SGER: Exemplar-based Modeling of Spanish Language Phenomena

点击翻译按钮获取中文摘要

The goal of this research is two-fold. The first is to test the adequacy of exemplar-based models of language processing. Two exemplar models will be utilized, Analogical Modeling of Language (AML) and Tilburg Memory-based Learner (TiMBL). To this end, these models will be tested on several linguistic phenomena in Spanish: stress placement, gender assignment, diminutive formation, and selection of change of state verb. A database containing 800-2000 instances of each phenomenon will be extracted from available frequency dictionaries and Spanish language corpora. Each instance will be encoded as a set of linguistic variables that are pertinent to predicting the phenomenon at hand. The task will be to predict the behavior of each instance by using the remaining database items as possible analogs. For example, the gender of a given noun will be predicted on the basis of its similarities to other nouns in the database.Since AML and TiMBL utilize radically different algorithms, the second aim is to determine which model makes better predictions. Each model will be compared on the above-mentioned language tasks in terms of its ability to correctly predict the linguistic behavior in question. For some of the phenomena, studies involving language acquisition, production of nonce words, and slips-of-the-tongue, are available. The ability of the exemplar-based models to mirror the data from these studies may provide additional tests on which they may be assessed.