Is un stylo sharper than une épée? Investigating the interaction of sound symbolism and grammatical gender in English and French speakers.

While the arbitrariness of language has long been considered one of its defining features, there is growing evidence that non-arbitrariness also plays an important role. Here we investigated two sources of non-arbitrariness: systematicity (via grammatical gender) and iconicity (via shape sound symbolism). We manipulated these two elements orthogonally, allowing us to examine the effect of each. In Experiment 1, we found that French speakers associated nonwords containing feminine (masculine) endings with round (sharp) shapes. French speakers also associated nonwords containing round-sounding (sharp-sounding) phonemes with round (sharp) shapes. This was repeated using auditory presentation with both an English-speaking (Experiment 2a) and French-speaking (Experiment 2b) sample. As predicted, the English speakers showed no effects of grammatical gender, while the French speakers did. These results demonstrate that speakers of a language with grammatical gender associate different properties to words belonging to different genders. The results also show that sound symbolism can emerge in stimuli with existing associated information (i.e., endings indicative of grammatical gender, and the association that they evoke). Finally, while previous studies have looked at effects of arbitrary and non-arbitrary mappings contained in a single stimulus, this is the first study to demonstrate that different kinds of non-arbitrary mappings can have an effect when appearing in the same stimulus. Together these results add to our understanding of the importance of non-arbitrariness in language.

Introduction

Arbitrariness, iconicity and systematicity

Does the form of a word provide cues to its meaning? Moreover, are individuals sensitive to these cues? These are fundamental questions about the nature of language that cut across both linguistics and cognitive psychology. One possibility is that form and meaning are arbitrarily related, such that form provides no reliable cue to meaning. Indeed, this has often been taken as the default. For instance, Charles Hockett suggested that the arbitrary relationship between form and meaning (i.e., one lacking any “physical or geometric resemblance”, ([1], p. 8) was one of the universal features of language. Dingemanse, Blasi, Lupyan, Christiansen and Monaghan defined arbitrariness as cases where “apart from a social convention to use word A for meaning B, there is no connection between the sound of a word and its meaning” ([2], p. 604).

More recently, arbitrariness has come to be understood as one possible type of form-meaning mapping (for a review see [2]). Indeed, there seem to be other types of form-meaning mappings that are non-arbitrary. In this paper we use the framework suggested by Dingemanse et al. [2], in which there are two (non-exclusive) types of non-arbitrary mappings: iconicity and systematicity. In iconicity, aspects of form map onto aspects of meaning via resemblance. Take for instance the word ding, whose abrupt onset and fading offset resemble aspects of its meaning (i.e., these same properties in the sound of a bell; [3]). This would be considered an instance of onomatopoeic iconicity: instances in which the sound of a word directly resembles some auditory aspect of its meaning.

It is also possible to have non-onomatopoeic iconicity, in which the sound of a word resembles non-auditory aspects of its meaning indirectly. One way in which this can happen is through the sound symbolic associations of phonemes: various perceptual or semantic features that individuals associate with particular phonemes (for reviews see [4, 5]). One example is the tendency to associate certain phonemes with visual roundness, and others with visual sharpness (i.e., the maluma/takete effect; [6]). In particular, studies indicate that individuals associate sonorant consonants (e.g., /l/, /m/ and /n/), voiced stop consonants (e.g., /b/, /d/, /g/; though to a lesser extent), and back (often rounded) vowels (e.g., /u/), with roundness. Conversely, individuals associate voiceless stop consonants (e.g., /p/, /t/ and /k/), and unrounded front vowels (e.g., /i/) with sharpness [7, 8]. To illustrate how these associations might contribute to non-onomatopoeic iconicity, consider the word balloon. It refers to a round object and contains phonemes that are sound symbolically associated with roundness.

Individuals appear to be sensitive to iconic form-meaning mappings. There is evidence that iconic words are easier for both infants [9] and adults [10] to learn. In addition, neuroimaging studies with adults have suggested that iconic words are processed differently than non-iconic words, leading to observable differences using both fMRI (e.g., [11]) and EEG (e.g., [10, 12, 13]). There is also behavioural evidence of a processing benefit for iconic words [14, 15]. Lastly, many studies with nonwords have shown that they are responded to differently when presented with iconically congruent vs. incongruent stimuli (e.g., [16-18]).

Despite this work, the extent to which non-onomatopoeic iconicity can affect processing is still somewhat unclear. Studies that have directly examined sound symbolic associations such as the maluma/takete effect in existing language have been equivocal [5, 18–20] with some authors speculating that when linguistic stimuli are associated with existing semantic information, this might diminish effects of sound symbolism (see [18, 21]). That is, when a word has meaning associated with it, the activation of that meaning during processing could interfere with sound symbolic associations evoked by its phonemes. One goal of the present study was to extend this previous work by examining the extent to which the maluma/takete effect would emerge in nonwords with associated information from word endings that are indicative of grammatical gender. In a departure from previous studies, this associated information is implicit and not situated in word meanings but rather associated with the grammatical gender categories to which the words ostensibly belong.

The other possible type of non-arbitrary form-meaning mapping is systematicity [2]. This refers to statistical regularities among groups of words belonging to the same syntactic (or even semantic) category. For instance, there are systematic differences in the forms of nouns and verbs [22-24]. Farmer et al. [22] examined the phonological forms of 3,158 nouns and verbs, and found that nouns were more similar to each other than they were to verbs (and vice versa for verbs). Moreover, individuals were sensitive to these cues, responding faster to a phonologically typical noun or verb when the sentential context led to the expectation of a noun or a verb, respectively. Systematicity could even apply to broad semantic categories. For instance, studies have demonstrated that there are differences in the forms of concrete and abstract nouns, and that participants are more accurate when judging the concreteness of phonologically typical vs. atypical, concrete and abstract nouns [26]. Beyond these examples, Dingemanse et al. noted that derivational morphology represents “another pervasive source of systematicity” ([2], p. 607). For instance, the word teacher reveals an association with the word teach.

Importantly, words do not fall wholly into the categories of arbitrary, iconic or systematic. Rather, words can contain each kind of mapping to varying degrees. In fact, studies have begun to demonstrate that individuals are sensitive to different kinds of cues appearing in the same word. For instance, Monaghan, Christiansen and Fitneva [26] found that there are advantages to having both arbitrary and systematic elements within individual words. Participants were better able to learn nonwords’ meanings when they contained both arbitrary and systematic elements, as compared to those that were wholly arbitrary or systematic. However, whether speakers can be sensitive to both systematic and iconic cues (when separate; cf. [27]), within the same stimulus, remains to be seen. The present study examined this in the context of grammatical gender.

Grammatical gender

In many languages, nouns belong to particular classes commonly referred to as grammatical genders [28]. These affect the forms of other elements of the sentence (e.g., the forms of articles or verbs might then have to agree with the gender of the noun; [29]). Despite their name, grammatical genders do not necessarily delineate based on gender per se (e.g., the Georgian language makes a distinction based on animacy). Nevertheless, many languages–such as French–do classify nouns into categories designated as feminine or masculine. In French, nouns that have a biological sex are typically assigned to a category based on that sex (e.g., a male singer is denoted by the masculine noun un chanteur, while a female singer is denoted by the feminine noun une chanteuse). However, delineations are not purely based on gender, as nouns that do not have a biological sex are still assigned to one of these two categories. This assignment does not necessarily reflect semantic features; for instance, chair is feminine (une chaise) while the semantically similar stool is masculine (un tabouret).

Speakers of a language with grammatical gender are constantly required to attend to the gender of nouns, in order to use articles and pronouns that agree with those nouns grammatically. This leads to the possibility that a given noun’s representation may be influenced by its grammatical gender (see [30]). Vigliocco, Vinson, Paganelli and Dworzynski [31] proposed that because words for females and males belong to feminine and masculine grammatical gender categories, individuals may come to associate either grammatical gender category with properties typically associated with females and males. Once this happens, nouns belonging to either grammatical gender category–even nouns without a biological sex themselves–may become associated with typically female and male traits (for other perspectives on the development of such effects, see [32-34]).

Indeed, there is evidence of individuals extending gender stereotypical features to words based on their grammatical gender. Boroditsky et al. [32] compiled a list of objects that had contrasting genders in German and Spanish. They then asked German and Spanish speakers to describe each object with the first three adjectives that came to mind. Independent raters judged these adjectives as being feminine or masculine. By including objects that belonged to different genders in German and Spanish, the authors were able to demonstrate that the same object was described by stereotypically female- or male-associated adjectives, depending on its gender in a language. For instance, bridge is feminine in German, and was described by German speakers as “beautiful”, “elegant” and “fragile”. It is, however, masculine in Spanish, and was described by Spanish speakers as “big”, “dangerous” and “strong” (see also [35]). Furthermore, studies using a semantic differential technique (i.e., ratings on scales anchored by antonyms such as pleasant/unpleasant) have found that a given noun is rated higher in potency (i.e., strong vs. weak) when it is grammatically masculine vs. feminine [36-38].

Grammatical gender categories often show signs of systematicity. That is, there are statistical regularities among groups of words in each gender category. For instance, Lyster [39] examined a corpus of 9,961 French nouns, and discovered that 81% of feminine nouns, and 80% of masculine nouns, have orthographic endings that are predictive of their grammatical gender. This is a prevalent example of systematicity in language. Studies have capitalized on this to examine whether these cues in isolation (i.e., predictive endings attached to nonword stems) can lead to the sort of effects observed by Boroditsky et al. [32]. This approach has the advantage of controlling for possible cultural effects in the use of real words. For instance, Boroditsky et al. [32] supposed that “the way objects are personified in fairy tales or poetry may depend on the grammatical genders of their names”; this could affect their representation indirectly.

Ervin [40] found that Italian speakers rated nonwords with a feminine ending (-a) as being prettier, weaker, and smaller than those with a masculine ending (-o). In a more recent study, Vuksanović, Bjekić and Radivojević presented Serbian-speaking participants with nonwords that were “indicative [of] two different grammatical genders” ([41], p. 387) in Serbian. These were ostensibly the names of instruments, and participants were asked to rate those instruments on a variety of dimensions. As expected, instruments labeled with feminine (masculine) nonwords were judged to be higher on scales typically associated with females (males). This was even observed when participants were shown an image of each instrument.

However, examining the nonwords used by Vuksanović et al. [41] reveals a potential confound. Besides differences in endings, the feminine and masculine nonwords differed systematically in their stems. Every feminine nonword stem contained a sonorant consonant, while none of them contained a voiceless stop. Conversely, only two of the eleven masculine stems contained a sonorant, while nine of them contained a voiceless stop. Given the different sound symbolic associations of these different groups of phonemes (e.g., [42, 43]), it is possible that these drove the gendered associations rather than the endings themselves. In addition, sound symbolism may have played a role in the study by Ervin [40]. While nonwords ending with -o were rated as larger than those ending with -a, recall that /o/ is sound symbolically associated with large objects [44, 45]. This presents the intriguing possibility that sound symbolism may be an as yet unmeasured contributor to grammatical gender effects.

Indeed, grammatical gender effects have proven to be susceptible to a variety of moderating factors including: age (in some studies the effect only emerges in older infants; e.g., [35, 46]); language (some studies suggest the effect to be stronger in languages with two vs. more than two grammatical genders, perhaps because the former makes gender association clearer; see [31]), animacy (in some studies the effect emerges only for animate nouns; e.g., [31]), gender (stronger effects for feminine targets; e.g., [47]), and whether the stimuli are words or pictures (in some studies the effect only emerges for words; e.g., [48]). Thus, another purpose of the present study was to test for grammatical gender effects while also accounting for potential confounds of phonology.

The present study

As noted earlier, certain phonemes are sound symbolically associated with either round or sharp shapes. Furthermore, in a study examining whether the maluma/takete effect would extend to existing first name labels, Sidhu and Pexman [49] found that female (male) names were more likely to be associated with round (sharp) shapes. This has since been replicated with a French-speaking population [50]. The results of these studies suggest that gender is also associated with round and sharp shapes. Thus, shape provides an excellent opportunity to test the coexistence of iconic cues (i.e., shape-sound symbolism) and systematic cues (i.e., endings indicative of grammatical gender), within individual linguistic stimuli. As mentioned earlier, while arbitrariness, iconicity and systematicity are believed to be able to coexist, even at the level of single words, this has yet to be fully examined for iconicity and systematicity. In the present study, we examined whether individuals are able to attend to, and be affected by, both kinds of cues in a single stimulus.

Experiment 1

Method

Ethics statement

Experiments 1 and 2b were approved by the Université de Moncton research ethics board; Experiment 2a was approved by the University of Calgary research ethics board. An informed consent form was used in all experiments.

Participants

Participants were 50 undergraduate students (40 female; M Age = 19.26, SD = 1.60) at the Université de Moncton who received course credit. Note that the age of one participant was not collected. All participants reported French fluency and normal or corrected to normal vision.

Materials and procedure

The linguistic stimuli were 24 nonwords created in the following manner. Half of the nonwords contained a round-sounding stem (e.g., mon-), and half contained a sharp-sounding stem (e.g., tip-). Round-sounding stems consisted of voiced stops (/b/), sonorant consonants (/l/, /m/ and /n/) and back and/or rounded vowels (/ɑ/, /u/, /y/, /œ/, /ɔ/ and /o/); sharp-sounding stems consisted of voiceless consonants (/p/, /t/ and /k/) and front unrounded vowels (/i/ and /e/). These types of phonemes have been shown to be associated with round and sharp shapes, respectively, in previous studies [8, 9]. Half of the nonwords with each type of stem were attached to an ending that typically occurs (≥ 95.00%; [51]) in feminine French nouns, and half to an ending that typically occurs in masculine French nouns (≥ 94.00%; [51]); see Table 1. Thus, there were six each of: round-feminine, round-masculine, sharp-feminine, and sharp-masculine nonwords. Notably, none of these endings contained round- or sharp-sounding consonants. Within each group of nonwords, two to four nonwords had an ending with a round vowel. Since this was not perfectly balanced, the vowel content of endings was included as a control variable in the analyses. Stimuli lists for all experiments are available at the following OSF repository: https://osf.io/yqvd9/.

Table 1

Grammatical gender endings used in Experiment 1.

Feminine Endings	Masculine Endings
-arde	-age
-esse	-ard
-euse	-eau
-ier	-eur
-oise	-eux
	-ier
	-is

We first ran a pilot study in order to verify that nonword gender would be apparent based on nonword endings. In this study, 20 additional French-speaking participants (19 female; M age = 18.91; SD = 1.45) assigned a gender to each of the nonwords by selecting un or une as an appropriate article. All nonwords were responded to accurately (i.e., with an article that matched the intended gender) at least 80% of the time (M = 95.83, SD = 6.63). Accuracy was not significantly different for nonwords with feminine (M = 94.44, SD = 7.55) vs. masculine (M = 97.22, SD = 5.55) endings, t(22) = 1.03, p = .32; nor for those with round-sounding (M = 97.22, SD = 5.55) vs. sharp-sounding stems (M = 94.44, SD = 7.55), t(22) = 1.03, p = .32.

The shape stimuli consisted of 48 shapes created using Adobe Photoshop. Half of these were comprised of round edges and the other half were comprised of sharp edges. The sets of round and sharp shape stimuli were matched in terms of area, height and width. These shapes are typical of those used in studies on the maluma/takete effect (e.g., [19, 52]). See Fig 1 for examples.

Fig 1

Examples of round and sharp stimuli used in each experiment.

Each trial began with a fixation cross for 1000 ms. This was then replaced by a visually presented nonword that remained onscreen for 2000 ms, which the participants were instructed to read to themselves silently. After this, participants were shown a pair of shapes: one round and one sharp, one on the left and one on the right side of the screen. Alignment of the round and sharp shapes for particular nonwords was counterbalanced across participants. Participants chose the shape that they thought best matched the nonword via keyboard press. Their response triggered a 500 ms blank screen, after which the next trial began. There was one practice trial, followed by 24 trials in the experiment proper.

Results

Analyses for all experiments consisted of mixed effects logistic regression models. We used the packages "lme4" [version 1.1-18-1] [53], "afex" [0.23–0] [54], and "RePsychLing" [0.0.4] [55] to perform our statistical analysis in R [3.5.1] [56]. We took a confirmatory approach (for a discussion see [57]) and fit models including all fixed effects of interest. We developed each model’s random effects structure using the approach suggested by Bates, Kliegl, Vasishth, and Baayen [58]. In brief, we began by fitting the model with all random slope terms for each fixed effect and removed correlations among random effects if this did not converge. We then performed a principal components analysis on the random effects and simplified the structure based on the suggested number of components. This was accomplished by iteratively removing the random slope for the highest order effect with the lowest amount of variance [55]. We also tested the inclusion of correlations among random effects, and the effects themselves, using likelihood ratio tests. The detailed procedure for model selection, along with code used for the entire process, can be found in the OSF repository. We only report the results of the model containing the final random effects structure in the text. Note that models always included random subject and item intercepts to deal with non-independence.

In the present experiment, we used this approach to examine the effects of nonword type (round-sounding [1] vs. sharp-sounding [-1]) and nonword gender (feminine [1] vs. masculine ending [-1]) on shape selection, with the likelihood of selecting the round shape as the dependent variable. We also included an interaction between these predictors. The presence of a rounded vowel in the nonword’s ending (present [1] vs. absent [-1]) was also included as a control variable. See Table 2 for the resulting model. Results indicated that, compared to the average across all factors, participants were 2.82 times more likely to select the round (sharp) shape for round-sounding (sharp-sounding) nonwords (p < .001). They were also 1.23 times more likely to select the round (sharp) shape for nonwords with a feminine (masculine) ending (p = .02). The interaction between these predictors was not significant (p = .43). See Fig 2. Note that, for this and the subsequent experiment, we also ran a version of the analyses in which endings containing the phoneme /w/ preceding a vowel were also considered round. This did not change the pattern of results.

Table 2

Logistic mixed effects regression model predicting round shape choices in Experiment 1.

Fixed Effect	B	SE	OR	Wald’s Z	p
Intercept	0.18	0.10	1.20	1.88	.06
Type	1.04	0.14	2.82	7.21	< .001***
Gender	0.21	0.09	1.23	2.29	.02*
Round Vowel in Ending	-0.05	0.10	0.95	-0.46	.64
Type x Gender	-0.07	0.09	0.93	-0.78	.44
Random Effect				s2
Subject Intercept				0.00
Subject Type Slope				0.56
Item Intercept				0.09

Notes.

* p < .05

** p < .01

*** p < .001

Fig 2

Proportions of round shapes chosen, by nonword type and gender, for each participant in Experiment 1.

Boxplots represent medians and quartiles; violin plots represent overall distribution of the data.

Discussion

We observed the maluma/takete effect, with participants associating round- (sharp-sounding) nonwords with round (sharp) shapes. It is notable that this emerged in nonwords containing informative endings (i.e., indicative of grammatical gender), in contrast to typical studies on the effect in which ending does not carry specific information. This suggests that sound symbolism can have an effect even in the context of nonwords with existing associated information. In addition, participants associated nonwords with a feminine (masculine) ending with round (sharp) shapes, suggesting an effect of grammatical gender. Importantly, this result is not explainable as being due to a confound with nonword phonology.

In the next experiment, we tested both English and French speakers, in order to isolate the effects of endings as being due to their grammatical gender per se and not to some uncontrolled factors. That is, if French grammatical gender endings have an effect because of the gender they imply, then we should not see an effect in English monolinguals. In addition, we presented stimuli auditorily, both to ensure that English speakers processed the same phonology as French speakers, and to examine the effects under a different set of conditions (e.g., after removing potential effects of orthography; see [59]).

Experiment 2a

Methods

Participants were 74 undergraduate students at the University of Calgary who received course credit. Our stopping rule was as follows: participants were tested until data were collected from 50 participants (44 female; M Age = 20.88, SD = 5.32) who did not demonstrate French proficiency, defined a priori as accuracy on a French semantic categorization task (does this word refer to a living or a non-living thing?) that did not exceed what would be expected by chance (p < .01; [58]). All participants reported English fluency and normal or corrected to normal vision.

The linguistic stimuli were identical to those used in Experiment 1, except that three nonwords with the ending -esse were replaced by nonwords with a different feminine ending. This was done because the suffix -ess is used in English to create female forms of words (e.g., actress, goddess), and we wished to avoid endings with any English gender associations. Once again, these endings did not contain round- or sharp-sounding consonants, and their vowel content was controlled for in the analyses. In addition, nonwords were presented auditorily. The audio files were created using the voice synthesizer of the Acapela group https://www.acapela-group.com/fr/. For the French pronunciation, we used the voice of Claire from France, and for the English pronunciation, we used the voice of Karen from the United States. Audio recordings were created with Sound Tap Streaming Audio Recorder version 2.31, and sound files were edited with WavePad Sound Editor version 5.96. We created two versions of each nonword: one with a French and one with an English pronunciation. Importantly, there were differences between the French and English pronunciations of vowels. For instance, the nonword boularde was pronounced /bəlɑrd/ in English, and /bulaʁdø/ in French. Note that after these changes, round-associated vowel phonemes still only occurred in the stems of round-sounding nonwords.

Each trial began with a blank screen for 500 ms. After this, participants were shown a pair of shapes: one round and one sharp, one on the left and one on the right side of the screen. Alignment of the round and sharp shapes for particular nonwords was counterbalanced across participants. Participants were simultaneously presented with the nonword auditorily via headphones. They chose the shape that they thought best matched the nonword via button press on a response box. There was one practice trial, followed by 48 trials in the experiment proper. Participants heard each of the 24 nonwords twice: once with a French pronunciation and once with an English pronunciation. These stimuli in each block were presented in a random order.

Participants then took part in a semantic categorization task with French real word targets, in order to assess their French proficiency [60]. Stimuli consisted of 100 French nouns, half of which referred to living things (e.g., agneau: lamb) and half of which referred to non-living things (e.g., acier: steel). Each trial began with a blank screen for 1000 ms. Following this, participants were visually presented with a French word to categorize as living or non-living via button press on a response box. Words were presented in a random order.

The data from 50 participants, who did not demonstrate French proficiency (where proficiency was defined as above chance performance on the French living/non-living categorization task), were analyzed using the analysis approach described previously. In addition, this analysis also included nonword pronunciation (French [1] vs. English [-1]), as well as additional two-way interactions and a three-way interaction. See Table 3 for the resulting model. Results indicated that, compared to the average across all factors, participants were 2.05 times more likely to select the round (sharp) shape for round-sounding (sharp-sounding) nonwords (p < .001). They were also 1.53 times more likely to select the round (sharp) shape for nonwords pronounced with a French (English) accent (p < .001). Notably, nonword gender did not reach significance (p = .26), nor did any of the interactions. See Fig 3.

Table 3

Logistic mixed effects regression model predicting round shape choices in Experiment 2a.

Fixed Effect	B	SE	OR	Wald’s Z	p
Intercept	0.06	0.09	1.06	0.62	.54
Type	0.72	0.10	2.05	7.52	< .001***
Gender	-0.08	0.07	0.93	-1.12	.26
Accent	0.43	0.11	1.53	3.94	< .001***
Round Vowel in Ending	0.07	0.06	1.07	1.06	.29
Type x Gender	-0.02	0.07	0.98	-0.33	.74
Type x Accent	-0.09	0.05	0.91	-1.94	.052
Gender x Accent	0.02	0.05	1.02	0.50	.62
Type x Gender x Accent	0.02	0.05	1.03	0.53	.60
Random Effect		s2
Subject Intercept		0.18
Subject Type Slope		0.22
Subject Accent Slope		0.46
Item Intercept		0.05

Notes.

* p < .05

** p < .01

*** p < .001

Fig 3

Proportions of round shapes chosen, by nonword type, gender and accent, for each participant in Experiment 2a.

Boxplots represent medians and quartiles; violin plots represent overall distribution of the data.

While monolingual English participants showed a typical maluma/takete effect, they did not show any effect of grammatical gender. This supports the interpretation that the gender results in Experiment 1 were due to the information conveyed by nonword endings.

An interesting outcome of the present experiment was that nonwords pronounced with a French accent, and realized using French phonology, were judged to go with round shapes more often than those with an English accent/phonology. This was unexpected and we can only speculate as to its source. One relevant point is that French contains more rounded vowels than English, and that nonwords realized using French phonology contained a higher percentage of rounded vowels (M = 43.75, SD = 31.40) than those realized using English phonology (M = 32.64, SD = 31.65), t(23) = 2.71, p = .01. However, the effect of accent remains after adding in the percentage of round vowels present in each nonword as a predictor. Note that this was based on an admittedly narrow transcription of the sound files. There was some uncertainty regarding the presence of the vowels /ə/ and /ø/ at the ends of some of our nonwords. Another thing to note is that the same phoneme can be articulated differently by speakers with different accents (see [61]), which could also make them more or less round- or sharp-sounding. For instance, examining the formants of the vowel /u/ in the nonword bouleuse revealed that it was pronounced farther back (i.e., with a lower second formant) in the French vs. the English pronunciations. The same was true for other nonwords that shared an /u/ in their French and English pronunciations (i.e., lumarde and mubard). Recall that vowel backness is associated with roundness.

We next ran the same experiment in a French-speaking sample.

Experiment 2b

Participants were 50 undergraduate students at the Université de Moncton who received course credit. Demographic information for these participants was not collected. All participants reported French fluency and normal or corrected to normal vision. None of them took part in Experiment 1 or the pilot study.

Materials and procedure were identical to Experiment 2a except that participants made their responses via keyboard press and did not take part in the semantic categorization task.

Results

The data were analyzed in the same manner as Experiment 2a. See Table 4 for the resulting model. Results indicated that, compared to the average across all factors, participants were 2.54 times more likely to select the round (sharp) shape for round-sounding (sharp-sounding) nonwords (p < .001). They were also 1.56 times more likely to select the round (sharp) shape for nonwords pronounced with a French (English) accent (p < .001). Additionally, there was a significant interaction between name type and gender (p = .004). We followed this up using the “emmeans” [1.2.3] package in R [62] to examine estimated marginal means, and found a significant effect of gender for sharp-sounding nonwords (p = .007) but not for round-sounding nonwords (p = .30). No other interactions reached significance. See Fig 4.

Table 4

Logistic mixed effects regression model predicting round shape choices in Experiment 2b.

Fixed Effect	B		SE	OR	Wald’s Z	p
Intercept	0.11		0.09	1.11	1.24	.22
Type	0.93		0.11	2.54	8.28	< .001***
Gender	0.09		0.08	1.09	1.08	.28
Accent	0.44		0.09	1.56	5.16	< .001***
Round Vowel in Ending	0.02		0.07	1.02	0.33	.74
Type x Gender	-0.19		0.07	0.82	-2.89	.004**
Type x Accent	0.09		0.05	1.10	1.86	.06
Gender x Accent	0.03		0.05	1.03	0.63	.53
Type x Gender x Accent	-0.03		0.05	0.97	-0.54	.59
Random Effect		s2
Subject Intercept		0.11
Subject Type Slope		0.40
Subject Gender Slope		0.09
Subject Accent Slope		0.25
Item Intercept		0.05

Notes.

* p < .05

** p < .01

*** p < .001

Fig 4

Proportions of round shapes chosen, by nonword type, gender and accent, for each participant in Experiment 2b.

Boxplots represent medians and quartiles; violin plots represent overall distribution of the data.

Discussion

We once again observed a robust maluma/takete effect, even when using nonwords with informative endings. We also observed the same effect of accent as in Experiment 2a, such that nonwords in a French (English) accent/phonology were more likely to be associated with round (sharp) shapes. It is notable that we found this in a French speaking population, suggesting that the effect of accent in Experiment 2a was not a product of the novelty of the French accent to English speakers.

We observed an effect of grammatical gender only for the sharp nonwords, such that nonwords with a feminine (masculine) ending were associated with round (sharp) shapes. One possibility is that the cue to roundness from round-sounding stems overshadowed any effects of grammatical gender. Indeed, there is some evidence of round sound symbolism being more robust than sharp sound symbolism [7, 63, 64]. This might be why the effect of grammatical gender was observed with sharp-sounding stems but not round-sounding stems.

General discussion

While arbitrariness has historically been considered a fundamental property of language [2], there are important ways in which language can be non-arbitrary. This includes iconicity and systematicity. Importantly, words need not fall wholly into one category or another. Instead, words can contain arbitrary, iconic and systematic properties [3]. Here we tested whether iconic and systematic cues occurring in the same nonword could affect its interpretation. In particular, iconic cues involved phonemes that were sound symbolically associated with either roundness or sharpness. Systematic cues were endings indicative of grammatical gender in French.

In Experiment 1, using a visual presentation, we found that French speakers associated round-sounding (sharp-sounding) nonwords, and those with feminine (masculine) endings, with round (sharp) shapes. In Experiment 2a, using auditory presentation, we found that English speakers (who had little to no knowledge of French) associated round-sounding (sharp-sounding) nonwords with round (sharp) shapes, but as predicted, were unaffected by nonword ending. Interestingly, this experiment also revealed an effect of accent, such that participants associated nonwords presented in a French (English) accent with round (sharp) shapes. Finally, in Experiment 2b, we found that French speakers also associated round-sounding (sharp-sounding) nonwords, and those presented with a French (English) accent, with round (sharp) shapes. These participants also showed an effect of grammatical gender for sharp-sounding nonwords, associating such nonwords with feminine (masculine) endings with round (sharp) shapes.

These findings demonstrate that grammatical gender affects interpretation of the meanings of linguistic stimuli. This is in line with the results of a previous study by Vuksanović et al. [41], which also found an association between feminine (masculine) nonwords and roundness (sharpness) when using Serbian endings. However, the present work goes beyond Vuksanović et al. [41] by ensuring that grammatical gender effects were not due to any sound symbolic properties of feminine or masculine endings. We did this by directly manipulating the sound symbolism of nonword stems; and avoiding sound symbolic consonants and controlling for sound symbolic vowels, in the nonword endings. That is, we isolated effects of iconicity (i.e., sound symbolic associations) and systematicity (i.e., grammatical gender). In this context, we still observed effects of grammatical gender, demonstrating that effects of grammatical gender are not explainable via sound symbolism. As suggested by Vilgiocco et al. [31], it seems that traits associated with one gender in the real world can come to be associated with that grammatical gender. Indeed, previous work has shown an association between female (male) names and roundness (sharpness; [49, 50]), although many aspects of these associations (e.g., effects of participant gender) remain unexplored.

This adds to a growing appreciation that the forms of words can impact the meaning individuals will attribute to them (see [2]). Indeed, endings that are predictive of grammatical gender represent a prevalent instance of systematicity, and thus non-arbitrariness in language. Previous work has demonstrated that systematic cues for syntactic class [22] and abstractness [25] affect language processing. This work demonstrates that cues for grammatical gender also affect interpretation.

We also observed a maluma/takete effect across all three experiments. This underscores the ubiquity of the effect, as it emerged in speakers of two languages and in nonwords read with two different accents. Moreover, this finding is notable because we believe it is the first time that the maluma/takete effect has been demonstrated using nonwords with grammatically informative endings. It has been shown that nonwords with inflectional suffixes evoke various associations of the categories they suggest (e.g., [40, 41]). Thus, these results demonstrate that the maluma/takete effect, and sound symbolism more broadly, will emerge even when examined using stimuli with existing associated information. This is notable as some have speculated that when linguistic stimuli are associated with existing semantics, effects of sound symbolism may be attenuated (for discussions see [18, 21]).

While we did observe a maluma/takete effect, if one compares the coefficients observed here to those from a similar study when stimuli did not have existing associated information, there does seem to be anecdotal evidence of attenuation. For example, in Experiment 1b, Sidhu and Pexman [65] visually presented participants with 40 trials of round- or sharp-sounding nonwords to be paired with a target shape. These stimuli did not have grammatically informative endings. When analyzing the first half of trials (i.e., the same number as used in the present experiments) in Sidhu and Pexman [65], the coefficient for the effect of shape was 2.13 (SE = 0.32, p < .001). When the data from the present Experiment 1 are analyzed in the same manner as in Sidhu and Pexman ([65]; i.e., with shape dummy coded and as the sole fixed effect, with only random intercepts included) the coefficient for the effect of shape is 1.83 (SE = 0.19, p < .001). Certainly there are other differences between these two studies (i.e., English- vs. French-speaking participants) to consider. Nevertheless, this may suggest that while sound symbolic effects can emerge in the presence of existing information, there may be some attenuation of sound symbolic effects in the face of that existing information. Further evidence of this can be found in Experiment 1a from Sidhu and Pexman [49], in which stimuli were 20 existing round- or sharp-sounding names, to be matched with round or sharp silhouettes. Analyzed in this manner, the coefficient for the effect of shape was 1.18 (SE = 0.26, p < .001), again suggesting some attenuation. Of course, future research directly comparing stimuli with and without existing information would be needed for any concrete conclusions to be drawn.

One of the goals of the present paper was to examine whether iconic (i.e., sound symbolic) and systematic (i.e., grammatical gender) cues could both have an effect on language users when they appear in a single stimulus. The results of Experiment 1 suggested that this was the case. Previous work had shown that arbitrary and systematic cues within a single stimulus could affect language processing [26]. However, these results are the first demonstration that separate iconic and systematic cues (cf. [27]) can have an effect within a single stimulus. This is an important point, as real words can contain both kinds of cues. For instance, ball is a monomorphemic word (i.e., a systematic cue for concreteness; [25]) while containing phonemes that are sound symbolically associated with roundness (i.e., an iconic cue).

Future research should examine the interplay of iconic and systematic cues. There may be some insight to glean from the fact that sound symbolism and grammatical gender interacted in Experiment 2b, with an effect of grammatical gender only appearing in sharp-sounding nonwords. It could be that the association between round-sounding phonemes and round shapes is stronger than that of sharp-sounding phonemes and sharp shapes. Indeed, there is some evidence of this in the literature (see [7, 63, 64]; cf. [66]). It may be that round shape sound symbolism overwhelmed any effects of grammatical gender, but that sharp shape sound symbolism did not. This suggests that the sources of non-arbitrariness are not always distinct but may interact in the interpretation of a linguistic stimulus. Note, however, that in these studies both cues related to the single dimension on which a decision was being made. This will not always be the case in real language processing. It is also important to point out that iconic cues came before systematic cues in our stimuli, which may have affected how participants weighted the cues.

Notably, iconicity and systematicity are not mutually exclusive. Systematicity refers to large scale statistical patterns in the forms of words belonging to a certain category. In some cases the forms that are indicative of a given category could have an iconic link with that category. For instance, this would be the case if endings predictive of feminine or masculine words contained phonemes that were sound symbolically associated with femininity or masculinity, respectively. However, we assume that even in this case, there are separate forces at work. That is, iconicity still would play a role through form-meaning resemblance, while systematicity might would play a role through regularity in the language. As another example, Imai and Kita [67] theorized that certain sound symbolic associations are language-specific, and are only apparent to individuals when they are present in the iconic words of a speaker’s language (e.g., if many nouns referring to round objects contained sonorants). This might be considered an instance of systematic iconicity.

An unexpected finding in Experiments 2a and 2b was that nonwords pronounced with a French (English) accent and with French phonology were associated with round (sharp) shapes. This may have been the cause for the marginal interactions observed between nonword type and accent, in which nonword type had a larger effect in English-accent nonwords. It may be that sharp-sounding nonwords were made somewhat round-sounding with a French accent, thus diluting the effect. As to the cause of the accent effect, one thing to note is that certain orthographic strings were realized with unrounded vowel phonemes in English but rounded vowel phonemes in French (e.g., the first vowel in bonoise was /ɑ/ in English but /o/ in French). However, the effect of accent remained after controlling for the percentage of round vowels in each nonword. Nevertheless, this highlights the interesting fact that languages contain different amounts of rounded vowels. For instance, French received pronunciation contains eight rounded vowel monophthongs (i.e., /y, ø, œ, u, o, ɔ, œ̃, and ɔ̃; [68]), compared to three in Standard American Newscaster English (i.e., /u, ʊ, and ɔ/; [69]). It would be informative for future research to explore the effects of differences such as these on the sound symbolic associations of speakers of different languages.

In addition to having different inventories of vowels, the vowels which two languages have in common can be implemented differently in each (e.g., [61]). As mentioned, the vowel /u/ was pronounced farther back (i.e., with a lower second formant) in French vs. English pronunciation. Recent work has demonstrated that sound symbolic effects are moderated by the presence or absence of phonemes in a participant’s phonological inventory [70]. The experiments reported here demonstrate that accent could be another important and overlooked moderator of sound symbolism effects. That is, even a simple nonword string like /kiki/ could be implemented differently in different accents, despite containing the same phonemes. It is also important to note that though stimuli with each accent were generated using the same website, they were unavoidably pronounced by different speakers. There may have been differences between the two that contributed to the effect.

These differences also hint at an interesting topic for the field of sound symbolism going forward with regard to cross language comparisons: examining whether these differences are acoustic or cognitive in nature (or, using linguistic terms, whether they arise from phonetics or phonology, respectively). At present there is some anecdotal evidence supporting both possibilities. Sapir [45] theorized that the reason English speakers did not rate the French phoneme /e/ as small as might be expected (according to size sound symbolism), is that they may have assimilated it to the English dipthong /eɪ/. Additionally, Styles and Gawne [70] theorized that the failure to replicate the maluma/takete effect in speakers of certain languages may have been due to experimental conditions preventing assimilation. These suggest the possibility of categorization overriding the acoustic properties of the phoneme itself. On the contrary, Fischer-Jørgensen [71] found that Danish speakers rated several pairs of allophones (i.e., different acoustic realizations of the same phoneme) differently on semantic differential scales. This is an instance of acoustic properties overriding effects of categorization. While future work would be needed to tease out the locus of the accent effect we observed, the possibility that it arises from different realizations of the same phoneme (i.e., different realizations of /u/) supports an effect of acoustic properties. Notably, a supplementary analysis combining Experiments 2a and 2b found that participant language did not interact with the effect of accent (b = -0.01, p = .92), suggesting that participants’ phonological inventories did not play a role in the effect.

While the arbitrariness of language had long been considered one of its defining features, there is mounting evidence that non-arbitrariness also plays an important role. Here we showed that both iconicity and systematicity play a role in the processing of linguistic stimuli. Individuals are sensitive to both types of cues, and consider them when making judgments about linguistic stimuli and shape.

10 Sep 2019

PONE-D-19-19956

Is un stylo sharper than une épée? Investigating the interaction of sound symbolism and grammatical gender in English and French speakers

PLOS ONE

Dear Mr. Sidhu,

Thank you for submitting your manuscript to PLOS ONE. After careful consideration, we feel that it has merit but does not fully meet PLOS ONE’s publication criteria as it currently stands. Therefore, we invite you to submit a revised version of the manuscript that addresses the points raised during the review process

We would appreciate receiving your revised manuscript by Oct 25 2019 11:59PM. When you are ready to submit your revision, log on to https://www.editorialmanager.com/pone/ and select the 'Submissions Needing Revision' folder to locate your manuscript file.

If you would like to make changes to your financial disclosure, please include your updated statement in your cover letter.

To enhance the reproducibility of your results, we recommend that if applicable you deposit your laboratory protocols in protocols.io, where a protocol can be assigned its own identifier (DOI) such that it can be cited independently in the future. For instructions see: http://journals.plos.org/plosone/s/submission-guidelines#loc-laboratory-protocols

Please include the following items when submitting your revised manuscript:

A rebuttal letter that responds to each point raised by the academic editor and reviewer(s). This letter should be uploaded as separate file and labeled 'Response to Reviewers'.

A marked-up copy of your manuscript that highlights changes made to the original version. This file should be uploaded as separate file and labeled 'Revised Manuscript with Track Changes'.

An unmarked version of your revised paper without tracked changes. This file should be uploaded as separate file and labeled 'Manuscript'.

Please note while forming your response, if your article is accepted, you may have the opportunity to make the peer review history publicly available. The record will include editor decision letters (with reviews) and your responses to reviewer comments. If eligible, we will contact you to opt in or out.

We look forward to receiving your revised manuscript.

Kind regards,

Marcus Perlman, Ph.D

Academic Editor

PLOS ONE

Journal Requirements:

When submitting your revision, we need you to address these additional requirements.

1. Please ensure that your manuscript meets PLOS ONE's style requirements, including those for file naming. The PLOS ONE style templates can be found at http://www.journals.plos.org/plosone/s/file?id=wjVg/PLOSOne_formatting_sample_main_body.pdf and http://www.journals.plos.org/plosone/s/file?id=ba62/PLOSOne_formatting_sample_title_authors_affiliations.pdf

Additional Editor Comments (if provided):

I have sent the paper to two expert reviewers, and I have also read through it carefully myself. The first reviewer generally thought that the study was well done and that the report was clearly written, but also offered some critical comments focused on the framing of the paper. They suggested that major revisions were needed. The second reviewer was highly positive about the paper and made just a few minor suggestions for points where further detail or literature could be added. In my own assessment, I agree with the reviewers that the study was well done, the analysis sound, and the report clearly written. In sum, I thought it was an interesting study that makes a clear and important contribution to the study of iconicity, specifically towards understanding the functions of iconicity and systematicity in words.

The main point of concern of Reviewer 1 relates to the treatment of iconicity and systematicity. In general, I found the way the paper treated iconicity and systematicity to be sensible (following in line with, e.g. Dingemanse et al 2015, TICS), and it seems to me that the experiment with English speakers is successful at teasing these properties apart. However, considering the reviewer's thoughtful comments on this point, I believe the paper could be improved by addressing these carefully. A related issue that I found was in the definition of ‘iconicity’ in Line 64: “One possibility is that the mapping can be iconic, with aspects of form mapping onto aspects of meaning.” Without referring to ‘resemblance’ in the mapping, I don’t think this definition actually distinguishes an iconic mapping from an arbitrary one.

A second point raised by both reviewers is that the literature review may be overstating the originality of the current work in focusing on iconicity and systematicity in words. For example, Lines 211-213: “while arbitrariness, iconicity and systematicity are believed to be able to coexist, even at the level of single words, this has not been tested for iconicity and systematicity.” Please consider the suggestions on this by the reviewers. Additionally, you might consider work on universal vs. language-specific universal sound symbolism, such as reviewed by Imai & Kita (2014, Phil Transactions of the Royal Society). I think it is worthwhile to be thorough in reviewing what work there is on this topic.

Considering this all on balance, my official decision for the manuscript is ‘minor revision’.

In your revision, I highly encourage you to carefully consider all the points made by the reviewers in addition to those I have highlighted. Please also see my minor comments below. You should not feel constrained to implement all the suggestions in your paper; but if you do not, I would expect to see a clear justification for why you did not.

Minor comments:

Line 90. “Notably, much of this work has been conducted with onomatopoeic words.” My impression is that this statement is not true for several of the works cited in the previous paragraph.

Line 146. “ontology”. Is ‘ontogeny’ the word that is meant here? It seems the paper is discussing how these effects develop.

Lines 162-165. “Lyster (39) examined a 162 corpus of 9,961 French nouns, and discovered that 81% of feminine nouns, and 80% of masculine nouns, have orthographic endings that are

predictive of their grammatical gender. A predictive ending was defined as one that occurs in nouns of a certain gender at least 90% of the time.” This is confusing me. It seems that the endings are predictive 81% or 80% of the time, not 90%?

Line 475. “We observed an effect of grammatical gender only for the sharp nonwords.” Please write out what the effect was.

A final note: The subject headings, e.g. Discussion, are organized such that there are several main Discussion sections, when I think these should be subordinated to their respective Experiments.

[Note: HTML markup is below. Please do not edit.]

Reviewers' comments:

Reviewer's Responses to Questions

Comments to the Author

1. Is the manuscript technically sound, and do the data support the conclusions?

The manuscript must describe a technically sound piece of scientific research with data that supports the conclusions. Experiments must have been conducted rigorously, with appropriate controls, replication, and sample sizes. The conclusions must be drawn appropriately based on the data presented.

Reviewer #1: Yes

Reviewer #2: Yes

**********

2. Has the statistical analysis been performed appropriately and rigorously?

Reviewer #1: Yes

Reviewer #2: Yes

**********

3. Have the authors made all data underlying the findings in their manuscript fully available?

The PLOS Data policy requires authors to make all data underlying the findings described in their manuscript fully available without restriction, with rare exception (please refer to the Data Availability Statement in the manuscript PDF file). The data should be provided as part of the manuscript or its supporting information, or deposited to a public repository. For example, in addition to summary statistics, the data points behind means, medians and variance measures should be available. If there are restrictions on publicly sharing data—e.g. participant privacy or use of data from a third party—those must be specified.

Reviewer #1: Yes

Reviewer #2: Yes

**********

4. Is the manuscript presented in an intelligible fashion and written in standard English?

PLOS ONE does not copyedit accepted manuscripts, so the language in submitted articles must be clear, correct, and unambiguous. Any typographical or grammatical errors should be corrected at revision, so please note any specific errors here.

Reviewer #1: Yes

Reviewer #2: Yes

**********

5. Review Comments to the Author

Please use the space provided to explain your answers to the questions above. You may also include additional comments for the author, including concerns about dual publication, research ethics, or publication ethics. (Please upload your review as an attachment if it exceeds 20,000 characters)

Reviewer #1: Review for “Is un stylo sharper than une épée? Investigating the interaction of sound symbolism and grammatical gender in English and French speakers”

General comments:

I think that this is an interesting paper which could be accepted for publication, provided especially that some more content is added to the discussion part. It is overall clearly written and well-articulated, with a sound experimental design and a good statistical approach. Data are available as well as the R code of the statistical analysis, which allows reproducibility.

Beyond minor points here and there throughout the manuscript, I have a few main concerns and suggestions:

- The authors consider iconicity and systematicity as two possibilities for non-arbitrary form-meaning mapping, which I find a bit problematic, since the first concept indeed points to the nature of the relationship between a form and a meaning, but the second one points rather to the fact that something occurs with a high frequency within a specific grammatical category. It seems to me that either non-arbitrary or arbitrary mappings could be systematic. I would rather keep the two notions on separate levels, which does not impact the core hypothesis of the paper

- More or less along the same line, the notion of linguistic category is a bit too vague to me, and this impacts the very notion of systematicity. Does linguistic means here grammatical, or it is more general?

- Both in the introduction and in the conclusion, the question is implicitly raised whether sound symbolism involves the phonological level and/or the phonetic level. This could perhaps be investigated a bit more explicitly.

- It could be recalled that the notion of gender system rests on syntactic evidence, namely agreement between the nouns and other elements of the sentence (Greville G. Corbett. 2013. Number of Genders. In: Dryer, Matthew S. & Haspelmath, Martin (eds.) The World Atlas of Language Structures Online. Leipzig: Max Planck Institute for Evolutionary Anthropology. (Available online at http://wals.info/chapter/30, Accessed on 2019-08-17.))

- As for the statistical analysis, some interactions are close to being significant predictors, and this deserves some attention, especially since statisticians have been criticizing the .05 threshold. Also, I wish to see more precisions regarding the contrasts and/or the use of ANOVA to assess the overall significance of interactions of categorical variables.

- I find the discussion(/conclusion) a bit lacking in depth. The first part is to me a bit too repetitive with respect to earlier sections (although it is good to summarize the findings and their context), and although there are some interesting points, I wish to get more “food for thought”. One suggestion I have, which I think could be analyzed statistically, would be to assess whether some subjects are much more sensitive to sensitivity than to iconicity, and vice-versa.

- A strong majority of participants to the experiments were female, which resonates with the fact that gender was at the center of the study. Although it is likely inconsequential, this could/should be discussed.

- Although the authors strongly insist on the novelty of combining “iconic” and “systematic” cues within the stimuli, previous authors have discussed the possibility that different iconic cues occurring in the same linguistic stimuli could “push in different directions”. Given the earlier comment on the nature of systematicity, and on its relationship with iconicity, one could argue that the approach is not radically new, and perhaps not original and wide-reaching enough for publication in a major scientific journal like PLOS ONE.

More detailed comments:

Line 40: “the effects of arbitrariness and non-arbitrariness in single stimuli”: it is understandable, but maybe it could be said a bit better. Maybe include a short discussion about languages with three genders like German or Serbian (not much is said about it in Serbian, although an experiment in this language is reported). Also, languages with many genders/nominal classes, e.g. a number of Bantu languages, could be mentioned.

Lines 55: the ‘need not’ is interesting, but could you provide a bit more details about why such an interpretation of Saussure’s dictum is acceptable (this dictum is very famous indeed, but finer details of Saussure’s ideas may not be familiar to all readers)

Lines 90-91…: maybe add something here about cross-cultural and cross-linguistic differences

Line 94: maybe provide an illustration or two of the concept of “existing semantic information” as soon as you mention it for the first time

Line 100: I am not convinced by the juxtaposition between iconicity and systematicity, since you may have systematic and non-systematic (non-onomatopoeic) iconicity… In other words, systematicity seems to exclude iconicity

Line 102: The concept of “linguistic category” would require more precision here. Do you mean “grammatical category”, something having to do with the structures of the language under study, or does this include semantic categories for example?

Line 108: following the previous comment, what is the distinction between concrete and abstract nouns: is it semantic? Some languages could perhaps grammatically encore the distinction between concrete and abstract nouns (I am not sure if anyone does), but there are definitely many languages which do not.

Line 112: maybe you can mention “derivational morphology” here

Line 113: once again, the juxtaposition of arbitrary, iconic or systematic is not really convincing to me.

Lune 117-118: this raises to me the question whether some subjects would be more sensitive to arbitrariness or systematicity. In turn, this raises the question whether some of your participants were more sensitive to iconicity than to systematicity, or vice-versa. This could be investigated with your data, but you did not do it. I would be interested in knowing more about that issue. It would also enrich your discussion (see my later comments about the latter)

Line 121: I would say that the articulation between the two sentences with however is not so good. Before, you are dealing with the opposition between arbitrary and systematic elements. Then you shift to iconic and systematic cues but also get rid of the memory requirement. That’s two differences, and therefore the transition is not so obvious to me

Line 128: that could be another place where to mention rich systems of genders/nominal classes, with all the accompanying intricacies.

Line 158: could you be more specific regarding the semantic differential technique?

Line 165: widespread example: maybe rephrase

Line 170: I only partially get it, could you be more explicit about the ‘portrayal in fiction’, with a possible example or two?

Line 199: words: either written or orally presented? Have these two options been contrasted?

Line 201: is it necessarily phonology, or could it be (also) phonetics? This makes sense especially with respect to some elements of your general discussion, when you insist on accents.

Line 215: stimuli rather than stimulus?

Line 221: Could the strong imbalance in favor of female participants have an impact in a task manipulating gender? Could you comment on that point somewhere? Would the comparison between males and females be possible, despite the small number of male participants?

Lines 232 to 234: could you provide here the endings you used?

Line 243: could the fact than 95% of the participants were female have had an influence on the classification? Maybe you could comment on this.

Lines 247 to 249: the same values are repeated. Could it be a mistake?

Line 266: it feels like a single practice trial was maybe not enough. Could you comment on that?

Line 270: maybe you could cite emmeans here with the other packages, rather than to cite it later

Line 272: please provide a reference for the “confirmatory approach”

Line 277: can you explain more precisely how you simplified the structure based on the suggested number of components?

Line 323: maybe you could mention some papers which have questioned the possible influence of written letters, notably Cuskley, Simner & Kirby 2015

Line 328: again, the participants were mostly female

Line 330: can you give some more details about the semantic categorization task here (you provide some later (line 368), but maybe better do it here for the sake of clarity)

Line 341 & 342: was the choice of two female voices in any way connected to the fact that the large majority of subjects were female?

Line 344: can you explain what kind of editing you did?

Line 345: I get what you did, but the connection between “one each” and the previous “two files” and “each nonword” is a bit confusing to me.

Line 382 – Table 2: the interaction between Type and Accent is nearly significant if one sticks to the infamous 5% threshold. I would suggest to pay attention to this, especially since you interpret the main effects of Type and Accent. Additionally, shouldn’t main effects be rather analyzed once non-significant interactions have been dropped from the model? And more, did you try refitting the model without the non-significant triple interaction, to check whether you would not then get a significant Type x Accent interaction?

Line 408: “that that”

Line 419: with respect to former comments, do you at least know whether the majority of participants were once again female? If the participants were more male this time, do you think one should pay attention to it?

Line 459 – Table 3: once again, the Type x Accent interaction is not far from significance – I have thus the same comments as for Table 2.

Line 530: To assess this idea of attenuation, you would have needed a third experiment with nonwords without endings suggestive of gender, then measures of effect size (OR is a measure of effect size for contingency tables or logistic regression if I am right, but the values are not that easy to interpret, compared to a measure between 0 and 1 like Cramer’s V – which we don’t readily have for complex logistic regressions with random effects) for your predictors to compare the experiments and observe a possible attenuation. Maybe that’s a perspective you could mention – in addition to the weak argument that “the results we observed speak against this notion, at least in its extreme form”.

Line 550: there have been arguments about the salience of angles and thus of sharp shapes, which seem to run counter to the hypothesis you mention. You may check De Carolis et al, 2018, in PLOS ONE, but given that I am one of the authors, this feels like I am trying to get citations for the paper – that’s not the case.

Line 579: with respect to an earlier comment, this point to the influence of phonetics in sound symbolism, and raises the question of the respective weights of phonology and phonetics in that domain. This could be an interesting elaboration.

Tables 1, 2, 3…: from what I get, you considered sum contrasts rather than treatment contrasts for your model summaries. If yes, maybe mention it explicitly. Also, I am a bit confused about the reported interactions: since you provide beta and OR, this must mean that you assess the difference between one condition (let’s say round sounding & masculine for the Type x Gender interaction, which would then mean that the base levels were sharp sounding and feminine) and the mean. This therefore does not tell you about the overall significance of the interaction, which is something a type-III ANOVA would tell you. Therefore, wouldn’t it be meaningful to (rather) report the output of Type-III ANOVAs for your models in order to better assess the significance of the interactions? (but maybe I am missing something because of the contrasts)

Reviewer #2: This very well written article uses a number of experiments to explore the interaction of systematicity (gender marking) and iconicity in determining the types of non-artbirary associations that both french and english speakers make between non-words and images varying in their curviness (the takete-maluma effect)

The article is well written, structured clearly, and interesting. It's nice that the authors don't get too bogged down in the exact same introduction that most articles about sound-symbolism/iconicity have, and instead focus on explaining the novel aspects of this work and how they relate to the aspects of language not typically discussed in this context

Overall I have almost no specific comments about the methodology or findings - they are well situated in other research and clear, and the authors do not overreach in their interpretation of those data or the claims that they make based on their findings. The authors do however state that this is the first work exploring systematicity and iconicity at the same time, which is probably not quite true. Minimally, Nielsen (2016), in his PhD thesis presents the results of an experiment where systematicity and iconicity were manipulated together. I believe Jonas Nolle from Edinburgh also has some experimental work exploring this possibility, although I am not certain if they have been published anywhere.

In the discussion, I think it would be great to hear the authors expand a bit more on whether systematicity and iconicity can be directly related to one another, rather than both being able to act as independent forces on predicting or determining word meanings.

**********

6. PLOS authors have the option to publish the peer review history of their article (what does this mean?). If published, this will include your full peer review and any attached files.

If you choose “no”, your identity will remain anonymous but your review may still be made public.

Do you want your identity to be public for this peer review? For information about this choice, including consent withdrawal, please see our Privacy Policy.

Reviewer #1: Yes: Christophe Coupé

Reviewer #2: No

[NOTE: If reviewer comments were submitted as an attachment file, they will be attached to this email and accessible via the submission site. Please log into your account, locate the manuscript record, and check for the action link "View Attachments". If this link does not appear, there are no attachment files to be viewed.]

While revising your submission, please upload your figure files to the Preflight Analysis and Conversion Engine (PACE) digital diagnostic tool, https://pacev2.apexcovantage.com/. PACE helps ensure that figures meet PLOS requirements. To use PACE, you must first register as a user. Registration is free. Then, login and navigate to the UPLOAD tab, where you will find detailed instructions on how to use the tool. If you encounter any issues or have any questions when using PACE, please email us at figures@plos.org. Please note that Supporting Information files do not need this step.

28 Oct 2019

Dear Dr. Perlman:

Thank you for your decision letter of September 10th, 2019. We were very grateful for the helpful suggestions from you and the reviewers contained therein. We have carefully considered these comments and made every effort to address them in this revised version of our manuscript “Is un stylo sharper than une épée? Investigating the interaction of sound symbolism and grammatical gender in English and French speakers.”. I am resubmitting this manuscript on behalf of my co-authors, for your consideration for publication in PLOS ONE.

Below we outline the steps we have taken in this regard.

Sincerely,

David Sidhu

Editor:

Major comments:

The main point of concern of Reviewer 1 relates to the treatment of iconicity and systematicity. In general, I found the way the paper treated iconicity and systematicity to be sensible (following in line with, e.g. Dingemanse et al 2015, TICS), and it seems to me that the experiment with English speakers is successful at teasing these properties apart. However, considering the reviewer's thoughtful comments on this point, I believe the paper could be improved by addressing these carefully.

To try and address any possible misunderstanding, we now explicitly state that we are working within the paradigm set out in Dingemanse et al. (2015). We also note in the General Discussion that our methods were designed to isolate iconicity and systematicity.

A related issue that I found was in the definition of ‘iconicity’ in Line 64: “One possibility is that the mapping can be iconic, with aspects of form mapping onto aspects of meaning.” Without referring to ‘resemblance’ in the mapping, I don’t think this definition actually distinguishes an iconic mapping from an arbitrary one.

Thank you for pointing this out, we have now added “via resemblance” to the definition.

A second point raised by both reviewers is that the literature review may be overstating the originality of the current work in focusing on iconicity and systematicity in words. For example, Lines 211-213: “while arbitrariness, iconicity and systematicity are believed to be able to coexist, even at the level of single words, this has not been tested for iconicity and systematicity.” Please consider the suggestions on this by the reviewers. Additionally, you might consider work on universal vs. language-specific universal sound symbolism, such as reviewed by Imai & Kita (2014, Phil Transactions of the Royal Society). I think it is worthwhile to be thorough in reviewing what work there is on this topic.

Thank you for pointing this out. We now have a paragraph in the General Discussion talking about how these could interact, including Imai and Kita’s theory that some sound symbolic associations are only apparent to individuals if they are attested to in the iconic words of that speaker’s language. This is presented as an example of iconicity and systematicity coexisting. In addition, we have rephrased the sentence you mention, to read: “…this has yet to be fully examined”, to make a less strong claim about the originality of this work.

Minor comments:

Line 90. “Notably, much of this work has been conducted with onomatopoeic words.” My impression is that this statement is not true for several of the works cited in the previous paragraph.

Thank you for catching this. We have removed the line in question and have rephrased the section that follows slightly: “Despite this work, the extent to which non-onomatopoeic iconicity can affect processing is still somewhat unclear. Studies that have directly examined sound symbolic associations such as the maluma/takete effect in existing language have been equivocal”

Line 146. “ontology”. Is ‘ontogeny’ the word that is meant here? It seems the paper is discussing how these effects develop.

Thank you, have switched to the simpler “development”.

Lines 162-165. “Lyster (39) examined a corpus of 9,961 French nouns, and discovered that 81% of feminine nouns, and 80% of masculine nouns, have orthographic endings that are

predictive of their grammatical gender. A predictive ending was defined as one that occurs in nouns of a certain gender at least 90% of the time.” This is confusing me. It seems that the endings are predictive 81% or 80% of the time, not 90%?

We have removed the definition of what constitutes a predictive ending (i.e., an ending is predictive if 90% of the words in which it occurs are feminine (or masculine)), which should clear this up. Just for interest’s sake, to clear up the confusion:

An ending is predictive if 90% of the words in which it occurs are feminine (or masculine). 80% of feminine words have such endings.

Line 475. “We observed an effect of grammatical gender only for the sharp nonwords.” Please write out what the effect was.

We have done so.

A final note: The subject headings, e.g. Discussion, are organized such that there are several main Discussion sections, when I think these should be subordinated to their respective Experiments.

Have fixed this, thanks!

Reviewer #1:

General comments:

- The authors consider iconicity and systematicity as two possibilities for non-arbitrary form-meaning mapping, which I find a bit problematic, since the first concept indeed points to the nature of the relationship between a form and a meaning, but the second one points rather to the fact that something occurs with a high frequency within a specific grammatical category. It seems to me that either non-arbitrary or arbitrary mappings could be systematic. I would rather keep the two notions on separate levels, which does not impact the core hypothesis of the paper

We are using the framework suggested in Dingemanse et al. 2015, which is quite common (206 citations), though certainly there may be other ways to construe this. We have now made it clear that we are using this framework by saying “using the framework suggested by…”. Under this framework, systematicity is a form of non-arbitrariness, and so a systematic mapping could not be arbitrary. Systematic mappings could, however, be iconic, as these categories are not mutually exclusive. We have added discussion of this to the General Discussion.

- More or less along the same line, the notion of linguistic category is a bit too vague to me, and this impacts the very notion of systematicity. Does linguistic means here grammatical, or it is more general?

The majority of examples studied include syntactic categories (e.g., nouns vs verbs; open vs. closed class verbs) though Dingemanse et al. do note that there could even be form patterns correlating with semantic factors such as concreteness. Thus we now say “words belonging to the same syntactic (or even semantic) category”.

- Both in the introduction and in the conclusion, the question is implicitly raised whether sound symbolism involves the phonological level and/or the phonetic level. This could perhaps be investigated a bit more explicitly.

We have retained the term “phonology” in the intro, as we intend to use it as it seems to be used in psychology (i.e., referring to the auditory properties of a word, rather than its orthography or meaning). We have now added a paragraph to the General Discussion talking about the complementary effects of acoustic features and phonemic categories. Here we mention that these two concepts would be termed “phonetics” and “phonology” in linguistics.

- It could be recalled that the notion of gender system rests on syntactic evidence, namely agreement between the nouns and other elements of the sentence (Greville G. Corbett. 2013. Number of Genders. In: Dryer, Matthew S. & Haspelmath, Martin (eds.) The World Atlas of Language Structures Online. Leipzig: Max Planck Institute for Evolutionary Anthropology. (Available online at http://wals.info/chapter/30, Accessed on 2019-08-17.))

Have added this, thanks.

- As for the statistical analysis, some interactions are close to being significant predictors, and this deserves some attention, especially since statisticians have been criticizing the .05 threshold. Also, I wish to see more precisions regarding the contrasts and/or the use of ANOVA to assess the overall significance of interactions of categorical variables.

Thank you this suggestion. We are in full agreement that effects near the .05 cutoff should not be entirely disregarded. We now offer some interpretation of these effects in the General Discussion (i.e., increasing the roundness of sharp-sounding nonwords with a French accent may have attenuated effects of nonword type in that condition; thus a stronger effect of nonword type for English nonwords).

Additionally, we have looked into the question of whether an ANOVA is necessary to follow up on the interaction term in a glmer() model. We were directed to this paper which suggests it is not necessary. https://arxiv.org/abs/1807.10451

- I find the discussion(/conclusion) a bit lacking in depth. The first part is to me a bit too repetitive with respect to earlier sections (although it is good to summarize the findings and their context), and although there are some interesting points, I wish to get more “food for thought”. One suggestion I have, which I think could be analyzed statistically, would be to assess whether some subjects are much more sensitive to sensitivity than to iconicity, and vice-versa.

As mentioned, we have added some content on the effects of phonology vs. phonetics, as well as the interplay of iconicity and systematicity. We also discuss how these effects compare to similar experiments in the field when nonwords do not have grammatical endings.

Regarding the point of comparing iconicity and systematicity sensitivity, thank you for this very interesting suggestion! We think the ideal way to do this would be to compare each subject’s random slope for effects of gender and shape. These should be interpreted with caution, as the analytical approach we used suggests that inclusion of both random effects was not warranted. For this reason we do not add these results to the paper, though we include them here for the sake of interest.

Below you will find each subject’s random slope for shape and gender, for each of the experiments we ran. As you can see, nearly every participant showed a stronger effect of shape than gender. Though there were a few cases of the opposite, this seemed to be because of a smaller shape effect in a particular participant. Interestingly, the effect of gender ending seemed to behave in the opposite direction in the experiment with English speakers, supporting the position that they did not derive gender from these endings. It seems that on the whole, iconicity was a stronger cue than systematicity for nearly all participants.

(Graphs in uploaded cover letter)

- A strong majority of participants to the experiments were female, which resonates with the fact that gender was at the center of the study. Although it is likely inconsequential, this could/should be discussed.

In Sidhu & Pexman 2015 Experiment 1a, we had participants associate male and female names with round or sharp-sounding phonemes, with round or sharp silhouettes. We found an effect of both name sound and gender. Notably, in that experiment, we had a roughly even split of participant gender (23 males; 30 females) and so examined whether participant gender affected either effect. It did not. Thus there is some evidence of participant gender not affecting pairings between gendered targets and round/sharp shapes.

Unfortunately we did not have enough males in the current experiments to check whether the same would hold true here as well. Nevertheless, we have added a line in the General Discussion mentioning that future research could examine individual differences playing a role in grammatical gender associations.

- Although the authors strongly insist on the novelty of combining “iconic” and “systematic” cues within the stimuli, previous authors have discussed the possibility that different iconic cues occurring in the same linguistic stimuli could “push in different directions”. Given the earlier comment on the nature of systematicity, and on its relationship with iconicity, one could argue that the approach is not radically new, and perhaps not original and wide-reaching enough for publication in a major scientific journal like PLOS ONE.

We don’t intend to claim that our approach is radically new. However, after the clarification that we are working within a specific paradigm of arbitrariness/iconicity/systematicity, we do believe that examining the iconicity and systematicity within single stimuli is novel (at least in terms of how these concepts are defined in this paradigm). Nevertheless, we have tempered claims throughout the manuscript with regards to the novelty of this work.

More detailed comments:

Line 40: “the effects of arbitrariness and non-arbitrariness in single stimuli”: it is understandable, but maybe it could be said a bit better. Maybe include a short discussion about languages with three genders like German or Serbian (not much is said about it in Serbian, although an experiment in this language is reported). Also, languages with many genders/nominal classes, e.g. a number of Bantu languages, could be mentioned.

Thanks, we have tried to clarify with “…effects of arbitrary and non-arbitrary mappings contained in a single stimulus…”

We do mention evidence that this seems to be stronger in languages with two than three genders. We aren’t inclined to add descriptions of grammatical genders in other languages as it might disrupt the flow of the article.

Lines 55: the ‘need not’ is interesting, but could you provide a bit more details about why such an interpretation of Saussure’s dictum is acceptable (this dictum is very famous indeed, but finer details of Saussure’s ideas may not be familiar to all readers)

Rather than going into this, we have removed reference to Saussure’s arbitrariness, in part because we are sympathetic to the view expressed in Hutton (1989) that Saussure was talking about the relationship between mentalistic concepts rather than the phonological form of a word and its referent in the world. We now use Hockett’s (1963) notion of arbitrariness, which is much more straight forward.

Lines 90-91…: maybe add something here about cross-cultural and cross-linguistic differences

We have not included this here as it did not seem sufficiently relevant to the main focus. However in the General Discussion we mention differential effects of phonetics and phonology to cross-linguistic differences.

Line 94: maybe provide an illustration or two of the concept of “existing semantic information” as soon as you mention it for the first time

We have added a line to clarify we mean that when a word has meaning, retrieving this meaning could interfere with the activation of sound symbolic associations.

Line 100: I am not convinced by the juxtaposition between iconicity and systematicity, since you may have systematic and non-systematic (non-onomatopoeic) iconicity… In other words, systematicity seems to exclude iconicity

We fully agree that these are not mutually exclusive, and we now make that clear in our definition of each. In addition, we give two examples in which they might interact. Even in these cases, however, there are two independent forces at work: iconicity and systematicity. We believe it makes sense to treat them as separate forces that can combine. Here, however, we are examining them separately, because we isolate each effect (i.e., cues to grammatical gender are not iconic).

Line 102: The concept of “linguistic category” would require more precision here. Do you mean “grammatical category”, something having to do with the structures of the language under study, or does this include semantic categories for example?

We have clarified that we mean syntactic and semantic categories.

Line 108: following the previous comment, what is the distinction between concrete and abstract nouns: is it semantic? Some languages could perhaps grammatically encode the distinction between concrete and abstract nouns (I am not sure if anyone does), but there are definitely many languages which do not.

Yes this was based on concreteness and imageability ratings. We haven’t added this detail to the paper but would be happy to do so.

Line 112: maybe you can mention “derivational morphology” here

Added mention of this, thanks.

Line 113: once again, the juxtaposition of arbitrary, iconic or systematic is not really convincing to me.

See earlier comments.

Lune 117-118: this raises to me the question whether some subjects would be more sensitive to arbitrariness or systematicity. In turn, this raises the question whether some of your participants were more sensitive to iconicity than to systematicity, or vice-versa. This could be investigated with your data, but you did not do it. I would be interested in knowing more about that issue. It would also enrich your discussion (see my later comments about the latter)

See earlier comment.

Line 121: I would say that the articulation between the two sentences with however is not so good. Before, you are dealing with the opposition between arbitrary and systematic elements. Then you shift to iconic and systematic cues but also get rid of the memory requirement. That’s two differences, and therefore the transition is not so obvious to me

Have removed mention of the memory aspect.

Line 128: that could be another place where to mention rich systems of genders/nominal classes, with all the accompanying intricacies.

We attempt to address this to some extent in the section mentioning that effects of grammatical gender can vary based on specific features (e.g., the number of grammatical categories in a language).

Line 158: could you be more specific regarding the semantic differential technique?

Have added a description of this method.

Line 165: widespread example: maybe rephrase

We have changed this to “prevalent”.

Line 170: I only partially get it, could you be more explicit about the ‘portrayal in fiction’, with a possible example or two?

We have now added the specific example given by Boroditsky et al., about objects being personified differently in fairy tales or poetry based on the grammatical gender.

Line 199: words: either written or orally presented? Have these two options been contrasted?

We are aware of this in studies of sound symbolism, however not in the context of grammatical gender.

Line 201: is it necessarily phonology, or could it be (also) phonetics? This makes sense especially with respect to some elements of your general discussion, when you insist on accents.

See earlier comments.

Line 215: stimuli rather than stimulus?

We think stimuli is correct.

Line 221: Could the strong imbalance in favor of female participants have an impact in a task manipulating gender? Could you comment on that point somewhere? Would the comparison between males and females be possible, despite the small number of male participants?

See earlier comment.

Lines 232 to 234: could you provide here the endings you used?

Thank you for this suggestion, we have added the table.

Line 243: could the fact than 95% of the participants were female have had an influence on the classification? Maybe you could comment on this.

See earlier comment.

Lines 247 to 249: the same values are repeated. Could it be a mistake?

Thank you for flagging this, however it is not a mistake.

Line 266: it feels like a single practice trial was maybe not enough. Could you comment on that?

Because this is such a simple task, we think that a single practice trial is enough. The experimenter was also present during the trial in case the participant had any questions.

Line 270: maybe you could cite emmeans here with the other packages, rather than to cite it later

We considered this, but would rather keep it later so it’s clear what calculation that function is performing.

Line 272: please provide a reference for the “confirmatory approach”

We have added reference to a chapter discussing this.

Line 277: can you explain more precisely how you simplified the structure based on the suggested number of components?

We now explain that this was accomplished by removing the random slope for the highest order effect with the lowest amount of associated variance.

Line 323: maybe you could mention some papers which have questioned the possible influence of written letters, notably Cuskley, Simner & Kirby 2015

Thank you for suggestion, we have added this.

Line 328: again, the participants were mostly female

See previous comment.

Line 330: can you give some more details about the semantic categorization task here (you provide some later (line 368), but maybe better do it here for the sake of clarity)

Added mention that the task was about “does this word refer to a living or a non-living thing?”.

Line 341 & 342: was the choice of two female voices in any way connected to the fact that the large majority of subjects were female?

We selected voices based on those that were the most intelligible.

Line 344: can you explain what kind of editing you did?

Audio recordings were created with Sound Tap Streaming Audio Recorder version 2.31, and sound files were edited with WavePad Sound Editor version 5.96. Editing simply consisted of removing the silence before and after the nonword.

Line 345: I get what you did, but the connection between “one each” and the previous “two files” and “each nonword” is a bit confusing to me.

Thank you for flagging this, we have changed it to: “versions of each nonword: one with a French and one with an English pronunciation” which should clear it up.

Line 382 – Table 2: the interaction between Type and Accent is nearly significant if one sticks to the infamous 5% threshold. I would suggest to pay attention to this, especially since you interpret the main effects of Type and Accent. Additionally, shouldn’t main effects be rather analyzed once non-significant interactions have been dropped from the model? And more, did you try refitting the model without the non-significant triple interaction, to check whether you would not then get a significant Type x Accent interaction?

See earlier comment re: ANOVA.

Thank you for the suggestion of rerunning the model without the threeway interaction. The p value for the interaction is p = .0523 (after developing random effects in the way we describe elsewhere). Nevertheless, we agree it’s worth interpreting and now mention it in the section about accent, suggesting that the French accent may have made sharp words seem rounder and thus attenuated the effect. We continue with the model including all interactions. Since we have used effects coding, lower order effects can be interpreted in the presence of higher order interactions (as these effects will be at the average of all other terms).

Line 408: “that that”

Corrected.

Line 419: with respect to former comments, do you at least know whether the majority of participants were once again female? If the participants were more male this time, do you think one should pay attention to it?

While that would have been interesting, the researcher who ran this study believes they were again mostly female.

Line 459 – Table 3: once again, the Type x Accent interaction is not far from significance – I have thus the same comments as for Table 2.

This effect without the threeway interaction is now p = .071. See earlier comment about our interpretation.

Line 530: To assess this idea of attenuation, you would have needed a third experiment with nonwords without endings suggestive of gender, then measures of effect size (OR is a measure of effect size for contingency tables or logistic regression if I am right, but the values are not that easy to interpret, compared to a measure between 0 and 1 like Cramer’s V – which we don’t readily have for complex logistic regressions with random effects) for your predictors to compare the experiments and observe a possible attenuation. Maybe that’s a perspective you could mention – in addition to the weak argument that “the results we observed speak against this notion, at least in its extreme form”.

Thank you for this point. We now provide some anecdotal comparisons of the coefficients observed here to those of similar previous studies in which stimuli do or do not have other associated information. We also mention that while sound symbolism effects will emerge when stimuli have other existing information associated with them, there may be some attenuation. We mention that future research directly comparing these kinds of stimuli would be needed, however.

Line 550: there have been arguments about the salience of angles and thus of sharp shapes, which seem to run counter to the hypothesis you mention. You may check De Carolis et al, 2018, in PLOS ONE, but given that I am one of the authors, this feels like I am trying to get citations for the paper – that’s not the case.

Have added reference, thanks.

Line 579: with respect to an earlier comment, this point to the influence of phonetics in sound symbolism, and raises the question of the respective weights of phonology and phonetics in that domain. This could be an interesting elaboration.

See earlier comments.

Tables 1, 2, 3…: from what I get, you considered sum contrasts rather than treatment contrasts for your model summaries. If yes, maybe mention it explicitly. Also, I am a bit confused about the reported interactions: since you provide beta and OR, this must mean that you assess the difference between one condition (let’s say round sounding & masculine for the Type x Gender interaction, which would then mean that the base levels were sharp sounding and feminine) and the mean. This therefore does not tell you about the overall significance of the interaction, which is something a type-III ANOVA would tell you. Therefore, wouldn’t it be meaningful to (rather) report the output of Type-III ANOVAs for your models in order to better assess the significance of the interactions? (but maybe I am missing something because of the contrasts)

See earlier comment.

Reviewer #2:

The authors do however state that this is the first work exploring systematicity and iconicity at the same time, which is probably not quite true. Minimally, Nielsen (2016), in his PhD thesis presents the results of an experiment where systematicity and iconicity were manipulated together. I believe Jonas Nolle from Edinburgh also has some experimental work exploring this possibility, although I am not certain if they have been published anywhere.

We now cite Nielsen 2016 in intro, with the distinction being made that in this work the iconic and systematic cues are separate, unlike in Nielsen 2016. We have also tempered our claims about the originality of this work.

In the discussion, I think it would be great to hear the authors expand a bit more on whether systematicity and iconicity can be directly related to one another, rather than both being able to act as independent forces on predicting or determining word meanings.

Thank you for this suggestion, we have now added some discussion of this interplay.

11 Nov 2019

Is un stylo sharper than une épée? Investigating the interaction of sound symbolism and grammatical gender in English and French speakers

PONE-D-19-19956R1

Dear Dr. Sidhu,

We are pleased to inform you that your manuscript has been judged scientifically suitable for publication and will be formally accepted for publication once it complies with all outstanding technical requirements.

Within one week, you will receive an e-mail containing information on the amendments required prior to publication. When all required modifications have been addressed, you will receive a formal acceptance letter and your manuscript will proceed to our production department and be scheduled for publication.

Shortly after the formal acceptance letter is sent, an invoice for payment will follow. To ensure an efficient production and billing process, please log into Editorial Manager at https://www.editorialmanager.com/pone/, click the "Update My Information" link at the top of the page, and update your user information. If you have any billing related questions, please contact our Author Billing department directly at authorbilling@plos.org.

If your institution or institutions have a press office, please notify them about your upcoming paper to enable them to help maximize its impact. If they will be preparing press materials for this manuscript, you must inform our press team as soon as possible and no later than 48 hours after receiving the formal acceptance. Your manuscript will remain under strict press embargo until 2 pm Eastern Time on the date of publication. For more information, please contact onepress@plos.org.

With kind regards,

Marcus Perlman, Ph.D

Academic Editor

PLOS ONE

Additional Editor Comments (optional):

Thank you for your detailed revisions and thoughtful responses to the comments of the reviwers (and me). I am pleased to accept the paper for publication at PLOS ONE -- congratulations.

Reviewers' comments:

20 Nov 2019

PONE-D-19-19956R1

Is un stylo sharper than une épée? Investigating the interaction of sound symbolism and grammatical gender in English and French speakers

Dear Dr. Sidhu:

I am pleased to inform you that your manuscript has been deemed suitable for publication in PLOS ONE. Congratulations! Your manuscript is now with our production department.

If your institution or institutions have a press office, please notify them about your upcoming paper at this point, to enable them to help maximize its impact. If they will be preparing press materials for this manuscript, please inform our press team within the next 48 hours. Your manuscript will remain under strict press embargo until 2 pm Eastern Time on the date of publication. For more information please contact onepress@plos.org.

For any other questions or concerns, please email plosone@plos.org.

Thank you for submitting your work to PLOS ONE.

With kind regards,

PLOS ONE Editorial Office Staff

on behalf of

Dr. Marcus Perlman

Academic Editor

PLOS ONE