Variation, gender and perception: the social meaning of Japanese linguistic variables

1 Introduction

One of the fundamental goals of sociolinguistic inquiry is to understand speakers’ motivations to use one linguistic variant over another. Recent work in sociolinguistics, referred to as third wave research (Eckert 2005), has, in particular, explored this question by focusing on social meaning as a force that motivates speakers to use certain linguistic variants (Agha 2003, 2007; Campbell-Kibler 2007, 2008, 2009, 2011; Johnstone and Kiesling 2008; Levon 2011; Moore 2004; Moore and Podesva 2009; Podesva 2007, 2011a, 2011b; Podesva et al. 2015; Zhang 2005, 2007, 2008). Podesva et al. (2015: 60) summarizes this development succinctly, “third wave studies shift their focus from linguistic change to the social meanings that motivate speakers to use one linguistic variant over another.” Contrary to earlier work in sociolinguistics, coined first and second wave research, which examines the relationship between linguistic variation and social, or demographic, categories on both major and local scales, respectively, third wave researchers suggest that variables are available for speakers to use as a resource to construct identities, stances and personas. This claim builds upon the work of Silverstein (1976), who argued that associated social categories are indexed by variables to form meaning which is significant to some speakers, particularly those involved in a communicative event. For example, the released variant of word-final /t/ occurs at high rates among Orthodox Jewish men (Benor 2001, 2004). Benor concluded that stop releases not only indexed learnedness but that, in the examined cultural context, learnedness indirectly indexed masculinity. Therefore, in order to sound like a learned man, the third wave expectation is that Orthodox Jewish boys would release their word-final /t/s. In the present study, we sought to further this line of inquiry into social meaning by examining individuals’ attitudes towards Japanese linguistic variables that have been previously shown to vary with speaker sex in production. Ultimately, we argue on the basis of two experimental results that the correlation of a linguistic form with a social category is a necessary but not sufficient condition for social indexation.

The ability of speakers to use socially indexed meaning to construct identities, stances and personas rests on two foundations. The first is that social meaning must be indexed by the variable, such that the choice of the form hints at the social category of the speaker in addition to any other semantic contributions (e.g., context-free truth-conditional meaning). The second is that individuals are aware of the indexed social meaning. In order for the variable to be used for the purpose of identity, personae or stance construction, within a community, the social meaning of the variable must be shared knowledge. If individuals are not aware of the indexed meaning, they could still produce the form as a result of imitative social conditioning, but the intended social information would be unstable. For example, quotative like in English has been shown to correlate with the social categories of gender, age and socioeconomic status (Dailey-O’Cain 2000). If these categories have been indexed onto the variable in addition to its quotative meaning, it would imply that individuals are aware of this additional meaning, either as unconscious or conscious knowledge. Should the knowledge be overt and conscious to the individual, it would be possible for speakers to be able to use like to construct specific identities in relation to the indexed social meaning. Such sociolinguistic performance is not possible if individuals are unaware of the indexed meaning. Buchstaller (2006) shows the social categories of age and gender amongst British listeners to be identifiable from quotative like use. This finding suggests that British individuals used like to construct identities in relation to age and gender but not in relation to socioeconomic status which was not found to be indexed onto the variable, at least in an overt manner.

The first wave studies (Labov 1966a; Trudgill 1974; Wolfram 1969) that examined the systematicity of socially conditioned variation across major demographic categories, and the second wave studies (Eckert 2000; Milroy 1980; Rickford 1986) that focused on the relationship between variation and local, participant-designed categories, have shown consistently and reliably that variables correlate with social categories in production. It is this finding from production-based studies that has been used to support the third wave line of inquiry investigating the possibility that variables can index social meaning. Specifically, correlations in practice are suggested to reflect the recruitment of variables for the communication of social meaning. The logic of this claim rests on the process of indexicalization (Eckert 2008; Silverstein 1976, 2003), where meaning is indexed through the correlation between a signifier and the signified in space and time. Forms are capable of indexing additional meanings, further to semantic meaning or first-order meanings, leading to what Eckert (2008: 454) describes as “a field of potential meanings—an indexical field, or constellation of ideologically related meanings, any one of which can be activated in the situated use of the variable.” Third wave variationists thus suggest that the social meaning(s) of linguistic variables are fluid and flexible, and their interpretation depends on the situational context.

The process of indexicalization is in line with usage-based approaches to language learning. Exemplar models are one such usage-based approach which assumes that individual speech utterances are aggregated in memory as exemplar representations that contain rich linguistic and non-linguistic information (Bybee 2001; Foulkes and Docherty 2006; Goldinger 1997, 1998; Johnson 1997, 2006; Pierrehumbert 2001, 2002). This aggregation results in a mapping of relevant social categories, pertaining to the speaker, to each exemplar (Drager 2005; Foulkes and Docherty 2006; Hay et al. 2006a, 2006b; Johnson et al. 1999). Research has shown that once an exemplar representation is stored in an individual’s memory it can be activated during both the production and perception of speech (Hay et al. 2006a, 2006b; Johnson 1997; Lozito and Mulligan 2010; Pierrehumbert 2001). According to exemplar theory, speakers are thus able to produce forms that index correlating social categories and perceive the social categories that are indexed onto the representations. Using the example above, exposure to an Orthodox Jewish learned man’s patterns of released /t/ would be expected to create a mental representation of released /t/ and its associated social categories: namely, religion, education and gender. A speaker can then use this feature as a stylistic device to create a particular social persona. This is only effective, however, if the feature and its associated categories are recognized as an index of this social persona. Consequently, should this knowledge be overtly perceivable, we would expect individuals to be able to retrieve the social information that correlates with linguistic variables within a given speech community.

Regional dialect labeling experiments have provided evidence that demonstrates individuals’ overt awareness of correlating social categories that are indexed onto variables (Baker et al. 2009; Cramer 2010; Fuchs 2015; Kirtley 2011; Purnell et al. 1999; Suárez-Budenbender 2009). Clopper and Pisoni (2004) examined Indiana college students’ ability to accurately categorize six North American regional dialects. They found that while the listeners’ general identification accuracy was low, their responses were statistically above chance. Moreover, speakers who had lived in at least three different states were more accurate than those who had only lived in Indiana. Speakers who had lived in a given region also categorized talkers from that region more accurately than speakers who had not lived there. This finding was in line with exemplar theory expectations, suggesting that listener experience is an important factor in identifying a speaker’s region based on linguistic variables.

Social evaluation studies have also shown that listeners are able to identify socially indexed meanings from linguistic variables. Campbell-Kibler (2007, 2008, 2011 examined the effects of the sociolinguistic variable (ING) (e.g., walkin’ vs. walking) on listeners’ attitudes toward speakers. Manipulating the realization of the final nasals in (ING) influenced listeners’ judgments about the speaker. Specifically, however, the results differed from previous studies that examined the social stratification of (ING). Previous studies found that, in addition to the associated social categories identified in Campbell-Kibler’s research, the social categories of gender, socioeconomic status, dialect, age and race were also shown to correlate with (ING) (Fischer 1958; Labov 1966a, 1966b; Shopen 1978; Shuy et al. 1968; Trudgill 1974). The asymmetry between patterns in production and patterns in individuals’ awareness has also been identified for other linguistic variables including t/d deletion in English (Baugh 1979; Campbell-Kibler 2006; Guy and Boyd 1990; Labov 1972a; Rickford 1999; Staum Casasanto 2010; Wolfram 1969); quotative and focuser like (Buchstaller 2006; Dailey-O’Cain 2000), fundamental frequency (Kirtley 2011; Linville 1998; Smyth et al. 2003), and /ay/ monophthongization (Kirtley 2011; Plichta and Preston 2005; Rahman 2008). This asymmetry raises a number of questions pertaining to why individuals show an awareness of the association between some linguistic variants and social categories but not all.

The context of the variable has been investigated as an explanation for individuals’ inability to overtly identify social meaning that is expected to be indexed onto a variable. As speech is inherently a social act, usually performed between a speaker and an interlocutor, both of whom are active participants in the exchange, we can expect that interpretations of meaning are contingent upon the interactants’ experiences, social positions and goals. In the case of Campbell-Kibler (2007, 2008, 2011, listeners rated a speaker who used the alveolar variant as compassionate when they perceived the speaker to be Southern, while rating the same speaker as condescending when they perceived the speaker to be from elsewhere. Social information about the speaker has also been shown to influence how listeners perceive the speech of the individual (Hay and Drager 2010; Hay et al. 2006a, 2006b; Koops et al. 2008; Niedzielski 1999; Strand 1999). In Hay and Drager (2010), New Zealand English speakers were exposed to either stuffed toys associated with Australia (kangaroos and koalas) or toys associated with New Zealand (stuffed kiwis) during a vowel perception task. Participants showed a shift in their perception boundaries according to which set of toys they were exposed to, i.e., participants matched natural vowels with more Australian-like synthesized vowels when they were in the Australian “kangaroo” condition. Thus, it is clear that listeners’ beliefs about the speaker are one factor that influences their perception of speech and indexed social meaning; however, other contextual factors, such as factors relating to the situational context, remain to be investigated in this way.

Given the history of inquiry into social meaning, the fluid and flexible nature of sociolinguistic perception presents an interesting question. There is no denying the role of speaker-context in individuals’ perceptions, both in terms of speakers and socially indexed meaning; however, if linguistic production reflects the recruitment of variables for the communication of social meaning, why then are listeners’ overt judgments so asymmetrical in nature? This is especially important when the asymmetry creates a phenomenon that violates both foundations that are required for speakers to use socially indexed meaning to construct identities, stances and personas. That is, the meaning must be indexed and subsequently be retrievable by individuals. One possible explanation lies in the ease with which a form is perceived by a listener, that is, the form’s social salience. Labov (1972b) proposed a model of social salience which delineates three variable types, demarcated by speakers’ awareness of their existence. The first level are indicators, which show zero degrees of social awareness and are therefore difficult to detect for both linguists and native speakers. Markers are usually socially stigmatized forms characterized by sharp social stratification across groups and styles. The highest level of social awareness for variables is the stereotype category. Stereotyped forms display both social and stylistic stratification and are subject to explicit meta-commentary due to their overt level of social awareness in the speech community. The salience of a variable in the speech community could therefore be crucial to the success of a listener’s awareness of the form. That is, if the variable is non-salient, at the indicator level, its associated social meaning(s) may not be learned by the listener. This claim also fits with the social category in question. Categories may share a similar scalic nature, where some are more significant to certain speech communities compared to others. It could therefore be that listener evaluations of socially indexed meaning are contingent upon the salience of the linguistic variable and the importance of the social category in question. Thus, the present paper examines individuals’ evaluations within a language community where the social category is not only suggested to be indexed onto the variables through their correlations in production but the linguistic variants and category are overtly marked in the language system and culture.

2 Gender marking in Japanese

Because linguistic variables have been shown to index a number of social meanings, some of which do and do not pattern with correlations in production, the current study takes the focused approach of examining only the social category of gender. Specifically, we investigated the category of gender and variables which pattern with the category in production. Our interests lie in the comparison between social categories found in production and those that are identifiable by individuals in perception. We, therefore, contribute to the wealth of knowledge pertaining to third wave variationist work by beginning this line of research with a single category which lays the foundation for further study of additional and potential co-present categories in future work (cf. Okamoto and Shibamoto-Smith (2016) for a detailed account of potential co-present categories and sociolinguistic categorization in Japanese).

The social category of gender has been widely studied in the domain of sociolinguistics. We would like to note the distinction here between gender, a constructed ideology that depends on perception, and sex which is a biological category. The category of gender has been claimed to be as impactful to the constructions of identity as the dimensions of region and age (Podesva and Kajino 2014). It abstracts over a range of globally and locally constructed practices (Eckert and Labov 2017). One of the earliest studies to examine the correlations between gender and speech was performed by Fischer (1958), who found that girls consistently used more of the perceived standard form of the (ING) variable [ɪŋ] than boys, a pattern that was later discussed by Labov (2001) as a preference for women to use more standard varieties than men. In addition to prestige, a number of sociolinguistic variables have been studied in connection with gender, for example, the Northern Cities Chain Shift (Eckert 1989), high-rising terminals in Australian English (Guy et al. 1986) and in New Zealand English (Britain 1992), and glottal stops in British English (Milroy et al. 1994). Gender has also been studied in other languages within a sociolinguistic framework. Some examples include phonological, morphological and lexical differences between male and female speakers of Koasati (Haas 1944), monophthongs and diphthongs in the speech of Tunis women (Trabelsi 1991), and patterns of non-palatalized [l] in Crete (Mansfield and Trudgill 1994).

Japanese, in particular, is a key language of interest, given the ideology that surrounds the social construct of gender. During the Meiji period (1868–1912), male intellectuals pushed the notion of the “ideal” woman, leading to the construction of Japanese Women’s Language (Inoue 2002, 2004, 2006; Nakamura 2008). Among others, the use of feminine self-referential forms (e.g., atakushi ‘I’), beautifying prefixes o- and go- (e.g., o-sushi ‘sushi,’ go-han ‘rice’), honorific expressions, as well as the use of new sentence-final particles to be used by women in place of traditional particles used by speakers of both genders, were advocated and propagated as the appropriate way for females to speak (Kajino 2014). These features were overt in the speech community, and would therefore be considered to be at least markers, if not stereotypes, on a social salience scale. While women’s speech is no longer constrained by official policy, metapragmatic discourses, both in real-world situations and fictional works (Mizumoto 2006; Mizumoto et al. 2008; Nakamura 2013), continue to demarcate socially desirable representations of “good” or “appropriate” feminine speech (Okamoto and Shibamoto-Smith 2016). Thus, the overt distinction between what is considered to be women’s speech and men’s speech in Japanese culture lends itself as an ideal case study for examining individuals’ awareness of the association between the category of gender and linguistic variables.

The linguistic features that have been examined as stereotypical features that correlate with the gender of the speaker frequently address the use of polite expressions. In her work on politeness and women’s language in Japanese, Ide (1982) notes the variation between men’s and women’s speech in the case of personal pronouns and honorifics. The list in (1) presents the representative forms of first-person pronouns by gender. The forms are marked with asterisks to indicate the degree of honorification (two asterisks indicate the highest degree).

Almost all forms are clearly associated with one of the two genders by appearing in only one of the lists. In these cases, the speaker’s deference towards the status of their interlocutor is expressed through the level of honorific degree as well as their self-identification as a male or female speaker. The cases of watakushi and watashi however are exceptions. Watakushi is the politest first-person pronoun for both male and female speakers. Watashi, on the other hand, is a polite form in men’s speech and also a plain form in women’s speech. That is, watashi has a distinction at the level of politeness between the two genders, unlike watakushi. The gendered distribution of Japanese pronouns has been examined both in naturalistic speech and in written text (Hagino 2007; Kojima 2013; Miyazaki 2002, 2004; Nakamura 2009; Owada 2011). In a study examining the speech of Japanese university students, Hagino (2007) found a tendency when speaking for men to use the pronoun ore (75.6 % of total pronouns) and females to use either watashi (36 %) or uchi (42 %). A distinction in gendered pronoun use can also be seen in the speech and literature of Japanese children. Nakamura (2009) examined elementary-school textbooks of the Japanese language and found that all five included units where girls were encouraged to use the female first-person pronoun watashi and boys the male first-person pronoun boku. In addition, Miyazaki (2002, 2004 found that some junior high school girls use masculine self-referential terms (e.g., boku, ore) instead of feminine forms (e.g., watashi, atashi). The varying degree of use of the variables according to the sex of the speaker suggests there are two levels of distinction for the forms: namely, deterministic, where the forms are used almost exclusively by one gender (e.g., ore and atashi), and probabilistic, whereby the forms have a higher frequency of use by one gender but are also used by the other (e.g., boku and watashi). It is important to note that we do not use the term deterministic in the sense that the relationship is fixed, rather, we use deterministic to indicate that the probability of the variable occurring with one gender or the other is very high.

Sentence-final particles have also been linked to gender in Japanese. These particles are used to express the speaker’s stance and are found most frequently in informal speech. As with pronouns, sentence-final particles correlate with the gender of the speaker as there is a higher frequency of use by one of the sexes with certain forms (Ide 1990; Ide et al. 1992; Ide and McGloin 1990; McGloin 1991; Mizumoto 2006; Mizumoto et al. 2008). Ide and Yoshida (1999) discuss some of the sentence-final particles and their use by each sex in production. They note that some particles are used almost exclusively by one sex, while others only have a higher frequency of use by male or female speakers. For example, the particle ze has a 100 % proportion of use by male speakers, whereas the particle wayo has a 100 % use by female speakers. The particle ka, on the other hand, has an 84 % proportion of use by male speakers, and the particle wa has an 89 % proportion of use by female speakers. The particle wa, and other feminine sentence-final particles, are claimed by Ide and Yoshida to have two different functions. The first is to establish empathy between the speaker and the interlocutor and the second is to soften the statement. To soften the statement, in this case, is a politeness strategy, as it weakens the imposition of the statement upon the interlocutor. The particles that are either exclusively or have a higher frequency of use by males, such as zo, ze, yo, and na, do not indicate softening and instead convey self-confidence, assertion, or confirmation. While a detailed account of Japanese pronominal and sentence-final particle use is outside of the scope of the current paper, Nakamura (2014) and Okamoto and Shibamoto-Smith (2016) provide overviews of Japanese gendered language, highlighting the use of stereotyped norms and the differences in the use of forms in naturalistic conversations and mediatized texts.

Outside the linguistic features studied under the lens of women’s language, other Japanese sociolinguistic variables have also been shown to be used disproportionately with one sex compared to the other. The reduced variant of the Japanese potential verb suffix is a well-discussed example in the literature (Ito and Mester 2004; Katada 1998; Kinsui 2003). It occurs when the potential suffix -rare is realized as -re by deletion of the syllable -ra. Thus, the phenomenon is known as ranuki ‘ra-deletion’. The long form, -rare, is the older variant which is the conservative and prescribed form of the suffix (Katada 1998; Sano 2009). The short form -re is the more recent variant of the potential verb suffix, first observed in the early 20th century, circa 1920 in the Kanto region in Japan (Kinsui 2003), and it is stigmatized as sloppy and lazy Japanese (Fumio 1998; Ito and Mester 2004). Discussions of ra-deletion in the literature have revealed that a relationship exists between the variant and speakers’ demographic categories. The distribution of ra-deletion has been shown to correlate with gender (Matsuda 1993; Miller 2004; Sano 2009, 2011), sex (Fuji et al. 2008; Matsuda 1993; Sano 2009, 2011), region, education, formality, and spontaneity (Sano 2009, 2011). In terms of sex, specifically, the corpus results presented in Sano (2009, 2011 showed a higher distribution of ra-deletion among females (females = 9.5 % of ra-deletion in the overall use of potential suffixes; males = 5.1 %). Contrarily, the self-report findings in Sherwood (2015) showed a higher frequency of ra-deletion in males (males = 44 % of ra-deletion in the overall use of potential suffixes; females = 27.4 %). Interestingly, the natural data and the self-reported data showed correlations that differed in the direction of the association between the variant and the sex of the speaker. This mismatch can be attributed to the linguistic security of the speaker, whereby speakers’ reported language use often reflects the pattern which is deemed to be socially desirable, whether the pattern be perceived as correct or incorrect by the speech community (Labov 1966b; Trudgill 1972). The tendency of males to self-report a higher usage of ra-deletion suggests that the variant is both socially salient and regarded to be a feature of vernacular speech, as standard forms are often more common in female speech (Fischer 1958; Labov 2001).

More recently, ra-deletion has also been examined within a third wave framework. Sano (2017) argued that the productive use of ra-deletion indexically signals fine-grained stylistic information. He found that the distribution of ra-deletion differs according to the relationship between speakers and the setting of the utterance. Specifically, ra-deletion is used to signal interpersonal relationships demonstrating intimacy/solidarity, and settings associated with the purpose and the atmosphere of the interaction. While these findings have significantly contributed to our understanding of ra-deletion on a stylistic level, we do not yet know whether ra-deletion indexes social meaning pertaining to the background of the speaker, such as their gender. That is, we do not know if individuals are able to judge these correlating social categories as social meaning from a speaker’s use of the variant.

Therefore, the current study examines the possible indexical association between Japanese linguistic variables and the social category of gender. In two perception experiments, we investigate pronouns, sentence-final particles and suffixes. We begin by first examining whether Japanese speaker-listeners associate the gender of a speaker with linguistic variables that have been previously shown to correlate with the social category of gender in production. This aim directly targets the second of the two foundations that presuppose individuals’ ability to use socially indexed meaning to construct identities, as outlined in the section above (Introduction). Experiment 1 explores the role of awareness in the evaluation of social meaning. In Experiment 2, we explore the role of context in individuals’ evaluations. In order to build upon the studies mentioned above (Hay and Drager 2010; Hay et al. 2006a, 2006b; Koops et al. 2008; Niedzielski 1999; Strand 1999), we examined situational context compared to context in the sense of knowledge about the speaker. This decision was made to further unpack the role of context, specifically, whether the judgments of social meaning are altered by knowledge of the speech utterance compared to knowledge of the speaker.

3 Experiment 1

This first experiment aimed to test the hypothesis that the social category of gender would be identifiable by Japanese individuals from linguistic variables that pattern with speaker sex in production. This expectation was formed on the basis that all forms in question have an overt salience in the speech community (markers or stereotypes), and the social category itself, gender, has a high significance in the speech community. We sought to examine attitudes towards variables that have been shown to correlate with the sex of the speaker in production. Specifically, the first-person singular pronouns ore, atashi, boku and watashi, and sentence-final particles ze, wayo, nda and wa. Given that the distribution of the phenomenon of ra-deletion has been previously shown to correlate with the social category of sex and gender in both corpora (Matsuda 1993; Sano 2009, 2011) and self-reports (Sherwood 2015), we also selected the potential suffix allomorphs for our case study. It is worth noting, however, that unlike pronouns and sentence-final particles which have been strongly, and in many cases prescriptively, associated with the gender of the speaker, the proportion of ra-deletion in natural speech (<10 % of the overall use of potential suffixes) is significantly lower than proportions suggested in even the probabilistic degrees of sentence-final particle usage (>84 %). We, therefore, expected that, while ra-deletion was hypothesized to have a social salience at either marker or stereotype level, the effect size would be smaller than that of the other variables in question.

3.2 Experiment 1: results

Figure 1 shows the mean ratings, with maleness represented by lower numbers and femaleness represented by higher numbers. The mean responses are presented by condition, including the subgroups of the pronoun and sentence-final particle conditions. Higher scores indicate that participants judged the sentences as more likely to have been said by a female speaker, and lower scores show that participants thought that the sentence was more likely to have been said by a male. A score of 3 would suggest that participants do not associate the respective variable with the social category of gender. The items in the deterministic subgroups for both the pronoun condition and the sentence-final particle condition were clearly identified as more likely said by a male or female speaker. This is also consistent for the probabilistic subgroups for both the pronoun condition and the sentence-final particle conditions and the lexical condition. However, the difference is smaller for the probabilistic subgroups and the lexical condition. An ordinal logistic regression analysis was conducted to determine whether the judgment scores differed significantly for the ambiguity factor (deterministic vs. probabilistic) and condition type (pronoun, particle, lexical or suffix). However, neither the ambiguity of the variable was a significant predictor in the model, coefficient estimate Exp(B) = 0.839, p = 0.664, log-likelihood test χ² (1) = 0.189, p = 0.664; nor the condition type, coefficient estimate Exp(B) = 0.867, p = 0.348, log-likelihood test χ² (1) = 0.882, p = 0.348.

Figure 1:

Mean judgment score by condition. Judgment scores ranged from 1 – Male (M) to 5 – Female (F). Error bars represent 95 % confidence intervals.

While there was no significant difference between the factors of ambiguity and condition, the differences of the pronoun, sentence-final particle and lexical conditions were far stronger than in the ranuki condition. Surprisingly, no difference was observed for mean judgments of long-form, -rare (2.63), and short-form items, -re (2.68). This was despite the significant gender effects reported in both the corpus study results and the self-report results. Specifically, the corpus results (Sano 2009, 2011) showed a higher distribution of ra-deletion among females (females = 9.5 % of ra-deletion in the overall use of potential suffixes; males = 5.1; the self-reports (Sherwood 2015) on the other hand showed a higher frequency of ra-deletion in males (males = 44 % of ra-deletion in the overall use of potential suffixes; females = 27.4 %). To understand this discrepancy between the current result and those of previous studies, the result of the current study was further investigated by examining the distribution of responses.

To examine whether the overall speech community was not sensitive to the gender effect, or if there were some individuals who interpreted ra-deletion as indicative of a female or male speaker, a gender score was created by subtracting the participant’s mean -re score from their mean -rare score. Positive gender scores indicated that the participant judged -re items as more likely said by males, whereas negative gender scores suggested the participant judged the -re items as more likely said by females.

Figure 2 shows the frequency distribution of the gender scores for each participant. The majority of gender scores clustered around the mean gender score (−0.05), indicative of a normal unimodal distribution. This suggests that the majority of participants did not associate the gender of a speaker with the potential suffix allomorph variant. There were, however, individuals who associated potential suffix allomorphs with the gender of the speaker. Four participants had a negative gender score that was less than one standard deviation below the mean (<−0.6). And one participant had a positive gender score that was greater than one standard deviation above the mean. Table 2 provides the demographic breakdown for each of these participants. There were no conclusive patterns to suggest an underlying reason that might explain why these particular individuals were sensitive to the effect of gender on ra-deletion.

Figure 2:

Distribution of participants’ gender scores for ra-deletion. Positive gender scores indicate the participant judged ra-deletion sentences as more likely said by a male speaker.

Table 2:

Qualitative analysis of participants with a gender score greater than and less than one standard deviation from the mean.

Gender score	Age	Sex	Life stage	Birthplace	Raised in	Occupation
−0.7	18–25	Male	Student	Tokyo	Tokyo
−0.7	18–25	Female	Student	Miyagi	Miyagi
−0.6	46–55	Female	Worker	Osaka	Osaka	School administration
−1.2	36–45	Female	Worker	Tokyo	Tokyo	Salaryman
0.8	46–55	Female	Worker	Kanagawa	Tokyo	Housewife

4 Experiment 2

Experiment 2 tested the hypothesis that some categories that correlate with linguistic variables require activation from a relevant situational contextual category to be perceivable by individuals. We investigated whether Japanese individuals associate the gender of a speaker with certain linguistic variables within certain situational contexts. To examine this question, we conducted a perception study that was based on Experiment 1 with methodological revisions to include contextual information about the utterances.

4.2 Experiment 2: results

Figure 3 compares the results for both Experiment 1 (no context) and Experiment 2 (context). The higher mean judgment scores indicate that participants judged the sentences as more likely said by a female speaker, while lower mean judgment scores are more likely judged by the participants as being said by a male speaker.

Figure 3:

Mean judgment score by condition with no context and context. Judgment scores ranged from 1 – Male (M) to 5 – Female (F). Error bars represent 95 % confidence intervals.

As shown in Figure 3, the overall pattern was similar across the no-context and with-context rating conditions (dark grey and light grey bars were similar across the 12 variables). The results of the pronoun, sentence-final particle and lexical conditions are consistent with the pattern observed in Experiment 1. The items in the deterministic subgroups for both the pronoun condition and the sentence-final particle condition were again identified as more likely to have been said by a male or female speaker, respectively. The pattern was also observed for the probabilistic subgroups for both the pronoun condition and the sentence-final particle conditions and the lexical condition. The difference was smaller for the probabilistic subgroups and the lexical condition but clearly demonstrated that the participants registered a difference according to the gender of the speaker. However, there appeared to be no differences between the ranuki conditions, -rare and -re.

To test these possibilities, mean judgment scores were analyzed with a 2 × (12) ANOVA, with the between-subjects factor of context group (context vs. no-context) and the within-subjects factor of variable (probabilistic Pronoun (M) vs. probabilistic Pronoun (F) vs. deterministic Pronoun (M) vs. deterministic Pronoun (F) vs. probabilistic Particle (M) vs. probabilistic Particle (F) vs. deterministic Particle (M) vs. deterministic Particle (F) vs. Suffix -rare vs. Suffix -re vs. Lexical (M) vs. Lexical (F)). The normality assumption was violated for some variables which were moderately skewed. However, ANOVA is robust to such violations (Carifio and Perla 2007). Thus, following the recommendations of Brown and Forsythe (1974), we employed Levene’s test of equality of variances that used the median, which showed, crucially, that the homogeneity of variance assumption was met, as was the homogeneity of covariance assumption (nonsignificant Box’s M). The results revealed no significant difference between the context and no-context groups, F (1, 108) = 0.094, p = 0.759, η p 2  = 0.001, as shown in Table 4. There was a significant main effect of variable, F (11, 1,188) = 1,114.56, p < 0.001, η p 2  = 0.912, and a significant Context Group × Variable interaction, F (11, 1,188) = 1.855, p = 0.041, η p 2  = 0.017. We examined the interaction via a series of orthogonal planned comparisons, employing an adjusted alpha of 0.025 (Bird 2004). Providing a workplace context did however affect judgments of gender for ra-deletion. In a work context, -re is less gendered; that is, the short form is less likely to trigger a maleness judgment when it is used in the workplace t (1,108) = 1.278, p = 0.012. None of the other comparisons reached significance, p > 0.025, and all are reported in Table 4.

Table 4:

Differences between context and no-context mean judgment scores for each of the conditions.

Variable	Type	No context rating	With context rating	t (1, 108)	p
Probabilistic pronoun (M)	Pronoun	1.20	1.14	−1.034	0.153
Probabilistic pronoun (F)	Pronoun	3.86	3.69	−1.789	0.868
Deterministic pronoun (M)	Pronoun	1.10	1.01	−0.833	0.284
Deterministic pronoun (F)	Pronoun	4.64	4.68	−0.584	0.296
Probabilistic particle (M)	Particle	2.53	2.62	1.239	0.566
Probabilistic particle (F)	Particle	4.48	4.45	−0.403	0.489
Deterministic particle (M)	Particle	1.32	1.27	−0.475	0.555
Deterministic particle (F)	Particle	4.73	4.81	1.190	0.116
Suffix -rare	Suffix	2.63	2.75	1.243	0.050
Suffix -re	Suffix	2.68	2.80	1.278	0.012
Lexical (M)	Lexical	1.86	2.00	1.801	0.181
Lexical (F)	Lexical	4.03	3.86	−1.990	0.660

5 General discussion

In two experiments, we investigated the awareness of speaker gender as conveyed by linguistic variables that have a skewed distribution across speaker sex in naturalistic speech. We found that, for variants overtly linked to gendered Japanese language, specifically, pronouns and sentence-final particles, the participants identified the gender of the speaker from these variants. The participants did not do so for the potential suffix allomorphs, even though there is a significant gender-based distributional skew in production. Additionally, situating the sentences within a specific speech context had no significant effect on individuals’ judgments.

The fact that the situational context largely did not affect participant judgments in this study does not indicate that context as a whole doesn’t play a role in the evaluation of social meaning more generally. Previous work which has investigated the role of both speaker context and situational context has shown that context does have a significant place in understanding social meaning, particularly with regard to listener knowledge regarding the speaker (Hay and Drager 2010; Hay et al. 2006a, 2006b; Koops et al. 2008; Niedzielski 1999; Pharao et al. 2014; Strand 1999). However, the role of situational context could be specific to the language group or the variables being studied. We expected that for this study the probabilistic particles and pronouns would be affected by the given context of the workplace. The presence of the first-person singular pronoun watashi as a polite form in men’s speech, and a plain form in women’s speech is a key example. We expected that knowledge of the utterance taking place in a workplace environment, specifically, a white-collar workplace environment, would suggest to individuals that the variable was used in a more formal context and would therefore be more ambiguous and less likely to be spoken by a female in the context condition. The difference in means (no context, 3.86; context, 3.69) did trend with our expectations, but it was not statistically significant in our sample size. Previous research which has examined ra-deletion and the contextual category of social status (Sherwood 2015) found that individuals were able to judge the social status of a speaker’s interlocutor by the use of potential verb suffix allomorphs alone. Specifically, individuals associate the short form, -re, with interlocutors who have a close social distance to a speaker. When a long form is heard, individuals judge that the interlocutor has a larger social distance with the speaker, such as a superior. We can conclude two points from this finding. Firstly, ra-deletion does index social status as social meaning, and secondly, politeness and formality are significant within a Japanese workplace environment. The slight maleness judgment for ra-deletion in the current study when the workplace context is provided could be an interaction with social status, but further study is needed to explore potential indexical relationships between potential co-present social categories that are tied to the speaker and those tied to the situational context, with particular emphasis on situations pertaining to Japanese workplaces.

Of the many social categories which have been investigated in the field of sociolinguistics, the socially constructed category of gender frequently yields mismatches between correlations in production and perception. There is no denying that the category is significant culturally, specifically in Japanese culture. Gender has been claimed to be as impactful to the constructions of identity as the dimensions of region and age (Podesva and Kajino 2014). This claim and the results of numerous studies across a wide variety of speech communities demarcates that there are a wide variety of linguistic resources that potentially convey speaker gender. However, it is also a category with a high number of mismatches between production and perception based studies (e.g., Baugh 1979; Campbell-Kibler 2007; Guy and Boyd 1990; Kirtley 2011; Labov 1972a, 1972b; Plichta and Preston 2005; Rickford 1999; Staum Casasanto 2010; Wolfram 1969). These mismatches across studies were some of our primary motivations for examining the social category within a speech community where the category was highly overt, which is the case in Japanese. From the results of this study, we can find clear evidence to support the notion that linguistic variables and social categories carry a certain salience with speaker listeners. Japanese pronouns and sentence-final particle variables have a significant association with the social category of gender, and this association is identifiable by individuals in isolation. Ra-deletion on the other hand patterns according to gender in the speech community, but the association with gender is not salient to community members. It seems possible then that ra-deletion, which has been found to not index gender as social meaning, may be part of a correlation that is formed as a result of imitative social conditioning, rather than for the use of identity construction. As such, it may be that other variables are being used to index the social meaning of gender, or other stylistic systems, such as clothing and non-verbal communication (Eckert 2008; Mendoza-Denton 2014), rather than the variables that show a correlation with gender in production. As such, ra-deletion may be more of a “supportive” variable rather than a “defining” variable for the purpose of identity, persona and stance construction, particularly in the case of gender.

Whatever the specific reason behind only certain variables indexing gender, the focus on examining social meaning through perception-based research methods highlights a significant issue with inferring social meaning through production study findings. The results of the current study show that variables that are stereotyped in the speech community as being attributed to male or female speech convey social meaning. Contrastively, other variables which are nonetheless skewed across gender lines in production may not convey similar social meaning. We cannot, therefore, assume the suggestion that correlations in practice reflect the recruitment of variables for the communication of social meaning. Furthermore, we cannot assume that individuals are capable of drawing upon exemplar clouds for the perception of social meaning. Rather, the beliefs of the individual may be the underlying force that drives the perception of social meaning. Individuals may weigh variables and social categories on a scale similar to Labov’s (1972b) model of social salience. A variety of studies in the area of social psychology have demonstrated that individuals draw on pre-existing beliefs and attitudes about social categories when making judgments about an interlocutor (Higgins and Bargh 1987; Levon 2014; Macrae and Bodenhausen 2001). We consequently can benefit from a combination of production- and perception-based methods to better understand social meaning and its role in constructing identities, stances and personas. The results of production studies can lead us to identifying potential social meanings, and perception studies will allow us to test if these social categories are identifiable and may therefore be recruited for the purpose of identity construction.

6 Conclusions

The results of this study show that Japanese individuals draw associations with the gender of a speaker and certain linguistic variables. The examined pronouns and sentence-final particles were shown to have a strong association with the social category of gender, suggesting both the variables and the category have a significant weight in the speech community. However, we have shown that correlations in production between gender and a specific form are not enough to indicate social meaning, nor is information about the situational context, the speech environment, of the variable sufficient to activate this supposed relationship in perception. The awareness of social meaning appears to be contingent upon the beliefs of the individuals, that is, whether the relationship between a variable and the social category is salient in the speech community. Identifying this association is not achievable by production studies alone but by a combination of both production and perception studies. In this way, we can identify possible social meanings and ascertain which are identifiable, and by extension, are available for identity construction.