Description
Since 2016, a shared task on morphological reinflection has been organized and extended to
universal morphological reinflection (Cotterell et al., 2016, 2017).1 The task is to generate a target
inflected form based on given a lemma with morphological features.
For this homework, you are going to implement the task 1 of CoNLL–SIGMORPHON 2018 Shared Task: Universal Morphological
Reinflection. All the required files are under ~/dropbox/19-20/570/hw5/task1.
Rubric:
2pts hw.tar.gz submitted, it should contain following files:
• baseline improved.sh
2pts readme.txt or readme.pdf submitted
6pts All files and folders are present in expected locations
10pts Programs run to completion
5pts The output of programs on patas match submitted output
(75pts) Task 1: Type-level inflection
• Source form and target features:
– release V;V.PTCP;PRS
• Target form:
– releasing
Example of train:
reimburse reimburses V;3;SG;PRS
transliterate transliterate V;NFIN
tend tended V;PST
disallow disallowing V;V.PTCP;PRS
reconfirm reconfirmed V;V.PTCP;PST
The type of labels and their frequencies in train:
1https://sigmorphon.github.io/sharedtasks/2018/
2008 V;3;SG;PRS
2008 V;NFIN
2014 V;PST
1984 V;V.PTCP;PRS
1986 V;V.PTCP;PST
There are 77 in all and 9 languages in surprise.
• wc -l ../all/english-*
– 1000 ../all/english-dev
– 1000 ../all/english-test
– 100 ../all/english-train-low
– 1000 ../all/english-train-medium
– 10000 ../all/english-train-high
• python2 baseline.py
– english[task 1/low]: 0.772
– english[task 1/medium]: 0.908
– english[task 1/high]: 0.949
More detailed information available at
https://github.com/sigmorphon/conll2018/tree/master/task1/baseline
1. use English and at least one other language (specify in your readme
2. use only *-train-high for extracting rules
3. working as a group up to two students is possible (let us know before Oct. 30) — HW
submission will be individual
4. ./baseline improved.sh which will learn and evaluate your model (as in baseline.py)
5. your output will be similar to the original baseline.py
6. (50pts) your result will be hopefully better than the baseline result
7. (25pts) in readme, describe in detail what you change/improve alignment, span merging
and/or rule extraction
8. in readme, describe how to apply transformation rules
References
Cotterell, R., Kirov, C., Sylak-Glassman, J., Walther, G., Vylomova, E., Xia, P., Faruqui, M.,
K¨ubler, S., Yarowsky, D., Eisner, J., and Hulden, M. (2017). CoNLL-SIGMORPHON 2017
Shared Task: Universal Morphological Reinflection in 52 Languages. In Proceedings of the CoNLL
SIGMORPHON 2017 Shared Task: Universal Morphological Reinflection, pages 1–30, Vancouver.
Association for Computational Linguistics.
Cotterell, R., Kirov, C., Sylak-Glassman, J., Yarowsky, D., Eisner, J., and Hulden, M. (2016).
The SIGMORPHON 2016 Shared Task—Morphological Reinflection. In Proceedings of the 2016
Meeting of SIGMORPHON, pages 10–22, Berlin, Germany. Association for Computational Linguistics.