Tesis Doctoral "A computational model of Modern Standard Arabic verbal morphology based on generation"

Doctorando: Alicia González Martínez

Director: Antonio Moreno Sandoval (UAM)

Fecha: martes 29 de enero de 2013, 11h00

Lugar de defensa:  Sala de Vídeo del Departamento de Filología Árabe, Módulo II, Facultad de Filosofía y Letras, Universidad Autónoma de Madrid


The computational handling of non-concatenative morphologies is still a challenge in the field of natural language processing. Amongst the various areas of research, Arabic morphology stands out due to its highly complex structure.

We propose a model for Arabic verbal morphology based on a root-and-pattern approach, which satisfies both computational consistency and an elegant formalization. Our model defines an abstract representation of prosodic templates and a set of intertwined morphemes that operate at different phonological levels, as well as a separate module of rewrite rules to deal with morphophonological and orthographic alterations. Our verbal system model asserts that Arabic exhibits two conjugational classes. 

The computational system, named Jabalín, is focused on generation—the program generates a full annotated lexicon of verbal forms, which is subsequently used to develop a morphological analyzer and generator. The input of the system consists of a lexicon of 15,452 verb lemmas of both Classical Arabic and Modern Standard Arabic—taken from El-Dahdah (1991)—comprising a total of 3,706 roots. The output of the system is a lexicon of 1,684,268 verbal inflected forms. 

We carried out an evaluation against a lexicon of inflected verbs provided by the analyzer ElixirFM (Smrž, 2007a; 2007b), which we considered a Golden Standard, achieving a precision of 99.52%. Additionally, we compared our lexicon with a list of the most frequent verb lemmas—including the most frequent verbs from each conjugation—taken from Buckwalter and Parkinson (2010). The list includes 825 verbs which are all included in our lexicon and passed an evaluation test with 99.27% of accuracy.

Jabalín is available under a GNU license, and can be accessed and tested through an online interface. The Jabalín interface provides different functionalities: analyze a form, generate the inflectional paradigm of a verb lemma, derive a root, show quantitative data, and explore the database, which includes data from the evaluation.



RocketTheme Joomla Templates