Learnability and generalisation of Arabic broken plural nouns

Area of Science:

• Computational Linguistics
• Morphological Analysis
• Arabic Linguistics

Background:

• Modern Standard Arabic features a complex noun plural system.
• It involves suffixing 'sound' plurals and 31 non-concatenative 'broken' plural patterns.
• This complexity presents challenges for morphological learnability.

Purpose of the Study:

• To computationally analyze singular-plural pairs in Arabic.
• To identify statistically relevant linguistic information for morphological generalization.
• To evaluate the effectiveness of an analogical approach for predicting Arabic plurals.

Main Methods:

• Computational analysis of singular-plural pairs using the Corpus of Contemporary Arabic.
• Application of the generalized context model (an analogical approach).
• 10 rounds of cross-validation to assess model robustness.

Main Results:

• The generalized context model successfully predicts plural forms for given singulars.
• The CV (Consonant-Vowel) template carries most of the predictive power.
• The model demonstrates robustness and stability across cross-validation.

Conclusions:

• An analogical approach, particularly using the CV template, is highly effective for Arabic plural prediction.
• This supports the abstraction of prosodic templates in Arabic morphology.
• Linguistic information related to the CV template is key for morphological generalization.