Skip to main content

Wiktionary dump file parser and multilingual data extractor

Project description

Wiktextract

This is a utility and Python package for for extracing data from Wiktionary.

Version 1.99.3 is now on pypi and available for installation using pip (Python3). Think of it as a beta version for 2.0.0.

The release won't address everything. Especially linkages (hypernyms, hyponyms, etc.) and disambiguation have issues that I'm postponing to future releases. This release is still a major improvement and should handle non-English languages much better and be much more maintainable. Technically it is a nearly full rewrite and should now handle almost any templates as well as text and encoding generated by Lua modules.

Please report issues on github and I'll try to address them reasonably soon.

The current extracted version is available for browsing and download at: https://kaikki.org/dictionary/. I plan to maintain an automatically updating version of the data at this location. For most people the preferred way to get the extracted Wiktionary data will be to just take it from the web site.

Note: extracting all data for all languages from English Wiktionary takes about 7 hours on a modern 24-core desktop. You may want to download the pre-extracted data rather than run it yourself. If you run it yourself, be prepared to wait from several hours to a couple of days, depending on your computer. Expanding Lua modules is not cheap, but it enables superior extraction quality and maintainability!

Overview

This is a Python package and tool for extracting information from English Wiktionary (enwiktionary) data dumps. Note that the English Wiktionary contains extensive dictionaries and inflectional information for many languages, not just English. Only its glosses and internal tagging are in English.

One thing that distinguishes this tool from any system I'm aware of is that this tool expands templates and Lua macros in Wiktionary. That enables much more accurate rendering and extraction of glosses, word senses, inflected forms, and pronunciations. It also makes the system much easier to maintain. All this results in much higher extraction quality and accuracy.

This tool extracts glosses, parts-of-speech, declension/conjugation information when available, translations for all languages when available, pronunciations (including audio file links), qualifiers including usage notes, word forms, links between words including hypernyms, hyponyms, holonyms, meronyms, related words, derived terms, compounds, alternative forms, etc. Links to Wikipedia pages, Wikidata identifiers, and other such data are also extracted when available. For many classes of words, a word sense is annotated with specific information such as what word it is a form of, what is the RGB value of the color it represents, what is the numeric value of a number, what SI unit it represents, etc.

This tool extracts information for all languages that have data in the English wiktionary. It also extracts translingual data and information about characters (anything that has an entry in Wiktionary).

This tool reads the enwiktionary-<date>-pages-articles.xml.bz2 dump file and outputs JSON-format dictionaries containing most of the information in Wiktionary. The dump files can be downloaded from https://dumps.wikimedia.org.

This utility will be useful for many natural language processing, semantic parsing, machine translation, and language generation applications both in research and industry.

The tool can be used to extract machine translation dictionaries, language understanding dictionaries, semantically annotated dictionaries, and morphological dictionaries with declension/conjugation information (where this information is gavailable for the target language). Dozens of languages have extensive vocabulary in enwiktionary, and several thousand languages have partial coverage.

The wiktwords script makes extracting the information for use by other tools trivial without writing a single line of code. It extracts the information specified by command options for languages specified on the command line, and writes the extracted data to a file or standard output in JSON format for processing by other tools.

While there are currently no active plans to support parsing non-English wiktionaries, I'm considering it. Now that this builds on wikitextprocessor and expands templates and Lua macros, it would be fairly straightforward to build support for other languages too - and even make the extraction configurable so that only a configuration file would need to be created for a processing a Wiktionary in a new language.

As far as we know, this is the most comprehensive tool available for extracting information from Wiktionary as of December 2020.

If you find this tool and/or the pre-extracted data helpful, please give this a star on github!

Pre-extracted data

For most people, it may be easiest to just download pre-expanded data. Please see https://kaikki.org/dictionary/. There is a download link at the bottom of every page. You can download all data, data for a specific language, just a single word, or a list of related words (e.g., a particular part-of-speech or words relating to a particular topic or having a particular inflectional form). All downloads are in JSON format (each line is a separate JSON object). The bigger downloads are also available in compressed form.

Some people have asked for the data as a single JSON object. I've decided to keep it as a JSON object per line, because loading all the data into Python requires 40-50 GB of memory. It is much easier to process the data line-by-line, especially if you are only interested in part of the information. You can easily read the files using the following code:

import json
...
with open("filename.json", "r") as f:
    for line in f:
        data = json.loads(line)
        ... parse the data for this record

If you want to collect all the data into a list, you can read the file into a list with:

import json
...
lst = []
with open("filename.json", "r") as f:
    for line in f:
        data = json.loads(line)
        lst.append(data)

You can also easily pretty-print the data into more human-readable form using:

print(json.dumps(data, indent=2, sort_keys=True))

Here is a pretty-printed example of an extracted word entry for the word thrill as an English verb:

{
  "categories": [
    "Emotions"
  ],
  "derived": [
    {
      "word": "enthrill"
    }
  ],
  "forms": [
    {
      "form": "thrills",
      "tags": [
        "present",
        "simple",
        "singular",
        "third-person"
      ]
    },
    {
      "form": "thrilling",
      "tags": [
        "present"
      ]
    },
    {
      "form": "thrilled",
      "tags": [
        "participle",
        "past",
        "simple"
      ]
    }
  ],
  "heads": [
    {
      "template_name": "en-verb"
    }
  ],
  "lang": "English",
  "lang_code": "en",
  "pos": "verb",
  "senses": [
    {
      "glosses": [
        "To suddenly excite someone, or to give someone great pleasure; to electrify; to experience such a sensation."
      ],
      "tags": [
        "ergative",
        "figuratively"
      ]
    },
    {
      "glosses": [
        "To (cause something to) tremble or quiver."
      ],
      "tags": [
        "ergative"
      ]
    },
    {
      "glosses": [
        "To perforate by a pointed instrument; to bore; to transfix; to drill."
      ],
      "tags": [
        "obsolete"
      ]
    },
    {
      "glosses": [
        "To hurl; to throw; to cast."
      ],
      "tags": [
        "obsolete"
      ]
    }
  ],
  "sounds": [
    {
      "ipa": "/\u03b8\u0279\u026al/"
    },
    {
      "ipa": "[\u03b8\u027e\u032a\u030a\u026a\u026b]",
      "tags": [
        "UK",
        "US"
      ]
    },
    {
      "ipa": "[\u03b8\u027e\u032a\u030a\u026al]",
      "tags": [
        "Ireland"
      ]
    },
    {
      "ipa": "[t\u032a\u027e\u032a\u030a\u026al]",
      "tags": [
        "Ireland"
      ]
    },
    {
      "rhymes": "-\u026al"
    },
    {
      "audio": "en-us-thrill.ogg",
      "tags": [
        "US"
      ],
      "text": "Audio (US)"
    }
  ],
  "translations": [
    {
      "code": "nl",
      "lang": "Dutch",
      "sense": "suddenly excite someone, or to give someone great pleasure; to electrify",
      "word": "opwinden"
    },
    {
      "code": "fi",
      "lang": "Finnish",
      "sense": "suddenly excite someone, or to give someone great pleasure; to electrify",
      "word": "syk\u00e4hdytt\u00e4\u00e4"
    },
    {
      "code": "fi",
      "lang": "Finnish",
      "sense": "suddenly excite someone, or to give someone great pleasure; to electrify",
      "word": "riemastuttaa"
    },
...
    {
      "code": "tr",
      "lang": "Turkish",
      "sense": "slight quivering of the heart that accompanies a cardiac murmur",
      "word": "\u00e7arp\u0131nt\u0131"
    }
  ],
  "wikipedia": [
    "thrill"
  ],
  "word": "thrill"
}

Getting started

Installing

Preparation: on Linux (example from Ubuntu 20.04), you may need to first install the build-essential and python3-dev packages with apt update && apt install build-essential python3-dev python3-pip.

To install wiktextract, use pip (or pip3, as appropriate):

pip3 install wiktextract

Alternatively, you can get the latest development version from githup:

git clone https://github.com/tatuylonen/wiktextract.git
cd wiktextract && pip3 install -r requirements.txt && pip3 install -e .

This will install the wiktextract package and the wiktwords script.

This software requires Python 3.

Running tests

This package includes tests written using the unittest framework. They can be run using, for example, nose, which can be installed using pip3 install nose.

To run the tests, just use the following command in the top-level directory:

nosetests

(Unfortunately the test suite for wiktextract is not yet very comprehensive. The underlying lower-level toolkit, wikitextprocessor, has much more extensive test coverage.)

Expected performance

Extracting all data for all languages from English Wiktionary takes about 7 hours on a modern (year 2020) 24-core desktop (AMD EPYC 7402) or three hours on a 128-core dual AMD EPYC 7702 system. You may want to download the pre-extracted data rather than run it yourself unless you have special needs or want to modify the code. If you run it yourself, be prepared to wait from several hours to a couple of days, depending on your computer. While you can, you may not want to run this on a laptop. Expanding Lua modules is not cheap, but it enables superior extraction quality and maintainability! Also, the software needs 3-4 GB of memory per process. You can control the number of parallel processes to use with the --num-threads option; the default on Linux is to use the number of available cores/hyperthreads. On Windows and MacOS, --num-threads should currently be set to 1 (default on those systems).

Using the command-line tool

The wiktwords script is the easiest way to extract data from Wiktionary. Just download the data dump file from dumps.wikimedia.org and run the script. The correct dump file the name enwiktionary-<date>-pages-articles.xml.bz2.

An example of a typical invocation for extracting all data would be:

wiktwords --all --all-languages --out data.json enwiktionary-20201201-pages-articles.xml.bz2

If you wish to modify the code or test processing individual pages, the following may also be useful:

  1. To extract all pages from Wiktionary into separate files under pages/ and to create a cache file that you can use for quickly processing individual pages:
wiktwords --cache-file /tmp/wikt-cache --pages-dir pages enwiktionary-20201201-pages-articles.xml.bz2
  1. To process a single page, processing a human-readable output file for debugging:
wiktwords --cache-file /tmp/wikt-cache --all --all-languages --out outfile --page pages/Words/di/dictionary.txt

The following command-line options can be used to control its operation:

  • --out FILE: specifies the name of the file to write (specifying "-" as the file writes to stdout)
  • --all-languages: extract words for all available languages
  • --language LANGUAGE: extracts the given language (this option may be specified multiple times; by default, English and Translingual words are extracted)
  • --list-languages: prints a list of supported language names
  • --all: causes all data to be captured for the selected languages
  • --translations: causes translations to be captured
  • --pronunciation: causes pronunciation information to be captured
  • --linkages: causes linkages (synonyms etc.) to be captured
  • --redirects: causes redirects to be extracted
  • --pages-dir DIR: save all wiktionary pages under this directory (mostly for debugging)
  • --cache CACHE: save/use cache file(s) from this path
  • --num-threads THREADS: use this many parallel processes (needs 4GB/process)
  • --human-readable: print human-readable JSON with indentation (no longer machine-readable)
  • --override PATH: override a page or Lua module by this file (first line should be TITLE: pagetitle)
  • --help: displays help text (with some more options than listed here)

Calling the library

While this package has been mostly intended to be used using the wiktwords program, it is also possible to call this as a library. Underneath, this uses the wikitextprocessor module.

This code can be called from an application as follows:

from wiktextract import (WiktionaryConfig, parse_wiktionary, parse_page,
                         PARTS_OF_SPEECH)
from wikitextprocessor import Wtp, ALL_LANGUAGES

config = WiktionaryConfig(
             capture_languages=["English", "Translingual"],
             capture_translations=True,
             capture_pronunciation=True,
             capture_linkages=True,
             capture_compounds=True,
             capture_redirects=True)
ctx = Wtp()

def word_cb(data):
    # data is dictionary containing information for one word/redirect
    ... do something with data

parse_wiktionary(ctx, path, config, word_cb)

def parse_wiktionary(ctx, path, config, word_cb, capture_cb=None, phase1_only=False)

The parse_wiktionary function will call word_cb(data) for words and redirects found in the Wiktionary dump. data is information about a single word and part-of-speech as a dictionary and may include several word senses. It may also be a redirect (indicated by the presence of a "redirect" key in the dictionary). It is in the same format as the JSON-formatted dictionaries returned by the wiktwords tool.

Its arguments are as follows:

  • ctx (Wtp) - a wikitextprocessor processing context. The number of parallel processes to use can be given as the num_threads argument to the constructor, and a cache file path can be provided as the cache_file argument.
  • path (str) - path to a Wiktionary dump file (*-pages-articles.xml.bz2)
  • config (WiktionaryConfig) - a configuration object describing what to exctract (see below)
  • word_cb (function) - this function will be called for every word extracted from Wiktionary. The argument is a dictionary. Typically it will be called once for each word form and part-of-speech (each time there may be more than one word sense under "senses"). See below for a description of the dictionary.
  • capture_cb (function) - this can be None or a function to be called as capture_cb(model, title, text) for every page before extracting any words from it. It can be used to extract raw pages to disk. The model argument is wikitext for normal pages, Scribunto for Lua modules, and redirect for redirects (other values are also possible). title is page title and text is page content or page title to redirect to.
  • phase1_only - if this is set to True, then only a cache file will be created but no extraction will take place. In this case the Wtp constructor should probably be given the cache_file argument when creating ctx.

This call gathers statistics in config. This function will automatically parallelize the extraction. page_cb will be called in the parent process, however.

def parse_page(ctx, title, text, config)

This function parses text as if it was a Wiktionary page with the title title. The arguments are:

  • ctx (Wtp) - a wikitextprocessor context
  • title (str) - the title to use for the page
  • text (str) - contents of the page (wikitext)
  • config (WiktionaryConfig) - specifies what to capture and is also used for collecting statistics

PARTS_OF_SPEECH

This is a constant set of all part-of-speech values (pos key) that may occur in the extracted data. Note that the list is somewhat larger than what a conventional part-of-speech list would be.

class WiktionaryConfig(object)

The WiktionaryConfig object is used for specifying what data to collect from Wiktionary and is also used for collecting statistics during extraction.

The constructor is called as:

WiktionaryConfig(capture_languages=["English", "Translingual",
                 capture_translations=False,
                 capture_pronunciation=False,
                 capture_linkages=False,
                 capture_compounds=False,
                 capture_redirects=False,
                 capture_examples=False)

The arguments are as follows:

  • capture_languages (list/tuple/set of strings) - names of languages for which to capture data. It defaults to ["English", "Translingual"]. To capture all languages, one can use set(x["name"] for x in ALL_LANGUAGES) (with ALL_LANGUAGES imported from wikitextprocessor).
  • capture_translations (boolean) - set to True to capture translation information for words. Translation information seems to be most widely available for the English language, which has translations into other languages.
  • capture_pronunciation (boolean) - set to True to capture pronunciation information for words. Typically, this includes IPA transcriptions and any audio files included in the word entries, along with other information. The type and amount of pronunciation information varies widely between languages.
  • capture_linkages (boolean) - set to True to capture linkages between word, such as hypernyms, antonyms, synonyms, etc.
  • capture_compounds (boolean) - set to True to capture compound words containing the word.
  • capture_redirects (boolean) - set to True to capture redirects. Redirects are not associated with any specific language and thus requesting them returns them for all words in all languages.
  • capture_examples (boolean) - set to True to capture usage examples (XXX currently not implemented).

Format of extracted redirects

Some pages in Wiktionary are redirects. For these, word_cb will be called with data in a special format. In this case, the dictionary will have a redirect key, which will contain the page title that the entry redirects to. The title key contains the word/term that contains the redirect. Redirect entries do not have pos or any of the other fields. Redirects also are not associated with any language, so all redirects are always returned regardless of the captured languages (if extracting redirects has been requested).

Format of the extracted word entries

Information returned for each word is a dictionary. The dictionary has the following keys (others may also be present or added later):

  • word - the word form
  • pos - part-of-speech, such as "noun", "verb", "adj", "adv", "pron", "determiner", "prep" (preposition), "postp" (postposition), and many others. The complete list of possible values returned by the package can be found in wiktextract.PARTS_OF_SPEECH.
  • lang - name of the language this word belongs to (e.g., English)
  • lang_code - Wiktionary language code (e.g., en)
  • senses - list of word senses (dictionaries) for this word/part-of-speech (see below)
  • forms - list of inflected or alternative forms specified for the word (e.g., plural, comparative, superlative, roman script version). This is a list of dictionaries, where each dictionary has a word key and a tags key. The tags identify what type of form it is.
  • sounds - list of dictionaries containing pronunciation, hyphenation, rhyming, and related information. Each dictionary may have a tags key containing tags that clarify what kind of form that entry is. Different types of information are stored in different fields: ipa is IPA pronunciation, enPR is enPR pronunciation, audio is name of sound file in Wikimedia commons.
  • translations - non-disambiguated translation entries (see below)
  • synonyms - non-disambiguated synonym linkages for the word (see below)
  • antonyms - non-disambiguated antonym linkages for the word (see below)
  • hypernyms - non-disambiguated hypernym linkages for the word (see below)
  • holonyms - non-disambiguated linkages indicating being part of something (see below) (not systematically encoded)
  • meronyms - non-disambiguated linkages indicating having a part (see below) (fairly rare)
  • derived - non-disambiguated derived word linkages for the word (see below)
  • related - non-disambiguated related word linkages for the word (see below)
  • wikidata - non-disambiguated Wikidata identifer
  • wiktionary - non-disambiguated page title in Wikipedia (possibly prefixed by language id)
  • categories - list of non-disambiguated categories for the word
  • topics - list of non-disambiguated topics for the word
  • inflection - conjugation and declension entries found for the word, as dictionaries. These basically capture the language-specific inflection template for the word.
  • heads: part-of-speech specific head tags for the word. This basically just captures the templates (their name and arguments) as a list of dictionaries. Most applications may want to ignore this.

There may also be other fields.

Note that several of the field on the word entry level indicate information that has not been sense-disambiguated. Such information may apply to one or more of the senses. Currently only the most trivial cases are disambiguated; however, it is anticipated that more disambiguation may be performed in the future. It is also possible for the same key to be provided in a sense and in the word entry; in that case, the data in the sense has been sense-disambiguated and the data in the word entry has not (and may not be apply to any particular sense, regardless of whether the sense has some related sense-disambiguated information).

Word senses

Each word entry may have multiple glosses under the senses key. Each sense is a dictionary that may contain the following keys (among others, and more may be added in the future):

  • glosses - list of gloss strings for the word sense (usually only one). This has been cleaned, and should be straightforward text with no tagging.
  • nonglosses - list of gloss-like strings but that are not traditional glossary entries describing the word's meaning
  • tags - list of qualifiers and tags for the gloss. This is a list of strings, and may include words such as "archaic", "colloquial", "present", "participle", "plural", "feminine", and many others (new words may appear arbitrarily). Some effort has been put into trying to canonicalize various sources and styles of annotation into a consistent set of tags, but it is impossible to do an exact job at this.
  • senseid - list of textual identifiers collected for the sense. If there is a QID for the entry (e.g., Q123), those are stored in the wikidata field.
  • wikidata - list of QIDs (e.g., Q123) for the sense
  • wikipedia- list of Wikipedia page titles (with optional language code prefix)
  • categories - list of category names extracted from (a subset) of the Category links on the page
  • topics - list of topic names (kind of similar to categories but determined differently)
  • alt_of - list of words that his sense is an alternative form of; for example, for an abbreviation, this would typically be set to the full form
  • form_of - list of words that this sense is an inflected form of; for example, a participle form would typically set this to be the base form
  • translations - sense-disambiguated translation entries (see below)
  • synonyms - sense-disambiguated synonym linkages for the word (see below)
  • antonyms - sense-disambiguated antonym linkages for the word (see below)
  • hypernyms - sense-disambiguated hypernym linkages for the word (see below)
  • holonyms - sense-disambiguated linkages indicating being part of something (see below) (not systematically encoded)
  • meronyms - sense-disambiguated linkages indicating having a part (see below) (fairly rare)
  • derived - sense-disambiguated derived word linkages for the word (see below)
  • related - sense-disambiguated related word linkages for the word (see below)

Linkages to other words

Linkages (synonyms, antonyms, hypernyms, derived words, holonyms, meronyms, derived, related) are stored in the word's data if not sense-disambiguated, and in the word sense if sense-disambiguated. They are lists of dictionaries, where each dictionary can contain the following keys, among others:

  • word - the word this links to (string). If this starts with "Thesaurus:", then this entry is a link to a thesaurus page in Wiktionary. If this starts with "Category:", then this refers to a category page in Wiktionary.
  • sense - text identifying the word sense or context (e.g., "to rain very heavily").
  • tags: qualifiers specified for the sense (e.g., field of study, region, dialect, style). This is a list of strings.

Pronunciation

Pronunciation information is stored under the sounds key. It is a list of dictionaries, each of which may contain the following keys, among others:

  • ipa - pronunciation specifications as IPA strings
  • enpr - pronunciation in English pronunciation respelling
  • audio - name of a sound file in WikiMedia Commons
  • homophones - list of homophones for the word
  • hyphenation - list of hyphenations
  • tags - other labels or context information attached to the sense (e.g., regional variant)

Translations

Translations are stored under the translations key in the word's data (if not sense-disambiguated) or in the word sense (if sense-disambiguated). They are stored in a list of dictionaries, where each dictionary has the following keys (and possibly others):

  • lang the language name that the translation is for
  • code - Wiktionary's 2 or 3-letter language code for the language the translation is for
  • word - the translation in the specified language
  • sense - optional sense for which the translation is (this is a free-text string, and may not match any gloss exactly)
  • tags - optional list of qualifiers for the translations, e.g., gender

Related packages

The wikitextprocessor is a generic module for extracting data from Wiktionary, Wikipedia, and other WikiMedia dump files. wiktextract is built using this module.

The wiktfinnish package can be used to interpret Finnish noun declinations and verb conjugations and for generating Finnish inflected word forms.

Related tools

A few other tools also exist for parsing Wiktionaries. These include Dbnary, Wikiparse, and DKPro JWKTL.

Contributing and reporting bugs

Please report bugs and other issues on github. I also welcome suggestions for improvement.

Please email to ylo at clausal.com if you wish to contribute or have patches or suggestions.

License

Copyright (c) 2018-2020 Tatu Ylonen. This package is free for both commercial and non-commercial use. It is licensed under the MIT license. See the file LICENSE for details. (Certain files have different open source licenses)

Project details


Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Files for wiktextract, version 1.99.4
Filename, size File type Python version Upload date Hashes
Filename, size wiktextract-1.99.4-py3-none-any.whl (73.9 kB) File type Wheel Python version py3 Upload date Hashes View
Filename, size wiktextract-1.99.4.tar.gz (85.2 kB) File type Source Python version None Upload date Hashes View

Supported by

AWS AWS Cloud computing Datadog Datadog Monitoring DigiCert DigiCert EV certificate Facebook / Instagram Facebook / Instagram PSF Sponsor Fastly Fastly CDN Google Google Object Storage and Download Analytics Microsoft Microsoft PSF Sponsor Pingdom Pingdom Monitoring Salesforce Salesforce PSF Sponsor Sentry Sentry Error logging StatusPage StatusPage Status page