Analogy in morphology: modeling the choice of linking morphemes in Dutch

This study argues that a productive but not fully regular morphological phe nomenon, the choice of linking morphemes in Dutch nominal compounds, is bas ed on analogy. In Dutch, a linking -s- or -en- can appear between the const ituents of a nominal compound. We present production experiments that revea l strong evidence that the choice of linking morphemes in novel compounds i s analogically determined by the distribution of linking morphemes in what we call the "constituent families." A "constituent family'' is the set. of existing compounds that share the first (or second) constituent with the no vel compound. A further experiment shows that in the case of derived pseudo -words as first constituents, it is the family of the suffix that influence s the choice of the following linking morpheme. In addition to these experi ments, we present computational simulation studies in which the choices mad e by participants in our experiments are predicted with a high degree of ac curacy using a machine-learning algorithm for analogy. These studies suppor t the status of the constituent family as the primary basis for analogical prediction. Finally, we outline a psycholinguistic model for analogy in the mental lexicon that does not give up symbolic representations and at the s ame time, captures nondeterministic variation.