The meaning of morphomes: distributional semantics of Spanish stem alternations

Abstract

Abstract Romance stem alternations have been argued to represent exclusively morphological objects (or “morphomes”) independent from semantic and syntactic categories. This conclusion has been based on feature-value analyses of the inflected forms, and definitions of natural classes that are theoretically driven and about which no consensus exists. Individual examples of morphomes are thus frequently challenged, while their autonomously morphological nature has never been tested quantitatively or experimentally. This is the purpose of the present study. We use context-based embeddings to explore the semantic profile of Spanish verb stem alternations. At the paradigmatic level, our findings suggest that Spanish morphomes’ cells are characterized by significantly above-chance distributional-semantic similarity. At the lexical level, similarly, verbs that show more similar patterns of alternation have also been found to be closer in meaning. Both of these findings suggest that these structures may have an extramorphological function. Using gradient distributional-semantic similarity offers a way to objectively assess the degree of (un)naturalness of a set of forms and meanings, something which has been lacking from most discussions on the structure of features and the architecture of paradigms.