Elamur Research⌕
☰
home / papers / 2019 / Distance-Based Authorship Verification Across Modern Standard Arabic Genres

Distance-Based Authorship Verification Across Modern Standard Arabic Genres

Paper · 2019 · Hossam Ahmed

Summary

The paper addresses Authorship Verification (AV) for Modern Standard Arabic: given a few texts by one known author, decide whether a disputed text is also theirs. Earlier Arabic AV work covered only Classical Arabic and used a Manhattan-style metric. No study had compared distance measures for modern Arabic AV. The author builds a corpus of 125 short documents across five genres (fiction, non-fiction, economics, politics, opinion columns), with five authors per genre and five texts per author, drawn mostly from Egyptian sources. Features are n-grams of tokens, stems and POS tags (from MADAMIRA) and roots (from the ISRI stemmer). The known texts are merged into a single profile. The disputed text is accepted if its distance stays within a threshold estimated only from the author's own texts, so no negative training data is needed. Four measures are compared: Canberra, Manhattan, Cosine and Jaccard. Canberra performs best in most genres, reaching 97.8% accuracy, and plain token unigrams outperform the richer linguistic features. The result shows that a cheap bag-of-words setup with minimal preprocessing can clearly beat the prior Classical Arabic baseline.

Key points

  • Corpus design: 5 genres × 5 authors × 5 documents = 125 texts. Average lengths are roughly 746–1010 tokens. Book texts are sampled as about 1100 words from the middle of each epub.
  • Tools: MADAMIRA version 2.1 with standard settings supplies POS tags and stems. Roots come from the ISRI Stemmer in NLTK.
  • Best accuracy is up to 97.8%, using Canberra Distance with token unigrams. This is at least 10% above the Classical Arabic AV baseline (Ahmed, 2018).
  • The old best setup (stem bigrams with a Manhattan-based similarity) gives only 80–85% on MSA, compared with 87.1% on Classical Arabic.
  • The extraction tools differ sharply in quality. MADAMIRA reports 95.9% POS-tagging and 96.0% stemming accuracy, while ISRI reports precision and recall below 48%. This may handicap the root and stem features.

Interesting ideas

  • I1. The acceptance threshold is set from the spread of distances among the author's own documents, so verification works without counter-examples; this suits any setting where only positive data exist.
  • I2. Canberra Distance divides each difference by the size of the values involved, so rare features count more than in Manhattan; that turned out to be decisive.
  • I3. Long texts are sampled from the middle to skip boilerplate that repeats verbatim per author (prefaces, dedications), which would otherwise leak identity cheaply.
  • I4. The author tests whether richer features lose because the extraction tool is weak (ISRI's low precision) rather than because the features themselves are uninformative.
  • I5. A post-editing explanation is checked against controls: genres without editing, and a single-source genre. This rules out an editorial confound.

For Mova

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

  • I1 (learn): Silver trees could be flagged as suspicious when their distance from a profile of trusted parses exceeds the spread seen within that trusted set. This would need no hand-labelled bad parses.
  • I2 (tree): Comparing deprel or construction frequency profiles with Canberra could reveal UD dialect differences between treebanks that hinge on rare relations rather than frequent ones.
  • I3 (learn): When building silver or evaluation sets from books, sampling middle passages avoids rewarding the system for memorising repeated front matter.
  • I4 (learn): Any ablation that shows a feature is useless should also report the accuracy of the upstream tool. Otherwise a weak lemmatiser or tagger can masquerade as a negative finding.
  • I5 (learn): Evaluation of reader skills on news versus fables could include single-source subsets as controls. These would separate genre effects from publisher style.

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, topics and concepts

📚 Shelves 5The field of science this work belongs to (OpenAlex fields).Computer Science · 18,614 worksArtificial Intelligence · 17,965 worksArts and Humanities · 17,450 worksLanguage and Linguistics · 17,296 worksInformation Systems · 1,730 works

🧭 Topics 1Research subjects and languages the work is about.Semitic languages · 3,066 works

Cite

Cite as Hossam Ahmed. 2019. Distance-Based Authorship Verification Across Modern Standard Arabic Genres. Proceedings of the 3rd Workshop on Arabic Corpus Linguistics.

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

@inproceedings{ahmed2019distance,
  title = {{Distance-Based Authorship Verification Across Modern Standard Arabic Genres}},
  author = {Hossam Ahmed},
  booktitle = {Proceedings of the 3rd Workshop on Arabic Corpus Linguistics},
  year = {2019},
  url = {https://aclanthology.org/W19-5611/}
}

References

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

  1. Ahmed Abbasi and Hsinchun Chen. 2005a. Applying Authorship Analysis to Arabic Web Content. In Paul Kantor, Gheorghe Muresan, Fred Roberts, DanielD. Zeng, Fei-Yue Wang, Hsinchun Chen, and RalphC. Merkle, editors, Intelligence and Security Informatics SE - 15, volume 3495 of Lecture Notes in Computer Science, pages 183–197. Springer Berlin Heidelberg. find
  2. Ahmed Abbasi and Hsinchun Chen. 2005b. Applying authorship analysis to extremist-group Web forum messages. IEEE Intelligent Systems, 20(5):67–75. ↗
  3. Hossam Ahmed. 2017. Dynamic Similarity Threshold in Authorship Verification: Evidence from Classical Arabic. Procedia Computer Science, 117:145–152. find
  4. Hossam Ahmed. 2018. The Role of Linguistic Feature Categories in Authorship Verification. Procedia Computer Science, 142:214–221. find
  5. Malik H. Altakrori, Benjamin C. M. Fung, Steven H. H. Ding, Abdallah Tubaishat, and Farkhund Iqbal. 2018. Arabic Authorship Attribution: An Extensive Study on Twitter Posts. ACM Trans. Asian Low-Resour. Lang. Inf. Process, 18(1):51. find
  6. Nacer Eddine Benzebouchi, Nabiha Azizi, Monther Aldwairi, and Nadir Farah. 2018. Multi-classifier system for authorship verification task using word embeddings. 2nd International Conference on Natural Language and Speech Processing, ICNLSP 2018:1–6. find
  7. Steven Bird, Ewan Klein, and Edward Loper. 2009. Natural language processing with Python: analyzing text with the natural language toolkit. O’Reilly Media, Inc. find
  8. John Burrows. 2002. “Delta”: a measure of stylistic difference and a guide to likely authorship. Literary and Linguistic Computing, 17(3):267– 287. ↗
  9. David García-Barrero, Manuel Feria, and Maria Teresa Turell. 2013. Using function words and punctuation marks in Arabic forensic authorship attribution. In Rui Sousa-Silva, Rita Faria, Núria Gavaldà, and Belinda Maia, editors, Proceedings of the 3rd European Conference of the International Association of Forensic Linguists, pages 42–56, Porto, Portugal. Faculdade de Letras da Universidade do Porto. find
  10. Oren Halvani, Christian Winter, and Lukas Graner. 2017. Authorship Verification based on Compression-Models. ↗
  11. Oren Halvani, Christian Winter, and Anika Pflug. 2016. Authorship verification for different languages, genres and topics. Digital Investigation, 16:S33–S43. ↗
  12. Fatma Howedi and Masnizah Mohd. 2014. Text Classification for Authorship Attribution Using Naive Bayes Classifier with Limited Training Data. Computer Engineering and Intelligent Systems, 5(4):48–56. find
  13. Magdalena Jankowska, Evangelos Milios, and Vlado Kešelj. 2014. Author Verification Using Common N-Gram Profiles of Text Documents. Proceedings of COLING 2014, the 25th International Conference on Computational Linguistics:387– 397. find
  14. Siham Ouamour and Halim Sayoud. 2013. Authorship Attribution of Short Historical Arabic Texts Based on Lexical Features. In 2013 International Conference on Cyber-Enabled Distributed Computing and Knowledge Discovery, pages 144–147. find
  15. Siham Ouamour and Halim Sayoud. 2018. A Comparative Survey of Authorship Attribution on Short Arabic Texts. find
  16. Arfath Pasha, Mohamed Al-badrashiny, Mona Diab, Ahmed El Kholy, Ramy Eskander, Nizar Habash, Manoj Pooleery, Owen Rambow, and Ryan M Roth. 2014. MADAMIRA : A Fast , Comprehensive Tool for Morphological Analysis and Disambiguation of Arabic. Proceedings of the 9th Language Resources and Evaluation Conference (LREC’14):1094–1101.
  17. Fabian Pedregosa, Gaël Varoquaux, Alexandre Gramfort, Vincent Michel, Bertrand Thirion, Olivier Grisel, Mathieu Blondel, Peter Prettenhofer, Ron Weiss, and Vincent Dubourg. 2011. Scikit-learn: Machine learning in Python. Journal of machine learning research, 12(Oct):2825–2830. ↗
  18. Efstathios Stamatatos, Walter Daelemans, Ben Verhoeven, Martin Potthast, Benno Stein, Patrick Juola, Miguel A. Sanchez-Perez, and Alberto Barrón-Cedeño. 2014. Overview of the author identification task at PAN 2014. In CEUR Workshop Proceedings, volume 1180, pages 877– 897. find
  19. Kazem Taghva, Rania Elkhoury, and Jeffrey Coombs. 2005. Arabic stemming without a root dictionary. In Information Technology: Coding and Computing, 2005. ITCC 2005. International Conference on, volume 1, pages 152–157. IEEE. find

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/ahmed-2019-distance-based-authorship-verification-across/