Machine learning models have largely served NLP applications, especially parsing task. All kinds of approaches were applied: symbolic, stochastic and connectionist. Ensemble learning is the learning paradigm that allows the use of multiple methods belonging to different approaches in a same system. This paradigm helps to reduce limits of individual methods. In this paper, we present a theoretical approach for parsing Arabic texts in an ensemble classification manner. A set of classifiers learns the same data (pairs of sentence and its correspondent derivation tree) from a Treebank presented using the TAG formalism. Then, each of them will predict the appropriate elementary tree to attribute to a word in a specified context. Individual results of all the classifiers will be combined using a combination process. The construction of the whole syntactic structure of a sentence is incremental and deterministic. A hybrid approach seems to be beneficial for our deterministic parser.
展开▼
