Morphological Analysis and Disambiguation for Gulf Arabic: The Interplay between Resources and Methods

Proceedings of the 12th Conference on Language Resources and Evaluation (LREC 2020), pages 3895–3904 Marseille, 11–16 May 2020 c⃝ European Language Resources Association (ELRA), licensed under CC-BY-NC 3895 Morphological Analysis and Disambiguation for Gulf Arabic: The Interplay between Resources and Methods Salam Khalifa, Nasser Zalmout, Nizar Habash Computational Approaches to Modeling Language (CAMeL) Lab New York University Abu Dhabi {salamkhalifa, nasser.zalmout, nizar.habash}@nyu.edu Abstract In this paper we present the first full morphological analysis and disambiguation system for Gulf Arabic. We use an existing state-of-the- art morphological disambiguation system to investigate the effects of different data sizes and different combinations of morphological analyzers for Modern Standard Arabic, Egyptian Arabic, and Gulf Arabic. We find that in very low settings, morphological analyzers help boost the performance of the full morphological disambiguation task. However, as the size of resources increase, the value of the morphological analyzers decreases. Keywords: Gulf Arabic, Morphology, Disambiguation 1. Introduction Despite the many advances in the field in Natural Language Processing (NLP) for Arabic, many dialectal Arabic vari- eties are lagging behind. Modern Standard Arabic (MSA), the official language of the Arab world, is well studied in NLP and has an abundance of resources including corpora and tools. On the other hand, most Arabic dialects are considered under-resourced, with the exception of Egyp- tian Arabic (EGY). One of the main challenges for Arabic NLP is data spar- sity due to morphological richness and the lack of orthog- raphy standards. To address such challenges, recent NLP models use deep learning architectures that have access to character and subword-level information. Such models are well equipped to model some aspects of morphology im- plicitly as part of an end-to-end system without requiring explicit feature engineering. However, these models are very data-intensive, and do not scale down well in the case of low-resource languages. Moreover, some morphological behaviors can be complicated and irregular at times. This makes morphology more challenging to capture implicitly while modeling other tasks. This further highlights the im- portance of explicit morphological modeling for languages that are morphologically-rich and low-resource in particu- lar. Morphological analyzers are dictionary-like resources that provide all the potential analyses for a given word out-of-context. Ideal morphological analyzers are expected to return all possible analyses of a given word (modeling ambiguity), along with covering all the different inflected forms of a word lemma (modeling richness). The quality of the morphological analyzers varies drastically based on the method used to create them. Higher quality analyzers are carefully built using existing linguistic dictionaries, and are manually checked. On the other end of the spectrum, mor- phological analyzers created automatically (e.g., extracted from annotated text or seed dictionaries) can have a lower quality depending on the quantity and quality of data used or the methods employed in creating them. Morphological disambiguation is the process of provid- ing the most probable morphological analysis in context for a given word. This task is achieved by either ranking the output of a morphological analyzer or through an end- to-end system that …

Content truncated to 500 words