Abstract

This article presents empirical evidence of the high referential specificity of sound-symbolic words, based on a FrameNet-aided analysis of collocational data of Japanese mimetics. The definition of mimetics, particularly their semantic definition, has been crosslinguistically the most challenging problem in the literature, and different researchers have used different adjectives (most notably, “vivid,” since Doke 1935) to describe their semantic peculiarity. The present study approaches this longstanding issue from a frame-semantic point of view combined with a quantitative method. It was found that mimetic manner adverbials generally form a frame-semantically restricted range of verbal/nominal collocations than non-mimetic ones. Each mimetic can thus be considered to evoke a highly specific frame, which elaborates the general frame evoked by its typical host predicate and contains a highly limited set of frame elements, which correlate and constrain one another. This conclusion serves as a unified account of previously reported phenomena concerning mimetics, including the lack of hyponymy, the one-mimetic-per-clause restriction, and unparaphrasability. This study can be also viewed as a methodological proposal for the measurement of frame specificity, which supplements bottom-up linguistic tests.