Nâbra: Syrian Arabic Dialects with Morphological Annotations

What it is about

Nâbra is a morphologically annotated corpus of Syrian Arabic: about 6,000 sentences and 60,021 tokens from social media, film and series scripts, song lyrics, proverbs and poetry, covering ten localities. Nine annotators tagged each token with prefixes, stem, suffixes, POS, dialect lemma, MSA lemma, person, number, gender, aspect, gloss and synonyms, in the SAMA tagset extended with Syrian affixes. A spreadsheet-like portal offered earlier annotations for lookup. Six linguists re-reviewed overlapping tokens; agreement F1 is 0.980 for prefixes, 0.944 for suffixes, 0.868 for POS, 0.805 for stem, 0.780 for MSA lemma and 0.753 for dialect lemma, with kappa in the same order. The corpus is public on the Currasat portal.

Ideas

• I1. Use one tag string for analyses: prefix/POS + prefix/POS + stem/POS + suffix/POS, joined by +. Mova's tools/hbo output could use this string, so its analyses are comparable with SAMA-style corpora and with tools/macula compare.
• I2. Clean tags deterministically before scoring. The authors cluster tag variants by edit distance, review the clusters, and fix them with regular-expression scripts. The same clean-up should run on macula tag values before any comparison.
• I3. Report the per-feature table. The spread is wide (prefixes 0.98, dialect lemma 0.75), so one accuracy number hides where a segmenter fails. The tools/hbo evaluation should report precision, recall, F and kappa per feature, with lemmas reported apart, as Table 4 does.

For Mova

I1 is the output format of tools/hbo (segmenter and CoNLL-U analysis). I2 is a clean-up step in tools/macula before compare. I3 is the scoring table for tools/hbo. Dialectal Arabic is not a Mova target now; these ideas apply to Hebrew and Aramaic only.

Cite

The paper itself, not this card. BibTeX from the library's record of the work. DOI 10.18653/v1/2023.arabicnlp-1.2

@inproceedings{nayouf2023nabra,
  title = {{Nâbra: Syrian Arabic Dialects with Morphological Annotations}},
  author = {Nayouf, Amal and Hammouda, Tymaa and Jarrar, Mustafa and Zaraket, Fadi and Kurdy, Mohamad-Bassam},
  booktitle = {Proceedings of ArabicNLP 2023},
  year = {2023},
  doi = {10.18653/v1/2023.arabicnlp-1.2},
  url = {https://doi.org/10.18653/v1/2023.arabicnlp-1.2}
}

References

1 references, 0 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

1. Muhammad Abdul-Mageed, Chiyu Zhang, Houda Bouamor, and Nizar Habash. 2020. NADI 2020: The first nuanced Arabic dialect identification shared task. In Proceedings of the Fifth Arabic Natural Language Processing Workshop, pages 97–110, Barcelona, Spain (Online). Association for Computational Linguistics. find

This page in Markdown → https://md.mova.elamur.ai/free/en/docs/papers/aclbulk-2023-nayouf-nabra-syrian-arabic-dialects-morphological/