Comprehension of mimetics by adult native speakers of Japanese

Linguistics traditionally regards the relationship between a word’s sound and its meaning as arbitrary; words which systematically relate sound and meaning – ‘sound-symbolic’ or ‘mimetic’ words – have been regarded as peripheral (Imai and Kita, 2014); however, increasingly, research has found that languages such as Japanese have highly developed and grammatically integrated lexical strata devoted to mimetic words (Hamano, 1998; Tsujimura, 2001, 2005; Tsujimura and Deguchi, 2003). In Japanese, Akita (2010) has posited that among mimetics that denote internal states (‘psychomimes’), three categories can be identified based on semantic, morphosyntactic, and syntactic properties. With respect to syntax, Akita proposes that compatibility with locus noun phrases constitutes a syntactic constraint on these mimetic classes’ naturalness that can serve to discriminate the three classes. In this experiment, we sought to find empirical evidence for this claim by way of native speakers’ judgment of the naturalness of mimetics in sentences according to a five-point scale. Our results provided empirical support for Akita’s claim, indicating that her categorization might indeed be a psychological reality for native speakers of Japanese.


Introduction
1.1.Sound-symbolism in linguistics.A fundamental tenet of introductory linguistics is that the relationship between a sound and its meaning is arbitrary (see, for instance, Fromkin et al., 2010, Chapter 1, pp. 3-4).For example, all of the following words have the same basic meaning: dog (English), Hund (German), sobaka (Russian), perro (Spanish), cane (Italian), inu (Japanese), gǒu (Mandarin).Despite all meaning the same thing, 'dog', there are virtually no similarities between their sounds.Conversely, the same (or virtually the same) sequence of sounds -e.g., [gIft] -may have one meaning in one language ('gift, present' in English) and an entirely different meaning in another language ('poison' in German).Thus, linguists have long argued that the relationship between sound and meaning is arbitrary (Imai and Kita, 2014).This rule finds its exception in words called, among other things, mimetics, or sound-symbolic words.These are words in which the relationship between sound and meaning is non-arbitrary.For instance, in English, there are words such as bang, buzz, and thud; and in Japanese, there are corresponding words such as ban, bii-bii, and dosun.These words are conventionally called 'onomatopoeia', or, to use the technical term, 'phonomimes' (Hamano, 1998).
In languages like English (and indeed, virtually all Indo-European languages, according to Imai and Kita, 2014), this is essentially the extent of sound-symbolism: we have phonomimes, and occasionally some isolated, covert sound-symbolic effects.For example, Fromkin et al. (2010) identify the prevalence of the sequence [gl] in words pertaining to light, such as glow, gleam, glisten, glimmer, glare; however, there are counterexamples, such as gloom, which indeed is related to darkness not light, or glower or glutton which have nothing to do with either.Because of this marginality, modern linguistics has treated these items as 'peripheral' or 'primitive': more noise imitation than actual 'language', and therefore not deserving of linguistic treatment (Imai and Kita, 2014).
(3) Pera-pera MIM to ADV hanas-u speak-NPST 'chatter away incessantly' Sentence (3) is identical to the sentence in (1a) (with accent marking omitted), only now the adverbial function of the mimetic is explicitly marked by to (see Toratani, 2006, for more on the optionality of tomarking, which is beyond the scope of this paper).
Finally, mimetics in Japanese also have unique phonotactic properties, a fact which constitutes one of the prime arguments for their constituting a discrete stratum in the lexicon.The classic example of this is that mimetics allow singleton [p] (as opposed to geminate [pp]), but disallow clusters of a nasal followed by a voiceless consonant [NC o ].This distinguishes mimetics from the three other proposed strata of the lexicon, e.g., the native stratum, in which both singleton [p] and [NC o ] are disallowed; the Sino-Japanese stratum, in which singleton [p] is disallowed, but [NC o ] is not; and the loanword stratum, in which both singleton [p] and [NC o ] are allowed (Hamano, 1998;Tsujimura and Deguchi, 2003).
As the discussion above reveals, Japanese mimetics clearly have systematic grammatical properties; therefore, in Japanese at least, we cannot consider mimetic words to be 'peripheral' or 'non-linguistic' phenomena; they merit being given a proper linguistic treatment.Indeed, study of sound-symbolism not only is interesting per se, but also has provided insight into first language acquisition (where it has been proposed to play several powerful roles -see Imai and Kita, 2014, for more on this), and could potentially provide insight into such areas as language evolution (ibidem, and see also Frellesvig, 2010, Chapter 17, for the case of Japanese in particular) and auditory phonetics (e.g., Kanero et al., 2014, found that at least some Japanese mimetics are processed in different areas of the brain than non-sound-symbolic words).
Accordingly, this paper seeks to corroborate claims of Akita (2010), who attempted to characterize the semantics of mimetic words in Japanese on the basis of their interaction with syntax (i.e., a grammar-driven approach).This study will be reviewed in the next section.

Literature review
2.1.Background.Since the 1990s, a good deal of research has been done on Japanese mimetics, and much has been learnt about the unique role they play in the language.Hamano (1998), for instance, is a seminal work in the characterization of Japanese mimetics and describes in detail their phono-semantic, morphosyntactic, and lexical properties.
Despite such gains, however, much of the work on mimetics up to this point relies on authors' individual native speaker intuitions and sometimes corpus data to make theoretical or analytical judgments about mimetic properties.For instance, Hamano (1998) draws on corpus data to examine the distribution of phonemes within the moraic structure of mimetic words, and Hamano (1988) analyzes the effect of syntax on the iconicity of mimetic words.Work by Tsujimura (2001Tsujimura ( , 2005) ) and Kita (1997) focuses on a theoretical analysis of mimetics, focusing in particular on the semantic integratedness of mimetics in sentences.
With respect to experimental approaches, Imai and Kita (2014) is a major study that empirically examines how sound-symbolism may play a role in first language acquisition (principally in Japanese, but in other languages as well).The difference between the current study and those such as Imai and Kita (2014) is that the current study seeks to assess native speakers' judgment of the naturalness of mimetic words in sentential contexts, whereas previous studies have either examined mimetics in isolation (such as Akita, 2010, see below), or have examined them only secondarily (such as Imai and Kita, 2014, for whom mimetic words are crucial to their study but are not the object of the study).Akita (2010) empirically examines the semantics of Japanese mimetics, building off of a previous characterization.We mentioned above that in a language like English, sound-symbolism mostly presents in the form of phonomimes -true onomatopoeia whose phonemic contents mimic sounds directly.Japanese, however, has been identified as having two other categories of sound-symbolic words: phenomimes, in which the sound of the word mimics a manner of action or a texture (i.e., phenomimes are visual in the same way that phonomimes are aural); and psychomimes, in which the sound of the word mimics an internal state (Hamano, 1998;Akita, 2010).
Akita's theoretical framework is 'embodiment theory', a notion from cognitive psychology in which abstract or intangible concepts are comprehended by metaphorically or metonymically linking them to concrete or tangible ones (e.g., the English expression look downcast has the meaning of 'appearing sad' because downcast eyes are considered to be a concrete (visible) reflection of sadness).Her analysis based on this theory leads her to propose two subcategorizations of the 'psych-mimetics' shown in Figure 1, e.g., 'somatopsych-mimetics', which embody psychological states as physical sensations (e.g., waku-waku above); and 'visuopsych-mimetics', which embody psychological states as visible behaviors (e.g., kuyo-kuyo 'brooding, fretting').Combining these two types of psych-mimetics with the mimetics for bodily sensation, Akita's analysis posits three classes of psychomimes in total.Furthermore, Akita argues that these classes display distinct grammatical behavior in various areas.The most important of these areas, the one in which Akita claims all three classes behave differently from one another, is compatibility with what she calls a 'locus NP'.A locus NP is a noun phrase which serves to localize the state denoted by the psychomime.Specifically, she claims that for mimetics of bodily sensation, locus NPs are obligatory; for 'somatopsych-mimetics', locus NPs are optional; and for 'visuopsych-mimetics', locus NPs are illicit.This proposal is summarized in the following item (4), which is a slightly simplified copy of Akita's item (3) In order to simplify the names given to these classes, we will hereafter refer to mimetics of bodily sensation as Class 1 psychomimes, 'somatopsych-mimetics' as Class 2 psychomimes, and 'visuopsych-mimetics' as Class 3 psychomimes.In other words, Akita's proposal is an alternative mimetic typology to that given in Figure 1, and this alternative is presented below in Figure 2.

Akita's experiment.
In order to find empirical support for this classification, Akita presented 12 psychomimes (4 from each of her proposed classes) to 20 adult native speakers and asked them each to draw a picture to depict the meaning of each word.These pictures -240 in total -were then judged by both herself and one other, non-linguist native speaker of Japanese, each picture being assigned a binary 'yes' or 'no' as to whether or not it portrayed a 'bodily response' (which Akita exemplifies as shivering or putting the hands to the head) or a 'behavior' (such as sitting down); these answers were not mutually exclusive, meaning that a picture might receive a 'yes' for both, or even a 'no' for both (pp. 1211-1212).
Akita predicted that each class of psychomime would show a particular pattern of behavior in the results, e.g., Class 1 psychomimes, which denote bodily sensations, would elicit more 'bodily response' judgments than 'behavior' judgments; and that Class 3 psychomimes would elicit the converse pattern.The prediction for Class 2 pscyhomimes was identical to that of Class 1, since Class 2 psychomimes were posited to use bodily sensations as metaphorical/metonymical vehicles for denoting psychological states (p.1212).
Akita's results confirmed her predictions for Classes 1 and 3, but not for Class 2: in her experiment, Class 2 psychomimes actually resembled Class 3 more than Class 1, showing no preference for 'behavior' judgments and actually dispreferring 'bodily response' judgments.These results constituted empirical evidence for, at least, the distinction between Class 1 and Class 3 psychomimes (pp.1212-1214).

Akita's limitations and the present study.
While Akita was the first to provide empirical evidence for this distinction, her methodological approach has some limitations.Firstly, pictures are an open-ended means of tapping native speakers' judgments and are as such quite difficult to interpret; additionally, the difference between a 'behavior' and a 'bodily response' -on which the empirical data hinged -is not entirely clear.Secondly, mimetics, being sound-symbolic words, are inherently inclined to polysemy, and since the mimetics in Akita's experiment were presented free of context, this leaves a great deal of latitude to the participant in interpreting exactly what should be depicted, contributing to the open-endedness of the pictures.(Akita acknowledges this issue in her paper.)Thirdly, the open-ended picture drawing that Akita adopted as her experimental task did not allow for the discrimination of all three of her putative classes, since both Class 1 psychomimes and Class 2 psychomimes are semantically analyzed as denoting bodily sensations (as in Figure 2) and expected responses for those classes in picture drawing would therefore be indistinguishable.
This discussion leads us to the purpose of our current study.We aimed to corroborate Akita (2010)'s classification of Japanese psychomimes by testing the property that she claims distinguishes them clearly: their syntactic compatibility with a locus NP.That is, building off of Akita, who targeted the semantics of these classes through native speakers' interpretations of the mimetic words as revealed in pictures, we extended her approach and attempted to target these interpretations directly through the syntax.3Specifically, we created sentences which included psychomimes according to Akita's proposed typology (recall Figure 2), and then had Japanese native speakers judge the naturalness of these sentences along a five-point scale, providing us with a quantitative measure of their judgments.Thus, complementing Akita's experiment, which relied on qualitative interpretation of pictorial depictions of psychomimes presented in isolation, our experiment extended the scope of inquiry to a direct, quantitative measure of native speakers' comprehension of the naturalness of pscyhomimes in sentential context.
We did this in order to answer the following question: is there empirical evidence that Akita's proposed three-way categorization of psychomimes is a psychological reality for native speakers of Japanese?

Materials
In order to test Akita's three-way categorization of psychomimes, six representatives from each class were selected from the appendix of her paper to serve as stimulus items; we did our best to minimize the polysemy of the stimuli chosen, but with mimetics this is difficult to do perfectly.Table 1 below presents these stimuli, along with a rough translation and the locus NP Akita claims that they prefer.
Sentence pairs were then constructed around these items to give them context for grammaticality judgment; in each case, the sentences in the pair differed only by the presence or absence of a locus NP (e.g., the locus NPs given in Table 1, which were claimed by Akita to correspond to those mimetic words), which appeared immediately before the mimetic, which was verbalized with the verb suru 'do'.This resulted in 36 (6 mimetics × 3 classes × 2 sentences per mimetic) sentences in 18 pairs to serve as stimuli.
As representative examples, we present in (5-7) below the first pair of sentences within each class, with the mimetic (along with its equivalent in the translation) in bold and the locus noun underlined (we omit this latter feature in translation); in each case, the a sentence contains the locus NP, and the b sentence does not; unnatural sentences are marked by *.As shown in the sentences above, on Akita's analysis of these three classes of psychomimes, Class 1 mimetics, which require a locus NP, should be natural in the a sentence that contains one, and unnatural in the b sentence which does not; Class 2 mimetics, for which locus NPs are optional, should be natural in both a and b, with or without locus NPs; and Class 3 mimetics, for which locus NPs are illicit, should be unnatural in the a sentences that have them, and natural in the b sentences that do not.
An equal number of distractors were used which also targeted mimetics -specifically, non-psychomimic mimetics.Because the verb suru is not as useful with these mimetics (see, for instance Hamano, 1988Hamano, , 1998, for reasons why), it was decided that none of the distractors would use this verb; instead, each mimetic appeared as an adverb modifying a different verb appropriate to the context of the sentence.
In each pair of distractor sentences, an effort was made to make one sentence natural and the other unnatural; the locus of this match-mismatch was of two types: For the first 9 of the 18 pairs, which we call Type 1, the locus was the mimetic itself (i.e., the difference between the sentences in these pairs was which mimetic was used); and for the second 9 of the 18 pairs, which we call Type 2, the locus was something else in the main clause, either the subject or the verb (i.e., in these pairs, the mimetic was the same in both sentences, but a subject or verb was changed) -in one case, it was both subject and verb which were changed (see Appendix).
For example, examples (8) and ( 9) below are the first pairs in Type 1 and Type 2 each, given in the same order as above, with the mimetics and their English equivalents in bold (again, the unnatural member of the pair is marked with a *).All stimuli were checked for naturalness and consistency, and were matched for length and syntactic structure.The sentences were then divided into two lists according to a Latin Square design, in which the a and b sentences were assigned alternately to one list or the other.Within the three target classes, a sentences had locus NPs (i.e., a = [+NP]), while b sentences did not (i.e., b = [-NP]); and within the distractors, a sentences were constructed to be natural, and b sentences to be unnatural.

Methodology
4.1.Participants.Our participants were 7 adult native speakers of Japanese, 6 female and 1 male; four were native speakers of western dialects, while the other three were native speakers of eastern or northern dialects.All participants were paid $5 US for their participation.

Task.
Participants were given a response sheet with instructions written in Japanese (which were also given orally in English), along with a five-point scale from 'unnatural' (1) to 'natural' (5) (with (3) being neither) for each of the 36 sentences of a given list.They were instructed to, using only their native intuition and without thinking too deliberately, judge the naturalness of the sentences as they were presented to them one by one, and to not change their answers afterward.These sentences were presented individually, accompanied only by their order number, on PowerPoint slides with plain white backgrounds.Four participants were shown list 1, and three list 2.
After the participants had recorded their judgments for all 36 sentences, they were asked for a followup.In this phase, the experimenters took a sheet on which all of the sentences of the list were written in the order they had been presented, and, using the participants' collected response sheets, highlighted those sentences which the participants had judged as 1 or 2 -'unnatural' or 'slightly unnatural'.These followup sheets were then given to the participants, and they were asked to, only for the sentences that had been highlighted, provide a brief explanation for why they gave the unnatural judgment, either by circling the unnatural part of the sentence or by writing a short sentence (in Japanese or English) explaining what made it unnatural.

Predictions.
Our predictions were based on the hypothesis that Akita's analysis was correct, e.g., that for Class 1 psychomimes, for which locus NPs are claimed to be obligatory, a target sentences that contain locus NPs will be natural (and will receive a score towards 5), and b target sentences that do not contain locus NPs will be unnatural (and will receive a score towards 1); that for Class 2 psychomimes, for which locus NPs are claimed to be optional, both a and b sentences will be natural; and that for Class 3, psychomimes, for which locus NPs are claimed to be illicit, a sentences with locus NPs will be unnatural, and b sentences without them will be natural.These predictions are summarized in Table 2 below.

Results
We converted the raw scores of the naturalness judgment into z-scores, and calculated mean z-scores for each participant.We then conducted one-tail pairwise t-tests to compare mean z-scores between a sentences and b sentences individually for each Class.The subject-based analysis revealed a significant difference between a sentences and b sentences for Class 1 [t(6) = 2.394, p = 0.027] and for Class 3 [t(6) = −3.788,p = 0.005], but not for Class 2 [t(6) = 0.109, p = 0.458].The item-based analysis revealed a significant difference between a sentences and b sentences for Class 1 [t(5) = 3.816, p = 0.006], but not for Class 2 [t(5) = −0.358,p = 0.368] or for Class 3 [t(5) = −1.238,p = 0.136].Solely from the subject-based analysis, the predicted naturalness judgment patterns were borne out for all of the Classes; however, the item-based analysis revealed a non-significant difference for Class 3, against the prediction, which drew our attention to further focus on the across-item variance (discussed below).The results of the naturalness judgment task are summarized in Figure 3.

Discussion
6.1.General results.Before we discuss the results, it is important to note that due to the low number of participants (indeed, roughly half the number in Akita, 2010), these results should be regarded as preliminary.Nevertheless, it is equally important to note that, as we will describe below, the predicted trends are present, in two out of three cases statistically supported, and consistent.
That said, the fact that [+NP] sentences were judged as natural, that [-NP] sentences were judged as unnatural, and that the difference was statistically significant in Class 1, confirms our prediction based on Akita's analysis that Class 1 psychomimes require locus NPs.Likewise, the results for Class 2 -e.g., [+NP] and [-NP] were both natural and not statistically different -accord with our predictions based on Akita's analysis that for Class 2 psychomimes, locus NPs are optional.
The results for Class 3, however, did not accord with prediction, at least according to our item-based analysis: We predicted that in this class, [+NP] sentences would be rated as unnatural, [-NP] sentences would be rated as natural, and that the difference would be statistically significant, since Akita claims that for these psychomimes, locus NPs are illicit.Although this pattern was borne out in the subjectbased analysis, the result was not robust in the item-based analysis (in the item-based analysis, both [+NP] and [-NP] Class 3 sentences were judged as unnatural, and the difference between [+NP] and [-NP] was not statistically significant).Thus, overall, we argue that predicted patterns were borne out, but raise the question of why the trend for Class 3 was not robust.

Class 3.
With respect to the responses for Class 3, we examined whether there was variability in the responses to specific stimulus items or variability on the part of the individual participants.

Items.
Regarding the items, by examining the qualitative follow-up responses that our native speakers provided, in which they explained their assignment of unnatural ratings, we found that some items followed the predicted patterns more robustly than others, sometimes with rather wide differences.For instance, in Class 3, one mimetic, puri-puri (number 1 in Table 1) behaved perfectly as predicted: Every single participant rated it as either natural without a locus NP (if they saw the b sentence), or unnatural with one (if they saw the a sentence).But a second mimetic in that class, uzi-uzi (Table 1 number 2) did not behave at all as predicted: Not one of the three participants who saw it in an a sentence (with a locus NP, where it was predicted to be unnatural) judged it to be unnatural.
In examining our participants' follow-up responses more closely to discern any systematic patterns, we noted that two things accounted for the majority of unexpected unnatural judgments (i.e., unnatural judgments whose reason did not have to do with a locus NP): frequency and semantic naturalness.
Frequency was an evident issue mostly with the Class 3 mimetic kose-kose (Table 1 number 4): Two of our participants explicitly stated in their follow-up responses that they had never heard of this word before.We did not control for the frequency of the stimulus items in this experiment, and so it could be that some of the unnaturalness we encountered (in Class 3 in particular) was due to the words being so uncommon that they did not feel natural to our native speakers.
Interestingly, only three of our seven participants judged sentences with kose-kose as natural: Of the other four, two had never heard of the word, but the remaining two gave other reasons for finding the sentences unnatural, which invariably had to do with the mimetic itself (both of them replaced it with another mimetic expression as their suggestion for correcting the sentence).The three speakers who found the word natural were all westerners (though one western native speaker did judge it as unnatural, replacing it with another mimetic).What all of this might suggest is that kose-kose is a low-frequency word -possibly restricted to some degree to western dialects, an issue we will return to momentarily.
Semantic naturalness was a more recurrent issue that we observed in the stimuli.In several cases, our participants explained the unnaturalness of a sentence as being driven by some semantic mismatch between the mimetic and the context of the sentence.For instance, (10) below shows the sentence used for the Class 3 mimetic uzi-uzi (the locus NP is given in parentheses, with the notation indicating that, according to prediction, it would be unnatural if present; as before, we bold the mimetic and its English translation).The general idea that we intended in creating this sentence was that Midori, upon meeting the boy of whom she is extremely fond, becomes unable to interact with him.We built this sentence from the definition of uzi-uzi given by Akita, which is 'hesitating' (a definition supported by Kakehi et al., 1996Kakehi et al., , pp. 1222Kakehi et al., -1223)); however, although most of our participants were okay with this usage, two suggested that it did not match the context.One replaced the mimetic with dere-dere, another mimetic which more specifically describes someone who is besotted with someone to the point of being incapable of normal social interaction; but the other participant said that uzi-uzi described 'sulking' behavior specifically, and that it therefore did not match the context of interacting with someone one is fond of.This kind of fine-grained semantic mismatch turned out to be accountable for several of the unnatural judgments given for this and the other two classes, and it seemed -for reasons that elude us -particularly prevalent in Class 3.These two effects together -frequency and semantic naturalness -could have contributed to the overall reduction in naturalness of the Class 3 sentences to our participants.

6.2.2
Participants.We also examined potential variability on the part of the individual participants.We observed that 4 of our 7 native speakers displayed individual results with the predicted directionalitye.g., they rated [+NP] sentences for Class 3 as less natural than [-NP] sentences (the other 3 rated both as either natural or unnatural).This can be seen in Figure 4 below, which shows the individual mean raw scores for Class 3 for all 7 participants.In Figure 4, we can see that (if we count an a score of 3 'neutral' as fitting the pattern) participants J3, J4, J6, and J7 trend toward the predicted pattern.Of these four participants, all but J6 were speakers of western dialects, while of the three who did not trend toward the predicted pattern -e.g., J1, J2, and J5 -all but J5 were speakers of eastern or northern dialects.What this means is uncertain, but it is possible that the features of the mimetics themselves that we identified above -e.g., frequency and semantic naturalness (in the specific sentential context used) -could vary across dialects (as particular mimetics might be more or less common or have slightly different meanings in different dialects, as we speculated above for kose-kose), contributing to this pattern in participant judgments.Unfortunately, this claim is speculative because to our knowledge, there has been virtually no attempt that we know of in the current literature to catalog such dialectal variations in mimetic words in Japanese.

Class 2.
To conclude our discussion, there is one final aspect of Figure 3 that merits attention: Although Class 1 and Class 2 both behaved as predicted, the reader will note that the Class 2 sentences were not as natural as the Class 1-[+NP] sentences.That is, the magnitude of the natural judgment appears slightly lower for Class 2 than for Class 1-[+NP].Though we did not attempt to see if this difference was statistically significant, the presence of the numerical trend prompted us to look a bit more closely at the Class 2 data.This investigation revealed that these lower scores were due to a single item, saba-saba (number 6 in Table 1).Only one participant rated it as natural (5), and one as neutral (3), with the other five rating it as either 1 or 2. Of these five, three saw the [+NP] sentence, and two saw [-NP] sentence, and not one of the five mentioned locus NPs as the crux of their judgment: Two of them explicitly stated that the issue was that saba-saba did not describe an emotional state of feeling 'relieved' (or 'refreshed' to use Akita's word; p. 1216) at the dissolution of some source of stress, as we used it in our sentence (see Appendix below); but rather, a personality type (though neither defined what that personality was in their response).Interestingly, both of these definitions are given in Kakehi et al. (1996Kakehi et al. ( , pp. 1061Kakehi et al. ( -1062)).This issue appears to be yet another case of semantic naturalness, as in the case noted above for uzi-uzi.
Because saba-saba was receiving consistently unnatural judgments, we repeated our statistical tests on Class 2, this time excluding saba-saba.The result is shown in Figure 5 below.The magnitude issue that prompted our investigation is resolved, with Class 2 sentences -both [+NP] and [-NP] -being rated as natural with a magnitude on par with that of Class 1- [+NP]; and the difference between [+NP] and [-NP] is still not statistically significant (subject-based: t(6) = 2.394, p = 0.458; item-based: t(4) = 1.309, p = 0.13).
Kansas Working Papers in Linguistics | Vol.37 (2016) | pp.23-41 Thus, our conclusion above that the Class 2 sentences behaved as predicted still stands.

Conclusions and future directions
In summary, our predictions based on Akita's analysis of Japanese psychomimes were borne out overall; in Class 3, the predicted directionality was less robust.According to our detailed analysis of Class 3, possible contributing factors to this might include the frequency of the mimetic stimuli used, as well as their naturalness in the sentential context in which they were presented to our participants.
Most importantly, our results constitute empirical evidence that Akita's putative classification of psychomimes in Japanese may indeed be a psychological reality for native speakers, which was our immediate motivation for conducting this study.Also, recall that her picture drawing experiment did not distinguish Classes 1 and 2, which were both semantically analyzed as denoting bodily sensations (though they did end up being different in her results): Our study, however, has provided empirical evidence of this distinction, which combined with Akita's original evidence for the distinction between Classes 1 and 3, provides further solidity to the conclusion that these classes may indeed be psychological realities.
Clearly, the most immediate goal of future research will need to be to attempt to obtain robust statistical support for all three classes simultaneously, which will require reinvestigating Class 3 (assuming that such research also targets the locus NP distinction) while more tightly controlling factors such as frequency and semantic naturalness, as well as, of course, attempting to recruit larger sample sizes and use even more stimuli.It may also be of interest to examine whether dialect has any effect on results of future experimentation.
Beyond this, there are some other interesting possibilities.For one, since to our knowledge there is very little research into second language acquisition of mimetic words in Japanese (at least with English as the L1), it is worth asking whether learners of Japanese are sensitive to the syntactic constraint of locus NPs a priori (particularly with an L1 like English, which as we have noted, has only covert sound-symbolism); as well as whether, if not, they can be taught these constraints and subsequently acquire them a posteriori, which certainly has implications for not only linguistics per se, but also for Japanese pedagogy.Another (even broader) possibility is to investigate whether there are other syntactic or morphosyntactic properties which can serve to distinguish these putative classes and how these different properties interact; as well as whether similar distinctions can be made in the phono-and phenomimes.

Figure 2 :
Figure 2: Akita's proposed semantic typology of mimetics in Japanese.The original names she uses have been replaced for clarity, and have been bolded to highlight that they constitute three distinct categories of psychomimes.
-i-ru do-te-be-NPST 'Because Ken pulled an all-nighter last night, today, his head has been throbbing since morning.pulledan all-nighter last night, today, his head has been throbbing since morning.'Class 1 (locus NP = obligatory) cry-PAST 'Because Lisa accidentally stepped on its tail, the cat yowled repeatedly.' accidentally stepped on its tail, the cat grunted repeatedly.' went outside, it was pouring snow.' 4

Figure 3 :
Figure 3: Mean standardized scores by condition.Class 1 is given in red, Class 2 in blue, and Class 3 in green; the prediction for each condition is given in parentheses.The bracket and asterisk over Class 1 indicate that in that class, the difference between the a [+NP] and b [-NP] sentences was statistically significant; the lack of such marking for Classes 2 and 3 indicates that for those classes, said difference was not statistically significant, based on the item-based comparisons.
terribly fond of him, but whenever she meets him, she gets all tongue-tied.'

Figure 4 :
Figure 4: Class 3 mean raw scores by condition (a/b) for all 7 participants.Note that no bar appears for participant J1 in the a ([+NP]) condition because that participant's mean raw score for Class 3a sentences was 1.00.

Figure 5 :
Figure 5: Average standardized scores per condition (predictions excluded this time), excluding Class 2 Item 6 (saba-saba).Note that as compared with Figure 3, the magnitude of the naturalness of Class 2 sentences is now on par with that of Class 1a ([+NP]) sentences, and the difference between Class 2a and Class 2b ([-NP]) is still not statistically significant.
-i-ru do-te-be-NPST 'Because Ken pulled an all-nighter last night, today, his head has been throbbing since morning.-i-rudo-te-be-NPST 'It seems that Tom drank too much last night, and today his head is pounding from hangover.' 3. Iti-niti-zyuu one-day-throughout pasokon laptop no of mae front ni at suwat-te-i-tara, sit-te-be-when Anna-wa Anna-TOP *(me-ga) *(eye-NOM) syobo-syobo MIM si-te-ki-ta do-te-come-PAST'When she sat in front of a laptop all day, Anna's eyes started to feel bleary.' her future, Sumire's stomach tightened up painfully from excessive unease.' -ki-ta do-te-come-PAST 'It seems that the air pollution in this industrial area is severe, and so Satoshi's eyes came to feel irritated

Table 1 :
(Akita, 2010)ic stimuli used and the locus NPs that they select according to(Akita, 2010).

Table 2 :
Predictions for target sentences.The number following the prediction indicates the corresponding number in the five-point scale by which participants judged the sentences.

sick to his stomach.'
Whenever she thinks that summer break starts tomorrow, Sakura gets giddy with excitement.' Terry likes to drink, but whenever he drinks Japanese sake, he inevitably feels Whenever she shakes it, the small ceramic bell that Meg is holding {rings/*rumbles}.'Because she just ran five whole kilometers, Jessica is gasping {raggedly/*whiningly} for breath.'The mouse that appeared in Erik's shop fled, {scampering/*plodding} away to the underside of the shelf.' MIM si-te-i-ru do-te-be-NPST 'With her long-standing worries resolved, Sachiko feels a sense of relief.' 'Because he exercised too much yesterday, today K ōji is completely worn out.' Kansas Working Papers in Linguistics | Vol.37 (2016) | pp.