Enriching the NArabizi Treebank: A Multifaceted Approach to Supporting an Under-Resourced Language
Arij Riabi, Menel Mahamdi, Djamé Seddah
TL;DR
This work targets the scarcity of annotated data for the low-resource NArabizi dialect by delivering an enriched Narabizi Treebank (V2) with two new annotation layers for Named Entity Recognition (NER) and Offensive Language Detection, along with a thorough re-annotation of tokenization and UD layers. The authors conduct a multi-faceted evaluation using gold and non-gold tokenization across UD parsing, NER, and offensive language tasks, discovering that tokenization consistency and high-quality annotations substantially improve downstream performance. They also develop a new CharacterBERT pretraining corpus and compare subword models, showing state-of-the-art UD results on NarabiziV2 and demonstrating robust performance for noise-prone data. The dataset and models are publicly released, enabling better NLP tooling for NArabizi and serving as a resource for research into under-resourced dialects and user-generated content.
Abstract
In this paper we address the scarcity of annotated data for NArabizi, a Romanized form of North African Arabic used mostly on social media, which poses challenges for Natural Language Processing (NLP). We introduce an enriched version of NArabizi Treebank (Seddah et al., 2020) with three main contributions: the addition of two novel annotation layers (named entity recognition and offensive language detection) and a re-annotation of the tokenization, morpho-syntactic and syntactic layers that ensure annotation consistency. Our experimental results, using different tokenization schemes, showcase the value of our contributions and highlight the impact of working with non-gold tokenization for NER and dependency parsing. To facilitate future research, we make these annotations publicly available. Our enhanced NArabizi Treebank paves the way for creating sophisticated language models and NLP tools for this under-represented language.