The nature and origins of derivational polysemy

Abstract

Booij argued recently that the problem of Agentive-Instrumental polysemy among Dutch nominalizations ending in - er can be resolved without resorting to the separation of derivation from affixation (the Separation Hypothesis). He argues for a ‘semantic extension scheme’ whereby an instrumental nominalization like open-er ‘opener’ is explained as a relatively less ‘agent-like’ semantic extension of the category ‘Agentive’: spel-er ‘player’. This question is only part of the much larger issue of derivational polysemy: why do certain categories, e.g. Agent and Instrumental, tend to receive polysemous marking across languages when others, e.g. Agentive and Action (nominalization), do not. This paper offers an alternate explanation of Booij's problem in a framework which addresses the larger issue. It proposes that the same universal categories, the category functions of traditional grammars, determine both inflectional and derivational relations (Unitary Grammatical Function Hypothesis). Combined with the Separation Hypothesis, unitary grammatical functions can predict (1) which languages are more likely to exhibit polysemy and which, synonymy, and (2) which categories are more or less likely to be shared by a polysemous derivational affix. Even enriched with semantic extension schemes, purely sign-based frameworks make the wrong predictions.