AdaSL: An Unsupervised Domain Adaptation framework for Arabic multi-dialectal Sequence Labeling
作者：
Highlights：
• We present AdaSL, an unsupervised framework for dialectal Arabic sequence labeling.
• We introduce a sub-word pooling aggregator for the full representation of words.
• We apply AdaSL for multilingual and Arabic Transformer pre-trained language models.
• We validate AdaSL on Named Entity Recognition (NER) and Part-of-Speech (POS) tagging.
• We achieve new SOTA zero-shot performances for dialectal Arabic sequence labeling.
摘要
•We present AdaSL, an unsupervised framework for dialectal Arabic sequence labeling.•We introduce a sub-word pooling aggregator for the full representation of words.•We apply AdaSL for multilingual and Arabic Transformer pre-trained language models.•We validate AdaSL on Named Entity Recognition (NER) and Part-of-Speech (POS) tagging.•We achieve new SOTA zero-shot performances for dialectal Arabic sequence labeling.
论文关键词：Dialectal Arabic,Arabic natural language processing,Domain adaptation,Multi-dialectal sequence labeling,Named entity recognition,Part-of-speech tagging,Zero-shot transfer learning
论文评审过程：Received 15 December 2021, Revised 24 April 2022, Accepted 25 April 2022, Available online 6 May 2022, Version of Record 6 May 2022.
论文官网地址：https://doi.org/10.1016/j.ipm.2022.102964
