Trilled /r/ is associated with roughness, linking sound and touch across spoken languages

Abstract

Cross-modal integration between sound and texture is important to perception and action. Here we show this has repercussions for the structure of spoken languages. We present a new statistical universal linking speech with the evolutionarily ancient sense of touch. Words that express roughness-the primary perceptual dimension of texture-are highly likely to feature a trilled /r/, the most commonly occurring rhotic consonant. In four studies, we show the pattern to be extremely robust, being the first widespread pattern of iconicity documented not just across a large, diverse sample of the world's spoken languages, but also across numerous sensory words within languages. Our deep analysis of Indo-European languages and Proto-Indo-European roots indicates remarkable historical stability of the pattern, which appears to date back at least 6000 years.

Figure 1

How phonemes relate to roughness ratings. Left: The ten most predictive phonemes in a random forest analysis for (a) English and (b) Hungarian; the vertical black line corresponds to the absolute value of the least predictive phoneme, which is a heuristic cut-off rule for predictors that do not contribute. Right: Boxplots (whiskers = smallest, largest value within 1.5 × IQR) for words with and without /r/ in (a) English and (b) Hungarian.

Figure 2

Words for the concepts ‘rough’ and ‘smooth’ in 179 spoken languages that have a trilled /r/ in their phoneme inventory, showing a much higher percentage of /r/ in rough words than smooth words.

Figure 3

Left: Languages with only a trilled /r/ in their phoneme inventory (n = 179); Right: languages with other types of /r/ (n = 153). Results are aggregated by language family, with each colored dot representing a family (size of the circle = number of languages). Grey dots and whiskers show model predictions along with 95% Bayesian credible intervals. In languages with trilled /r/, words for ‘rough’ have a higher proportion of /r/ (posterior mean = 37%) than words for ‘smooth’ (10%); virtually no difference is observed for languages with other types of /r/.

Figure 4

(A) Across the Indo-European language family, the proportion of rough words with /r/ is much higher than the proportion of smooth words with /r/; (B) Each dot represents a language (size of the circle = number of words); whiskers show 95% Bayesian credible intervals corresponding to the mixed-effects Bayesian logistic regression analysis indicating that rough words have a much higher proportion of /r/ (posterior mean = 63%) than smooth words (posterior mean = 35%).