A3C: Arabic Anaphora Annotated Corpus
Made by : Abdelhalim Hafedh Dahou, Mohamed Abdelmoazz, Mohamed Amine CHERAGUI
Cite us : Dahou, A. H., Abdelmoazz, M., & Cheragui, M. A. (2021). A3C: Arabic Anaphora Annotated Corpus. ICNLSP 2021, 117.
The A3C corpus objectif
The annotated corpora proposed a large resource for researchers who are trying to create tools for Arabic resolution systems and evaluating the performance of their works. On the other hand, the available corpora in Arabic language can pay more attention to the data training and more complicated learning algorithms.
Using The A3C corpus
The texts in our corpus are taken from the Alriyadh newspaper (a daily Arabic newspaper) and they are structured in XML format. The A3C corpus is divided into five categories: culture, sports, politics, economy, and miscellaneous.
Anaphora Types
What makes the anaphora resolution mechanism complex in natural language processing in general and in Arabic, in particular, is the fact that it can manifest in different forms (linguistic categories: lexical and grammatical).In our case , we tackle the pronominal and the verbal types.
Resolution Approache
In our project we used a rule-based approache which based mainly on language knowledge(Morphological,syntactic and semantic) and depended on three basic steps : determining search space, applying contraints and applying preferences.
Annotation Scheme
The A3T will add automatically the following tags for the antecedent and the anaphora : the first will be marked with the Antecedent tag and the second with Anaphor . We also include the features listed above in each antecedent and anaphora tag. Finally , the A3T will generate an XML file that contains the text with anaphoric relationship tags..
The motivation behind building the A3C corpus
The objective behind the creation of this corpus is to fill the lack of resources concerning the resolution anaphora (especially pronominal and verbal) in the Modern Standard Arabic language, by creating a newly annotated corpus that we have called A3C with the anaphoric relations. This study discusses the problem of a lack of resources for verbal anaphora.
Resolving and Annotation using the A3T tool
To satisfy our objective, we created A3T, an anaphoric annotating tool that uses linguistic and statistical rules to automatically detect anaphors and their referents. After that, We resort to human specialists to verify and correct our A3T annotation's errors for the corpus's credibility.
Features of the A3C corpus
The A3C contains the links between two types of anaphora (Pronominal and verbal) and theirs referents in different texts genres.In addition, A3C holds Morphological and statistical information about the referent and the pronominal anaphora such as gender , number and frequency.
