Show summary
Iconicity is the perceived resemblance between aspects of the form and the meaning of linguistic signs. Iconicity seems to be widespread within and across languages and can be found in basic vocabulary for color, shape, texture, speed, and size, among others (Blasi et al., 2016; Johansson et al., 2020). The observed prevalence of iconicity stands in contrast to how languages commonly change: Regularly occurring phonological and semantic changes are blind to formfunction correspondences and are thought to erode iconicity (e.g. Flaksman, 2017). How then is iconicity maintained in spoken languages if regular change inevitably leads to a conventionalized lexicon? Flaksman (2017) suggests that lexical innovation, e.g., newly formed onomatopoeias, keep iconicity within the system. This view is in line with experimental evidence from iterated learning experiments showing that iconic forms can indeed emerge through imperfect lexical learning over iterations of signal transmission (Vinson et al., 2021). Iterated learning experiments on spoken/written language usually operate on the level of categorical change of segments as opposed to gradual phonetic change. However, at least within a conventionalized signaling space such as human languages, change is not mainly driven by lexical innovation, but by re-occurring variability of conventionalized forms that leads to gradual change of form and meaning over time (e.g. Ohala, 1993; Blevins, 2004). We introduce a novel take and the iterated learning paradigm to explore the hypothesis that iconicity gradually emerges in a conventionalized signaling space through gradual sound change. In our paradigm, the consonants of words are fixed and shown in written form displaying the consonantal grid of the words, but we mask the vowel quality (e.g. g#v#d). All the vowels of the initial stimuli are pronounced by a trained speaker who produces illicit schwas in vowel position. Schwa cannot appear in stressed syllables in the participants’ L1 (Norwegian), forcing participants to produce variants of the presented seed forms. Thereby we limit system changes to the gradual modulation of vowels and suprasegmental properties of speech such as duration and fundamental frequency. Participant dyads learn a miniature artificial language, consisting of disyllabic pseudowords referring to alien creatures which crucially differ systematically in size. In the training phase, participants listen to pre-recorded sentences containing the names and are shown the referents next to a boot used for size comparison (see Fig. 1). After the learning phase, participants play a communication game taking turns being speaker and guesser over three rounds. The speaker is shown a trial (Fig. 1, left) and tasked with producing the sentence. The guesser has to choose the correct referent out of four options which always contain two big and two small referents (Fig. 1, right). Both speaker and guesser receive feedback on whether the communication was successful or not. Participants’ productions are recorded and fed into the following generation of participants. Based on previous research on size iconicity (summarized in Winter, 2025), we expect pitch, vowel duration and quality and/or word length to potentially be used by participants to convey the size of the referent without compromising the fixed consonantal template for each form. The assumed emerging iconicity is moreover expected to improve the learnability of the artificial language over generations (Tamariz et al., 2018). Pilot results show that participants adapt well to the task, that form-meaning pairs are learned over repeated interactions, and that speakers rapidly move away from the initial (illicit) vowel quality, allowing us to test our predictions. In our poster presentation, we aim to present results from at least 10 iteration chains with 5 generations of speakers each.
