n9e6y/persian-dehkhoda-dictionary

Persian Dehkhoda Dictionary (لغت‌نامهٔ دهخدا)

The Loghatnameh of Ali Akbar Dehkhoda (لغت‌نامه علی‌اکبر دهخدا) is the largest encyclopedic dictionary of the Persian language. This dataset is a cleaned, structured version of it:

• 343,316 entries and 240,810 distinct headwords
• 566,654 senses, of which 83,849 belong to compound sub-entries
• 841,986 source citations from 855 distinct sources

Each entry is one JSON object. Its senses, sub-entries and cited sources are nested inside it.

لغت‌نامهٔ دهخدا در قالب داده‌ای ساخت‌یافته: هر مدخل با تلفظ، نوع دستوری، معانی جداگانه، ترکیبات و منابع.

A version split into linked CSV tables is on Kaggle

Quick start

from datasets import load_dataset

ds = load_dataset("n9e6y/persian-dehkhoda-dictionary", split="train")

entry = ds.filter(lambda e: e["headword"] == "جلو")[0]
for sense in entry["senses"]:
    print(sense["sense_no"], sense["subentry"] or "", sense["text"])

Without the datasets library:

import json

with open("data/dehkhoda.jsonl", encoding="utf-8") as f:
    for line in f:
        entry = json.loads(line)

Structure

One split, train, with one JSON object per line. Entries are sorted in Persian alphabetical order.

{
  "id": 50000,
  "order": 49865,
  "headword": "اویزه",
  "headword_plain": "اویزه",
  "pronunciation": "اَ زَ / زِ",
  "pos": "اِ",
  "is_proper_noun": false,
  "text": "[ اَ زَ / زِ ] (اِ) شراب انگور و می. (ناظم الاطباء) (برهان).",
  "raw_html": "[ اَ زَ / زِ ] <span style=\"color:orange\">(اِ)</span> شراب انگور و می. (ناظم \nالاطباء) (برهان).",
  "senses": [
    {
      "sense_no": 1,
      "subentry_no": null,
      "subentry": null,
      "pos": "اِ",
      "text": "شراب انگور و می. (ناظم الاطباء) (برهان).",
      "sources": [
        {"citation": "(ناظم الاطباء)", "source": "ناظم الاطباء"},
        {"citation": "(برهان)", "source": "برهان"}
      ]
    }
  ]
}

Entry fields

field	type	description
id	int	Entry id from the source dump. It is the unique key, because headwords repeat for homographs.
order	int	1-based position in Persian alphabetical order
headword	string	Headword as published, with diacritics
headword_plain	string	Headword without diacritics or tatweel, for search
pronunciation	string | null	Vocalization, e.g. اَ زَ / زِ; missing for ~9% of entries
pos	string | null	Part of speech or entry type, e.g. اِ (noun), ص (adjective), ع مص (Arabic infinitive), اِ مرکب (compound noun), اِخ (proper name)
is_proper_noun	bool	True when pos is اِخ: people, places, books, tribes and so on (~39% of entries)
text	string	The whole definition as clean text. Paragraphs are separated by \n.
raw_html	string	The original HTML fragment, unchanged
senses	list	The entry's senses, in order

Sense fields

field	type	description
sense_no	int	1-based order within the entry
subentry_no	int | null	1-based number of the compound sub-entry; null for senses of the headword itself
subentry	string | null	The compound phrase, e.g. از جلو کسی درآمدن under جلو
pos	string | null	Part of speech, when the sense declares its own
text	string	Definition text of the sense
sources	list	Citations in this sense: citation is the text as written, e.g. (از فرهنگ جغرافیایی ایران ج ۸), and source is the normalized title, e.g. فرهنگ جغرافیایی ایران

How it was made

The source is a public raw dump of the dictionary, scraped from the University of Tehran website , with one HTML fragment per entry. It was parsed as follows:

• Senses: split on the dictionary's || separators.
• Sub-entries: a dash at the start of a line, followed by phrase؛ definition. A dash together with || adds another sense to the same sub-entry.
• Pronunciation and part of speech: read from the leading [ … ] and the coloured ( … ) labels.
• Text cleaning:
  • The hard line wraps of the print layout are joined.
  • <br> becomes a newline.
  • Arabic ي/ك become Persian ی/ک.
  • Footnote markers become [۱].
• Sources: parenthetical citations whose normalized title appears at least 20 times and is not a grammatical or usage label. A leading از, volume and page numbers, and edition suffixes are stripped from the title.
• Order: entries are sorted in Persian alphabetical order, ignoring diacritics, spaces and ZWNJ, which matches Dehkhoda's own ordering. The site's ids are mostly alphabetical but not entirely; for example, a block of about 3,900 م entries appeared after ی.

Limitations

• Missing entries: 506 entries of the printed dictionary are absent from the source scrape.
• No footnotes: the website never published the footnotes, so only their markers remain.
• Poetry layout: joining the print wraps also joins the lines of quoted poetry. raw_html keeps the original line breaks.
• Heuristic parsing: sub-entry and source extraction are rule-based. Rare sources (fewer than 20 citations) are not listed in sources, though they remain in text. A small number of sub-entries and citations will be wrong.

License

CC BY-NC-SA 4.0

Acknowledgements

The dictionary is the life's work of Allameh Ali Akbar Dehkhoda (1879–1956) and his collaborators. The raw text was scraped and released into the public domain by the compiler of the original CC0 dump.