Skip to content

Files

Latest commit

shijie-wushijie-wu
Integrate pre-commit and make it happy
Nov 3, 2020
2a9e1f7 · Nov 3, 2020

History

History

sigmorphon2019-shared-tasks

README.md

SIGMORPHON 2019 Shared Tasks Baseline

Arya D McCarthy, Ekaterina Vylomova, Shijie Wu, Chaitanya Malaviya, Lawrence Wolf-Sonkin, Garrett Nicolai, Miikka Silfverberg, Sebastian J Mielke, Jeffrey Heinz, Ryan Cotterell, and Mans Hulden. The SIGMORPHON 2019 Shared Task: Morphological Analysis in Context and Cross-Lingual Transfer for Inflection. SIGMORPHON. 2019.

Task 1: crosslingual-inflection-baseline

Baseline Performance

Available for all baselines here.

Training from Scratch

dir=example/sigmorphon2019-shared-tasks
# 0-soft
sh $dir/run-task1-tag.sh soft adyghe--kabardian
# 0-hard
sh $dir/run-task1-tag.sh hard adyghe--kabardian
# 0-mono
sh $dir/run-task1-monotag.sh hmm adyghe--kabardian
# 1-mono
sh $dir/run-task1-monotag.sh hmmfull adyghe--kabardian

Decoding with Pretrained Model

dir=example/sigmorphon2019-shared-tasks
python src/sigmorphon19-task1-decode.py \
    --in_file $dir/sample/task1/adyghe--kabardian/kabardian-dev \
    --out_file decode/task1/adyghe--kabardian-dev-out \
    --lang kabardian \
    --model $dir/sample/task1/model/adyghe--kabardian.1-mono.pth

Task 2: contextual-analysis-baseline

Baseline Performance

Available for all baselines here.

Task 2 decoded files: https://www.dropbox.com/s/2kqkhsu0kil6rzu/BASELINE-DEV-00-2.tar.gz

Training from Scratch

We train the model with the jackknifing training data and at dev time, we decode the lemma with predicted tag.

dir=example/sigmorphon2019-shared-tasks
# 0-mono
sh $dir/run-task2.sh af_afribooms

Decoding with Pretrained Model

dir=example/sigmorphon2019-shared-tasks
python src/sigmorphon19-task2-decode.py \
    --in_file $dir/sample/task2/af_afribooms-um-dev.conllu.baseline.pred \
    --out_file decode/task2/af_afribooms-um-dev.conllu.output \
    --model $dir/sample/task2/model/af_afribooms.pth

Pretrained Models

Link: https://www.dropbox.com/sh/p4vu5imyn69wyyp/AAA-3bQeGJmnCex78xx7T0ZPa

Size of models:

3.4G	sigmorphon2019/public/task2/lemmatizer-model