Sensitivity to ideophones by child non-native speakers increases with age

Sound-symbolic effects have been claimed to facilitate word learning, because of the motivated relationship between form and meaning. According to some views, such effects result from human-specific perceptive and cognitive biases and can be thus universal. This paper tests the sensibility to ideophones (a type of sound-symbolic words) by children acquiring Spanish, a language lacking this word class. Results show that younger children are mostly insensitive to sound-symbolic effects as found in ideophones, but that this sensibility emerges and improves with age. Also, children were found to perform better in tasks involving verbal responses compared to visual tasks. By contrast, no effect of sex or the type of ideophone on success rates was observed. Overall, these results support the view that although we might be endowed with a special sensitivity to sound-symbolic effects, this sensitivity does not provide a very large advantage when learning a language and/or a minimal amount of exposure to the phenomena to be acquired is necessary to put it effectively into use.

Most words are symbols because the relationship between their form and their meaning is arbitrary. However, languages also contain iconic words, in which their form resembles their meaning to a variable degree. This is typically observed in onomatopoeias, which are purely imitative words, but there are other instances of sound-symbolic words, like those provoking synesthetic-like effects, that is, visual, tactile, or even emotional responses to sounds (see Köhler, 1929; Hinton et al., 1994; Monaghan et al., 2014; Dingemanse, 2015; Akita and Dingemanse, 2019 for general discussions). Iconic words have been argued to be easier to learn, both by children and second language learners, because of a purported sensitivity to such sound-symbolic effects that roots in an inborn ability to map and integrate multimodal inputs, which would be used to bootstrap language acquisition (see Imai et al., 2008; Imai and Kita, 2014; Nielsen and Dingemanse, 2021 for detailed discussions).

The world languages exploit these sound-symbolic effects to different degrees. Whereas all languages have onomatopoeia-like and synesthetic-like words, only a small fraction has a specific class of sound-symbolic words. This class is usually referred to as ideophones. Ideophones have been characterized by Dingemanse (e.g., 2019; 2023) as marked words that vividly depict, in conventionalized ways, sensory images or events. Besides their sound-symbolic nature, they usually exhibit several other

distinctive features. Phonologically, they tend to be enriched in sounds with broader sound-symbolic effects (like the widespread association of high vowels to small sizes), show a distinctive intonational pattern within utterances (i.e., they are typically uttered with a higher or lower pitch), and/or result from distinctive morphophonological mechanisms (typically, reduplication, or lengthening) (Dingemanse, 2015; Dingemanse and Akita, 2017). Morphologically, they are usually undecomposable, although they can be occasionally combined with other morphemes, mostly with derivative affixes (Dingemanse, 2015). Furthermore, they are poorly integrated in the sentence, frequently appearing at the edges (Dingemanse, 2017; Dingemanse and Akita, 2017); they can also exhibit some syntax-specific properties, like positive polarity (Tolskaya, 2011). Semantically, ideophones typically convey affective or imagistic meanings, evoke types of movements or sounds, or qualify verbs of perception, informing about textures, shapes, or tastes, with these functions being more or less important depending on the language family (see Kita, 1997; Dingemanse, 2012, 2018; for discussion and examples). Pragmatically, they are usually uttered together with a rich gestural and body language (Moshi, 1993) and contribute to reinforcing the perlocutive force of conversational exchanges (Baba, 2003; Clark, 2016).

As observed with iconic words more generally, ideophones have been claimed to be easier to learn than arbitrary words (Burling, 2005; Haiman, 2018; Nielsen and Dingemanse, 2021), not only because of the non-arbitrary form-meaning relationship, but also because of some others of their properties such as reduplication, marked prosody, and emotional content (Dingemanse, 2012; Nielsen and Dingemanse, 2021). This effect has been observed also with L2 learners. For instance, Lockwood et al. (2016) found that adult speakers of Dutch learn Japanese ideophones better than non-ideophonic adjectives, and specifically, that they can guess their meaning above chance in a two-alternative forced choice test.

In this paper, we tried to determine whether Spanish-speaking children aged 8–12 years show such a differential sensitivity to non-native ideophones. Spanish lacks a proper class of ideophones, although, as all known languages, it has several sound-symbolic words, mostly onomatopoeias. Certainly, the differences between ideophones and sound-symbolic words can be argued to be mostly quantitative in nature, not qualitative. Such a discussion is beyond the scope of this paper, but it could be enough for our objectives here to consider that ideophones are more numerous than onomatopoeias and that they exhibit all the properties mentioned above in more marked ways. Onomatopoeias, by contrast, are less abundant and they only involve sound-to-sound correspondences. All in all, having onomatopoeias but lacking ideophones means that one is less exposed to sound-symbolic effects. About the procedure, we also adopted a two-alternative forced choice test approach aimed at evaluating the child's ability to guess the meaning (presented either verbally or visually) of ideophones from five natural languages and referring to three different visual dimensions: color, shape, and motion.

Participants

We tested six groups of Spanish students (n = 64; 29 boys and 35 girls) from three different grades (two groups for each grade): 2nd grade (born in 2016), 4th grade (born in 2014) and 6th grade (born in 2012). Each group consisted of 10 or 11 children. A more detailed sociodemographic profile of the participants can be found in Supplementary file 1. Children attended the grade school “Amor de Dios,” located in Cádiz, southern Spain.

Tasks

We designed two two-alternative forced choice tests. Both comprised the same set of 30 ideophones from five natural languages from different language families (six ideophones per language): Siwu (Niger-Congo, Ghana), Ewe (Niger-Congo, Ghana), Korean (isolated, Korea), Japanese (isolated, Japan), and Semai (Austroasiatic, Malaysia). For each language, two ideophones conveyed information about color properties (e.g., ‘black'), two ideophones conveyed information about shape (e.g., ‘round'), and two ideophones conveyed information about motion (e.g., ‘walk fast'). Ideophones were retrieved from Dingemanse et al. (2016). The full list of ideophones used in the study, together with their meanings, can be found in Supplementary file 2. We designed two testing sheets to be fulfilled by children: in one of them, two opposite meanings for each of the ideophones comprising the task were presented visually (as two images contrasting in the feature conveyed by the ideophone). In the other sheet, the two possible meanings were presented verbally, as two Spanish antonyms contrasting in the meaning conveyed by the ideophone. In both cases, ideophones were arranged randomly, although the same order was followed in both tasks. The two testing sheets can be found in Supplementary files 3, 4, respectively.

Procedure

For both tasks, children listened twice to a recording of the ideophone uttered by a native speaker of the language. Recordings were retrieved from Dingemanse et al. (2016). Afterwards, children were asked to circle one of the two options in the sheet (a picture or a word, depending on the task). Before starting the test, children were informed in detail about the procedure. They also practiced with an example: the forms used in the bouba-kiki experiment by Köhler (1929). For each grade, we tested one of the two groups (A) using the version of the test with pictures, and the other group (B) using the version with written words. The trial task revealed that the youngest children were unable to understand the task involving written words, so we decided to use with them (i.e., group B from grade 2nd) the task with pictures too.

Statistical analyses

For analyzing the responses by the children, we conducted mixed-effects logistics regressions to know whether they were guessing the meaning of the ideophones above chance, but also to determine whether performance changed with age, depending on the type of task (words vs. pictures), or according to the meaning conveyed by the ideophone (motion vs. shape vs. color), while controlling for idiosyncratic variation in each trial (since some trials might be naturally harder than others). For each variable of interest, we conducted a separate mixed-effects logistics regression under each task condition, with the variable of interest being coded as the fixed effect, using the lme4 package (Bates et al., 2015) in R (R Core Team, 2024). For the random effects, we started out by prescribing the maximally complex random effect structure justified by the design of the experiment (Barr et al., 2013), and when the algorithm did not converge or had a singular boundary, we further simplified the random effect structure by removing random effects that correlated with others, as well as those that explained little variance. We ended up removing the random slopes and intercepts related to participants, since they all explained little variance. For testing whether participants were answering better than chance, the final formula we used, written in glmer syntax, was: correct ~ 0 + Grade + (1 | trial). For testing the potential effect of age on participants' performance, we normalized their age to bring the data closer to zero in range, and as a result, the oldest participant received 1 in his relative age, whereas the youngest participant got 0. The final formula we used, written in glmer syntax, was: correct ~ 1 + age_normalized + (1 + age_normalized | trial). For testing the potential effect of participants' sex on performance, the final formula we employed, written in glmer syntax, was: correct ~ 1 + sex + (1 + sex | trial). Finally, for testing the potential effect of the type of ideophone, the final formula we used, written in glmer syntax, was: correct ~ 1 + type + (1 + type | trial).

All the groups succeeded in completing the tests, with the exception, as noted, of one of the youngest groups (group B grade 2nd), who did not understand the test involving written words. As also noted, we used instead the test with pictures with them. Our results are presented in Figure 1. The raw results can be found in Supplementary file 5. The mixed-effects logistics regressions revealed that only the oldest children (from Grade 6th) were answering better than chance in both tasks (p = 0.0002 for the test with words; p = 0.017 for the test with pictures). With regards to the potential effect of age on the participants' performance (Figure 1A), we found a significant effect only for the task using words (p = 0.0053), but not for the task using pictures (p = 0.223). Concerning the potential effect of the participants' sex on performance (Figure 1B), we found no effect (p > 0.2 in both cases). Finally, regarding the potential effect of the type of ideophones on performance (Figure 1C), we found no significant differences between types either (p > 0.25 in all comparisons).

There is a vivid discussion around the universal nature of sound-symbolic associations and their potential bootstrapping effect on language acquisition and learning. Both possibilities have been tested using different types of sound-symbolic words (or pseudowords), and with speakers of different languages (for an overview, see e.g., Svantesson, 2017). Some of this research used, specifically, ideophones from natural languages. Still, at present, this approach has noteworthy limitations: it has mostly focused on babies or adult speakers, and it has examined a reduced set of source and target languages. For instance, Imai et al. (2008) observed that 2-year-old Japanese children are sensitive to sound-symbolic effects as found in Japanese ideophones, but also that adult speakers of English with no knowledge of Japanese show such a differential sensitivity. Oda (2000) detected a similar effect in his study with native speakers of English. Likewise, Lockwood et al. (2016) found that adult speakers of Dutch can guess and learn the meaning of Japanese ideophones better than the meaning of non-ideophonic words. This sensitivity to the sound-symbolic patterns of Japanese have been also observed in English-speaking children with no knowledge of Japanese (Kantartzis et al., 2011; Yoshida, 2012).

In our research, we have found that Spanish-speaking children with no knowledge of languages with ideophones are also able to guess the meaning of ideophones above chance, although not too high above chance. This is in line with previous claims (e.g., Nielsen and Dingemanse, 2021) that iconicity might provide some advantages but not very large advantages in language acquisition and learning. We also found that this sensitivity to sound-symbolic effects increases with age. This is in line with previous findings too, for example, with research by Nielsen and Dingemanse (2021), but it is at variance with results by other researchers, for example, Ozturk et al. (2013). To reconcile these apparently opposite findings, Lockwood and Dingemanse (2015) have suggested that child sensitivity to sound-symbolism might be dependent on the amount of input, whereas adults can make form-meaning associations from limited input even if they are endowed with a reduced sensitivity to sound-symbolism. This hypothesis could certainly explain the effect of age we have observed. But it could account as well for another of our findings, namely, that children perform better in tasks involving written words compared to pictures. As the child ages, she is exposed to a richer and more diverse vocabulary, and accordingly to a richer input, that can be expected to improve her ability to make sound-to-meaning associations, as observed in adults. Supporting this view, Spanish-speaking children have been observed to succeed in naming tasks involving the association of textures to non-words, even if they fail in pointing selection tasks (Falcón et al., 2019). One more general question, to be addressed by future research, is whether this effect of age on ideophone understanding (and ultimately, on our sensitivity to sound-symbolism) results just from the biological changes associated to growth or (also) to our socialization, as hypothesized for the differential sensitivity to sound-symbolic effects exhibited by adult males vs. adult females (see Gordon and Heath, 1998 for discussion).

Lastly, we did not find evidence that the type of meaning conveyed by an ideophone impacts on its intelligibility by non-native speakers. Previous research (e.g., Iwasaki et al., 2007) observed higher success rates for more transparent ideophones (i.e., those conveying sensory information, and particularly, sound-to-sound correspondences). One explanation for our finding could be that the meanings conveyed by all the ideophones in our sample pertain to the same sensory domain, namely, vision (i.e., color, shape, or motion), whereas we did not include ideophones conveying auditive information, or less transparent, more abstract meanings, like affective or imagistic dimensions.

To finish, we wish to acknowledge the limitations of our study. First, as noted above, more types of ideophones should be included in the testing sample, so that findings (and conclusions) are not circumscribed to the visual semantic domain. Moreover, a larger sample of children would be very welcome, one that encompasses more children, of more diverse ages, from more diverse socioeconomic and sociocultural backgrounds, and seemingly, from more diverse geographical origins. This should enable to avoid, specifically, the limitations resulting from the small sample size and the classroom-based grouping used in this study. Lastly, some of the explanations given to our findings also demand a more ambitious experimental design to be properly supported, like the possibility that sound-symbolic words are not easier to learn than arbitrary words, and particularly, the possibility that our sensibility to sound-symbolic effects changes as we grow and our brain develops. By all these reasons, this paper should be viewed as a pilot research that points to some of the questions that, if properly addressed, should clarify the nature of sound-symbolic effects and their role in language structure and language acquisition.