Abstract: We present a system for multiclass classification of simplified morphology of Spanish verbs within the framework of morphology generation for Statistical Machine Translation (SMT) from English into Spanish. In [1] it was proved that, when statistically translating from English into Spanish, the richness of morphology of the target language affects the translation models at training time by creating data sparseness. In order to determine the correct morphology of the Spanish translation we use a hierarchical set of classifiers through a Decision Directed Acyclic Graph (DDAG) structure, each decision-node operates with a classifier which is a Support Vector Machine (SVM). This structure is justified because it allows to introduce prior information about the difficulty of the task. The classification results are analyzed and commented.
Index Terms: morphology, machine learning, statistical machine translation.