Vowel but not consonant identity and the very informal English lexicon

Abstract

This paper studies the phonological properties of shitgibbons, a class of insulting English compounds made up of a monosyllabic obscenity followed by a trochaic innocuous noun. Our experimental data shows that in addition to these categorical prosodic requirements, there are gradient segmental requirements: native speakers judge shitgibbons as more wellformed when their two stressed vowels are identical (e.g. shit-whistle is better than fuck-whistle), but matching word-initial consonants do not improve wellformedness. A corpus study of English compounds shows that both vowel identity and initial consonant identity are overrepresented in the lexicon. Our explanation for the mismatch between the lexicon and the experiment relies on a typological asymmetry: vowels interact across intervening consonants in many languages, but consonants do not selectively interact across other intervening consonants in this way, e.g. the two matching [f]'s in fuck-frisbee cannot be compelled to match while ignoring the intervening coda [k]. The analysis is implemented as a MaxEnt grammar, with a locality bias that prevents assigning weight to the constraint that demands initial consonant matching.