Iconicity mediates semantic networks of sound symbolism^a)

Sound symbolism is an iconic, indexical, or systematic association between sound and meaning. In many cases of sound symbolism, a single speech sound is associated with more than one meaning. Numerous experiments have revealed that the high front vowel /i/ may evoke a small, bright, sharp, fast, or precise image, and voiced obstruents are more likely than voiceless obstruents to be associated with darkness and roundedness (Lockwood and Dingemanse, 2015). Winter (2019, p. 2) call this semantic property of sound symbolism “pluripotentiality,” arguing that “the same phonological pattern can have multiple distinct meanings, depending on the context” (see also Kawahara and Kumagai, 2021; Kumagai, 2020; Winter , 2021). In the current study, we dig into this issue using a nonword-based experiment in Japanese, Korean, Mandarin, and English. These languages have been in close cultural contact over centuries but are arguably not phylogenetically related to each other. We report on cases in which one speech sound is linked with more than one meaning, presumably via independent motivations.

The organization of this paper is as follows. Section II introduces previous studies on the pluripotentiality of sound symbolism. Section III describes the method of our cross-linguistic experiment. Section IV presents the results, which will be interpreted in terms of the pluripotentiality of sound symbolism. We discuss the sources of these multiple associations with reference to universal and language-specific factors in Sec. V, where we also make a methodological proposal for describing the semantic networks of sound symbolism for cross-linguistic comparison.

Pluripotentiality in sound symbolism has long been recognized (Joseph, 1994; among others) but probably often as a common feature that does not deserve special attention. Most experiments use specific semantic-differential scales, such as size, brightness, hardness, and cuteness, or visual images, such as rounded “bouba” shapes and spiky “kiki” shapes (Ramachandran and Hubbard, 2001). They typically make specific generalizations using these scales and images, such as “high front vowels are associated with smallness and brightness.” How these meanings are related to each other has been an open question.

Winter (2019) focus on two Korean interjections—khu [k^hɯ] and khya [k^hja]—both uttered after downing a shot of liquor. They asked American English speakers to choose which of the two auditorily presented interjections is more suitable for different types of liquor. They found that khya is more strongly associated with softness, smoothness, femaleness, sweetness, and pleasantness. Importantly, Winter (2019) note that “multiple cognitive accounts” would be possible for the observed pluripotentiality.

The abstract mapping account [Fig. 1(A)], which posits that different meanings are instances of a general meaning that is associated with a sound, reflects the abstract nature of sound-symbolic phenomena. As for the metaphor account [Fig. 1(B)], in which one meaning extends metaphorically to another, Kawahara (2015) experimentally show that the voiceless obstruents /t, k, s/ are associated with angular shapes and inaccessible types of personality. Angularity and inaccessible personality can be related to each other, even without the mediation of obstruents. In fact, a spiky/edgy personality stands for “an inaccessible personality,” and this metaphor is shared by the Japanese expressions tsuntsun shita hito /tsɯntsɯnɕitäçito/ “an edgy person” and toge no aru ii-kata /toɡenoaɾɯiːkätä/ “a stinging way of speaking.”¹

The two cognitive accounts are unlikely to be mutually exclusive. It appears to be more realistic that some cases of sound symbolism instantiate abstract sound–meaning mappings, whereas other cases are semantic extensions from primary sound–meaning associations. It may also be possible that one case of sound symbolism is a specific instance of an abstract sound–meaning association and, at the same time, a metaphorical (or metonymical) extension from another specific sound–meaning association [Fig. 1(C)]. For example, the obstruent–angularity–inaccessibility association might be motivated by a general link between voiceless obstruents and abruptness as well as by the “personality is an object” metaphor. Furthermore, it is logically possible that the different sound-symbolic meanings of a sound can exist on their own without the mediation of abstract mappings or metaphors/metonymies [Fig. 1(D)].²

This study contributes to this discussion using a cross-linguistic experiment. Presenting the same set of nonword stimuli to speakers of four languages, we investigate which sound–meaning associations are more common among these languages and how these and other associations may arise. Specifically, we identify cases of cross-linguistically shared and language-specific sound symbolism without evident horizontal, metaphorical/metonymical relationships [i.e., Fig. 1(A) or 1(D)]. Based on the results, we also discuss how to advance the study of pluripotential sound symbolism, drawing on the “semantic map,” a linguistic-typological framework for language comparison.

The current study adopts the design of the preliminary experiment by Akita and McLean (2021) in Japanese and English. We combine these data with our newly collected Korean and Mandarin data to reach broader generalizations.

The breakdown of the participants in the four languages is given in Table I. All participants are self-reportedly monolinguals who do not use their nonnative languages on a daily basis. The participants were rewarded for their participation with an Amazon Electronic Gift Card (Seattle, WA; 500 JPY, 5 USD), CUpon (5000 KRW), or bonus points in exams. Consent was obtained electronically.

As listed in Table II, 30 audio stimuli were created by systematically combining 5 vowels [i, e, u, o, ä] and six stop consonants [b, d, ɡ, p, t, k] in the initially stressed vowel-consonant-vowel (VCV) form. A female phonetician, whose native language is Japanese, pronounced all stimuli, and three Japanese speakers and three English speakers confirmed that the recordings sound as intended.

The experiment in all languages was conducted online using Google Forms (Mountain View, CA). We made two semi-randomized versions of the stimulus set. In version 1, the 30 words were randomized, each followed by three 7-point semantic-differential scales: size (0 “small” ↔ 6 “large”), darkness (0 “dark” ↔ 6 “bright”), and hardness (0 “soft” ↔ 6 “hard”). In version 2, the words were presented in the counterbalanced order of version 1, with the three semantic scales given in the reversed order: hardness, brightness, and size. In each language, half of the participants answered version 1, and the other half answered version 2. The first page of the Google Forms provided the following instructions in the language of the participant:

We focused on imaginary jewel names because they serve as a natural context for all three of the semantic scales.

The participants were also instructed to wear headphones or earphones, adjust the volume, and not change it during the experiment. They were allowed to listen to each word as many times as they preferred.

According to the previous experimental literature on sound symbolism, it is expected that the nonfront vowels [u, o, ä], lower vowels [e, o, ä], and voiced obstruents [b, d, ɡ] are associated with largeness (Lockwood and Dingemanse, 2015; Shinohara and Kawahara, 2010; Westbury , 2018), and the rounded vowels [u, o] with darkness and softness (Mok , 2019; see also Johansson , 2020a). Studies comparing basic word lists across languages also suggest that the voiceless non-labial obstruents [t, k] may be linked with hardness (Johansson , 2020b; Johansson, 2017). We replicate and refine these observations using the same auditory stimuli across three semantic dimensions and four groups of participants.

The experiment revealed several interesting sound-symbolic patterns in the four languages. We, first, present the descriptive results of the three ratings one by one and then analyze them with statistical models.

Figure 2 shows the distributions of size ratings in terms of the five relevant phonetic features: vowel frontness (front [i, e] vs nonfront [u, o, ä]), height (high [i, u] vs mid [e, o] vs low [ä]), lip rounding (rounded [u, o] vs unrounded [i, e, ä]), consonant voicing (voiced [b, d, ɡ] vs voiceless [p, t, k]), and place of articulation (bilabial [b, p] vs alveolar [d, t] vs velar [ɡ, k]). In accord with the previous findings, in all languages, nonfront, lower, and rounded vowels tended to be associated with larger images. The effect of obstruent voicing on size ratings was clearest in Japanese speakers. Place of articulation had little influence, if any, on size ratings.

The results of the darkness ratings are presented in Fig. 3. The overall results are, again, consistent with the previous findings. In all languages, except Korean, the nonfront vowels tended to be rated darker than the front vowels. The effect of lip rounding was shared across the four languages: the rounded vowels were darker, although this effect was weak in Korean. Voicing, once more, yielded a sharp contrast only in Japanese. The effect of place of articulation was limited, as with the size ratings.

Figure 4 shows the results of the hardness ratings. As compared with the size and darkness ratings, the results here look less clear and less consistent across languages. Nevertheless, in all languages, the rounded vowels tended to be rated softer than the unrounded vowels. The bilabial stops were also rated softer than the alveolars in all languages. Interestingly, the velar stops were perceived relatively hard by the Japanese and Mandarin speakers but relatively soft by the Korean and English speakers, to whom the alveolar stops sounded harder.

We used the ordinal package version 2022.11.16 (Christensen, 2022) to perform our cumulative link mixed model (CLMM) analysis in R version 4.3.1 (R Core Team, 2023). CLMMs predict ordinal response variables from fixed and random effects. We fit a separate model for each semantic scale in each language without a Bonferroni correction, given that the three rating tasks and four groups of participants constituted separate datasets to examine. Each model included vowel (i.e., [i, e, u, o, ä]) and two consonantal features (i.e., voicing and place of articulation) as fixed effects and participant and stimulus as random effects. We also included an interaction between voicing and place, anticipating different sound-symbolic values between [d, ɡ] and [t, k] (see D'Onofrio, 2014). [ä] and the voiced bilabials (i.e., [ˈäbä]) were set as a reference level. We did not include the three vocalic features (i.e., frontness, height, and rounding) in the models, as in our stimulus set the rounded vowels [u, o] are both nonfront and [ä] is the only low vowel. Using [ä] as a reference allows us to indirectly see the effects of all frontness ([ä] vs [i, e]), height ([ä] vs [i, e, u, o]), and rounding ([ä] vs [u, o]).

Table III presents the statistically reliable effects obtained in the size models in the four languages. The association between the front vowels and smallness was replicated in all languages. On the other hand, only Japanese speakers rated the voiceless stops to be smaller than the voiced stops. The smallness of the velar stops as compared with the alveolar stops was unique to Korean speakers, and the relative smallness of the rounded vowels was unique to Mandarin speakers.

Table IV presents the reliable effects obtained from the darkness models. The darkness of [u] was shared across all four languages, whereas that of the other rounded vowel [o] was not shared by the Korean speakers. [e] was also rated relatively dark in Japanese, Mandarin, and English. The Japanese speakers used voicing again, and this time, it was for dark images. The Mandarin speakers also rated [i] brighter than [ä].

Table V is a summary of the hardness models. The softness of the rounded vowels, especially [u], was shared by the Japanese, Mandarin, and English speakers. [e] was also rated relatively soft in Japanese and Mandarin. Place of articulation turned out to have different effects across languages. The Japanese and Mandarin speakers rated the velars harder than the bilabials, whereas the Korean and English speakers rated the alveolars relatively hard. We obtained a reliable interaction between voicing and place in English, indicating that [k] but not [ɡ] was rated relatively hard. The slopes of these effects are relatively low compared to those of the reliable effects in the size and darkness models, suggesting the weak linkage between sound and hardness.

The results of the current experiment revealed cross-linguistically shared and language-specific sound-symbolic perceptions. On one hand, speakers of all four languages associated the front vowels [i, e] with small jewels and the rounded vowel [u] with dark jewels. The softness of the rounded vowels was also relatively common across languages. The strongest associations were found in the domain of size, followed by darkness, while the associations with hardness were less strong. This last result may be partly attributed to the fact that we presented the stimuli as the names of jewels, which, as an anonymous reviewer pointed out, are by their nature on the hard side and may not be compatible with the concept of softness.

We also found some language-specific associations. Only the Japanese speakers used voicing effectively in their size and darkness ratings. Moreover, whereas the Japanese and Mandarin speakers linked the velars [ɡ, k] with hardness, the Korean and English speakers linked the alveolars [d, t] with this attribute.

It appears reasonable to assume that many, if not all, sound–meaning associations shared across many languages are clearly and strongly motivated by the sensory resemblance between the sound and meaning (i.e., iconicity; for related discussions, see Dingemanse , 2016; Johansson , 2020b; Saji , 2019; Shih , 2019; Thompson , 2021). For example, the frontness–smallness association is very common arguably because people can feel the iconic correspondence between the oral cavity size and the object's size, or they can instinctively or experientially associate high F₂ with small objects via the frequency code, according to which smaller vocalizers (e.g., mice) tend to make higher-pitched sounds than larger vocalizers (e.g., elephants; Ohala, 1994). The association between lip rounding and darkness is not so well researched but could be mediated by a more general crossmodal correspondence between pitch and brightness, in which lower frequency sounds are perceived as darker than high frequency sounds (Marks, 1975). The acoustic correlate of lip rounding is lower formant frequencies, especially F₃, which could lead to rounded vowels sounding dark. In the tactile domain, meanwhile, rounded vowels were linked to softness. It is possible that the softness of the lips matches a soft object or their rounded shape evokes a dull, rather than sharp, image that matches a soft surface (see, e.g., Köhler, 1929).³

In contrast, language-specific sound symbolism is more subject to individual languages' lexical patterns. Japanese speakers' sensitivity to obstruent voicing is a clear example of this. Voicing is the most important phonetic feature in the sound-symbolic system of Japanese ideophones, which very often form minimal pairs that only differ in initial voicing (e.g., kirakira /kiɾakiɾa/ “glittering” vs giragira /ɡiɾaɡiɾa/ “glaring”; porori /poɾoɾi/ “a light object dropping” vs borori /boɾoɾi/ “a heavy object dropping”; tonton /tonton/ “knocking” vs dondon /dondon/ “banging”; Hamano, 1998). The current results demonstrate that Japanese speakers use this phonosemantic pattern in their sound-symbolic perception of nonwords. Note that voicing of stop consonants is not phonemically contrastive in Korean and Mandarin, and Korean, instead, has a three-way distinction of stops as in ca-ta /tɕada / “sleep” (plain), cca-ta /tɕ'ada/ “be salty” (tense), and cha-ta /tɕ^hada/ “kick” (aspirated). These phonological backgrounds may partly explain why speakers of these languages were not sensitive to voicing symbolism. Recall that Korean speakers, instead, attributed smallness to the velar stops and Mandarin speakers attributed smallness to the rounded as well as front vowels.⁴

Similarly, the Korean speakers' sound-symbolic perception of the rounded vowels can be attributed to the phonological system of this language. Korean ideophones exhibit vowel harmony according to their unique two-tiered vowel system, which consists of dark [i, e, ɯ, ʌ, u] and “light” vowels [ɛ, ø, a, o] (Kim-Renaud, 1978; Sohn, 1999). Dark vowels typically co-occur with dark vowels (e.g., phwungten /p^huŋdʌŋ/ “a large object splashing”), and light vowels with light vowels (e.g., phongtang /p^hoŋdaŋ/ “a small object splashing”), expressing dark/heavy/large and bright/light/small meanings, respectively. Our Korean participants perceived [u] but not [o] as dark precisely because of this language-specific sound-symbolic system.

Language-specific sound symbolism may also arise from the analogy to specific ideophones or other lexical items. The Japanese speakers' association between the velars and hardness may be partly attributed to Japanese words for hard texture, such as katai /katai/ “be hard,” kachikachi /katɕikatɕi/ “stone-hard,” and gachigachi /ɡatɕiɡatɕi/ “hard-frozen” (for a related discussion, see Kumagai , 2022). The Korean speakers' perception of [t] as hard might come from the common texture adjective ttakttak-hata /t'akt'ak^hada/ “be hard.” The English speakers' associations between [d, t, k] and hardness might also be related to certain onomatopoeic words, such as dingdong, tap, crack, and crunch. Similarly, in Mandarin, the word ying4 /ɪŋ˥˩/ hard contains a velar.

The conflicting results in the English data, which linked [k] with hardness but not [ɡ], might also have a lexical explanation. The English phonestheme kn, in words such as knee, knob, knot, knuckle, and knoll, is associated with round things, which as discussed could, in turn, be related to softness. Crucially, etymologically, these words go back to Proto-Indo-European roots in which the velar stop was actually voiced (e.g., knee goes back to PIE *ĝenu “knee,” knot, knob, and knuckle go back to PIE *gen “to pinch, clench, ball up”). While there is no way to know for certain that this is the source of English speakers' tendency to consider [ɡ], in particular, as relatively soft, previous studies have found evidence for iconic associations in languages having a long history and even persisting long after sound change has obscured the original phonetic motivations for the associations (Carling and Johansson, 2014; Philps, 2023; Winter , 2022).

The current study identified a few cases of pluripotential sound symbolism. Here, we consider how the different meanings are related to each other.

The associations between the rounded vowels and dark and soft images were found most widely. Both sound–meaning associations have relatively clear and strong iconic motivations. As we discussed in Sec. V A, the low frequency of the rounded vowels may evoke a dark image, and the soft texture and rounded shape of the lips may mediate these vowels and softness. Crucially, it is difficult to draw a direct link between darkness and softness. In fact, we do not know any conventional metaphors or metonymies in the four languages in which a word for “dark” means “soft” or a word for “soft” means “dark” (a soft light would not mean “a dark light”). Without the mediation of the rounded vowels, it would even be possible that softness is associated with brightness rather than darkness, presumably, through the positive valence they have in common. For example, the Japanese words yawaraka-na /jawaɾakana/ “be soft” and hizashi /çizaɕi/ “sunshine” but not hikage /çikaɡe/ “shade” have relatively positive meanings, which allow yawaraka-na hizashi to mean “a gentle sunshine” as a so-called synesthetic metaphor (Winter, 2019).^5,6

Another example of pluripotentiality is the Japanese speakers' associations between voicing and size and darkness. Both sound–meaning correspondences have an iconic basis. Voiced obstruents in Japanese involve greater energy than their voiceless counterparts and also expand the oral cavity to lower the air pressure (Kawahara , 2018). These phonetic characteristics of voicing may motivate its association with large size. The voicing–darkness link can be attributed to the relatively low frequency of voiced obstruents just as with the rounding–darkness association. Thus, the two sound–meaning associations have their own reasons to exist. The two concepts, size and darkness, are not related to each other in the lexicon either. Big night would not mean “dark night,” and a dark house would not mean “a big house.”

Finally, the Mandarin speakers rated [i] as small and bright. As discussed in Sec. V A, the smallness of [i] is straightforwardly attributed to its articulatory and acoustic properties via iconicity (and perhaps indexicality). The association between [i], a vowel with high F₂, and brightness is another example of the pitch–brightness correspondence. It is, again, difficult to find a direct metaphorical or metonymical link between the two concepts. A small room would not mean “a bright room,” and a bright dog would not mean “a small dog.”

These results confirm that one sound can be iconically associated with different meanings without the mediation of metaphor or metonymy.^7,8 In fact, none of the above pairs of concepts are reported to “colexify” (i.e., cross-linguistically tend to be encoded in the same lexemes; Rzymski , 2020). However, it remains unclear whether these meanings instantiate an abstract, multisensory image [i.e., Fig. 1(A)] or exist independently [i.e., Fig. 1(D)].⁹ It is also worth investigating whether iconicity can “block” metaphorical extension in sound symbolism as reported for sign language (Meir, 2010) and ideophones (Akita, 2013; see also Winter, 2019). Furthermore, there is the issue of sound–meaning associations based on lexical analogy (e.g., the alveolar stops for hardness in English, Sec. V A). It remains to be investigated whether the sounds in these analogy-mediated associations are also pluripotential or restricted to specific semantic scales due to their strong relations to particular lexical items. To answer this question, future research will need to test more semantic scales.

We close our discussion with a visual summary of our findings on the pluripotentiality of sound symbolism using what we term “phonosemantic” maps (cf. McLean, 2021; Van Hoey, 2024). Our phonosemantic maps are inspired by semantic maps primarily found in typological investigations of grammatical morphemes, where they are used to track and compare cross-linguistic patterns of polysemy and semantic change. For example, Haspelmath (2003) draws a maximal cognitive space for dative morphemes across languages that covers DIRECTION, RECIPIENT, PURPOSE, BENEFICIARY, etc. Dative morphemes in individual languages, such as English to and French à, are mapped on this space, allowing us to make cross-linguistic generalizations. For example, all dative morphemes that can express BENEFICIARY [e.g., Ai ni keeki o yaku /ainikeːkiojakɯ/ “bake a cake for Ai” (Japanese)] can also express RECIPIENT (e.g., Ai ni keeki o ageru /ainikeːkioaɡeɾɯ/ “give a cake to Ai”) but not vice versa.

In a typical semantic map, the base of the map is the cognitive space, and it is onto this cognitive space that the functions of different grammatical morphemes are mapped. In our phonosemantic maps, on the other hand, the base of the map is, instead, the phonological (or phonetic) space, and it is onto this phonological space that phonosemantic associations are mapped. Whereas in a typical semantic map, the assumption is that semantic extension occurs primarily through conceptual links, in our phonosemantic maps, we assume that semantic extension occurs primarily via links mediated by sound symbolism. That is, it is the inherent properties of the sounds themselves that provide the natural links between concepts, not necessarily the semantic properties of the concepts (as in metaphor and metonymy). The phonosemantic maps for the results of our experiment are shown in Fig. 5, which is, again, based on the CLMMs whose reference levels are [ˈäbä] in most cases.

The phonosemantic map representation effectively shows how each sound behaves sound symbolically and how the behavior is shared across languages. For example, Fig. 5 suggests that the relationship between the three semantic scales is not uniform: the relationship between size and hardness might be weaker than the other two relationships. We will be able to test this hypothesis by increasing the amount of data (see Westbury , 2018, for a related discussion).

Studies on sound symbolism have accumulated experimental and descriptive evidence for numerous sound–meaning associations in numerous languages. Phonosemantic maps will serve as a cross-linguistic common ground for sound symbolism researchers to put the evidence together and draw general conclusions.

In this paper, we tested the sound-symbolic intuitions of speakers of four languages from different phyla on three semantic scales: size, brightness, and hardness. A few phonetic features were iconically associated with two scales: lip rounding for darkness and softness in Japanese, Mandarin, and English, Japanese voicing for large size and darkness, and Mandarin [i] for smallness and brightness. We analyzed these cases as pluripotential sound symbolism without metaphorical or metonymical mediation. In particular, darkness and softness symbolisms suggest that the semantic networks of sound symbolism can differ from those of (synesthetic) metaphor and metonymy. We summarized our findings with a phonosemantic map approach to the description of sound symbolism that is expected to facilitate further investigations into pluripotential sound symbolism that cover a wider range of languages and cultures (Ćwiek , 2021). We hope that future research will enrich the maps to illuminate the role of abstraction and metaphor/metonymy in sound symbolism.

Our sincere gratitude goes to Susanne Fuchs and the two anonymous reviewers of The Journal of the Acoustical Society of America (JASA), whose comments helped us significantly improve this paper. We also thank John L. A. Huisman for his advice in statistical analysis. This work was supported by Japanese Society for the Promotion of Science (JSPS) Grant-in-Aid for Scientific Research (C; Grant No. 20K00567).

The authors have no conflicts to disclose.

The experiment reported in this paper was approved by the Institutional Review Board at the Australian National University (Project No. 2018/544). All participants gave online informed consent and agreed on the publication of their anonymized data.

The data that support the findings of this study, all experimental stimuli, data, and R code are available through the Open Science Framework at https://osf.io/3vdre/.