Controlled Semi-automatic Annotation of Classical Ethiopic

What it is about

The paper describes the annotation workflow of the TraCES project for Classical Ethiopic (Gǝʿǝz), a very low-resource Semitic language with no usable lexicon and no corpus large enough for statistical tools. Annotation is done in two stages: a manual deep-annotated corpus first, built with a controlled semi-automatic step, then a learned tagger trained on that corpus. The tag set has about 30 parts of speech, several of them particles and pronouns with many subtypes. The tool, GeTa, stores objects by id: graphical units (text ending at the word separator), tokens, Ethiopic characters and transcription letters. Only the transcription letters are stored; the other labels are generated from a correspondence table. Transliteration is first done by rule, then corrected by hand, because gemination and the sixth-order vowel depend on the part of speech, which is not known at that point. No numbers are reported: the paper is a design description.

Ideas

I1. Two-stage transliteration with a rule first and a human pass second, for tools/hbo pointed input: the rule layer gives the default vowels of each consonant, and only the forms the rules cannot decide (the ones that depend on the word class) go to review.
I2. Store only the base units and derive the rest from a table: in tools/hbo, keep the consonants and the point marks as the stored layer and generate the consonantal, pointed and transliterated views from the table. This keeps the unpointed and pointed modes in one data model and avoids drift between them.
I3. Use an explicit fine tag set with key-value features on each token, so new features can be added without changing the tag: a useful check against the Hebrew tag set used by macula compare, to see which of the 30 classes have no counterpart in UD.

For Mova

I1 goes into 1. Script and pointing and tools/hbo (the pointed-input path). I2 goes into tools/hbo (data model) and tools/ocrfix (layers of a printed edition). I3 is a note for The lexicon and tools/macula (tag comparison). The Ethiopic script itself is not used by Mova.

Topics: Part-of-speech tagging

Cite

The paper itself, not this card. BibTeX from the library's record of the work. DOI 10.26615/978-954-452-059-5_004

@inproceedings{vertan2019controlled,
  title = {{Controlled Semi-automatic Annotation of Classical Ethiopic}},
  author = {Vertan, Cristina},
  booktitle = {Proceedings of the Workshop on Language Technology for Digital Historical Archives},
  year = {2019},
  doi = {10.26615/978-954-452-059-5\_004},
  url = {https://doi.org/10.26615/978-954-452-059-5_004}
}

References

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

ይ contains the Transcription-letter objects: y and ǝ Bollmann, Marcel and Petran, Florian and Dipper,Stefanie and Krasselt, Julia 2014: ‘ CorA: A web-based annotation tool for historical and other • ቤ contains the Transcription-letter objects: b and e nonstandard language data’, in:Kalliopi Zervanou and Cristina Vertan and Antal van den Bosch and Caroline Sporleder (Eds.), Proceedings of the 8th Workshop on Language Technology for Cultural Heritage, Social Sciences, and Humanities (LaTeCH) Gothenburg, Sweden April 2014, 86-90. find
ሎ contains the Transcription-letter objects: l and o Throughout the transliteration-tokenisation phase three Token-objects are built: wa, yǝbel, and o Finally, the initial GU-Object will have attached two labels: ወይቤሎ and wa-yǝbel- o. For synchronisation reasons we consider the word separator Druskat, Stephan and Vertan, Cristina 2017, ‘ Nachnutzbarmachung von Forschungsdaten und Tools am Beispiel altäthiopischer Korpora ’ , in Gerog Vogeler (ed.) Kritik der Digitaler Vernunft Konferenzabstracts, Köln 2018, 270-273 find
Eckart de Castilho, Richard and Mújdricza-Maydt, Éva and Yiman, Seid Muhie and Hartmann,Silvana and Iryna and Frank, Anette and Biemann, Chris 2016, ‘A Web-based Tool for the Integrated Annotation of Semantic and Syntactic Structures’, in Erhard W. Hinrichs and Marie Hinrichs,and Thorsten Trippel (eds.), Proceedings of the LT4DH workshop at COLING 2016, Osaka, Japan: 76-84. find
Hummel, Susanne and Wolfgang Dickhut 2016. ‘A part of speech tag set for Ancient Ethiopic’, in Alessandro Bausi and Eugenia Sokolinski, eds, 150 find
Years after Dillmann ’ s Lexicon: Perspectives and Challenges of Gǝʿǝz Studies, Supplement to Aethiopica, 5 (Wiesbaden: Harrassowitz Verlag, 2016), 17–29. find
Krzyżanowska, Magdalena 2017. ‘A Part-of-Speech Tagset for Morphosyntactic Tagging of Amharic’, Aethiopica, 20 (2017), 210–235. find
Maegaard, Bente and Krauwer, Steven and Choukri, Khalid and Jørgensen, Lars, 2006, ‘The BLARK concept and BLARK for Arabic’, in Proceedings of the LREC Conference 2006, http://lrec-conf.org/proceedings/lrec2006 ↗

This page in Markdown → https://md.mova.elamur.ai/free/en/docs/papers/aclbulk-2019-vertan-controlled-semi-automatic-annotation-classical/