Finite State Machine Pattern-Root Arabic Morphological Generator, Analyzer and Diacritizer

What it is about

The authors build FSAM, a rule-based finite-state system for undiacritized MSA. It has three layers: a templatic automaton (root plus pattern), a concatenative automaton (prefix, stem, suffix), and a rewrite transducer for orthography. The same network is used to analyse, to generate from a root, and to diacritize, since a finite-state transducer can be read in both directions. The system uses no stem lists and no training. It is evaluated on Tashkeela (a diacritized corpus), which gives 82% coverage (MADAMIRA 82%). Against the gold standard built from the intersection of PADT UD and Tashkeela, the root is found from a word with about 92% accuracy and generation from a root reaches 100%. Full diacritization reaches 84%, against 59% for MADAMIRA. The gold data have spelling inconsistencies that lower the diacritization score.

Ideas

• I1. Use the segmenter of tools/hbo as a validity filter: a root-pattern check. A segmentation is kept only when its consonants can fill a known pattern (binyan, or a noun pattern) in The verb and The Noun, the Adjective and the Numeral. The check can run after gparse and before CoNLL-U output, and it is deterministic.
• I2. Diacritize an unpointed Hebrew word by inserting vowels into a fixed consonant skeleton, with the pattern table as the only source. This is a finite-state insertion transducer, the reverse of the analyser. It fits the unpointed mode in 2. Cantillation: the accents (te'amim) and Phonology of Biblical Hebrew, and it can be measured against pointed Hebrew in the treebanks.
• I3. Report coverage, root accuracy and diacritization accuracy separately, and check the gold standard for spelling mismatches first. Use the same reporting for tools/hbo, and state the gold's own errors as a limit.

For Mova

I1 is a filter in tools/hbo, with the patterns written in The verb. I2 is a rule set for the unpointed mode in Phonology of Biblical Hebrew; the measure is the treebank comparison. I3 is the reporting rule for tools/hbo. The paper is Arabic, so no Arabic chapter is needed.

Cite

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

@inproceedings{alkhairy2020finite,
  title = {{Finite State Machine Pattern-Root Arabic Morphological Generator, Analyzer and Diacritizer}},
  author = {Alkhairy, Maha and Jafri, Afshan and Smith, David},
  booktitle = {Proceedings of the Twelfth Language Resources and Evaluation Conference},
  year = {2020},
  url = {https://aclanthology.org/2020.lrec-1.473/}
}

References

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

1. Al-Razi, M. b. A. B. (1981). Mukhtar al-sihah. Beirut: Dar al-Fikr. find
2. Allauzen, C., Riley, M., Schalkwyk, J., Skut, W., and Mohri, M. (2007). OpenFst: A general and efficient weighted finite-state transducer library. In International Conference on Implementation and Application of Automata, pages 11–23. Springer. find
3. Beesley, K. R. and Karttunen, L. (2003). Finite-state morphology: Xerox tools and techniques. CSLI, Stanford. find
4. Bird, S., Klein, E., and Loper, E. (2009). Natural language processing with Python: Analyzing text with the Natural Language Toolkit. O’Reilly Media, Inc. find
5. Dahdah, A. and Abdulmassih, G. M. (1981). A dictionary of Arabic grammar in charts and tables. Librairie du Liban. find
6. El-Dahdah, A., Matar, E., and Abdul-Massih, G. M. (1990). Majam qawaaid al-arabiiat al-aalamiiat (A Dictionary of Universal Arabic Grammar). find
7. Farghaly, A. A. S. (2010). Arabic computational linguistics. CSLI Publications, Center for the Study of Language and Information. find
8. Habash, N. Y. (2010). Introduction to Arabic natural language processing. Synthesis Lectures on Human Language Technologies, 3(1):1–187. find
9. Hajič, J., Smrz, O., Zemánek, P., Šnaidauf, J., and Beška, E. (2004). Prague Arabic dependency treebank: Development in data and tools. In Proc. of the NEMLAR Intern. find

This page in Markdown → https://md.mova.elamur.ai/free/en/docs/papers/lingbulk-2020-alkhairy-finite-state-machine-pattern-root/