Help

Difference between revisions of "How to create a frequency list?"

Words lists sorted by frequency are a very good way to cover one language methodically. After reading this page you will be able to find or create your own frequency list, clean and split it into easy-to-handle files.

 
(11 intermediate revisions by 3 users not shown)
Line 33: Line 33:
  
 
=== UNILEX's lists ===
 
=== UNILEX's lists ===
UNILEX is an Unicode Consortium project which curates 999 languages. As many frequency lists are available under GNU-like license. This data requires minor clean up, example with Igbo (<code>ig</code>) :
+
UNILEX is an Unicode Consortium project which curates 1001 languages. As many frequency lists are available under GNU-like license. This data requires minor clean up, example with Igbo (<code>ig</code>) :
 
<pre>
 
<pre>
 
mkdir -p ./clean                                                              # create a folder
 
mkdir -p ./clean                                                              # create a folder
google-chrome github.com/lingua-libre/unilex/tree/master/data/frequency       # open in web-browse to browse available languages
+
google-chrome https://github.com/unicode-org/unilex/tree/main/data/frequency # open in web-browse to browse available languages
 
iso=ig                                                                        # defined your target language
 
iso=ig                                                                        # defined your target language
curl https://raw.githubusercontent.com/lingua-libre/unilex/master/data/frequency/${iso}.txt | tail -n +5 | sort -k 2,2 -n -r | cut -d$'\t' -f1 | sed -E 's/^/# /g' > ./clean/${iso}-all.txt
+
curl https://raw.githubusercontent.com/unicode-org/unilex/main/data/frequency/${iso}.txt | tail -n +5 | sort -k 2,2 -n -r | cut -d$'\t' -f1 | sed -E 's/^/# /g' > ./clean/${iso}-all.txt
 
# download, remove first 5 lines, sort by 2nd column numerical value descendant, cut and keep first field, add # to make a list, print all to file.
 
# download, remove first 5 lines, sort by 2nd column numerical value descendant, cut and keep first field, add # to make a list, print all to file.
 
split -d -l 5000  --additional-suffix=".txt" ./clean/${iso}-all.txt ./clean/${iso}-words-by-frequency-
 
split -d -l 5000  --additional-suffix=".txt" ./clean/${iso}-all.txt ./clean/${iso}-words-by-frequency-
Line 66: Line 66:
 
Some research centers are curating the web to provide large corpura to linguists and netizens alike.  
 
Some research centers are curating the web to provide large corpura to linguists and netizens alike.  
  
* Corpus Crawler is an open-source crawler for building corpora in 1000+ languages (easy to add more, send pull requests); https://github.com/googlei18n/corpuscrawler
+
{| class="wikitable sortable"
* P. Lison and J. Tiedemann (2016), ''"OpenSubtitles2016: Extracting Large Parallel Corpora from Movie and TV Subtitles"'', http://stp.lingfil.uu.se/~joerg/paper/opensubs2016.pdf . In Proceedings of the 10th International Conference on Language Resources and Evaluation (LREC 2016)
+
! Project introduction || Type || Languages (2024) || Portal all || Language specific || Download link || Comments
** "The corpora is made freely available to the research community on the OPUS website" − Lison and Tiedemann (2016).
+
 
** http://opus.nlpl.eu/OpenSubtitles2018.php
+
|-
 +
| [https://github.com/googlei18n/corpuscrawler Google Corpus Crawler] is an open-source crawler to building corpora || crawler (Python)<br>sentences (if run)<br>frequency list || 1000+ languages || [https://github.com/googlei18n/corpuscrawler home] || n.a. ||[https://www.gstatic.com/i18n/corpora/wordcounts/aai.txt aai] (freq) || Python 3 project. Easy to add a crawler, send pull requests.
 +
 
 +
|-
 +
| OpenSubtitles 2016/2018<br> || Subtitles<br>Parallel sentences<br>Monolingual sentences || 75 || [https://opus.nlpl.eu/OpenSubtitles/corpus/version/OpenSubtitles Portal] || [https://opus.nlpl.eu/OpenSubtitles/br&en/v2018/OpenSubtitles `br&en`] || [https://object.pouta.csc.fi/OPUS-OpenSubtitles/v2018/mono/bre.txt.gz bre] (mono) || '''Source:''' * P. Lison and J. Tiedemann (2016), ''"OpenSubtitles2016: Extracting Large Parallel Corpora from Movie and TV Subtitles"'', http://stp.lingfil.uu.se/~joerg/paper/opensubs2016.pdf .
 +
'''Licence:''' unclear, "The corpora is made freely available to the research community on the OPUS website" − Lison and Tiedemann (2016).  
 +
 
 +
|-
 +
| Wortschatz by Leipzig || Sentences<br>Monolingual || 290+ || || [https://wortschatz.uni-leipzig.de/en/download/bre bre] || [https://downloads.wortschatz-leipzig.de/corpora/bre_wikipedia_2021_100K.tar.gz bre]: 100k sentences, WP 2021 || List of sentences corpora : [https://api.wortschatz-leipzig.de/ws/swagger-ui/index.html#/Corpora/getAvailableCorpora API reference] > https://api.wortschatz-leipzig.de/ws/corpora
 +
 
 +
|-
 +
| CC-100 || Sentences<br>Monolingual || 115 || [https://data.statmt.org/cc-100/ Portal] || n.a. || [https://data.statmt.org/cc-100/br.txt.xz br] (mono) || « No claims of intellectual property are made on the work of preparation of the corpus. »
 +
|}
  
 
==== Wiki(p)edia dumps ====
 
==== Wiki(p)edia dumps ====
Line 76: Line 88:
 
* [https://github.com/google/corpuscrawler/issues/78 Discussion on harvesting Wikipedia]
 
* [https://github.com/google/corpuscrawler/issues/78 Discussion on harvesting Wikipedia]
  
== From corpus to frequency data `{occurences} {item}` ==
+
== From corpus to frequency data `{occurrences} {item}` ==
 
Main tools will be <code>[https://ss64.com/bash/grep.html grep]</code> to grab the text strings, <code>[https://ss64.com/bash/awk.html awk]</code> to count them, <code>[https://ss64.com/bash/sort.html sort]</code> to sort and rank them.
 
Main tools will be <code>[https://ss64.com/bash/grep.html grep]</code> to grab the text strings, <code>[https://ss64.com/bash/awk.html awk]</code> to count them, <code>[https://ss64.com/bash/sort.html sort]</code> to sort and rank them.
  
Line 84: Line 96:
 
</pre>
 
</pre>
  
==== Space-separated Words frequency (+sorted): ====
+
==== Space-separated Words frequency (+sorted) ====
 
<pre class="console">
 
<pre class="console">
 +
# Spaces or punctuation separated words
 
$ grep -o '\w*' longtext.txt | awk '{a[$1]++}END{for(k in a)print a[k],k}' | sort -n -r -t' ' -k1,1  &gt; sorted-words.txt
 
$ grep -o '\w*' longtext.txt | awk '{a[$1]++}END{for(k in a)print a[k],k}' | sort -n -r -t' ' -k1,1  &gt; sorted-words.txt
# or  
+
# Space or punctuation separated words, except if punctuation is : '-
 +
cat longtext.txt | sed 's/^\(.\)/\L\1/' | sed -E "s/((['-]*\w+)*)/\1\n/gi" | sed -E "s/ //g" | awk '{a[$1]++}END{for(k in a)print a[k],k}' | sort -n -r -t' ' -k1,1  &gt; sorted-words.txt
 +
# Space or line jump separated words
 
$ cat longtext.txt | awk '{a[$1]++}END{for(k in a)print a[k],k}' RS=" |\n" | sort -n -r -t' ' -k1,1 &gt; sorted-words.txt
 
$ cat longtext.txt | awk '{a[$1]++}END{for(k in a)print a[k],k}' RS=" |\n" | sort -n -r -t' ' -k1,1 &gt; sorted-words.txt
 
</pre>
 
</pre>

Latest revision as of 19:49, 17 February 2024

Reminder : to start a recording session you need
  1. One LinguaLibre user,
  2. One willing speaker, and
  3. One list of items to record with one item by line.
    One item can be any easy to read sign, word, sentence or paragraph. The most common use-case is to record a comprehensive words list for your target language.

Reusing open license frequency lists

Hermite Dave's lists

Hermite Dave created 61 frequency lists from OpenSubtitle data, covering most major languages under CC license. This data requires minor clean up, example with Korean (ko) :

mkdir -p ./clean                                                              # create a folder
google-chrome github.com/hermitdave/FrequencyWords/tree/master/content/2018   # open in web-browse to browse available languages
iso=ko                                                                        # defined your target language
curl https://raw.githubusercontent.com/hermitdave/FrequencyWords/master/content/2018/${iso}/${iso}_50k.txt | sort -k 2,2 -n -r | cut -d' ' -f1 | sed -E 's/^/# /g' > ./clean/${iso}-all.txt
# download, sort by 2nd column numerical value descendant, cut by space then keep first field, add # to make a list, print all to file.
split -d -l 5000  --additional-suffix=".txt" ./clean/${iso}-all.txt ./clean/${iso}-words-by-frequency-
# split in files of 5000 items

On LinguaLibre.org, create your lists as List:{Iso3}/words-by-frequency-00001-to-5000, etc. Ex. List:Pol/words-by-frequency-00001-to-02000.

After creating the list on LinguaLibre, add the following to its talkpage:

==== Source ====
{{Hermite Dave}} 

UNILEX's lists

UNILEX is an Unicode Consortium project which curates 1001 languages. As many frequency lists are available under GNU-like license. This data requires minor clean up, example with Igbo (ig) :

mkdir -p ./clean                                                              # create a folder
google-chrome https://github.com/unicode-org/unilex/tree/main/data/frequency  # open in web-browse to browse available languages
iso=ig                                                                        # defined your target language
curl https://raw.githubusercontent.com/unicode-org/unilex/main/data/frequency/${iso}.txt | tail -n +5 | sort -k 2,2 -n -r | cut -d$'\t' -f1 | sed -E 's/^/# /g' > ./clean/${iso}-all.txt
# download, remove first 5 lines, sort by 2nd column numerical value descendant, cut and keep first field, add # to make a list, print all to file.
split -d -l 5000  --additional-suffix=".txt" ./clean/${iso}-all.txt ./clean/${iso}-words-by-frequency-
# split in files of 5000 items

On LinguaLibre.org, create your lists as List:{Iso3}/words-by-frequency-00001-to-5000, etc. Ex. List:Pol/words-by-frequency-00001-to-02000.

After creating the list on LinguaLibre, add the following to its talkpage:

==== Source ====
{{UNILEX License}} 

Subtlex's lists

The Subtlex movement, a group of academic frequency list studies based on open subtitles, also provides about 10 of the highest quality frequency lists. Items are better cleaned up, etc. These resources are published under various licenses. Their usage on LinguaLibre must be on a case by case basis.

Corpus

Requirements for relevant corpus :

  • Size: 2M+ words.
  • Type: raw text.
  • Language: monolingual or close to be.

Download a corpus

You can download available corpuses in your language or collect your own corpus via some datamining. Corpura are easily available for about 60 languages. Corpuses for rare language are likely missing, you will likely have to do some data mining.

Some research centers are curating the web to provide large corpura to linguists and netizens alike.

Project introduction Type Languages (2024) Portal all Language specific Download link Comments
Google Corpus Crawler is an open-source crawler to building corpora crawler (Python)
sentences (if run)
frequency list
1000+ languages home n.a. aai (freq) Python 3 project. Easy to add a crawler, send pull requests.
OpenSubtitles 2016/2018
Subtitles
Parallel sentences
Monolingual sentences
75 Portal `br&en` bre (mono) Source: * P. Lison and J. Tiedemann (2016), "OpenSubtitles2016: Extracting Large Parallel Corpora from Movie and TV Subtitles", http://stp.lingfil.uu.se/~joerg/paper/opensubs2016.pdf .

Licence: unclear, "The corpora is made freely available to the research community on the OPUS website" − Lison and Tiedemann (2016).

Wortschatz by Leipzig Sentences
Monolingual
290+ bre bre: 100k sentences, WP 2021 List of sentences corpora : API reference > https://api.wortschatz-leipzig.de/ws/corpora
CC-100 Sentences
Monolingual
115 Portal n.a. br (mono) « No claims of intellectual property are made on the work of preparation of the corpus. »

Wiki(p)edia dumps

One possibility is to harvest Wikipedia's contents. See:

From corpus to frequency data `{occurrences} {item}`

Main tools will be grep to grab the text strings, awk to count them, sort to sort and rank them.

Characters frequency (+sorted)

$ grep -o '\S' longtext.txt | awk '{a[$1]++}END{for(k in a)print a[k],k}' | sort -n -r -t' ' -k1,1 > sorted-letters.txt

Space-separated Words frequency (+sorted)

# Spaces or punctuation separated words
$ grep -o '\w*' longtext.txt | awk '{a[$1]++}END{for(k in a)print a[k],k}' | sort -n -r -t' ' -k1,1  > sorted-words.txt
# Space or punctuation separated words, except if punctuation is : '-
cat longtext.txt | sed 's/^\(.\)/\L\1/' | sed -E "s/((['-]*\w+)*)/\1\n/gi" | sed -E "s/ //g" | awk '{a[$1]++}END{for(k in a)print a[k],k}' | sort -n -r -t' ' -k1,1  > sorted-words.txt
# Space or line jump separated words
$ cat longtext.txt | awk '{a[$1]++}END{for(k in a)print a[k],k}' RS=" |\n" | sort -n -r -t' ' -k1,1 > sorted-words.txt

Loop on all .txt, recursively within folders

find -iname '*.txt' -exec cat {} \; | grep -o '\w*' | awk '{a[$1]++}END{for(k in a)print a[k],k}' | sort -n -r -t' ' -k1,1 > sorted-words.txt

Output

39626 aš
35938 ir
33361 tai
28520 tu'21th
26213 kad'toto
...

Cleaning up frequency lists

Most sources provide wordlists with {number_of_apparitions}{separator}{item} or its mirror {item}{separator}{number_of_apparitions}, already sorted from most frequent to less ones. We what to keep the field {item} and drop the {separator} and {number_of_apparitions}.

Input data we have Output data we want
$ cat frequency-list.txt
39626 aš
35938 ir
33361 tai
28520 tu'21th
26213 kad'toto
...
$ cat words-list.txt
# aš
# ir
# tai
# tu'21th
# kad'toto
# ...
Command
cut frequency-list.txt -d$' ' -f2 | sed -E 's/^/# /g' > words-list.txt
# load file line by line, cut by space then keep field 2, replace start of line by # on all lines, print in file.

This final result is what you want for LinguaLibre Help:Create your own lists.

Additional helpers

Sort command

See man sort for details.

-n: numeric sort
-r: reverse (descending)
-t: changes field separator to ' ' character
-k: as -k:1,1, sort key starts on field 1 and ends on field 1

Counting lines of a file

wc -l filename.txt       # -l : lines

See sample of a file

head -n 50 filename.txt       # -n : number of line

Splitting a very long file

split -d -l 2000 --additional-suffix=".txt" YUE-words-by-frequency.txt  YUE-words-by-frequency-

Words-lists files generally are be over 10k lines long, thus not convenient to run recording sessions. Given 1000 recordings per hour via LinguaLibre and 3 hours sessions being quite good and intense, we recommend sub-files of :

  • 1000 lines, so you use 1, 2 or 3 files per session
  • 3000 lines, so you use 1 file per session and kill it off like a warrior ... if your speaker and yourself survives.

See How to split a large text file into smaller files with equal number of lines in terminal?

Convert encoding

iconv -f "GB18030" -t "UTF-8" SUBTLEX-CH-WF.csv -o $iso2-words.txt

Create frequency list from en:Dragon

See also 101 Wikidata/Wikipedia API via JS
curl 'https://en.wikipedia.org/w/api.php?action=query&titles=Dragon&prop=extracts&explaintext&redirects&converttitles&callback=?&format=xml' | tr '\040' '\012' | sort | uniq -c | sort -k 1,1 -n -r > output.txt

How to compare lists ?

[Section status: Draft, to continue.] (example).
 comm - compare two sorted files line by line

See also

Lingua Libre Help pages
General help pages Help:InterfaceHelp:Your first recordHelp:Choosing a microphoneHelp:Configure your microphoneHelp:TranslateHelp:LangtagsLinguaLibre:Language codes systems used across LinguaLibreLinguaLibre:List of languages
Linguistic help pages Help:Add a new languageHelp:HomographsHelp:List translationHelp:Ethics
Lists help pages Help:Create your own listsHelp:How to create a frequency list?Help:Why wordlists matter?Help:Swadesh listsHelp:ListsHelp:Create a new generator
Events, Outreach Lingualibre:EventsLingualibre:RolesLingualibre:WorkshopsLingualibre:HackathonLingualibre:Interested communitiesLingualibre:Events/2022 Public Relations CampaignLingualibre:MailingLingualibre:JargonLingualibre:AppsLingualibre:CitationsService civique 2022-2023
Strategy Lingualibre 2022 Review (including outreach)2022-2023 Lingualibre wishlist • {{Wikimedia Language Diversity/Projects}} • Speakers map • Voices gender • StatsLingua Libre SignIt/2022 report • {{Grants}}