SIGMORPHON 2021 Shared Task on Morphological Reinflection: Generalization Across Languages

• Published in: Proceedings of the 18th SIGMORPHON Workshop on Computational Research in Phonetics, Phonology, and Morphology
• Main Authors: Pimentel, Tiago, Ryskina, Maria, Mielke, Sabrina J., Wu, Shijie, Chodroff, Eleanor, Leonard, Brian, Nicolai, Garett, Ate, Yustinus Ghanggo, Khalifa, Salam, Habash, Nizar, El-Khaissi, Charbel, Goldman, Omer, Gasser, Michael, Lane, William, Coler, Matt, Oncevay, Arturo, Montoya Samame, Jaime Rafael, Silva Villegas, Gema Celeste, Ek, Adam, Bernardy, Jean-Philippe, Shcherbakov, Andrey, Bayyr-ool, Aziyana, Sheifer, Karina, Ganieva, Sofya, Plugaryov, Matvey, Klyachko, Elena, Salehi, Ali, Krizhanovsky, Andrew, Krizhanovsky, Natalia, Vania, Clara, Ivanova, Sardana, Salchak, Aelita, Straughn, Christopher, Liu, Zoey, Washington, Jonathan, Ataman, Duygu, Kieraś, Witold, Woliński, Marcin, Suhardijanto, Totok, Stoehr, Niklas, Nuriah, Zahroh, Ratan, Shyam, Tyers, Francis M., Ponti, Edoardo M., Aiton, Grant, Hatcher, Richard J., Prud'hommeaux, Emily, Kumar, Ritesh, Hulden, Mans, Barta, Botond, Lakatos, Dorina, Szolnok, Gábor, Ács, Judit, Raj, Mohit, Yarowsky, David, Cotterell, Ryan, Ambridge, Ben, Vylomova, Ekaterina
• Other Authors: Department of Computer Science
• Format: Conference Object
• Language: English
• Published: 2022
• Subjects: 113 Computer and information sciences; Aymara; Evenki; Sakha; Chukchi; Itelmen; karelian
• Online Access: http://hdl.handle.net/10138/340752

This year's iteration of the SIGMORPHON Shared Task on morphological reinflection focuses on typological diversity and cross-lingual variation of morphosyntactic features. In terms of the task, we enrich UniMorph with new data for 32 languages from 13 language families, with most of them being under-resourced: Kunwinjku, Classical Syriac, Arabic (Modern Standard, Egyptian, Gulf), Hebrew, Amharic, Aymara, Magahi, Braj, Kurdish (Central, Northern, Southern), Polish, Karelian, Livvi, Ludic, Veps, Võro, Evenki, Xibe, Tuvan, Sakha, Turkish, Indonesian, Kodi, Seneca, Asháninka, Yanesha, Chukchi, Itelmen, Eibela. We evaluate six systems on the new data and conduct an extensive error analysis of the systems' predictions. Transformer-based models generally demonstrate superior performance on the majority of languages, achieving >90% accuracy on 65% of them. The languages on which systems yielded low accuracy are mainly under-resourced, with a limited amount of data. Most errors made by the systems are due to allomorphy, honorificity, and form variation. In addition, we observe that systems especially struggle to inflect multiword lemmas. The systems also produce misspelled forms or end up in repetitive loops (e.g., RNN-based models). Finally, we report a large drop in systems' performance on previously unseen lemmas. Peer reviewed