I argue for lexical representations of Arabic verbs in which consonants and vowels are represented together (Berent et al. 2007), as in all other languages. I claim that the vowel choice of imperfective verbs is determined by the consonants surrounding the imperfective vowel. In addition, I argue that native speakers learn the impact of consonants on the imperfective vowel selection and consequently extend their knowledge to nonce verbs. Existing analyses focusing on Semitic languages such as Modern Standard Arabic (Brame 1970, Gafos 2003, Shekeri & Scheer 2004, 2005, Rosenthall 2006) claim that such languages require a particular structure of lexical representations; they require a consonantal root. Instead of arguing in favor of a special morphemic structure exclusive to Semitic languages, I propose universal representations. In order to test the hypothesis, I created a corpus of Urban Hijazi Arabic verbs belonging to four verb types, monosyllabic [aa], monosyllabic [a], disyllabic [a], and disyllabic [i]. Studying the lexical trends of the imperfective of Hijazi Arabic, four natural classes, emphatics, liquids, nasals and gutturals, seem to influence the choice of imperfective vowel. In order to check whether the patterns can be learned and generalized, I employ a rule-based learner, the Minimal Generalization Learner (Albright & Hayes 2002, 2003, 2006). Findings from lexical decision and production experiments with auditory stimuli support the stem hypothesis; the presence of emphatics, for instance, affects the imperfective vowel choice of all four verb types. Although the MGL successfully learns the trends of the imperfective vowel, its predictions follow the lexicon too closely relative the experimental results. I present a constraint-based analysis using the MaxEnt learner. With its default settings, the learner gives expected probabilities similar to the predictions of the MGL. I show that unlike the MGL, the MaxEnt grammar has the flexibility necessary to account for variability in the imperfective choice through adding biases against source oriented constraints. This improvement to the model is one of the main contributions of this work.
اصطلاحهای موضوعی کنترل نشده
اصطلاح موضوعی
Linguistics
نام شخص به منزله سر شناسه - (مسئولیت معنوی درجه اول )