Combinatorial Optimization Approach for Arabic Word Recognition Based on Adaptive Simulated Annealing

Abstract

The present paper proposes an approach based on a combinatorial optimization technique for Arabic word recognition, distinguished by its flexional nature and significant topological variability. We treat a large vocabulary of Arabic decomposable words, which we choose to factorize them by their roots and schemes. We adopt a structure that resembles a molecular cloud. This design rhymes well with the Arabic linguistic philosophy of constructing words from their roots. Each sub-vocabulary, corresponding to a sub-cloud, embodies neighboring words, which are derived from one root and follow different schemes and forms of derivation, flexion, and agglutination (proclitic and enclitic). Therefore, we propose to use the metaheuristic simulated annealing (SA) method, as a recognition approach, in this wide cloud. It’s an algorithm based on elastic comparisons between their structures and primitives. As an extension of previous works, we opt to implement the SA algorithm by integrating linguistic knowledge. Preliminary experiments were conducted on Arabic word corpus including samples and agglutinated words from APTI database and yielded interesting outcomes.