The sound symbolism bootstrapping hypothesis for language acquisition and language evolution

Abstract

Sound symbolism is a non-arbitrary relationship between speech sounds and meaning. We review evidence that, contrary to the traditional view in linguistics, sound symbolism is an important design feature of language, which affects online processing of language, and most importantly, language acquisition. We propose the sound symbolism bootstrapping hypothesis, claiming that (i) pre-verbal infants are sensitive to sound symbolism, due to a biologically endowed ability to map and integrate multi-modal input, (ii) sound symbolism helps infants gain referential insight for speech sounds, (iii) sound symbolism helps infants and toddlers associate speech sounds with their referents to establish a lexical representation and (iv) sound symbolism helps toddlers learn words by allowing them to focus on referents embedded in a complex scene, alleviating Quine's problem. We further explore the possibility that sound symbolism is deeply related to language evolution, drawing the parallel between historical development of language across generations and ontogenetic development within individuals. Finally, we suggest that sound symbolism bootstrapping is a part of a more general phenomenon of bootstrapping by means of iconic representations, drawing on similarities and close behavioural links between sound symbolism and speech-accompanying iconic gesture.

1. Sound symbolism is not peripheral in modern language

Since de Saussure's [1] highly influential work, the arbitrary relationship between form and meaning in words has been considered to be one of the ‘design features’ of language [2] in traditional linguistics. This is supported by the fact that different languages assign different sounds to the same concept (e.g. ‘tree’ in English versus ‘arbre’ in French) [1]. Though de Saussure acknowledged that onomatopoeias (words that imitate sounds, e.g. ‘bowwow’ for dogs’ bark) are exceptions to the arbitrariness principle, they were considered to be a marginal phenomenon in language. This view has been inherited in more recent writing [3]. For example, Newmeyer [4] writes that ‘the number of pictorial, imitative, or onomatopoetic nonderived words in any language is vanishingly small’ (p. 758).

Indeed, for most words in the lexicons of all languages, mapping between sound and meaning may seem arbitrary. However, sound symbolic words—those that have an inherent non-arbitrary link between sound and meaning—are more abundant than typically assumed, as we will review below.

Sound symbolism can be seen as a form of iconicity. Perniss & Vigliocco [5] define iconicity as resemblance between properties of a linguistic form and the sensori-motor and/or affective properties of referents. At least some types of sound symbolism show clear resemblance between properties of speech sounds and properties of their referents. For example, reduplication in Japanese mimetic words indicates repetition in the referent events (e.g. ‘goron’ is a heavy object rolling once, ‘gorogoro’ is a heavy object rolling repeatedly).

Many languages of the world have a large grammatically defined class of sound symbolic words (called ‘ideophones’, ‘expressives’ or ‘mimetics’) in which the iconic relation between sounds and meaning is apparently felt by native speakers of the language and sometimes even by people who do not speak that language. Such a specialized word class exists in most of the East Asian languages (Japanese [6,7], Korean [8], Cantonese [9]), many of the Southeast Asian languages [10,11], most of the sub-Saharan African languages [12,13], some of the Australian Aboriginal languages [14,15], some of the South American languages [16,17] and some non-Indo-European languages of Europe (Finnish and Estonian [18], Basque [19]). Sound symbolic word classes may contain thousands of words; for example, one dictionary of Japanese mimetics lists 4500 entries [20].

Sound symbolism can also be observed in conventionalized words that are not considered as ‘specialized sound symbolic words’ (e.g. mimetics, expressives and ideophones). Berlin [21] noted that there is sound symbolism in names of two species of birds, rails and tinamous, in 17 languages spoken in indigenous tribes in South America. As seen in figure 1, the shape of rails (figure 1a) is angular and sharp, whereas that of the tinamous (figure 1b) is round. In these languages, the words naming rails include a stop consonant [t] or [k], whereas the words naming tinamous birds tend to have nasals, which connote slowness, roundness, fatness, softness and heaviness.

Figure 1.

Wood Rail, Aramides calopterus (a) and Great Tinamow, Tinamus major (b), which are allied species in Berlin [21]. Drawn by Joseph Smit [22,23]. (Online version in colour.)

Systematic sound–meaning correspondence exists in English, which does not have a special class of sound symbolic words. Some clusters of words in English with similar meanings have the same sounds at the beginning or the end (called phonoesthemes [24,25]). For example, several words beginning with ‘gl’ have meanings related to light: ‘glitter’, ‘glare’, ‘glow’ and ‘glistening’.

The psychological reality of sound symbolism has been well established, not only for speakers of languages with a grammatically defined class of sound symbolic words but also for those without such a word class. For example, Köhler [26] noted that certain sound–shape correspondences were judged to be a good match: when novel words ‘maluma’ and ‘takete’ are presented as labels for a rounded versus a spikey object, speakers of different languages judged ‘maluma’, to be more appropriate for the rounded object and ‘takete’ for the spikey object (see also [21,26–30]). The size symbolism pointed out by Sapir [31]—‘mil’ is judged to be more appropriate for the small object and ‘mal’ for the large object—has also been empirically established across speakers of different languages as well [21,32]. It has also been shown that English speakers' automatic lexical processing is affected by sound symbolism. For example, in a lexical decision task, English-speaking adults were faster to reject non-words when they sound-symbolically matched the shape of the frame in which the words appeared in (e.g. ‘kide’ in a spikey shape, i.e. Köhler's shape sound symbolism [26]) than when they did not match [33].

Furthermore, in a large-scale corpus analysis, Monaghan et al. [34] demonstrated that sound–meaning mappings in the English lexicon are more systematic than would be expected by chance. That is, subtle sound symbolism, which people do not consciously detect, may exist throughout the conventional (i.e. non-mimetic) lexicon [34,35].¹ Thus, form–meaning correspondence is not entirely arbitrary, even in a language like English.

Monaghan et al.'s [34] finding further suggests that sound symbolism is continuous rather than dichotomous, as some sound symbolic sound–meaning relations may be so subtle that people do not consciously detect them under usual circumstances, while others can be more apparently detectable (e.g. sound symbolism in words like ‘thump’ and ‘bump’ [25]). The degree of iconicity varies within specialized sound symbolic words as well. For example, Japanese mimetic words for sound (i.e. ‘wan-wan’ for dog barking) are more iconic than mimetic words for perceptible motions (e.g. doshi-doshi) and object properties (e.g. tsuru-tsuru), which are, in turn, more iconic than those for mental states and emotions (e.g. uki-uki), according to Akita's analysis [36].

The prevalence of sound symbolism may vary across different conceptual categories of words, reflecting the pressure from two directions—one towards expanding the vocabulary to accommodate needs to make finer contrasts, and the other towards maintaining iconicity (cf. [34,37])—as well as reflecting how easy it is to represent the referents by sound features. While numerous mimetics in Japanese make fine-grained distinctions in manners of actions, manners of physical sensations and certain properties of objects (e.g. texture), few mimetics denote objects. This tendency has also been observed in other languages with specialized sound symbolic vocabulary [10,38]. One exception to this tendency is the mimetics used in Japanese infant-directed speech, where caretakers extensively use onomatopoeias as object names (e.g. ‘wan-wan’, dogs’ barking, to refer to dogs) [39] in the early stage of language development, during which time children's vocabularies are still small [40].

Object names are underrepresented among mimetics, in comparison to action, relation or property names. This may be because the need for fine discrimination among similar concepts is greater for object names than for actions, relations and object properties. While sound symbolism might be useful, it could impede word learning when there is a need to make fine discriminations among similar concepts [41]. One reason why fine discrimination is necessary for object names may be that nouns are more open to new entries, compared with predicate terms such as verbs and adjectives [42].

Object names may well have been iconic in their historical origin. For example, names of birds such as ‘karasu’ (crow) and ‘gan’ (goose) that are conceived as non-mimetic conventional lexical words in modern Japanese have onomatopoeic origins (e.g. mimicry of bird vocalizations) [43]. As the object name lexicon has grown over historical time, however, the original iconicity may have become obscure because preserving iconicity became disadvantageous as the words denoting similar objects were continuously added to the lexicon.

Importantly, however, Monaghan et al. [34] reported that the degree of sound–meaning systematicity was not any greater for nouns than for verbs in English. Thus, the distribution of sound symbolic words may be different for languages with a specialized sound symbolic vocabulary, like Japanese, and for those without such vocabulary, like English. How languages stand on the balance between iconicity and arbitrariness is an extremely important issue for future research.

In any case, recent findings from cognitive psychology, cognitive neuroscience, developmental psychology, cognitive and anthropological linguistics converge on the view that iconicity plays a core role for philogenesis and ontogenesis of language as well as for online language processing, as pointed out by Perniss & Vigliocco [5]. As such, sound symbolism is not a marginal phenomenon, but is an integral part of language.

The primary goal of this article is to discuss why the role of sound symbolism is important for language, with a special focus on the role of sound symbolism in language development, and to propose the sound symbolism bootstrapping hypothesis, which claims that sound symbolism provides a scaffolding mechanism for children in various stages of language development. In what follows in this article, we extensively discuss the details of the sound symbolism bootstrapping hypothesis. We then explore how sound symbolism could have played a role in our ancestors' protolanguage.

2. The induction problem in word learning and sound symbolism bootstrapping hypothesis

During the first year of life, infants start to map speech sounds onto meaning, and in subsequent years they acquire a vast number of words that build a lexicon. How children acquire language—its ontogenesis and the developmental path thereafter—is still not entirely understood. How do they come to know, for example, that the sounds people make with their mouths are ‘words’ and that they are names for objects, actions or properties? How do they learn the meanings of words? We will argue below that sound symbolism can help children overcome these challenges.

When a child hears a word in an everyday situation, the visual information they receive from the world is very rich, as well as unsegmented. Imagine that a child hears his mother say ‘Oh, look at the dog walking over there!’ when they pass by a dog with a leash walking with its owner. How would he identify the referent of ‘dog’ or ‘walking’ out of the extremely rich perceptual information that is changing from moment to moment? How can he be prevented from thinking that ‘dog’ means a thing with four legs moving together with a person?

Even when a novel word is explicitly associated with an object in an unambiguous manner, and the child successfully connects the word to the referent in that scene, this success still does not allow children to use the word in new situations. For example, to be able to use the word ‘dog’, the correct visual identification of a particular dog in the observed scene is not sufficient for a child to judge what other things can also be called a ‘dog’. The child somehow has to have a visual as well as conceptual representation of ‘dog’ to be able to determine whether other objects that carry similarity to the original referent (e.g. other dogs of a different kind, other small four-legged animals, etc.) can also be considered as a referent of ‘dog’ (cf. [44]). Likewise, to be able to use the verb ‘to walk’, whose meaning is usually considered to be very concrete, one needs to know that it can be applied to a wide range of motion events, including toddlers tottering, a woman sashaying, an athlete walking very fast for speed walking or a horse walking with four legs, but not to visually similar events with the same agents such as a human running or a horse galloping.

In other words, to be able to use a word—be it a noun, verb or adjective—children need to find the invariance in the contexts in which the word has appeared, working at first from a small number of exemplars. This induction is logically not possible when a child encounters a new word for the first time because there are too many possible ways of generalization from a single exemplar [45]; however, this is exactly what children face when they learn words [46–48].

A large body of research has addressed how children get around this problem. Young children recruit constellations of cues—conceptual, social, pragmatic and distributional—to constrain the inference of word meanings [47–50]. For example, children know that, in order to identify the referent, the speaker's eye-gaze or other social cue is useful [51,52]. They also know that words appearing in different positions in a sentence and in different forms are mapped to different kinds of concepts such as objects, substances, actions or properties [53,54]. In generalizing object names, for example, children know that words (count nouns) are extended on the basis of shape [55–57].

Not all cues are available from the earliest stages of lexical development. For example, word leaning biases such as the shape bias and the mutual exclusivity bias are likely to emerge through experience of word learning [56,58]. For verb learning, it is not known whether children at the initial stage of lexical development can exploit the knowledge of the relation between the argument structure and the verb meaning (i.e. the syntax–semantics mapping) when inferring the meaning of a novel word. Furthermore, although this cue is helpful for mapping the verb to a rough, macro-level concept (e.g. whether it should be mapped to a caused motion or a spontaneous motion [59–61]), it would not help children to find the differences among words that appear in the same argument structure (e.g. walking versus running versus hopping).

Thus, finding the meaning of a word is challenging for children, especially for infants who have few words in their vocabulary and cannot yet take advantage of cues that have to be learned through experience. What other cues are available for such infants? Here, we propose that a biologically endowed ability to detect sound symbolism provides one such cue.

We argue that sound symbolism helps children learn the meaning of words at different stages of early lexical development. In particular, this sound symbolism bootstrapping hypothesis consists of several related claims.

• (1) Children, even pre-verbal infants, are sensitive to sound symbolism, due to a biologically endowed ability to map and integrate multi-modal input, as suggested by Ramachandran & Hubbard [30] and Spector & Maurer [62].

• (2) Young children are sensitive to a wider range of possible sound symbolic correspondences than adults, but this sensitivity gets pruned and reorganized as they learn more words in their native language.

• (3) Sound symbolism helps infants who have just started word learning to gain the insight that speech sounds refer to entities in the world (i.e. the referential insight for speech sounds).

• (4) Sound symbolism helps infants associate speech sounds and their referents and establish a lexical representation.

• (5) Sound symbolism helps toddlers identify referents embedded in a complex scene, alleviating Quine's problem [45].

Here, we review evidence for these claims. We further explore how the ontogenesis of lexical development might mirror the evolution of language in our ancestors and how sound symbolism relates to iconicity in sign language and speech-accompanying gesture.

3. Children's sensitivity to sound symbolism

In order for sound symbolism to bootstrap lexical development, children have to be able to detect it. Cross-linguistially recognized sound symbolism may have an especially natural correspondence between sounds and meanings. Thus, such sound symbolism may be the best place to explore whether children have sensitivity to sound symbolism at the start of, or even prior to, lexical development.

Toddlers’ sensitivity to universal sound symbolism has been demonstrated by Maurer et al. in two-way forced choice tasks [29], using Köhler's shape sound symbolism [26]. Canadian toddlers were presented with a novel word (e.g. ‘kay-kee’ or ‘boo-baa’) and two line drawings, one of which had a rounded shape and the other had a spikey/jagged shape. They were able to pick the shape and the matching novel word at levels above chance [29].

Toddlers’ sensitivity to cross-linguistically recognizable sound symbolism was also shown with Japanese 25-month-olds [63] in the domain of manner of motion. The children were presented with a novel sound symbolic words and two video clips, each showing a person walking in a specific manner. They were asked to select the manner of walking to which the word referred. Only one of the videos sound-symbolically matched the word, after being established through the results of prior experiments with Japanese-speaking and English-speaking adult participants (see figure 2 for an example). The children were able to select the correct video at levels above chance.

Figure 2.

An example of manner of walking, used by Imai et al. [63], which sound-symbolically matches the novel word, ‘nosunosu’. In a pretest, Japanese- and English-speaking adults judged the novel word ‘nosunosu’ to sound-symbolically match this heavy and slow manner of walking. (Online version in colour.)

Japanese 3-year-olds also use their sensitivity to cross-linguistic sound symbolism in generating novel sound symbolic words. When describing events involving rolling and jumping, the toddlers produced novel mimetics, along with conventional ones. English-speaking adults with no knowledge of Japanese were able to guess which novel mimetics were used for which type of event (rolling or jumping) at above chance levels of accuracy [64]. This indicates that the novel mimetics produced by Japanese toddlers included cross-linguistically recognizable sound symbolism.

Young infants can also detect cross-linguistically shared sound symbolism. Spanish-reared three-month-olds are sensitive to the sound symbolism of vowels and size [65], that is, the association of high/mid-frontal vowels (/i/, /e/) and low/mid-posterior vowels (/o/, /a/) with small objects and large objects, respectively [31]. In a two-way preferential looking paradigm, infants were presented with a syllable (e.g. ‘di’ versus ‘do’, or ‘de’ versus ‘da’) with two geometric objects (e.g. ovals), which differed only in size and were presented side by side (thus one object was a sound symbolic match, and the other was a mismatch). The infants looked at the sound-symbolically matching object (e.g. for ‘di’, the smaller object; for ‘do’, the larger object) longer than the mismatching object.

The results are mixed for very young infants' sensitivity to detect Kölher's sound symbolism [26] for shapes. In a study by Ozturk et al. [66], American four-month-olds showed sensitivity to this sound symbolism. Using an infant-controlled sequential preferential looking paradigm, infants were presented with one of two words (‘bubu’ and ‘kiki’) and one of two shapes (a rounded shape and a spikey shape). Results showed that infants looked longer at the sound-symbolically mismatching shape than the matching one. By contrast, Forte et al. [67] conducted a series of experiments with French-learning five- and six-month-olds, who failed to show sensitivity to Köhler's sound symbolism [26]. Here, in a simultaneous preferential looking paradigm, infants were presented with a novel word (e.g. ‘buba’ or ‘kite’) along with two shapes side by side, one of which was rounded and the other spikey. There was no significant difference in looking times for the two objects. It is not clear why Ozturk et al.'s [66] and Fort et al.'s [67] results differed, but the discrepant results suggest that the effect of this type of sound symbolism in this age group may be fragile.

4. Development of language-specific versus universal sound symbolism

There has been an assumption in the literature that sound symbolism is universal; if a certain sound–meaning correspondence is identified by speakers of one language, this should be generalizable to speakers of any other languages. This assumption has been supported by the fact that speakers of many different languages sense Köhler's shape sound symbolism [26] in the same way, as reviewed earlier (for English [32], Japanese [68,69], Himba [27], Kitwonge-Swahili bilinguals [70]). Furthermore, some aspects of sound symbolism in words in a given language can successfully be decoded by speakers of another language (Japanese sound symbolic words for laughing/smiling and for pain by English speaker [71,72]; see also Imai et al.'s result [63] for motion sound symbolism). The finding that people can correctly match a pair of antonyms in a foreign language to the corresponding pair of words in their native language [73–75] also endorses this view.

However, some sound symbolic words in a given language are opaque to adult speakers of other languages. Iwasaki et al. [71,72] found that adult English speakers' judgements of conventional Japanese mimetic words for laughing and walking tended to converge with those of Japanese speakers on semantic dimensions concerning the magnitude (of size and sound), while they were quite different on evaluative dimensions (e.g. beauty and pleasantness), supporting the idea that some aspects of sound symbolism are universal, while others are language-specific.

Details of cross-linguistically common and language-specific aspects of sound symbolism were explored in a study in which English- and Japanese-speaking adults generated novel words for various manners of motion. Saji et al. [76] presented various locomotion videos to Japanese and English speakers and asked them to generate a word that would sound-symbolically match each action, then rate that action on five semantic dimensions (size, speed, weight, energeticity and jerkiness). Results showed that certain sound–meaning links were common across the two languages.

For example, English- and Japanese-speakers shared voicing-speed and nasality-speed mappings (i.e. voiced = slow, voiceless = fast; nasal = slow, non-nasal = fast). This may be accounted for by the longer duration of vocal cord vibration in voiced and nasal stops (see [7,77] for similar discussion). As voiced consonants have a shorter voice onset time, vocal cord vibration starts earlier and voicing is sustained for a longer duration than their voiceless counterparts. Nasal consonants can be prolonged without a change in quality. The longer voicing duration involved in these consonants and their non-turbulent nature appear to be readily mapped to the long duration of slow motion. Thus the sound–meaning associations shared by Japanese and English may be accounted for by our common bodily experience with articulation (see also [78]) or audition.

These similarities do not mean that English and Japanese speakers always mapped sounds and meanings in the same way, however. In fact, even though the two languages used the same sound properties on the same semantic dimensions, the directionality of the sound–meaning mapping was sometimes reversed (see [75] for a relevant finding). For example, the affricate manner of articulation (e.g. the palato-alveolar affricate [t∫]) was associated with light motions in Japanese, but with heavy motions in English. These disagreements may be explained by the cross-linguistic differences in the phonological status of these sounds. For example, in Japanese, the phone [t§] often appears secondarily, as a result of the palatalization process (in a context such as /ty/), whereas this is not the case in English. In any case, this result implies that language-specific sound symbolism exists and that cross-linguistically shared and language-specific parts of sound symbolism are intricately intertwined within each language.

How, then, does the sensitivity to language-specific and universal sound symbolism develop in children? One possibility is that young children first detect only sound symbolism that is shared universally, and later learn language-specific sound symbolism through learning of their native language. An alternative possibility, however, is that young children are sensitive to all possible sound symbolic correspondences that could appear in any language of the world, but only a subset of these correspondences are compatible with the phonological inventory and the existing words in the language the children are learning. As they grow up, the sensitivity to the incompatible correspondences wanes, and adults maintain only the sensitivity to the compatible correspondences. Thus, each language draws from a universal inventory of possible sound symbolic correspondences. Some sound symbolic correspondences (e.g. Köhler's shape sound symbolism and Sapir's size–sound symbolism) appear in many languages, perhaps because they are strongly supported by iconic relationships between articulatory and/or acoustic features of speech sounds and the referents.

Evidence for the latter possibility is found in a study of sound symbolic sensitivity in English- and Greek-speaking adults and 3-year-olds [79]. In a pretest, adult speakers of Japanese, English and Greek rated the degree of sound symbolic match between novel words and various manners of walking, similar to the ones in figure 2. Based on the ratings, three types of items were selected: universal items (rated as a good sound symbolic match by the speakers of all three languages), English-specific items (rated as a good sound symbolic match by English speakers, but not by Greek and Japanese speakers) and distractor items (rated as a poor sound symbolic match by the speakers of all three languages). Then, adult and child speakers of English and Greek (adult speakers were different from the ones in the pretest) were presented with a novel word and a pair of manners of walking. In the universal condition, one manner of walking was a universal item (target) and the other was a distractor item. In the English-specific condition, one manner of walking was English-specific and the other was a distractor.

Not surprisingly, adult and child English-speakers correctly chose the target in both the universal and English-specific conditions. Adult Greek speakers correctly chose the target at levels above chance only in the universal condition. Crucially, Greek-speaking children could correctly choose the target video in both universal and English-specific conditions. That is, Greek-speaking children were sensitive to a wider range of sound symbolic correspondences than Greek-speaking adults. This suggests that Greek-speaking adults had lost sensitivity to sound symbolic correspondences supported by properties of English but not by properties of Greek.

This narrowing of sound symbolic sensitivity during development may be analogous to the way language-specific phonemic contrasts are acquired. Up to 10–12 months old, infants are sensitive to phonemic contrasts in a foreign language that their carer have long before lost sensitivity to [80]. For example, infants growing up in an English-speaking environment can distinguish Hindi contrasts (i.e. dental and alveolar stops) that are not contrastive in English and thus English-speaking adults cannot distinguish.

How do children become tuned into the system of sound symbolism in their native language? Here, it is relevant to note that sound symbolic words in a specialized word class, such as Japanese mimetics, are very difficult for adult second language learners to master [71,72]. To acquire a native speaker's sensitivity and productive competence—that is, to be able to comprehend and use conventional and novel mimetics productively and creatively—may require extensive exposure to mimetics used in real contexts. This suggests that, even though some aspects of sound symbolism may be biologically grounded, it is crucial to have intensive exposure to a specific language in early stages of development. Through statistical learning [81], young children may be able to extract patterns of form–meaning co-occurrences in the words they have learned [34] and abstract out language-specific aspects of sound symbolism. Such learning experiences may result in sound symbolic words that have both universal and language-specific components.

The studies reviewed so far provide evidence for sound-symbolic sensitivity, but these studies do not provide evidence that sound symbolism is directly related to the ontogenesis of language. The following infant EEG study addresses this issue.

5. Neural response to sound symbolism in 11-month-old infants

Sound symbolism may arise from the sense of similarity between speech sounds and other types of information through naturally occurring cross-modal mapping. Owing to dense connectivity across different sensory brain regions, infants may spontaneously map perceptual experiences across different modalities onto speech sounds (for a review, see [82]). Human infants can already map information in different modalities in the way adults would. For example, they can map size and numerosity [83], or acoustic properties of speech and non-sounds onto properties of visually presented objects [84,85]. This cross-modal mapping ability may not be limited to humans, as chimpanzees can map auditory pitch and luminance [86]. This perceptual ability may later develop into a more abstract system of symbols embodied in language [87,88] both ontogenetically and phylogenetically.

Is infants’ ability for cross-modal mapping directly linked to language processing, and if so, how? Asano et al. [68] (see also [89]) investigated this question with Japanese-reared 11-month-olds in an EEG study. In each trial, infants were presented with a picture of a shape (spikey or rounded) followed by a novel word (‘kipi’ or ‘moma’). The word–shape pairs were either sound-symbolically matching or mismatching (Köhler's shape sound symbolism [26]).

The recorded EEG was submitted to averaging (ERP, ‘event-related potentials’) and large-scale phase synchronization analyses to explore (i) whether 11-month-old infants detect sound symbolism and treat sound-symbolically mismatching words as semantically unexpected and (ii) how different regions of the infant brain communicate while sound-symbolically matching and mismatching words are processed. Growing evidence indicates that large-scale synchronous neural oscillations play an important role in dynamically linking multiple brain regions in adults, something that presumably reveals functional communication among these regions [90–97].

Concerning the ERP pattern, infants responded differently to the sound-symbolically matching and mismatching word–shape pairs. The timing and topography were similar to the typical N400 effect, with a stronger negative deflection for the mismatching pairs at about 400 ms after the stimulus onset [98], an index of semantic integration difficulty [98–101]. Second, phase synchronization of neural oscillations (phase locking value) increased (as compared with the baseline period) significantly more in the mismatch condition than in the match condition. This effect was observed in the β-band and most pronounced over left-hemisphere electrodes during the time window of the N400 (301–600 ms). The time course of large-scale synchronization suggests that cross-modal binding was achieved quickly in the match condition, but sustained effort into the time range of the N400 effect was required in the mismatch condition. An additional brain oscillation analysis showed an increase of early (less than 200 ms latency) γ-band oscillations in the match condition compared with the mismatch condition. A number of adult studies have revealed that early γ-band activity is related to multi-sensory integration (see [102] for a review).

In a different study [103], when adult participants were presented with real words and non-words in isolation, real words elicited strong EEG coherence in the β-band in the left hemisphere, in comparison to the resting state, but non-words did not do so. The stronger inter-regional communication in the left hemisphere in Asano et al.'s infants [68] thus may indicate that the sound–shape pairings were processed in the language-processing network (in the left hemisphere) in 11-month-old infants.

Taken together, the results from ERP and phase synchronization analyses suggest that 11-month-olds could clearly detect Köhler's shape sound symbolism [26], and further suggests that sound symbolic associations fosters multi-sensory integration and semantic processing. When infants at this age are presented with a spoken word and a visual stimulus, they already attempt to integrate the two stimuli and establish a lexical representation. This process requires substantial effort when the word and visual stimuli do not sound-symbolically match. However, when the two stimuli match sound-symbolically, the sound–vision integration comes naturally due to the iconicity between the sound and visual stimuli, leading to a nascent representation of the word meaning without effort. This may help infants to realize that words stand for concepts; that is, it may provide infants with referential insight for speech sounds.

It is difficult to know whether the 11-month-old infants in Asano et al.’s study had already gained this insight in some form. Given that they showed an N400 response, it could be that at 11-months, these infants already assumed that the two speech sounds referred to the objects, and thought that the round shapes and the angular shapes were anomalous as referents of the words ‘kipi’ and ‘moma’, respectively. However, even if 11-month-olds already had the referential insight in a nascent form through previous word learning experiences, they would need much further experience of word learning to solidify it. If they are able to identify the referents of a word due to sound symbolism and infer the meaning of words using the spontaneous multi-modal binding ability, this should provide important bootstrapping experiences for infants who scarcely have any words in their vocabulary.

6. Sound symbolism scaffolds acquisition of word meaning

(a). Establishing word–referent associations

One of the key claims of the sound symbolism bootstrapping hypothesis is that sound symbolism facilitates children's word learning. Previous research has shown that at 11 months, the language-processing network in the infant's brain responded to sound symbolism [68], but it is not clear whether they were able to establish the word–referent associations and retain it in memory.

We demonstrated that sound symbolism facilitates word learning in 14-month-old Japanese-speaking infants [69]. The infants were tested in a word-learning task that combined habituation and preferential looking. They were repeatedly presented with two word–shape pairs. For half of the infants, word and shape sound-symbolically matched (‘moma’ for a round shape and ‘kipi’ for a spiky shape); for the other half, they mismatched (‘moma’ for a spiky shape and ‘kipi’ for a round shape) [26]. After infants had been habituated, they heard either ‘kipi’ or ‘moma’ and saw the two shapes side by side. Infants looked at the shape that had been associated with the word during habituation faster and longer in the sound-symbolically matching condition than in the mismatch condition. This suggests that sound symbolism helps children to learn word–referent associations at 14 months of age.

(b). Helping children find the invariance for generalization

As discussed earlier, the establishment of word–referent associations is not sufficient for the acquisition of the meaning of the word. For children to be able to use a word in new situations, they need to extract invariance across referents (i.e. to create a word meaning representation). However, this is extremely difficult to achieve from a single or a limited number of exposures, because what the child sees is very rich and contains a great deal of information that is not part of the meaning of the word.

This problem is particularly serious for verb learning as compared with the learning of object names. Unlike objects, actions are ephemeral and difficult to individuate [42,104], so it is not obvious when the action referred to by a given word starts and when it ends. Finding a referent in a spatio-temporally changing scene itself is not easy, and indeed, young children become able to associate a verb to the referent (i.e. the action) later than for nouns in experimental settings [105,106]. To be able to use the verb in different situations, children further need to understand which specific aspect of the action is invariant for the verb and which aspects can vary across different situations in which the verb is used [104,107–109].

As reviewed earlier, it is difficult for preschool age children to extend a novel verb to a new scene in which the action is the same, but the agent (or the theme object, or the instrument of the action) in the original scene is replaced with a new one [104,107,110–112]. Imai et al. [63] and Kantartzis et al. [113] tested whether sound symbolism would help Japanese- and English-speaking 3-year-olds find the invariance for a newly taught verb in action events for successful generalization. Children were assigned to one of three conditions and were taught novel verbs while observing a person walking in different manners. In the experimental condition, a novel verb, which had been created by modifying an existing Japanese mimetic word (i.e. a sound-symbolic word) was paired with a manner of walking that sound-symbolically matched. For example, for a fast walk with small steps, the novel mimetic choka-choka was created from the existing Japanese mimetic choko-choko and presented as a verb (‘choka-choka-shiteru' in Japanese and ‘doing choka-choka’ in English) (see also figure 2 for another example). In the first control condition, the mimetic-based nonsense word was paired with a different motion that did not sound-symbolically match. In the second control condition, a nonsense word that resembled a typical monosyllabic verb in Japanese and English (e.g. neke-tteiru or fepping, respectively) was paired with the same motion from the experimental condition; as a non-mimetic word, it did not sound-symbolically match.

Replicating results from previous studies [104,107,111,112], in the two control conditions with novel words that did not sound-symbolically match their referent actions, both Japanese and English 3-year-olds failed to generalize the newly taught verb to the identical action performed by a different actor. However, when the novel verb sound-symbolically matched the action, not only Japanese 3-year-olds but also English-reared 3-year-olds (who were not familiar with the sound symbolic system of Japanese mimetics) were able to use this cue to generalize the verb to a new event (see also Yoshida's study [114] for similar findings).

Thus, regardless of the language they were acquiring, sound symbolism helped the children to find the relevant invariance in the scene for the verbs. Here, as noted earlier, young children are sensitive to a broader range of sound symbolism [29,79], including sound symbolism that adults speaking the same language might not detect. Thus, they may be more likely than adults to take advantage of sound symbolism in word learning.

7. Sound symbolic words in child-directed speech

Another support for the sound symbolism bootstrapping hypothesis comes from studies investigating sound symbolic words in child-directed speech (CDS). Caretakers often use sound symbolic words with CDS for young children in a way that is appropriate for the children's language comprehension ability. In a classic study, Fernald & Morikawa [115] noted that Japanese mothers used sound symbolic words such as onomatopoeia/mimetics frequently when talking to Japanese infants.

Saji & Imai [116] further studied how Japanese caretakers of 2- and 3-year-old children use mimetics when describing action events in semi-experimental settings. A mother and her child were presented with animated videos of everyday actions that could be expressed either by a conventional verb or a mimetic (e.g. clapping hands, cutting a piece of paper with scissors), and the mother was asked to describe the video to her child. After that, the mother was also asked to describe the video to the experimenter. The mother used mimetics more often for younger listeners, that is, most often to 2-year-olds, the second most often to 3-year-olds, and least often to the adult experimenter.

The caretakers also adjusted the way they used the mimetics according to their child's comprehension ability. For example, the mothers in Saji & Imai's study [116] used mimetics in different syntactic frames depending on their listener's age. When they were talking to 2-year-olds, they used a mimetic word by itself without incorporating it in a sentence, exactly when the motion started in the animation, often with accompanying gestures depicting the motion. By contrast, when they were talking to 3-year-olds, they tended to incorporate mimetics into a sentence; they either attached the light verb –suru (‘to do’) to the mimetic, or they used it adverbially to modify a conventional non-mimetic verb. When describing the action video to the experimenter, they mostly used conventional non-mimetic verbs.

The same pattern was also found in a longitudinal corpus study [117] of a father's input to a boy between the age of eight months and 36 months. In addition to the syntactic frame change seen in Saji & Imai's study [116], the father adjusted the choice of words for action reference along with the child's development. As the child developed from no use of mimetics to productive use of mimetics, the father's input shifted from mostly mimetics alone, to mimetics plus semantically similar conventional verbs, and finally to mostly conventional verbs alone. These studies suggest that Japanese caretakers adjust their input closely according to their child's level of lexical development, and they used mimetics as a tool for the adjustment.

In parallel to caretakers’ more frequent use of mimetics with younger children, younger children produce more mimetics than older children and adults. When Japanese 3-year-olds, 5-year-olds and adults described motion events, the manner of motion was described either by a conventional verb or with a mimetic. The proportion of descriptions using mimetics was higher for younger participants [64].

In the production–elicitation studies by Saji & Imai [116], when caretakers used onomatopoeic/mimetic words without embedding them in a sentence to 2-year-olds, the onomatopoeic/mimetic words were often uttered exactly when the referent action took place. Previous studies demonstrated that parents use various devices to help children connect the word and the referent. For example, parents often point to the object [118,119], and children tend to learn the name better when the referent object was pointed at in the past [120]. Temporal co-occurrence of the referent and labelling is also important [121,122].

However, although pointing and temporal occurrence can help children note the here-and-now word–referent relation, they do not necessarily help them to use the word beyond that particular context. Sound symbolism, in contrast, can directly link the word form and the word meaning, and thus it can help children extract invariance of the word meaning, which is crucial for using the word in new situations. In this sense, sound symbolism would be a more powerful cue for word meaning than pointing and temporal co-occurrence of a word and its referent.

Caretakers also use different types of sound symbolic words, depending on the child's stage of language development [116]. As noted earlier, some mimetic words (e.g. onomatopoeias) are considered to be more iconic than others (e.g. mimetics denoting object properties, manners of action, emotion) (cf. [36]). When their child has a very small number of words in their vocabulary, caretakers mostly use onomatopoeias—direct mimicry of sounds (e.g. ‘chirin-chirin’, mimicry of a bicycle bell when referring to a bicycle) [115]. When children become more experienced in word learning, caretakers no longer use such onomatopoeic words for labelling objects and instead use conventional object names. However, caretakers continue to rely on sound symbolism when their child still shows difficulty in extracting invariance of word meaning. When children become advanced word learners, they no longer use mimetic words in the contexts in which adults would use non-mimetic, conventional words [117].

8. Does sound symbolism always facilitate word learning?

Monaghan et al. [41,123] argue that sound symbolism can inhibit one-to-one mappings when each mapping competes with a similar mapping, but can help make class-to-class associations (e.g. words with certain phonological features refer to actions, and words with other phonological features refer to objects); hence, sound symbolism is beneficial only for grouping words into a large cluster (e.g. identifying grammatical category of individual words) and it impedes learning of individual words. However, many of infants' first words are basic-level terms (e.g. ‘bird’) that refer only to category prototypes (robins, but not penguins). Thus, young infants are not likely to face the kind of situations in which sound symbolism is detrimental to word learning, that is, situations in which phonologically similar words have to be mapped onto similar shape categories, as suggested by Monaghan et al. [34]. Sound symbolism may instead help infants identify the particular part of an ambiguous scene as the referent: when hearing kipi, infants may selectively attend to pointiness in visual shapes, which may help to guide them to the correct referent.

We agree that sound symbolism does not always help or sometimes even impedes word leaning. Learning the meaning of a new word may be impeded if another similar sounding word with similar meaning is activated in children's minds. Also, if sound symbolism is too powerful and children always map a word onto only the element in the scene that is most salient due to sound symbolism, then it would be difficult to build a large lexicon. Eventually, children need to learn to pay attention to other cues as well, and thus the relative influence of sound symbolism may decrease as their lexical development proceeds and children become able to access many other cues for the inference of word meanings, as discussed in the previous section.

However, the novel verb learning studies [63,113,114] indicate that sound symbolism could be helpful for word learning even for preschool age children. The facilitative role of sound symbolism for individual word learning may differ across different classes of words and across different developmental stages. That is, sound symbolism could impede learning of nouns when the noun vocabulary is sufficiently large, but it could still facilitate learning of verbs. This is because verbs do not compete with each other as much because verbs tend to carve up the world in less fine-grained way than nouns, and consequently, languages typically have fewer verbs than nouns. The same facilitative effect may also extend to names of properties (adjectives and adverbs), as learning a property name involves mapping a word to a single property of an object out of the multitude of properties (size, texture, colour, weight, speed, abruptness, etc.), which also poses a challenge for young children [124].

An important unanswered question is how children learn words that are not sound symbolic and yet make use of sound symbolism when, and only when, the word carries sound symbolic properties. It is possible that, through experience in language learning, children quickly learn that words are not always sound symbolic and are willing to form word–referent associations even when they do not detect sound symbolism between the word and the referent, especially for object names. However, when children do detect sound symbolism in learning a novel word, they take advantage of it, and this additional cue is especially helpful for the learning of names for actions and properties, which are especially challenging for children.

9. Implications for theories of language evolution

Researchers have often discussed the possibility that the process of language development in modern-day children mirrors how language was started by our distant ancestors and evolved through history. Some have even speculated that sound symbolism in a modern-day language may be a vestige of protolanguage that was mostly sound symbolic [125,126] and hypothesized that symbolic use of cross-modal mapping is one of the key steps in language evolution.

The emergence of sound symbolic words and how they began to be used as symbols by our ancestors has been debated by researchers. One possibility is that the motor system played a key role. It has been argued that a critical foundation for language evolution was humans’ ability to mimic the external world [127]. In the course of evolution, sound symbolism may have arisen as mimicking of events and object properties in the external world with movements of lips and the tongue [21,62]. For example, the size–sound symbolism for vowel heights may be based on the oral cavity size mimicking the referent object size [31]. In contrast to these motor-based accounts of sound symbolism, some argued that cross-modal mapping between audition and other modalities is the key to sound symbolism. For example, the size–sound symbolism of vowel quality (higher front = small, lower back = big) [31] and consonants (e.g. voiceless = small, voiced = big) could be explained based on acoustic frequency of speech (more acoustic energy in higher frequency = small, in lower frequency = big) [128]. These accounts of sound symbolism do not have to be mutually exclusive; instead, they could be thought of reflecting different forms of iconicity [5]. Be it motor-based or audition-based, sound symbolism may have set the foundation for using speech sound to systematically refer to concepts.

How might sound symbolism have played a role in the bootstrapping process for emergence of a modern-day voluminous lexicon in language evolution? First, the very idea that speech sound could have been used to refer to objects and events in the world may have arisen due to intrinsic and biologically endowed association between speech and information in other modalities (e.g. vision). The awareness could have further brought ‘referential insight’ to our ancestors—that is, the insight that oral sounds can be used to symbolize things that are not the oral sounds themselves, including things that are not present at the speech event (i.e. ‘displacement’ in Perniss & Vigliocco [5]). Second, sound symbolic associations may have helped our ancestors quickly build a shared lexicon that can be intuitively understood by members of the community, which would have promoted the use of ‘words’ (speech sounds) for their primary medium of communication [125].

Sound symbolism may have played a further role in the emergence of combinatoric structure in language [125]. Because phonetic features and other units smaller than a word can carry sound symbolic meanings, a word can have a complex meaning that combines the meanings of parts of the word. To illustrate this in existing Japanese sound symbolic words, the word ‘gorogoro’ refers to a heavy object rolling repeatedly. This word contrasts with the following related words: ‘goron’ = a heavy object rolling once, ‘korokoro’ = a light object rolling repeatedly, ‘koron’ = a light object rolling once, ‘guruguru’ = a heavy object rotating around an axis repeatedly, ‘gurun’ = a heavy object rotating around an axis once, ‘kurukuru’ = a light object rotating around an axis repeatedly, ‘kurun’ = a light object rotating around an axis once. Here, the sequence of a velar stop plus /r/ indicates rotational movement. The voicing of the initial consonant indicates a heavy object. The word's final ‘n’ indicates that the event is completed after a single rotation and has been analysed to symbolize reverberation [77]. The reduplication indicates a repetitive event. That is, the meanings of these words are a combination of component meanings. This compositional nature of sound symbolic words may have facilitated a transition from a ‘holophrase’, a single unanalysable (monomorphemic) word with complex meaning [129], to a complex word with a morphological structure and combinatoric semantics. This principle of combinatoric semantics can subsequently be extended to words in a sentence [130]. This development would further expand the expressive power of protolanguage.

As discussed earlier, language is shaped by two competing forces: one towards arbitrariness and the other towards iconicity [37]. When the size of the lexicon becomes large and different words are used to make fine-grained contrasts for similar concepts, it will become difficult for language to maintain sound symbolism for all words and it could impede growth of the lexicon, as pointed out by Monaghan et al. [41]. Thus, in modern languages, sound symbolism may often be a very subtle tendency in the lexicon, which may not be consciously detectable even by naive speakers, unless a language has a clearly defined word class dedicated to sound symbolic words (e.g. mimetics, expressives, ideophones in various languages of the world).

10. Relation to the iconicity in gestures and sign language

The idea that motivated form–meaning relations facilitate lexical development can be extended to gesture and sign language. In British Sign Language, signs that are judged as iconic by adult raters (the sign form resembles the meaning, e.g. bringing a C-shaped hand closer to the mouth for the sign ‘drink’) were learned earlier than non-iconic arbitrary ones [131]. Speech-accompanying iconic gestures can also guide word learning in hearing children. When 3-year-old English-reared children were presented with a novel verb and a complex action scene, along with an iconic gesture, children interpreted the verb's referent to be the part of the scene depicted by the iconic gesture [132]. That is, iconic gestures guided children to pick out a particular part of a complex scene as the referent of a novel verb.

Sound symbolism has a direct link to iconic gestures. When Japanese speakers produce mimetics during description of motion and action, they tend to produce a co-expressive iconic gesture at the same time [6,133] (see also [134–136] for further discussions). This tendency is stronger in children (3-year-olds) than in adults [64]. Such a link suggests that a common imagistic representation underlies both sound symbolic words and iconic gestures [6].

The close tie between sound symbolic words and iconic gestures has implications for theories of language evolution. The common-underlying-representation view suggests that sound symbolic words and iconic gestures emerged together in the course of language evolution [64]. As discussed earlier, the ability to use cross-modal non-arbitrary mappings to create symbols may have given rise to a communication system that consisted mainly of tightly linked sound symbolic words and iconic gestures. This contrasts with a gestural origin theory of language evolution, which states that protolanguage based on iconic gestures (without speech) preceded spoken protolanguage [137–139]. The gesture-first-theory has difficulty in explaining the close tie between sound symbolic words and iconic gestures in modern humans.

The use of motivated form–meaning mapping may be an important foundation for the human symbolic ability. Be it sound symbolism or other types of iconicity in gestures and signs, children are equipped with an ability to readily take full advantage of it to crack the code of language.

11. Conclusion

Contrary to the traditional view that sound symbolism is a peripheral phenomenon in language, sound symbolism is widely observed in languages in the world. We have argued that sound symbolism facilitates lexical development in children. We reviewed the following lines of evidence for the sound symbolism bootstrapping hypothesis. Pre-verbal infants detect sound symbolism in unfamiliar words and process them as if they were real words, which may lead them to (or solidify) the realization that speech sounds have meanings. Sound symbolism further scaffolds word learning from infancy to early childhood, helping children to establish word–referent associations and also to extract the word meaning invariance from rich and unsegmented perceptual information children observe when they hear a word.

The impact of sound symbolism in children's language development may be surprising, given that most words are not apparently sound symbolic. We suggested that sound symbolism is a vestige of a protolanguage that was mostly sound symbolic. Sound symbolism may have helped our ancestors to develop their lexicon and also combinatoric nature of language. Furthermore, because of the tight link between sound symbolic words and co-speech iconic gestures, we also suggested that sound symbolism and iconic gestures, both of which involve non-arbitrary cross-modal mapping, evolved together. Children today still maintain the ability to take advantage of sound symbolism in word learning. Once children break into the system of linguistic symbols and start building lexicons with the help of iconic relationship between sound and meaning, they come to realize that many words do not have apparent form–meaning resemblance. When their vocabulary becomes substantially large, they may no longer expect sound symbolism in words even though there may be covert sound symbolism all across the lexicon [34].

To summarize, sound symbolism provides key insights into how language develops in children and how language evolved in human history. It should no longer be considered to be a peripheral phenomenon in language.

Funding statement

This research was supported by MEXT KAKENHI (no. 15300088, no. 22243043, grant-in-aid for Scientific Research on Innovative Areas no. 23120003) to M.I. and BBSRC Research Development Fellowship (BB/G023069/1) to S.K.