In recent years, we witness a rise of interest in accounting for probabilistic generalizations in linguistic patterns using formal grammatical theories. In this context, Maximum Entropy Harmonic Grammar (MaxEnt HG) has been shown to be a useful analytical tool in modeling probabilistic patterns in various linguistic domains. Some recent studies (Kawahara 2020c, 2021) have proposed to extend the scope of MaxEnt HG by applying it to the analysis of sound symbolism, systematic associations between sounds and meanings. These studies examine what Hayes (2020) refers to as the quantitative signature of MaxEnt HG, the set of typical probabilistic patterns that MaxEnt HG is predicted to generate. The particular quantitative signature that these studies found is wug-shaped curves, which consist of multiple sigmoid curves. Inspired by these studies, the current experiment examined whether we can identify yet another instance of wug-shaped curves in sound symbolism, with the empirical target being the judgment of Pokémon characters’ evolution status by native speakers of Japanese. The current experiment shows that as name length increases, the post-evolution responses increase in a sigmoidal manner, and that this sigmoidal curve is shifted depending on the vowel quality of the stimuli, resulting in a wug-shaped curve. To model the results, an MaxEnt HG analysis, equipped with OT-style constraints, is developed as an example of a generative phonological analysis of sound symbolism.