---
title: "Classification of Closely Related Sub-dialects of Arabic Using Support-Vector Machines — Wray 2018"
description: "The author builds a Twitter dataset from tweets located by geo-tags or by user-written locations… 65% accuracy. 27 references, cited by 2 papers here."
authors: ["Samantha Wray"]
year: 2018
venue: "Proceedings of the Eleventh International Conference on Language Resources and Evaluation (LREC 2018)"
source: "https://aclanthology.org/L18-1580/"
licence: "CC BY-NC 4.0 (ELRA, LREC proceedings)"
licence_url: "https://creativecommons.org/licenses/by-nc/4.0/"
summary_licence: "Elamur Cosmic License"
canonical: "https://research.elamur.ai/free/en/papers/wray-2018-classification-closely-related-sub-dialects/"
---
[Elamur Research](https://md.research.elamur.ai/free/en/) · [this page in HTML](https://research.elamur.ai/free/en/papers/wray-2018-classification-closely-related-sub-dialects/)

[home](https://md.research.elamur.ai/free/en/) / [papers](https://md.research.elamur.ai/free/en/papers/) / [2018](https://md.research.elamur.ai/free/en/papers/2018/page/4/) / Classification of Closely Related Sub-dialects of Arabic Using Support-Vector Machines

# Classification of Closely Related Sub-dialects of Arabic Using Support-Vector Machines

**Authors:** Samantha Wray · **Year:** 2018 · **Venue:** Proceedings of the Eleventh International Conference on Language Resources and Evaluation (LREC 2018) · **Licence:** [CC BY-NC 4.0 (ELRA, LREC proceedings)](https://creativecommons.org/licenses/by-nc/4.0/)

**Links:** [Publisher's page](https://aclanthology.org/L18-1580/) · [PDF](https://aclanthology.org/L18-1580.pdf) · [OpenAlex](https://openalex.org/W2807654876)

**Citations:** 7 citations (OpenAlex, 2026-10-14) · [2](https://md.research.elamur.ai/free/en/papers/wray-2018-classification-closely-related-sub-dialects/cited-by/) in our library, of the 140,884 papers whose bibliographies we read

## Summary
The paper addresses Arabic dialect identification at a finer level than usual. Most prior work separates MSA from a few broad dialect groups. Here the task is to distinguish among four varieties within Levantine Arabic, namely those of Jordan, Lebanon, Palestine and Syria. These are hard to separate because they share a script, much vocabulary and many features. The author builds a Twitter dataset from tweets located by geo-tags or by user-written locations matched against GeoNames. Part of it comes from Mubarak and Darwish (2014), and the rest was newly harvested in 2014–2015. The data was cleaned with several heuristics. A linear multiclass SVM uses SRILM-derived character n-gram (1–5) and word n-gram (1–3) features. It is compared with a classifier built on dialect-unique word lists taken from two LDC Levantine telephone corpora, and with a combination of both. The n-gram model reaches 65% accuracy in 10-fold cross-validation and beats both alternatives. The study shows that sub-dialect labels can be obtained from spontaneous social media text. It also proposes using such a classifier to estimate the sub-dialect composition of existing mixed-dialect resources.

## Key points
- Data: 10,000 tweets per Levantine country (Jordan, Palestine, Syria, Lebanon), about 100,000 words in total by the paper's count, partly drawn from Mubarak and Darwish (2014).
- Tweets were collected via the Twitter API using the Twython library. Location came from geo-tags (only about 2% of tweets carry one) or from user-provided locations (about 70% of tweets), matched against GeoNames name variants.
- Cleaning steps:
  - tweets were dropped if their geo-tag and user location disagreed;
  - tweets with vowel diacritics were dropped, since these were mostly Quran verses;
  - emojis, mentions, punctuation and links were stripped;
  - collection ran across most of a year to dampen seasonal effects such as Ramadan greetings.
- Unique-word vocabularies came from Appen LDC2007T01 and Fisher Levantine LDC2007S02. Words occurring in only one sub-dialect were kept as binary features.
- Accuracy under 10-fold cross-validation:

| Model | Accuracy |
|---|---|
| n-gram SVM | 65% |
| Vocabulary pruning only | 54% |
| n-gram + pruning combined | 46% |

## Interesting ideas
- Two independent but noisy location signals (geo-tag and self-declared location) are cross-checked, and only items where they agree are kept, which trades data volume for label reliability.
- Vocabulary taken from a different register (professionally transcribed phone speech) did not transfer to tweets, and merging it with good features actually lowered accuracy. Resources should be tested separately before being combined.
- A cheap surface cue (diacritics) was used to detect and remove formulaic quoted material that does not represent the target variety.
- The paper notes that prompted translation corpora contain contributor-specific insertions, such as pan-Arabic exclamations. These insertions can become spurious class cues unrelated to the actual variety.
- A trained classifier is reused to audit the hidden composition of existing mixed resources instead of trusting their nominal labels.

## For Mova

How these ideas could serve Mova, our small open language model — its files, stages and tests:

- **I1** (learn): Silver data could likewise keep only sentences where independent witness parsers agree on a structure, discarding disagreements rather than trying to resolve them.
- **I2** (learn): Every new lexical resource added to training (for example, from another genre) should first be checked alone and in combination, with the drop-in-score case treated as a built-in negative control.
- **I3** (tokens): The tokenizer could flag quoted, scriptural or boilerplate spans by surface cues so that later stages treat them as embedded material rather than ordinary narrative text.
- **I4** (back): When evaluating translation or text → graph → text output, check for added material that has no source in the meaning graph, and do not let such insertions be learned as style signals.
- **I5** (tree): A genre or dialect classifier could profile the treebanks used for the perceptron parser and gparse, revealing their real mix before UD dialect conversion or training.

*Our summary of the paper, in our own words — not its text. The work belongs to its authors; read it at the links above.*

Fields: [Linguistics & languages](https://md.research.elamur.ai/free/en/shelves/linguistics/)

Topics: [🐪 Semitic languages](https://md.research.elamur.ai/free/en/topics/semitic-languages/)

## Institutions

- [New York University Abu Dhabi](https://md.research.elamur.ai/free/en/universities/new-york-university-abu-dhabi/) — University, Abu Dhabi, United Arab Emirates · 45 works · 2,055 citations

## Cite

Cite as: Samantha Wray. 2018. Classification of Closely Related Sub-dialects of Arabic Using Support-Vector Machines. Proceedings of the Eleventh International Conference on Language Resources and Evaluation (LREC 2018).

The paper itself, not this page. BibTeX from the library's record of the work.

```bibtex
@inproceedings{wray2018classification,
  title = {{Classification of Closely Related Sub-dialects of Arabic Using Support-Vector Machines}},
  author = {Samantha Wray},
  booktitle = {Proceedings of the Eleventh International Conference on Language Resources and Evaluation (LREC 2018)},
  year = {2018},
  doi = {10.63317/5knmymeqsxge},
  url = {https://doi.org/10.63317/5knmymeqsxge}
}
```

- [← previous Towards JointUD: Part-of-speech Tagging and Lemmatization using Recurrent Neural Networks](https://md.research.elamur.ai/free/en/papers/arakelyan-2018-jointud-part-speech-tagging-lemmatization/)
- [next → GenDR: A Generic Deep Realizer with Complex Lexicalization](https://md.research.elamur.ai/free/en/papers/lareau-2018-gendr-generic-deep-realizer-complex/)
- [↑ back to papers of 2018](https://md.research.elamur.ai/free/en/papers/2018/page/4/)

## References

27 references, 6 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

1. Almeman, K. and Lee, M. (2013). Automatic building of Arabic Multi Dialect text corpora by bootstrapping dialect words. In Communications, Signal Processing, and their Applications (ICCSPA), 2013 1st International Conference on, pages 1–6. IEEE. [find](https://scholar.google.com/scholar?q=Almeman%2C+K%2E+and+Lee%2C+M%2E+%282013%29%2E+Automatic+building+of+Arabic+Multi+Dialect+text+corpora+by+bootstrapping+dialect+words%2E+In+Communications%2C+Signal+Processing%2C+and+their+Applications+%28ICCSPA%29%2C+2013+1st+)
2. Darwish, K., Sajjad, H., and Mubarak, H. (2014). Verifiably Effective Arabic Dialect Identification. EMNLP-2014. [find](https://scholar.google.com/scholar?q=Darwish%2C+K%2E%2C+Sajjad%2C+H%2E%2C+and+Mubarak%2C+H%2E+%282014%29%2E+Verifiably+Effective+Arabic+Dialect+Identification%2E+EMNLP%2D2014%2E)
3. Dimalen, D. M. and Roxas, R. E. O. (2007). AutoCor: A Query Based Automatic Acquisition of Corpora of Closely-related Languages. In PACLIC. [find](https://scholar.google.com/scholar?q=Dimalen%2C+D%2E+M%2E+and+Roxas%2C+R%2E+E%2E+O%2E+%282007%29%2E+AutoCor%3A+A+Query+Based+Automatic+Acquisition+of+Corpora+of+Closely%2Drelated+Languages%2E+In+PACLIC%2E)
4. Eisenstein, J. (2013). What to do about bad language on the internet. In HLT-NAACL, pages 359–369. [our card](https://md.research.elamur.ai/free/en/papers/eisenstein-2013-what-do-about-bad-language/)
5. Eldesouki, M., Dalvi, F., Sajjad, H., and Darwish, K. (2016). QCRI@ DSL 2016: Spoken Arabic Dialect Identification Using Textual. VarDial 3, page 221. [find](https://scholar.google.com/scholar?q=Eldesouki%2C+M%2E%2C+Dalvi%2C+F%2E%2C+Sajjad%2C+H%2E%2C+and+Darwish%2C+K%2E+%282016%29%2E+QCRI%40+DSL+2016%3A+Spoken+Arabic+Dialect+Identification+Using+Textual%2E+VarDial+3%2C+page+221%2E)
6. Elfardy, H. and Diab, M. T. (2013). Sentence Level Dialect Identification in Arabic. In ACL (2), pages 456–461. [find](https://scholar.google.com/scholar?q=Elfardy%2C+H%2E+and+Diab%2C+M%2E+T%2E+%282013%29%2E+Sentence+Level+Dialect+Identification+in+Arabic%2E+In+ACL+%282%29%2C+pages+456%E2%80%93461%2E)
7. Huck, J., Whyatt, D., and Coulton, P. (2012). Challenges in geocoding socially-generated data. [find](https://scholar.google.com/scholar?q=Huck%2C+J%2E%2C+Whyatt%2C+D%2E%2C+and+Coulton%2C+P%2E+%282012%29%2E+Challenges+in+geocoding+socially%2Dgenerated+data%2E)
8. Joachims, T. (1999). Svmlight: Support vector machine. SVM-Light Support Vector Machine [http://svmlight.joachims](http://svmlight.joachims). org/, University of Dortmund, 19(4). [↗](http://svmlight.joachims)
9. Ljubešic, N., Fišer, D., and Erjavec, T. (2014). Tweet-CaT: a Tool for Building Twitter Corpora of Smaller Languages. In Proceedings of the Ninth International Conference on Language Resources and Evaluation (LREC’14), Reykjavik, Iceland. European Language Resources Association (ELRA). [find](https://scholar.google.com/scholar?q=Ljubes%CC%8Cic%2C+N%2E%2C+Fis%CC%8Cer%2C+D%2E%2C+and+Erjavec%2C+T%2E+%282014%29%2E+Tweet%2DCaT%3A+a+Tool+for+Building+Twitter+Corpora+of+Smaller+Languages%2E+In+Proceedings+of+the+Ninth+International+Conference+on+Language+Resources+and+E)
10. Madsen, R. E., Sigurdsson, S., Hansen, L. K., and Larsen, J. (2004). Pruning the vocabulary for better context recognition. In Pattern Recognition, 2004. ICPR 2004. Proceedings of the 17th International Conference on, volume 2, pages 483–488. IEEE. [find](https://scholar.google.com/scholar?q=Madsen%2C+R%2E+E%2E%2C+Sigurdsson%2C+S%2E%2C+Hansen%2C+L%2E+K%2E%2C+and+Larsen%2C+J%2E+%282004%29%2E+Pruning+the+vocabulary+for+better+context+recognition%2E+In+Pattern+Recognition%2C+2004%2E+ICPR+2004%2E+Proceedings+of+the+17th+Internation)
11. Malmasi, S., Refaee, E., and Dras, M. (2015). Arabic dialect identification using a parallel multidialectal corpus. In Proceedings of the 14th Conference of the Pacific Association for Computational Linguistics (PACLING 2015), Bali, Indonesia, pages 209–217. [find](https://scholar.google.com/scholar?q=Malmasi%2C+S%2E%2C+Refaee%2C+E%2E%2C+and+Dras%2C+M%2E+%282015%29%2E+Arabic+dialect+identification+using+a+parallel+multidialectal+corpus%2E+In+Proceedings+of+the+14th+Conference+of+the+Pacific+Association+for+Computational+L)
12. Malmasi, S., Zampieri, M., Ljubešic, N., Nakov, P., Ali, A., and Tiedemann, J. (2016). Discriminating between similar languages and Arabic dialect identification: A report on the third DSL shared task. VarDial 3, page 1. [our card](https://md.research.elamur.ai/free/en/papers/malmasi-2016-discriminating-between-similar-languages/)
13. Mansour, S., Al-Onaizan, Y., Blackwood, G., and Tillmann, C. (2014). Automatic Dialect Classification for Statistical Machine Translation. In Proc. of the Conf. of the Assoc. for Machine Translation in the Americas (AMTA), Vancouver, BC, Canada. [find](https://scholar.google.com/scholar?q=Mansour%2C+S%2E%2C+Al%2DOnaizan%2C+Y%2E%2C+Blackwood%2C+G%2E%2C+and+Tillmann%2C+C%2E+%282014%29%2E+Automatic+Dialect+Classification+for+Statistical+Machine+Translation%2E+In+Proc%2E+of+the+Conf%2E+of+the+Assoc%2E+for+Machine+Translation+i)
14. Mubarak, H. and Darwish, K. (2014). Using Twitter to collect a multi-dialectal corpus of Arabic. ANLP 2014. [find](https://scholar.google.com/scholar?q=Mubarak%2C+H%2E+and+Darwish%2C+K%2E+%282014%29%2E+Using+Twitter+to+collect+a+multi%2Ddialectal+corpus+of+Arabic%2E+ANLP+2014%2E)
15. Rangel, F., Rosso, P., Potthast, M., and Stein, B. (2017). Overview of the 5th author profiling task at pan 2017: Gender and language variety identification in twitter. Working Notes Papers of the CLEF. [find](https://scholar.google.com/scholar?q=Rangel%2C+F%2E%2C+Rosso%2C+P%2E%2C+Potthast%2C+M%2E%2C+and+Stein%2C+B%2E+%282017%29%2E+Overview+of+the+5th+author+profiling+task+at+pan+2017%3A+Gender+and+language+variety+identification+in+twitter%2E+Working+Notes+Papers+of+the+CLE)
16. Refaee, E. and Rieser, V. (2014). Subjectivity and sentiment analysis of Arabic twitter feeds with limited resources. In Workshop on Free/Open-Source Arabic Corpora and Corpora Processing Tools. [find](https://scholar.google.com/scholar?q=Refaee%2C+E%2E+and+Rieser%2C+V%2E+%282014%29%2E+Subjectivity+and+sentiment+analysis+of+Arabic+twitter+feeds+with+limited+resources%2E+In+Workshop+on+Free%2FOpen%2DSource+Arabic+Corpora+and+Corpora+Processing+Tools%2E)
17. Stolcke, A. et al. (2002). SRILM - An extensible language modeling toolkit. In Proceedings of the 7th International Conference on Spoken Language Processing (ICSLP 2002), volume 2002, pages 901–904. [find](https://scholar.google.com/scholar?q=Stolcke%2C+A%2E+et+al%2E+%282002%29%2E+SRILM+%2D+An+extensible+language+modeling+toolkit%2E+In+Proceedings+of+the+7th+International+Conference+on+Spoken+Language+Processing+%28ICSLP+2002%29%2C+volume+2002%2C+pages+901%E2%80%93904%2E)
18. Tillmann, C., Al-Onaizan, Y., and Mansour, S. (2014). Improved sentence-level Arabic dialect classification. In Proceedings of the VarDial Workshop, pages 110–119. [find](https://scholar.google.com/scholar?q=Tillmann%2C+C%2E%2C+Al%2DOnaizan%2C+Y%2E%2C+and+Mansour%2C+S%2E+%282014%29%2E+Improved+sentence%2Dlevel+Arabic+dialect+classification%2E+In+Proceedings+of+the+VarDial+Workshop%2C+pages+110%E2%80%93119%2E)
19. Wray, S. (2016). Decomposability and the effects of morpheme frequency in lexical access. University of Arizona. [find](https://scholar.google.com/scholar?q=Wray%2C+S%2E+%282016%29%2E+Decomposability+and+the+effects+of+morpheme+frequency+in+lexical+access%2E+University+of+Arizona%2E)
20. Zaidan, O. F. and Callison-Burch, C. (2011). The Arabic Online Commentary Dataset: an annotated dataset of informal Arabic with high dialectal content. In Proceedings of the 49th Annual Meeting of the Association for Computational Linguistics: Human Language Technologies. Association for Computational Linguistics. 8. [our card](https://md.research.elamur.ai/free/en/papers/zaidan-2011-arabic-online-commentary-dataset-annotated/)
21. Language Resource References Almeman, K. and Lee, M. (2013). Automatic building of Arabic Multi Dialect text corpora by bootstrapping dialect words. In Communications, Signal Processing, and their Applications (ICCSPA), 2013 1st International Conference on, pages 1–6. IEEE. [find](https://scholar.google.com/scholar?q=Language+Resource+References+Almeman%2C+K%2E+and+Lee%2C+M%2E+%282013%29%2E+Automatic+building+of+Arabic+Multi+Dialect+text+corpora+by+bootstrapping+dialect+words%2E+In+Communications%2C+Signal+Processing%2C+and+their+App)
22. Appen. (2007). Levantine Arabic Conversational Telephone Speech LDC2007T01. Linguistics Data Consortium. [find](https://scholar.google.com/scholar?q=Appen%2E+%282007%29%2E+Levantine+Arabic+Conversational+Telephone+Speech+LDC2007T01%2E+Linguistics+Data+Consortium%2E)
23. Bouamor, H., Habash, N., and Oflazer, K. (2014). A Multidialectal Parallel Corpus of Arabic. In LREC, pages 1240–1245. [our card](https://md.research.elamur.ai/free/en/papers/bouamor-2014-multidialectal-parallel-corpus-arabic/)
24. Maamouri, M., Buckwalter, T., Graff, D., and Jin, H. (2007). Fisher Levantine Arabic Conversational Telephone Speech LDC2007S02. LDC. [find](https://scholar.google.com/scholar?q=Maamouri%2C+M%2E%2C+Buckwalter%2C+T%2E%2C+Graff%2C+D%2E%2C+and+Jin%2C+H%2E+%282007%29%2E+Fisher+Levantine+Arabic+Conversational+Telephone+Speech+LDC2007S02%2E+LDC%2E)
25. Mubarak, H. and Darwish, K. (2014). Using Twitter to collect a multi-dialectal corpus of Arabic. ANLP 2014. [find](https://scholar.google.com/scholar?q=Mubarak%2C+H%2E+and+Darwish%2C+K%2E+%282014%29%2E+Using+Twitter+to+collect+a+multi%2Ddialectal+corpus+of+Arabic%2E+ANLP+2014%2E)
26. Zaidan, O. F. and Callison-Burch, C. (2011). The Arabic Online Commentary Dataset: an annotated dataset of informal Arabic with high dialectal content. In Proceedings of the 49th Annual Meeting of the Association for Computational Linguistics: Human Language Technologies. Association for Computational Linguistics. [our card](https://md.research.elamur.ai/free/en/papers/zaidan-2011-arabic-online-commentary-dataset-annotated/)
27. Zbib, Rabih and Malchiodi, Erika and Devlin, Jacob and Stallard, David and Matsoukas, Spyros and Schwartz, Richard and Makhoul, John and Zaidan, Omar F and Callison-Burch, Chris. (2012). Machine translation of Arabic dialects. [our card](https://md.research.elamur.ai/free/en/papers/zbib-2012-machine-translation-arabic-dialects/)