Universal Dependencies for Amharic

What it is about

The first UD treebank of Amharic, an Ethio-Semitic language: 1,096 sentences, 8,025 tokens (clitics not counted), taken mainly from grammar books, plus fiction, biographies, religious texts and news, hand-corrected for spelling. The hard step is clitic segmentation: one orthographic word can be a whole clause (a negated verb with subject and object pronouns), clitics fuse with their host in the syllabic script (ባንድ "in one" = በ + አንድ, so cutting off the first syllable leaves nonsense), assimilation hides morphemes (the passive prefix disappears in jussive and relative forms), and some strings are ambiguous between a split and an unsplit reading (ከሱ "from him" vs "they lost weight"). Unlike Hebrew and Arabic, the definite and accusative markers are suffixes that attach to the first element of the noun phrase, not necessarily the head, so the authors segment them and treat them as syntactic dependents (det, case). Subject and object agreement clitics are segmented too: with an overt nominal they get expl, without one they carry nsubj/obj (pro-drop). Converbs get advmod (co-extensive) or compound:svc (serial and consecutive), and light verbs are compounds headed by the existential/copular verb. Agreement was kappa 0.862 for spotting complex words, 0.585 for their segmentation, 0.622 for POS and 0.488 for relations, followed by linguists' review.

Ideas

• I1. Recover the hidden form before splitting: when a prefix fuses with its host in writing, the segmenter must restore the host's underlying first letter (Amharic ባንድ → በ + አንድ). The Hebrew analogue is the article swallowed by a preposition (בָּאָרֶץ = בְּ + הַ + אֶרֶץ), which tools/hbo already marks with H; Aramaic ד and ל fusions are the next cases to test the same way.
• I2. Agreement clitics as expl when the nominal is present and as the argument when it is absent: a uniform rule for pro-drop that the frame reader can use — the argument slot is filled by the overt noun if any, else by the clitic. Hebrew and Aramaic object suffixes on verbs fit the same rule.
• I3. Measure segmentation agreement in two parts, as here: whether a word is complex at all, and how it splits. The two kappas differ a lot (0.86 vs 0.59), so a single number would hide where the difficulty is; use the same two-part report for any hand-segmented Hebrew or Aramaic gold set.
• I4. A list of fused/ambiguous forms that may or may not split (the paper's ከሱ case) is a ready negative-control test: the segmenter must keep both readings and not split blindly.

For Mova

I1 and I4 are tests for tools/hbo and a future Aramaic segmenter (Aramaic in Mova: what it would need); I2 is a rule for prag frames and a note in Pronouns, particles and prepositions; I3 is a reporting rule for any gold segmentation set. The treebank itself is in treebanks/am.

Topics: Universal Dependencies

Cite

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

@inproceedings{seyoum2018universal,
  title = {{Universal Dependencies for Amharic}},
  author = {Seyoum, Binyam Ephrem and Miyao, Yusuke and Mekonnen, Baye Yimam},
  booktitle = {Proceedings of the Eleventh International Conference on Language Resources and Evaluation (LREC 2018)},
  year = {2018},
  url = {https://aclanthology.org/L18-1350/}
}

References

19 references, 3 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

1. Binyam, E., Miyao, Y., and Baye, Y. (2016). Morphosyntactically Annotated Amharic Treebank. In Proceedings of Corpus Linguistics Fest (CLiF 2016), June 6-10, 2016 (pp. 48–57). find
2. Binyam, G. G. (2010). Part of Speech Tagging for Amharic. University of Wolverhampton. find
3. de Marneffe, M.-C., Dozat, T., Silveira, N., Haverinen, K., Ginter, F., Nivre, J., and Manning, C. D. (2014). Universal Stanford Dependencies: A cross-linguistic typology. In Proceedings of the Ninth International Conference on Language Resources and Evaluation (LREC14) (pp. 4585–4592). find our card
4. de Marneffe, M.-C., and Manning, C. D. (2008). Stanford typed dependencies manual. In Proceedings of COLING 2008 Workshop on Cross-framework and Cross-domain Parser Evaluation. find
5. Demeke, G. a., and Getachew, M. (2006). Manual annotation of Amharic news items with part-of-speech tags and its challenges. Ethiopian Languages Research Center Working Papers (Vol. 2). find
6. Desalegn, A. (2016). The Inceptive Construction and Associated Topics in Amharic and Related Languages. Stockholm University. find
7. Gambäck, B., Olsson, F., Argaw, A. A., and Asker, L. (2009). Methods for Amharic part-of-speech tagging. Proceedings of the First Workshop on Language Technologies for African Languages. find
8. Gamback B. (2012). Tagging and Verifying an Amharic News Corpus. In Proceedings of the Eighth International Conference on Language Resources and Evaluation. (pp. 79–84). European Language Resources Association. find
9. Goldberg, Y., Elhadad, M., and Gurion, B. (2009). Hebrew Dependency Parsing: Initial Results. In Proceedings of the 11th IWPT09 (pp. 129–133). Paris. Retrieved from http://www.aclweb.org/anthology/W09-3819 ↗ our card
10. Martha, Y. T., Solomon, T. A., and Besacier, L. (2011). Part-of-Speech Tagging for Under-Resourced and Morphologically Rich Languages – The Case of Amharic. In Conference on Human Language Technology for Development (pp. 50–55). Alexandria, Egypt. find
11. Meyer, R. (2011). The Converb in Amharic. Journal of Semitic Studies, (1980), 165–192. find
12. Nivre, J. (2015). Towards a Universal Grammar for Natural Language Processing. In A. Gelbukh (Ed.), Proceeding of the 16th International Conference of Computational Linguistics and Intelligent Text Processing, CICLing 2015 (Vol. 9041, pp. 3–16). Springer International Publishing Switzerland. find
13. Nivre, J., Agić, Ž., Ahrenberg, L., Aranzabe, M. J., Asahara, M., Atutxa, A., … Zhu, H. (2017). Universal Dependencies 2.0. Http://universaldependencies.org/. find
14. Nivre, J., de Marneffe, M.-C., Ginter, F., Bullet, M., Ginter, F., Goldberg, Y., … Zeman, D. (2016). Universal Dependencies v1: A Multilingual Treebank Collection. In Proceedings of the Tenth International Conference on Language Resources and Evaluation (LREC 2016) (pp. 1659–1666). find our card
15. Petrov, S., Das, D., and Mcdonald, R. (2012). A Universal Part-of-Speech Tagset. In Proceedings of the International Conference on Language Resources and Evaluation, LREC 2012 (pp. 2089–2096). find
16. Rychlý, P., and Suchomel, V. (2016). Annotated Amharic Corpora. In 19th International Conference on Text, Speech and Dialogue (TSD 2016) (Vol. LNAI 9924, pp. 295–302). Brno, Czech Republic: Springer International Publishing Switzerland. find
17. Sisay, F. A. (2005). Part of Speech tagging for Amharic using Conditional Random Fields. In Proceedings of the ACL workshop on computational approaches to semitic languages (pp. 47–54). Association for Computational Linguistics. find
18. Zeman, D. (2008). Reusable Tagset Conversion Using Tagset Drivers. In Proceedings of the 6th International Conference on Language Resources and Evaluation (LREC) (pp. 213–218). find
19. Zeman, D., Mareček, D., Popel, M., Ramasamy, L., Štěpánek, J., Žabokrtský, Z., and Hajič, J. (2012). HamleDT: To Parse or Not to Parse? In Proceedings of the 8th International Conference on Language Resources and Evaluation (LREC) (pp. 2735–2741). find

This page in Markdown → https://md.mova.elamur.ai/free/en/docs/papers/ud-2018-seyoum-universal-dependencies-amharic/