Abstract Minimal generalization of Dutch diminutives

A comparative study was conducted about the acquisition of diminutive forms in the Dutch language. A former study, using the C4.5 algorithm, is discussed and contrasted with the implemented Minimal Generalization model by Albright and Hayes. In addition, the model is also compared to a conducted behavioral study using wug words that follow the Dutch phonetic rules, but do not exist in the language. The Minimal Generalization model is very good in creating the correct diminutive forms from lemmas. In addition, the model corresponds quite well with the behavioral data. A notable exception are the wug words intended to correspond to the rule with suffix kje, for them the Minimal Generalization model does not give good estimates. An explanation for this result is given. The authors believe that the model’s method of learning the necessary rules for this task displays characteristic similarities to the way humans learn these rules