Search

This thesis addresses Arabic word generation from semantic descriptions. The purpose of this study is two-fold. Firstly, to analyse semantic specifications of Arabic derivation and, secondly, to build a computational model that performs Arabic lexical choices using a knowledge-base that is motivated by the semantics of derivation. Lexical choice in a highly derived language, such as Arabic, is non-trivial in MT and NLG systems where the semantic input cannot be biased directly towards the words that the language provides. On the linguistic side, the major contribution of this study is a new framework that organises semantic interactions in Arabic derivation in a two-layer semantics: an inner layer accounts for core meanings realised by Arabic consonantal roots and an outer layer accounts for semantic features associated with derivational moulds. The meaning of derived words can be established using controlled interactions between the two layers. The interactions, accordingly, state the link between semantic concepts on the one hand and their surface realisations on the other. Furthermore, derivational conditions imposed on the interactions account for many derivational gaps in Arabic. On the computational side, semantic generalisation motivated by the two layer semantics are organised into a taxonomic knowledge-base. Following the tradition in the KL-ONE family, the knowledge-base adopts a separation between definitional and assertional components. The former is used to represent semantic specifications of Arabic derived concepts while the latter is used to describe their linguistic properties such as roots and moulds. The knowledge-base, as a whole, is built around structured hierarchies that represent an integration of knowledge resources such as default and multiple inheritance, and subsumption. This integration allows not only for an expressive representation of Arabic derivation but also for the mapping of semantic inputs into Arabic derived lexical items by means of automatic classification.