Repeated imitation makes human vocalizations more word-like

Abstract

People have long pondered the evolution of language and the origin of words. Here, we investigate how conventional spoken words might emerge from imitations of environmental sounds. Does the repeated imitation of an environmental sound gradually give rise to more word-like forms? In what ways do these forms resemble the original sounds that motivated them (i.e. exhibit iconicity)? Participants played a version of the children's game 'Telephone'. The first generation of participants imitated recognizable environmental sounds (e.g. glass breaking, water splashing). Subsequent generations imitated the previous generation of imitations for a maximum of eight generations. The results showed that the imitations became more stable and word-like, and later imitations were easier to learn as category labels. At the same time, even after eight generations, both spoken imitations and their written transcriptions could be matched above chance to the category of environmental sound that motivated them. These results show how repeated imitation can create progressively more word-like forms while continuing to retain a resemblance to the original sound that motivated them, and speak to the possible role of human vocal imitation in explaining the origins of at least some spoken words.

Keywords: iconicity; language evolution; transmission chain; vocal imitation.

Figure 1.

Vocal imitations collected in the transmission chain experiment. Seed sounds (16) were sampled from four categories of environmental sounds: glass, tear, water, zipper. Participants imitated each seed sound, and then the next generation of participants imitated the imitations, and so on, for up to eight generations. Chains are unbalanced due to random assignment and the above-mentioned exclusion criteria.

Figure 2.

Change in perception of acoustic similarity over generations of iterated imitation. Points depict mean acoustic similarity ratings for pairs of imitations in each category. The predictions of the linear mixed-effects model are shown with ± s.e. (Online version in colour.)

Figure 3.

Repeated imitations retained category resemblance. (a) Three types of matching questions. True seed and category match questions contained choices from different sound categories. Specific match questions pitted the actual seed against the other seeds within the same category. (b) Accuracy in matching vocal imitations to original seed sounds. Curves show predictions of the generalized linear mixed-effects models with ±1 s.e. of the model predictions. (c) Accuracy in matching transcriptions of the imitations to original seed sounds (e.g. ‘boococucuwich’ to a water-splashing sound). Circles show mean matching accuracy for the vocal imitations that were transcribed for comparison.

Figure 4.

Repeated imitations made for better category labels. (a) Mean RTs for correct responses in the category-learning experiment with ±1 s.e. (b) Cost of generalizing to new category members with ±1 s.e. (Online version in colour.)