TY - CONF A1 - Glocker, Kevin A1 - Georges, Munir T1 - Hierarchical Multi-Task Transformers for Crosslingual Low Resource Phoneme Recognition BT - Proceedings of the 5th International Conference on Natural Language and Speech Processing (ICNLSP 2022) N2 - This paper proposes a method for multilingual phoneme recognition in unseen, low resource languages. We propose a novel hierarchical multi-task classifier built on a hybrid convolution-transformer acoustic architecture where articulatory attribute and phoneme classifiers are optimized jointly. The model was evaluated on a subset of 24 languages from the Mozilla Common Voice corpus. We found that when using regular multi-task learning, negative transfer effects occurred between attribute and phoneme classifiers. They were reduced by the hierarchical architecture. When evaluating zero-shot crosslingual transfer on a data set with 95 languages, our hierarchical multi-task classifier achieves an absolute PER improvement of 2.78% compared to a phoneme-only baseline. KW - speech recognition KW - multilingual KW - zero-shot KW - multi-task learning Y1 - 2022 UR - https://opus4.kobv.de/opus4-haw/frontdoor/index/index/docId/3054 UR - https://aclanthology.org/2022.icnlsp-1.21 UR - https://nbn-resolving.org/urn:nbn:de:bvb:573-30544 SN - 978-1-959429-36-4 SP - 187 EP - 192 PB - Association for Computational Linguistics CY - Stroudsburg ER -