junaidaslam/gojri-devanagari-nastaliq-lexicon

Gojri Devanagari-Nastaliq lexicon

A lexicon that maps Gojri headwords from Devanagari to Nastaliq (ISO 639-3: gju).

The language of this dataset is Gojri, not Urdu or Hindi. Each row maps a Devanagari headword from a published Gojri dictionary to its Gojri Nastaliq (Perso-Arabic) form. Optional Roman transliteration and a short English gloss come from the same source line.

Source attribution

This lexicon is derived from headwords in:

Anjum, Rafique, and Ruksana Sadiq. Concise Gojri-English-Hindi Dictionary. Educational Publishing House / J K Anjuman Taraqi Gojri Adab, New Delhi. ISBN 978-81-19225-29-3. Compilation: Prof. (Dr.) Rafique Anjum (D.Phil). Composition and vetting: Ms. Ruksana Sadiq (M.Phil). © All rights reserved by the original rights holders.

Local PDF file used for extraction: anjumshanasi/Gojri-Hindi-English-Dictionary.pdf (458 pages; MD5 dbcf6814f1b19a111f8bb3955e341b11 in the project manifest).

Cite both this dataset and the original dictionary in any publication or redistribution that uses these headwords. See Citation and License.

Dataset files

File	Rows	Description
gojri_lexicon_final.tsv	1,862	Full lexicon (default split)
gojri_lexicon_human_verified.tsv	50	Native-speaker verified subset
LICENSE.md	-	Derivative rights notice
README.md	-	Dataset card

Language

Field	Value
Language	Gojri (Gujari)
ISO 639-3	gju
Source script	Devanagari
Target script	Perso-Arabic Nastaliq

Gojri is an Indo-Aryan language. It shares script and some vocabulary with Urdu, but it is a distinct language. Spellings in this lexicon follow Gojri print practice.

Columns

Column	Description
entry_id	Stable ID, e.g. p0019:L14 (source page and line)
page	Page number in the source PDF
line	Line number in the page extract
devanagari	Headword in Devanagari
roman	Optional Latin form from the dictionary
nastaliq	Gojri Nastaliq headword (main field)
english_gloss	Short English gloss snippet when present
quality	Trust tag

Quality tags

quality	Count	Meaning
human_verified	50	Approved by a native Gojri speaker
pass2_refined	299	Refined in pass 2
pass1_kept	1,513	Kept from pass 1 as correct Gojri Nastaliq

50 rows are human-verified so far. Remaining rows are automated conversions and require spot-checking before training-critical use.

How it was built

1. Extracted clean Unicode text from the Devanagari dictionary PDF.
2. Pass 1: Ran Aksharamukha Devanagari -> Urdu locally and removed short vowels.
3. Pass 2: Applied Gojri orthography rules:
   • Retained pass 1 when already correct.
   • Mapped न and ण to standard ن (U+0646) instead of ݨ (U+0768).
   • Preferred Gojri word-final -و endings where required.
   • Corrected Perso-Arabic etymology for Arabic and Persian loanwords.
4. Human verification: 50 seed entries (pages 19, 31, 167) were reviewed by a native Gojri speaker.
5. Initial heh (November 2026): the Gojri print writes a word that starts with /h/ with the two-eyed heh ھ (448 of 475 word-initial heh forms in the printed Gojri-English Dictionary text). Aksharamukha writes ہ there, so every Arabic-script word in nastaliq that started with ہ or ه now starts with ھ. This changed 112 words in 98 rows. Human-verified rows were not touched. scripts/lexicon_initial_heh.py in the project repository applies the rule.

The source PDF has no Nastaliq text. It holds Devanagari, Roman, English and Hindi only, so the nastaliq column is machine-made (rules and a language model) and only the 50 human-verified rows were checked by a speaker. Some rows are example phrases or sentence fragments from the dictionary, not single headwords.

Versions

Tag v1.0 is the October 2026 release. Tag v2.0 (November 2026) starts every Arabic-script word in the nastaliq column with ھ where it started with ہ (112 words in 98 rows; see build step 5). Load a version with load_dataset("junaidaslam/gojri-devanagari-nastaliq-lexicon", revision="v1.0").

Known verified correction: Devanagari अंगणू -> Nastaliq انگنو (not انگݨو).

Pipeline code is in the gojri-ai repository (scripts/dict_translit_*.py, prompts/devanagari_to_gojri_nastaliq_v1.txt).

Loading the dataset

from datasets import load_dataset

ds = load_dataset("junaidaslam/gojri-devanagari-nastaliq-lexicon")
print(ds["train"][0])

# Verified subset only
gold = load_dataset(
    "junaidaslam/gojri-devanagari-nastaliq-lexicon",
    name="human_verified",
)

Direct TSV loading:

from datasets import load_dataset

ds = load_dataset(
    "csv",
    data_files="gojri_lexicon_final.tsv",
    delimiter="\t",
)

Intended use

• Devanagari to Nastaliq headword lookup for Gojri.
• Seed data for orthography and transliteration tools.
• Vocabulary verification for OCR and ASR systems.

Limitations

• Contains headwords and brief glosses rather than full dictionary definitions.
• Includes single-letter alphabet entries from early dictionary pages.
• Most rows are not yet human-verified.
• Does not contain full page text or example sentences from the copyrighted book.

License

See LICENSE.md.

• The original dictionary remains copyright of the original rights holders (Anjum, Sadiq, Educational Publishing House, J K Anjuman Taraqi Gojri Adab).
• This release is a script-conversion derivative of headwords for language preservation and research.
• Users must retain attribution to the original authors and to this dataset.

Citation

@dataset{gojri_devanagari_nastaliq_lexicon_2026,
  title     = {Gojri Devanagari-Nastaliq Lexicon},
  author    = {Aslam, Junaid},
  year      = {2026},
  publisher = {Hugging Face},
  url       = {https://huggingface.co/datasets/junaidaslam/gojri-devanagari-nastaliq-lexicon},
  note      = {Script-conversion derivative of headwords from Anjum and Sadiq,
               Concise Gojri-English-Hindi Dictionary; 50 rows human-verified}
}

Original source dictionary:

@book{anjum_sadiq_gojri_dictionary,
  title     = {Concise Gojri-English-Hindi Dictionary},
  author    = {Anjum, Rafique and Sadiq, Ruksana},
  publisher = {Educational Publishing House / J K Anjuman Taraqi Gojri Adab},
  address   = {New Delhi},
  isbn      = {978-81-19225-29-3},
  note      = {Compilation by Prof. (Dr.) Rafique Anjum; composition and vetting
               by Ms. Ruksana Sadiq}
}

Contributions

Contributions to correct nastaliq forms or add verified rows are welcome via issues or pull requests on GitHub.

Contact

• Dataset maintainer: junaidaslam
• Repository: jaslam94/gojri-ai