AlphaBrains at WojoodNER shared task: Arabic Named Entity Recognition by Using Character-based Context-Sensitive Word Representations

Abstract

This paper presents Arabic named entity recognition models by employing single-task and multi-task learning paradigms. The models were developed by using character-based contextualized Embeddings from Language Model (ELMo) in the input layers of the Bidirectional Long-Short Term Memory (BiLSTM) networks. The ELMo embeddings are quite capable of learning the morphology and contextual information of tokens in word sequences. The single-task learning model outperformed the multi-task learning model, achieving micro F 1 -scores of 0.8751 and 0.8884, respectively, ranking 10 th and 7 th in the shared task for flat and nested NER.