BabyFST - Towards a Finite-State Based Computational Model of Ancient Babylonian

What it is about

The paper describes BabyFST, a finite-state morphological analyser for Babylonian (the southern dialect of Akkadian) that works on transcribed input. Verb stems are not built dynamically: the authors enumerate about 352k verbal stems for 1,410 lemmas from 1,341 templates over 42 conjugation classes, and they add noun, adjective and particle lexicons taken from Oracc, about 404k entries in all. Twenty-six rewrite rules handle assimilation, metathesis and spelling variation, and the compiled transducer is 6.2 MB. On the transcribed Oracc texts it reaches coverage of 95.8–97.4% and token recall of 90–94% for lemma and POS, but precision is only about 40%, because only 20.1% of tokens have one analysis. A first weighting of the transducer from partly validated corpus data raises the share of tokens whose top-ranked analysis is right to 57.4%.

Ideas

• I1. Weight the analyses from frequency: take the counts of analyses in a partly validated corpus and rank the analyses of each token by them, as the paper does. The measure is top-1 correctness on a held-out split, with the split fixed and the run repeated to check it is identical.
• I2. Template-generated stems: store one root with a class and vowel pattern, generate the stem list once at build time, and let the compiled automaton share the common parts. The authors show this scales to 352k stems. The same idea fits a Hebrew or Aramaic stem table generated from root and binyan/stem templates rather than built during parsing.
• I3. Compatibility as flags: suffix sets that are only allowed on some stems (the middle-weak ♂/♀ marks in the paper) are controlled by a flag on the stem, not by duplicating the lexicon. This is a good model for weak verbs whose endings depend on the final radical.
• I4. Report by variety: the paper gives coverage, recall and precision separately for the five dialect sub-corpora. Do the same for each Aramaic variety, so that a weakness in one (for example, Late Babylonian) is not hidden in an average.

For Mova

I1 is a weighting table for tools/hbo, measured against held-out tokens. I2 and I3 apply to The verb and The weak verbs (stem generation and flags for weak roots), and to The verb in Biblical Aramaic. I4 is a reporting rule for languages/arc-x-targum/, languages/tmr/ and languages/jpa/.

Cite

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

@inproceedings{sahala2020babyfst,
  title = {{BabyFST - Towards a Finite-State Based Computational Model of Ancient Babylonian}},
  author = {Sahala, Aleksi and Silfverberg, Miikka and Arppe, Antti and Lindén, Krister},
  booktitle = {Proceedings of the Twelfth Language Resources and Evaluation Conference},
  year = {2020},
  url = {https://aclanthology.org/2020.lrec-1.479/}
}

References

26 references, 2 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

1. Allauzen, C.; Riley, M.; Schalkwyk, J.; Skut, W. & Mohri, M. (2007). OpenFst: A General and Efficient Weighted Finite-State Transducer Library . In Conference on Implementation and Application of Automata. find
2. Arppe, A.; Cox, C.; Hulden, M.; Lachler, J.; Moshagen, S. N.; Silfverberg, M. & Trosterud, T. (2017). Computational Modeling of Verbs in Dene Languages: The Case of Tsuut’ina. In Jaker, A. (ed.), Working Papers in Athabaskan (Dene) Languages , 51-69. Alaska Native Language Center Working Papers 13. Fairbanks: Alaska Native Language Center. find
3. Bamman, D. (2012). 11-712 NLP Lab Report. https://github.com/dbamman/akkadian-morph-analyzer/blob/master/doc/latex/whitepaper.pdf ↗
4. Barthelémy, F. (1998). A Morphological Analyzer for Akkadian Verbal Forms with a Model of Phonetic Transformations. In Computational Approaches on Semitic Languages . find our card
5. Beesley, K, and Karttunen, L. (2003). Finite State Morphology. P alo Alto, California: CSLI Publications. find
6. Black, J. A., George, A. R. and Postgate J. N. (2000). A Concise Dictionary of Akkadian . Wiesbaden: Harrassowitz Verlag. find
7. Buccellati, G. (1996). A Structural Grammar of Babylonian . Wiesbaden: Harrassowitz Verlag. find
8. Harrigan, A.; Schmirler, K.; Arppe, A.; Antonsen, L.; Moshagen, S. N.; Trosterud, T. & Wolvengrey, A. (2017). Learning from the Computational Modeling of Plains Cree Verbs. Morphology , 27(4), 565–598. find
9. Hulden, M. (2009). Foma: a Finite-State Compiler and Library. In Proceedings of the 12th Conference of the European Chapter of the Association for Computational Linguistics (EACL): Demonstrations Session, 29–32. Athens: Association for Computational Linguistics. find
10. Kataja, L. and Koskenniemi, K. (1988). Finite-State Description of Semitic Morphology: A Case Study of Ancient Akkadian. find
11. Coling Budapest 1988 Volume 1: International Conference on Computational Linguistics. find
12. Kouwenberg, N. J. C. (2010). The Akkadian Verb and its Semitic Background. Languages of the Ancient Near East 2. Winona Lake: Eisenbrauns. ↗
13. Kouwenberg, N. J. C. (2011). Akkadian in General. In Semitic Languages. An International Handbook. S. Weninger, G. Khan, M. P. Streck & J. C. E. Watson (Eds.), De Gruyter Mouton, pp. 330–340. find
14. Lindén, K., Axelson, E., Hardwick, S., Pirinen, T. A. & Silfverberg, M. (2011). HFST-Framework for Compiling and Applying Morphologies. In Systems and Frameworks for Computational Morphology. Mahlow, C. & Piotrowski, M. (eds.). Springer-Verlag, pp. 67-85 19 p. (Communications in Computer and Information Science; vol. 100). find
15. Macks, A. (2002). Parsing Akkadian Verbs with Prolog. SEMITIC '02 Proceedings of the ACL-02 workshop on Computational approaches to Semitic languages. pp 1–6. find our card
16. Parpola, S. and Whiting, R. M. (2007). Assyrian-English-Assyrian Dictionary . Helsinki: The Neo-Assyrian Text Corpus Project. find
17. Reiner, E. (1966). A Linguistic Analysis of Akkadian. Janua Linguarum, Series Practica 21. The Hague: Mouton. ↗
18. Sahala, A. (2014). Babylonian diskontinuatiivisen morfologian ohjelmallinen jäsentäminen . MA Thesis, University of Helsinki. find
19. Soden, von, W. (1995). Grundriss der akkadischen Grammatik , 3., ergänzte Auflage (Analecta Orientalia 33/47) Roma: Pontificio Istituto Biblico. find
20. Streck, P. M. (2011). Babylonian and Assyrian. In Semitic Languages. An International Handbook. S. Weninger, G. Khan, M. P. Streck & J. C. E. Watson (Eds.), De Gruyter Mouton, pp. 359–398. find
21. Tinney, S. (2018). L2: How It Works. Oracc: The Open Richly Annotated Cuneiform Corpus. http://oracc.org/doc/help/lemmatising/howl2works/Language ↗
22. Resource References ARCHIBAB. (2008). Archives babyloniennes (XX e -XVII e siècles) http://www.archibab.fr/ ↗
23. CDLI. (2000). The Cuneiform Digital Library Initiative. https://cdli.ucla.edu/ ↗
24. Oracc. (2014). The Open Richly Annotated Cuneiform Corpus. http://oracc.org ↗
25. SAA(o). (1986). State Archives of Assyria (online). http://www.helsinki.fi/science/saa/ ↗
26. SEAL. (2008). Sources for Early Akkadian Literature. https://www.seal.uni-leipzig.de/ ↗

This page in Markdown → https://md.mova.elamur.ai/free/en/docs/papers/lingbulk-2020-sahala-babyfst-finite-state-based-computational/