polski w 145 3.55e-2 49664 91.6%
pt Portuguese português de 132 4.79e-2 33123 94.3%
ro Romanian română de 128 5.25e-2 43279 91.8%
ru Russian русский в 137 4.27e-2 59604 86.9%
sh Serbo-Croatian srpskohrvatski / српскохрватски je 136 4.37e-2 54698 88.7%
sk Slovak slovenčina a 147 3.39e-2 59414 88.9%
sl Slovenian slovenščina je 133 4.68e-2 53882 90.6%
sv Swedish svenska är 148 3.31e-2 30515 94.0%
ta Tamil தமிழ் ஒரு 199 1.02e-2 64756 86.2%
th Thai ไทย ที่ 170 2.00e-2 39680 71.0%
tr Turkish Türkçe ve 163 2.34e-2 63130 86.9%
uk Ukrainian українська в 160 2.51e-2 46286 90.0%
ur Urdu اردو کے 137 4.27e-2 22442 96.0%
vi Vietnamese Tiếng Việt là 174 1.82e-2 10519 95.4%
zh Chinese 中文 的 121 6.17e-2 35723 90.7%
合计 43 种语言 — — — — 1,712,418 —

first word 是列表中最高频的词 —— 快速确认文件语言正确。cb 是该词所在的 centibel 频段（上游将词频四舍五入到整数 centibel，频段 b ≈ 10−b/100 token 概率；float 即榜首词的该概率值）。words 是清洗后列表大小。cumulative coverage 按各频段词数 × 10−b/100 求和估算覆盖比例 —— 距 100% 的差主要是 10⁻⁶ 收录门槛。末行 合计 为各语种 words 之和；其余列是各语种的独立事实，不可相加。

本数据集包含的语种代码：

  • ar
  • bg
  • bn
  • ca
  • cs
  • da
  • de
  • el
  • en
  • es
  • fa
  • fi
  • fil
  • fr
  • he
  • hi
  • hu
  • id
  • is
  • it
  • ja
  • ko
  • lt
  • lv
  • mk
  • ms
  • nb
  • nl
  • pl
  • pt
  • ro
  • ru
  • sh
  • sk
  • sl
  • sv
  • ta
  • th
  • tr
  • uk
  • ur
  • vi
  • zh

数据格式

每个文件是 gzip 压缩的 MessagePack 数组，沿用上游 wordfreq 的 cBpack 编码（与 wordfreq 库读取的文件相同，仅做了原地清洗）：结构为 [header, band_0..band_599]，频段 b 装有词频四舍五入到 b centibel 的词。

import gzip, msgpack

with gzip.open('data/small_en.msgpack.gz', 'rb') as f:
    header, *bands = msgpack.unpackb(f.read())

# header == {'format': 'cB', 'version': 1}
# 频段 b 装有词频四舍五入到 b centibel 的词（约 10**(-b/100)）；
# 按顺序连接频段即得到按词频排序的词表。
words = [(w, 10 ** (-b / 100)) for b, band in enumerate(bands) for w in band]
# words[0] ≈ (b'the', 5.37e-2) —— 第 1 位，约占 token 的 5.4%

这是上游 wordfreq 的 "cBpack" 编码：每语种 600 个频段，词频四舍五入到最近的 centibel。多数场景只需排名（频段顺序）即可；频段中点概率（10**(-b/100)）供并列打破与阈值过滤使用。各文件的大小与 SHA-256 见 files.json。

版本历史

  • v4（2026-10-12）：新增清洗规则 digit_mix、possessive、mark_only、diacritic_single、stroke_single、jamo_single；meta.json 计数改为跨就地修订累计（total = kept + Σremoved）。
  • v3（2026-10-10）：新增 numeric 规则（数字起头代码、中文数字与数字短语不是词条）；meta.json 重组为 languages[lang] = {total, removed, kept}。
  • v1–v2：初始发布 —— wordfreq 3.0.2 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； 看过的内容还能送到 vocabradar.com，自动生成相关习题与游戏，把词真正记住。
Downloads last month
654
Total file size:
7.92 MB