It has often been noted that the phoneme inventories of human language are subject to considerations of distinctness (e.g., Martinet 1955, Liljencrants and Lindblom 1972, Flemming 2002). In this talk, I show that it is not only inventories of phonemes but also a language’s inventory of words (i.e., the lexicon) that is subject to the very same perceptual pressures. Words, just like phonemes, preferentially rely on highly perceptible contrasts for distinctness. I present evidence for this from the cross-linguistic distribution of minimal pairs.
Using a corpus of phonetically transcribed dictionaries of 61 different languages from 25 major language families, I show that the number of words disambiguated solely by place contrasts in intervocalic (a[p]end:a[t]end; V_V), prevocalic (s[p]y:s[t]y; C_V), postvocalic (swee[p]s:swee[t]s; V_C), and interconsonantal (cor[p]se:cour[t]s; C_C) context decreases as a function of the perceptibility of cues to place in those contexts (Repp 1978, Fujimura et al. 1978, Ohala 1990, Wright 1996, Kochetov 2004). Using log-linear mixed-effects models, I show that these asymmetries go beyond what is expected from the frequency and relative distribution of individual sounds in different languages.
In a second study I show that analogous results obtain for laryngeal contrasts in different languages. There are more minimal pairs like a[p]ace:a[b]ase (V_V) than pairs like dis[p]erse:dis[b]urse (C_V) than pairs like sta[p]le:sta[b]le (V_C) than pairs like am[p]le:am[b]le (C_C), mirroring the availability of cues to voicing (Raphael 1981, Slis 1986, Lisker 1986). Again, I show that results hold while controlling for sound-specific distributions.
A third study finds that the number of minimal pairs based on consonantal differences in English increases significantly with the perceptual distinctness of the consonants distinguishing them. There are more pairs like pop:shop relying on the highly distinct /p/:/ʃ/ contrast, than pairs like thought:fought relying on highly similar /θ/:/f/ in the English lexicon.
I argue that these results have important implications for theories of the kind of knowledge speakers have access to in learning and using their language (cf Hayes and Steriade 2004 and Blevins 2004 for differing views). While many cross-linguistic patterns in phoneme inventories are amenable to explanations in terms of what Ohala (1990) refers to as innocent misapprehension, dispersion of words is not. If speakers simply misperceived certain contrasts more, then these contrasts should drop out regardless of whether they distinguish among words or not. For the patterns identified above to arise, language usage or learning must be sensitive to the perceptual similarity of words.