The SIGMORPHON 2022 Shared Task on Morpheme Segmentation

Abstract

The SIGMORPHON 2022 shared task on morpheme segmentation challenged systems to decompose a word into a sequence of morphemes and covered most types of morphology: compounds, derivations, and inflections. Subtask 1, word-level morpheme segmentation, covered 5 million words in 9 languages (Czech, English, Spanish, Hungarian, French, Italian, Russian, Latin, Mongolian) and received 13 system submissions from 7 teams and the best system averaged 97.29% F1 score across all languages, ranging English (93.84%) to Latin (99.38%). Subtask 2, sentence-level morpheme segmentation, covered 18,735 sentences in 3 languages (Czech, English, Mongolian), received 10 system submissions from 3 teams, and the best systems outperformed all three state-of-the-art subword tokenization methods (BPE, ULM, Morfessor2) by 30.71% absolute. To facilitate error analysis and support any type of future studies, we released all system predictions, the evaluation script, and all gold standard datasets. 1

Proceedings of the 19th SIGMORPHON Workshop on Computational Research in Phonetics, Phonology, and Morphology

“…We demonstrate that subword-level modeling does work for morpheme segmentation through our submissions to the SIGMORPHON 2022 Shared Task on Morpheme Segmentation (Batsuren et al, 2022). Our subword-level model, an entmax transformer with sampled ULM tokenizations, outperforms our character-level submissions and wins the word-level subtask.…”

“…With the goal of advancing research in this direction, we present a morpheme segmentation shared task and provide large-scale datasets over nine languages, evaluation metrics, and morphological annotations of five million word formations. In this, we rely on the latest release of UniMorph (Batsuren et al, 2022) which has introduced morpheme segmentations and derivational data from MorphyNet (Batsuren et al, 2021b). The resulting shared task is a follow-up to past morphological segmentation shared tasks such as "MorphoChallenge" (Kurimo et al, 2007(Kurimo et al, , 2008(Kurimo et al, , 2009 or "Multilingual parsing" (Zeman et al, 2017, where lemmatization as segmentation is a subtask).…”

“…We also collected base forms. For derivational and inflectional morphology, we have used the segmentation data from UniMorph 4.0 (Batsuren et al, 2022) and 2 https://github.com/sigmorphon/ 2022SegmentationST/tree/main/evaluation MorphyNet (Batsuren et al, 2021b). UniMorph contains inflectional paradigms collected from linguistic sources as well as Wiktionary, while Mor-phyNet represents derivations scraped from various editions of Wiktionary.…”

“…Previous approaches to morpheme segmentation include unsupervised methods (Creutz and Lagus, 2007), and more recently there have been neural approaches (Wang et al, 2016). This paper is a submission to the SIGMOR-PHON 2022 shared task on morpheme segmentation, which aims to benefit the NLP community with improvements for subword-based tokenization through morpheme segmentation (Batsuren et al, 2022). The shared task includes word-level and sentence-level morpheme segmentation subtasks for various development languages.…”

“…This work presents the AUUH (Aalto University -University of Helsinki) team submission to the SIGMORPHON 2022 Shared Task on Morpheme Segmentation (Batsuren et al, 2022). In this shared task, the imbalance of training data persists.…”