One of the more robust findings in sociolinguistics is that speakers shift their pronunciation across different addressees or speaker models (see, e.g. Giles 1973; Giles and Powesland 1975; Natale 1975; Bell 1984; Goldinger 1998; Pardo 2006; Delvaux and Soquet 2007). Many researchers have argued that such shifting is affected (and/or motivated) by speaker attitudes, usually emphasizing the role that positive attitudes towards the interlocutor/model have in facilitating convergence towards the interlocutor/model (Giles et al. 1973; Giles and Powesland 1975; Bell 1984; Yu et al. 2013; Weatherholtz et al. 2014). However, others have shown that negative attitudes can also result in divergence away from the model (Bourhis and Giles 1977; Bourhis et al. 1979), such that a speaker ends up being more dissimilar to the model than they were prior to hearing them. Babel (2012) has argued that convergence in speech, analogous to non-speech convergence (like walking in step), is the default product of an automatic alignment system, but social factors may inhibit such alignment.
Just as speakers can shift their production towards an interlocutor/model, it appears that listeners also perceptually adapt towards different speakers. Experimental work has established that listeners’ perception is affected by the perceived age (Hay et al. 2006b; Koops et al. 2008; Drager 2011), gender (Strand and Johnson 1996; Strand 1999; Johnson et al. 1999), ethnicity (Rubin 1992; Kang and Rubin 2009; Staum Casasanto 2010; McGowan 2015; Babel and Russell 2015), and regional dialect (Niedzielski 1999; Hay et al. 2006a) attributed to a speaker, in ways that are consistent with (real or imagined) production trends of speakers who share similar social attributes. Furthermore, listeners do not need to believe they are actually listening to a certain speaker in order to make these shifts; it appears that simply being exposed conceptually to a dialect region, through words or visuals (Hay et al. 2006a; Hay and Drager 2010; Walker 2014), can also result in perceptual adaptation towards a dialect region.
Unlike the speech accommodation literature, the role of the listener attitudes in perceptual accommodation is underexplored, though researchers have appealed to the role of affect in interpreting the behavior of listeners. Kang and Rubin (2009) show that students who see non-native speech as less socially attractive (seeing non-native speakers as unfriendly and dishonest) not only rate a teacher who they believe is Asian as being less effective than a teacher they believe is Caucasian, but they actually suffer comprehension loss. Hay and Drager (2010) observe a gender difference in whether New Zealand participants shift towards or away from Australian English when primed with stuffed kangaroo and koala toys, such that women shift towards Australian English, but men, relatively, shift more towards New Zealand English. They suggest that this gender difference may have its roots in the intense sporting rivalry between Australia and New Zealand, which men are typically more engaged with. 1 Men then, either due to strong anti-Australian feelings or invoked pro-New Zealand feelings, perceptually diverge from the primed dialect region. Walker (2014) finds that after reading word lists with either English or American themes, American expatriates, but not English expatriates, do better with English/American speakers in a transcription in noise task, and argues that the asymmetry may be due to the relative prestige and resulting attitudes surrounding the two dialects. Sumner and Kataoka (2013) find that Californian listeners have more false memories for a speaker with a New York accent compared to speakers with either a General American or Standard British English accent, and Sumner et al. (2015) argue that this may be due to the social weightings given to different types of speakers.
This paper sets out to explicitly investigate the role of affect in speech perception, by attempting to manipulate participant affect across conditions. Specifically, we try to manipulate the attitudes of New Zealanders towards Australia by exposing them to positive or negative facts about Australia, and measure whether this affects how much they perceptually adapt their vowels towards (or away from) Australian English. This study is preceded by two production studies that explicitly investigate vocalic accommodation towards Australian English in the speech of New Zealanders as a function of attitudes towards Australia(ns). Babel (2010) tested how much New Zealanders’ vowels moved towards an Australian speaker based on their attitudes towards Australia (gleaned from an Implicit Association Task), and based on whether they were told the speaker liked or did not like New Zealand (condition). She found some evidence that a New Zealand speaker’s implicit attitudes toward Australia were linked with their degree of accommodation to the Australian model, but no evidence of an effect of condition. In Drager et al. (2010), we presented an analysis of recordings of participants from the current study reading the Australian facts and wordlists that were used as primes in the current study (discussed in Section 2). The analysis revealed an interaction between condition (i.e., good versus bad facts about Australia) and whether the participant reported being a fan of international sport, which we hypothesized would influence behavior on the task because of the strong sporting rivalry between the two countries. If similar mechanisms are behind shifts in perception and shifts in production, these two studies suggest we should similarly find an effect of affect on speech perception.
2 Methodology and preliminaries
The experimental paradigm used in this study closely follows Niedzielski (1999). In that study, participants in Detroit were presented with a sentence visually and auditorally, with a key word underlined. After hearing the sentence, participants were then played a synthesized continuum of the target (underlined) vowel, and asked to pick which synthesized vowel best matched the target vowel in the sentence. Their choice was then modelled as a function of the condition they were in.
For our instantiation of this paradigm, we used identical stimuli as Hay and Drager (2010): recordings of a male New Zealander reading sentences (Appendix A), 20 of which include a target word that contained kit. 2 kit is a highly salient marker of Australian vs. New Zealand English (Bayard 2000), with a high, front variant in Australian English, and a mid, central variant in New Zealand English. To avoid participants focusing on the target vowel, 20 more sentences, 10 with the target vowel dress and 10 with the target vowel trap, were also included in the experiment. The approximate positions of these three vowels for New Zealanders and Australians are shown in Figure 1.
Sentences were presented to participants orthographically with the target words underlined. Participants were asked to listen to the recordings over headphones, paying special attention to the underlined word, and then were asked to match the pronunciation of the vowel in the underlined word to one from a 6 step synthesized continuum of the appropriate vowel. Following Niedzielski (1999), the participants were told that the purpose of the experiment was to test the adequacy of synthesized voices.
The tokens from the vowel-appropriate continuum were played after each sentence, and all auditory tokens were played only once per question. The continuum was always played in order, from 1 to 6. Each token from the continuum was preceded by a recording of a different male New Zealander saying the token number. This was done in order to help participants match the synthesized auditory token with the corresponding number on the answer sheet. The target vowels occurred sentence medially in half of the examples and sentence finally in the other (matched) half. An example sentence is given in (1), in its sentence-medial and sentence-final form. The full set of sentences is given in Appendix A.
During the colder months John and I both wanted to stay fit so we decided to join a yoga class.
During the colder months John and I both wanted to stay fit.
The continua were created by resynthesizing speech from the same speaker who read the sentences, using the program Klattworks (McMurray in prep). The kit continuum was designed to range from a high, front Australian-like kit to a lower, central NZ-like kit. Only F1 and F2 were manipulated. For dress and trap, the continua ranged from lower, more Australian-like tokens to higher, more New Zealand-like tokens. Across the continua, token 1 relates to an Australian variant and token 6 to a central NZ-like token that is more centralized than is likely to be produced by most New Zealanders (the actual vowels produced by the speaker averaged around token 4). The continuum is identical to the continuum used in Hay et al. (2006a), and Hay and Drager (2010).
In our previous studies, we included two conditions – a New Zealand condition and an Australian condition – but did not compare an individual’s primed and unprimed response on the task. In this study we gathered additional data as a control for each subject, so we could gauge the actual degree of shift from their baseline perception as a response to the experimental manipulation. Therefore, each participant completed the perception task twice, once at the beginning of the experimental session and once after the production task. The production task included reading positive facts about Australia in the good condition and negative facts about Australia in the bad condition. There was also a third control condition, in which the set of facts did not include any facts about Australia. The facts were presented to participants as a reading task and were entitled “Some interesting facts…”. In all conditions, the sheet began with the same series of sentences about zebras and orchids. In the control group there were no further sentences, but in the good and bad conditions there were an additional four facts about Australia. Amongst the good (2) or bad (3) facts (listed below), there was the same neutral fact about the population of the smallest town in Australia. It should be noted that the facts are not all equal in terms of how valenced they are, and some might conceivably engender stronger reactions than others.
In 2005, Australia was awarded the title of “The World’s Friendliest Nation” by the Anholt Nation Brands Index.
The Great Barrier Reef in Australia can be seen from outer space and has been labelled one of the seven natural wonders of the world.
The Australian government’s donation of $1 billion dollars to the Tsunami relief effort was the biggest made by any country, including those with considerably bigger populations.
As of 2005, Australia was the world’s largest emitter per capita of greenhouse gases, and has still not signed the Kyoto Protocol.
The 10 most venomous snakes in the world live in Australia, and there are around 3,000 snake bites reported annually.
Between 1900 and 1969 at least 100,000 Aborigine children were removed from their parents by the Australian government. The official report showed that many were forcibly taken and that some parents were even told that their children had died when this was not the case.
Our hypothesis (following work on speech accommodation in production outlined above) was that positive attitudes toward Australia would trigger a shift in the perception of kit toward more Australian-like variants, whereas negative attitudes toward Australia would trigger a shift in perception away from Australian-like variants. Our original hypothesis was that the good facts would lead to a more positive attitude than the bad facts.
The experiment was run in three iterations. In the first iteration, the experimenter was a non-blind New Zealander (NZ1) and additional participant information was sought: participants read a wordlist and answered questions concerning their level of experience with Australia and their reported attitudes towards Australia. The analysis of the wordlist production results has previously been presented in Drager et al. (2010).
Preliminary analysis of the NZ1 perception data revealed no significant effect on performance of answers to the attitude or experience questions. However, there was a significant difference between participant groups on the initial baseline task. We therefore conducted two further iterations of the perception experiment, with a simplified data-collection regime. Iteration two was run by a US experimenter (US), who was not blind to the conditions that participants were in but presumably had less likelihood of subconsciously adjusting her own speech in response to Australian primes. Iteration 3 was run by a fully blind New Zealander (NZ2). As exposure and attitude data were not found to predict perception patterns in iteration 1 (see Section 4.3), the questionnaires were not presented in the latter two iterations of the experiment.
For clarity the order of the tasks performed by participants in the NZ1 iteration is given below. Participants run by NZ2 and US completed a simpler subset of tasks: those involved in our key manipulation: (i), (iii) and (vi).
Listening task (baseline)
Word-list reading repeat
“Facts” Reading repeat
Listening task repeat (test)
Australian Integration Index (AII) and Australian Attitude Index (AAI) questionnaires.
Additionally, in iteration 1 (NZ1), the orthographic sentences were presented on a sheet of paper and the stimulus order was not varied across participants. In contrast, in iterations 2 and 3, token order was randomized across participants and responses were collected via the computer software DirectRT, thus minimizing interaction between the participant and the experimenter. The analysis presented in this paper considers all three sets of data together, comparing responses across conditions and experimenters. The distribution of the participants across conditions and the three experimenters is shown in Table 1. In order to address the uneven distribution of participant sex that is evident across experimenters and conditions, we conducted statistical analysis of both the full data set and a subset of the data with only responses from female participants who are more evenly distributed across iterations.
Logistic mixed-effects regression models were used to analyze the data. We use logistic rather than linear or ordinal models because we noticed that most people used only a small part of the 6-point response scale, with responses centered around the middle. To demonstrate this, Figure 2 shows the individual response patterns to the kit vowel, pooling responses to both block 1 and block 2. Inspecting the individual responses demonstrates that the most common response pattern spreads responses across step 3 and 4 of the continuum. This shows both that participants were fairly accurate on the task (since the speaker’s actual productions were closest to token 4), and that using linear regression would be an inappropriate tool for this data set.
Participants did the same task before and after being primed with positive and negative facts about Australia. We expected that the pre-manipulation patterns would be the same across conditions but that the post-manipulation results would differ across conditions. However, our preliminary analysis of the data from iteration 1 (NZ1) revealed that responses in the first, pre-manipulation block appeared to vary across conditions, even though the participants had not yet been exposed to the facts. We hypothesized that this difference might be due to the first, non-blind experimenter (NZ1) shifting her own speech and inadvertently influencing perception as a result. The difference across conditions in the initial block is what motivated further data collection with different experimenters.
The initial difference of responses to kit across conditions is shown in Figure 3, for each experimenter. These plots are based on the raw data. Comparison of the distribution of responses across the different experiments demonstrates very little difference across conditions for the US and NZ2 experimenters, with both showing a largely bimodal distribution. In contrast, there is a distinct difference between conditions for NZ1, with opposing distributions for the good and bad conditions and a bimodal distribution for the neutral condition that resembles those for the second two experimenters.
In order to test whether there was a significant effect of the initial condition that participants were sampled into, we fit a series of binary logistic models to whether the initial response was 4 or higher, as opposed to 3 or lower, as the distributions shown in Figure 2 show that a linear response scale would not be appropriate. This is a simple approach to establish whether different participants occupy different parts of the scale. The analysis revealed that there was a significant interaction between experimenter and condition for kit responses, reflecting the pattern shown in Figure 3. However, analysis of trap also revealed a significant interaction between experimenter and condition, driven by a higher distribution of responses in the good condition for NZ2. In order to determine whether these reflected overall response biases of the participants, we fit models to the data from each experimenter, testing for an interaction between vowel and condition. Significant interactions were found for NZ1 (driven by differences within kit) and for NZ2 (driven by differences within trap).
It is possible that the NZ1 pattern was in some way caused by the experimenter’s behavior; indeed, this was our initial suspicion. However, NZ2 was blind to both the hypothesis and which condition was being run and, therefore, could not have unwittingly affected participants’ responses. The fact that the data she collected also shows variation across conditions indicates that, across the experimenters, we have randomly sampled participants with different perceptual patterns or response biases into the conditions.
Thus, the patterns we observe show that significant population-level variation in perception patterns and/or response biases can lead to significant differences between sampled groups that arise entirely by chance. Including a baseline task – which we have done here but was not done in earlier, related work (Hay et al. 2006a; Hay and Drager 2010) – is therefore crucial in order to document and control for pre-existing biases. To test the hypotheses, we should investigate not the difference in responses across conditions but the difference in degree of change between a baseline and a test condition. Therefore, in our main analysis (described in the next section), we model the shift between the baseline and the test as our dependent variable.
4.1 The vowel
We now turn to our main analysis, which investigates the post-manipulation responses. As discussed, the response pattern shown in Figure 2 suggests that a logistic model would be the most appropriate technique for modeling the responses. However, for this analysis, a simple binary split between steps 3 and 4 would also not be appropriate for a non-trivial number of respondents whose responses were focused on other parts of the scale. Splitting the data between steps 3 and 4 would, for example, group all of participant 3’s responses with the lower category and all of participant 62’s responses in the upper category, despite the fact that both participants also occupy multiple points on the scale. For this analysis, we are not interested in differences across participants in terms of which part of the scale they occupy. Instead, we are interested in whether the manipulation has shifted their responses within whatever part of the scale they occupy. Further, we know that for some subsets of the data, there are some initial differences across conditions in terms of the responses, and we are interested in assessing whether participants significantly differ across the conditions once these initial differences are controlled.
To account for differences across participants, we created a binary variable that captures the direction of change across blocks, for each individual, for each stimulus. To create the binary variable, we first subtracted each participant’s response to a stimulus in the test condition from their response to the same stimulus item in the control condition: for a participant who first answered “4” in the control, and then switched to “3” in the test, this would result in 1, and it would represent a shift to a more Australian variant. From this we create a binary value: responses yielding a positive value are coded as having shifted toward an Australian variant whereas those yielding a negative value or indicating no shift are coded as not having shifted toward an Australian variant. 3
We fit a mixed effects logistic regression model to the kit tokens, modeling the likelihood of shifting to an Australian variant. We include participant and stimulus as random intercepts and, as a fixed effect, we include the participant’s response to the stimulus in the first block as a control, since this directly impacts the possibility of the likelihood of a shift: responses that were already toward the Australian end of the continuum have less room to move toward an Australian variant than when the initial response was toward the New Zealand end. We also tested for main effects of condition, experimenter and sex, and interactions between condition and sex, and condition and experimenter. We do not have enough male participants across all experimenters to test an interaction between sex and experimenter.
There were no significant effects of participant sex, and this was dropped from the model. The tested interactions were also not significant and were dropped. The final model yields significant main effects of the first response, the experimenter, and the condition. We attempted to include random slopes for all main effects. The only slope that led to a convergent model was a by-item slope for condition and the model reported here (Table 2) includes this slope. 4 As past work on this task has yielded sex differences, we wanted to ensure that any significant results were not an artefact of the uneven distribution of males across conditions and experimenters. We therefore refit the model on the female participants only and observed the same results as those reported here.
The model shown in Table 2 indicates that the more New Zealand-like the first response was, the higher the likelihood of a shift to an Australian variant. We interpret this as simply reflecting the fact that how much and/or the direction in which a participant can shift is heavily constrained by their original choice. The model also shows a main effect of experimenter, with the lowest probability of a shift towards Australian variants occurring with the second NZ experimenter, and the highest shift with the first (non-blind) NZ experimenter.
Most importantly for our research questions, we find a significant effect of condition (Figure 4). Participants in the bad condition show the greatest shift towards Australian variants, while participants in the good condition show the smallest shift. Participants in the neutral condition sit between each group, significantly different from neither, though trending closer to participants in the good condition.
4.2 Filler items
We repeated the same model-fitting technique with the trap tokens. The overall model contained a significant experimenter x condition interaction, with the condition having a different effect for every experimenter. In order to check that this wasn’t carried by the uneven distribution of male participants across conditions and experimenters, we attempted to refit the model on just the female participants. The interaction was no longer significant, indicating that it was carried by the uneven distribution of males and females across experimenters and conditions. We therefore report the more conservative model, in which there is a significant effect of experimenter and no effect of condition. This effect is significant over the whole data set, as well as just within the female participants. This model is given in Table 3.
As with the kit variable, the more NZ-like the first response, the more likely it was that the participant would shift toward responding with a more Australian-like variant. There was also a significant effect of experimenter, with the US experimenter leading to the greatest shifts in the Australian direction. However if we regroup the ‘no change’ responses with the Australian shifted variants, instead of the NZ-shifted variants, the experimenter effect does not reach significance, indicating that this is more fragile than the effects reported above for KIT. Finally, an analysis of the dress tokens yielded no significant effects, beyond the significant effect of the participant’s first response.
4.3 The effect on reported attitude
The participants who completed the task with NZ1 also completed two questionnaires: an Australian Integration Index (AII) questionnaire, designed to test exposure to Australians and their dialect, and an Australian Attitude Index (AAI) questionnaire, designed to elicit attitudes toward Australia. The AAI questionnaire consisted of 10 questions covering issues of solidarity, politics and sporting relations between New Zealand and Australia (see Appendix B). Participants were given statements and indicated whether they agreed with the statement, using a four-point continuum so that neutrality was not an option.
We first tested the binary factor of whether the participant “follows international sport” (following Drager et al. (2010)) in the models reported in Sections 4.1 and 4.2, in interaction with condition. It was not significantly predictive.
We then conducted a principal components analysis on the attitudinal questionnaire, orienting the scales so the higher numbers were always associated with positive responses. The resulting first two components are plotted in Figure 5. Participants are shown in the plot according to their condition (b=bad, g=good, c=control). The first component was broadly associated with positive attitudes toward Australia (e.g., “NZers and Australians are very similar”, “I would enjoy living in Australia”). Interestingly, this component is negatively associated with the sports questions, meaning that participants who indicated that they would like to live in Australia were also the most likely to support NZ teams when playing against Australia (sportsP) and the least likely to support Australia if NZ was not playing (sports2). The second principal component is mainly associated with a question about human rights; participants with high loadings on this component believe that Australia has a bad human rights record.
Individual loadings for PC1 and PC2 were tested in a model of the NZ1 participants’ kit responses, both in isolation and in interaction with condition. They were not significantly predictive. However, inspection of the distribution of the loadings themselves reveals that they are not randomly distributed across conditions. Participants in the good condition were more likely to have high loadings on PC1 than other participants (Wilcoxon test p<0.06) and participants in the bad condition were more likely to have high loadings on PC2 (Wilcoxon test p<0.03). This finding suggests that our manipulation somewhat shifted participants’ self-reported attitudes towards Australia(ns).
The design and analysis of our experiment represents a considerable methodological improvement on past work using this type of paradigm. First, we include a pre-test baseline task, and analysis of behavior in this task reveals the importance of this step; even prior to our manipulation, there were considerable differences across some groups and this was true even for the iteration where the experimenter that was completely blind. It was only by including this task and analyzing the degree of shift from these initial responses that we could be sure that we were capturing responses to our manipulation and not random variation across groups. Second, we have closely inspected the individual response strategies in the perception task, and established that most participants only occupy a small part of the scale. This makes linear models (as used by previous published work using this task) inappropriate. A binary model capturing the direction of shift provides us with an appropriate technique that enables us to control for initial responses. Third, we test for a possible role of experimenter identity in our models and find significant effects. This reinforces the conclusion of Hay et al. (2010) that one should try and control the experimenter or, failing that, control for potential experimenter effects statistically.
Our analysis technique allows us to show that, regardless of the reason for the initial differences between the three groups, the observed pre-prime differences are not responsible for the shift observed between blocks. We modelled the direction of shift from the initial response, while holding the absolute value of the initial response constant. The fact that we do not have any interactions between experimenter and condition in this shift provides a strong indication that the differences observed in the participants’ initial responses are appropriately controlled for in the analysis of the shift.
The results of this analysis demonstrate that New Zealanders’ perception of vowels is influenced by exposure to good and bad facts about Australia: listeners shifted toward hearing a more Australian-like kit vowel when exposed to the negative facts. Contrary to earlier experiments that also investigated perceptual adaptations by New Zealanders to Australian English (Hay et al. 2006a; Hay and Drager 2010), there is no interaction between gender and condition. One interpretation of this is that the facts effectively influenced sentiment, overriding differing default biases that men and women tend to bring to the task. That exposure to the facts influenced the participants’ attitudes toward Australia is further supported by the principle components analysis of responses to the Australian Attitude Index questionnaire: participants in the good condition were more likely to be associated with a higher value for PC1, indicating that they, broadly, responded more positively about Australia than participants in the other conditions.
Following exposure to the facts, listeners showed the least shift toward Australian variants following the good facts about Australia and the largest shift following the bad facts. The direction of this shift is in the opposite direction to our initial hypothesis. There are several possible explanations for why this might be.
The first possibility is that the differences in either the quality or quantity of valence across the conditions is driving the effect. Negatively valenced words have been shown to be more quickly accessed (Palma de Figueiredo 2015) and detected (Dijksterhuis and Aarts 2003). Furthermore, a positive-negative asymmetry effect leads negative information to carry more weight in affecting people’s evaluations than positive information (Peeters and Czapinski 1990). Baumeister et al. (2001) reviews a wide range of literature looking at the processing of positive and negative information, concluding that they “have found bad to be stronger than good in an almost disappointingly relentless pattern” (Baumeister et al. 2001: 362). Thus, the negatively valenced primes may have more strongly primed the concept of Australia. While such an effect may well be playing a role, it cannot be the whole story. This is because the difference in the shift between the good and bad conditions is significant, but the neutral condition, in which Australia was not mentioned at all, sits in between. If different levels of priming were the only reason for the difference between the good and bad conditions, then we would have expected the smallest shift to be with the neutral condition.
Another possibility – one which we find more compelling – is that the participants in our experiment had a defensive reaction to the positive facts about Australia, invoking a sense of New Zealand pride and a “but we’re better than them!” reaction. This interpretation is consistent with our interpretation of the production results, where we report an analogous effect in production where the participants who are sports fans produce a more New Zealand-like variant in the condition with the good facts about Australia (Drager et al. 2010: 49). Informal canvassing of a non-random sample suggests that at least some New Zealanders have a negative, overt reaction to one of the good facts in particular: “In 2005, Australia was awarded the title of “The World’s Friendliest Nation” by the Anholt Nation Brands Index”, commenting on the unlikelihood of Australians being friendlier than New Zealanders. This particular fact, then, rather than engendering positive emotion toward Australia, seems to instead result in a strengthened identity as a New Zealander. The degree of rivalry between New Zealand and Australia is strong (Smith 2004), and it seems possible that good facts about Australia can elicit defensiveness and trigger Kiwi pride, and through this, Kiwi vowels. In contrast, the bad facts may reinforce what is already felt about Australia and therefore simply prime Australia.
Such an interpretation may, at first glance, appear to be contradicted by the overt attitude ratings from the NZ1 participants, reported in Section 4.3. Recall that participants in the good condition report more positive attitudes towards Australia than other participants, yet they shift toward hearing more NZ-like kit vowels. It is important to consider that some of the attitude questions directly tapped into factors related to the facts themselves; it is not surprising, for example, that participants exposed to a fact about Aboriginal children being removed from their homes give low ratings to the statement “Australia has a good human rights record”. Similarly, hearing that Australia has been voted the world’s friendliest nation could very directly affect ratings on “Australia is a place I would like to go for a holiday” or “I would enjoy living in Australia” while still increasing one’s own sense of New Zealand identity. Hearing that Australians are deemed friendly can increase the likelihood of wanting to spend time there, while simultaneously increasing the resolve that New Zealanders are even friendlier. In other words, perhaps the effectiveness of our manipulation is not directly related to attitudes toward Australia, but rather, is more directly related to changes that we invoked in how strongly a participant identified as a New Zealander.
The results presented here suggest that emotional reactions to stimuli can have effects on speech perception that are analogous to effects of divergence in production. While these results show that attitudinal factors mediate perceptual shifts, they also highlight – together with Babel (2010) – the difficulty of experimentally manipulating attitudes. The results are not straightforward, and it’s clear that we need much more work in this area to fully understand the effects that positively and negatively valenced stimuli have on speech accommodation and perceptual adaptation, as well as how the affective manipulations interact with different types of linguistic variables and different types of listener biases.
The authors would like to acknowledge the valuable feedback from the reviewers and the time of our participants. This project was funded in part by a Rutherford Discovery Fellowship awarded to the second author.
Bayard, Donn. 2000. New Zealand English: Origins, relationships, and prospects. Moderna Språk 94(1). 8–14. Google Scholar
Bourhis, Richard Y. & Howard Giles. 1977. The language of intergroup distinctiveness. In Howard Giles (ed.), Language, ethnicity and intergroup relations, 119–135. London: Academic Press. Google Scholar
Bourhis, Richard Y., Howard Giles, Jacques-Phillipe Leyens & Henri Tajfel. 1979. Psycholinguistic distinctiveness: Language divergence in Belgium. In Howard Giles & Robert N. St Clair (eds.), Language and social psychology, 158–185. Oxford: Basil Blackwell. Google Scholar
Cox, Felicity & Sallyanne Palethorpe. 2008. Reversal of short front vowel raising in Australian English. Interspeech 1. 342–345. Google Scholar
Drager, Katie, Jennifer Hay & Abby Walker. 2010. Pronounced rivalries: Attitudes and speech production. Te Reo 53. 27–53. Google Scholar
Giles, Howard. 1973. Accent mobility: A model and some data. Anthropological Linguistics 15. 87–105. Google Scholar
Giles, Howard & Peter F. Powesland. 1975. Speech style and social evaluation. London: Academic Press. Google Scholar
Giles, Howard, Donald M. Taylor & Richard Y. Bourhis. 1973. Towards a theory of inter- personal accommodation through language: Some Canadian data. Language in Society 2. 177–192. CrossrefGoogle Scholar
Hay, Jennifer, Margaret Maclagan & Elizabeth Gordon. 2008. New Zealand English. Edinburgh: Edinburgh University Press. Google Scholar
Hay, Jennifer, Aaron Nolan & Katie Drager. 2006a. From fush to feesh: Exemplar priming in speech perception. The Linguistic Review 23. 351–379. Google Scholar
Kang, Okim & Donald Rubin. 2009. Reverse linguistic stereotyping: Measuring the effect of listener expectations on speech evaluation. Journal of Language & Social Psychology 28. 441–456. Web of ScienceCrossrefGoogle Scholar
Koops, Christian, Elizabeth Gentry & Andrew Pantos. 2008. The effect of perceived speaker age on the perception of PIN and PEN vowels in Houston, Texas. University of Pennsylvania Working Papers in Linguistics 34(2). 93–101. Google Scholar
McMurray, Bob. in prep. Klattworks: A [somewhat] new systematic approach to formant-based speech synthesis for empirical research. Manuscript in preparation. Google Scholar
Natale, Michael. 1975. Convergence of mean vocal intensity in dyadic communication as a function of social desirability. Journal of Personality and Social Psychology 32. 790–804. CrossrefGoogle Scholar
Peeters, Guido & Janusz Czapinski. 1990. Positive-negative asymmetry in evaluations: The distinction between affective and informational negativity effects. European Review of Social Psychology 1. 33–60. CrossrefGoogle Scholar
Smith, Adrian. 2004. Black against gold: New Zealand-Australia sporting rivalry in the modern era. In Dilwyn Porter & Adrian Smith (eds.), Sport and national identity in the post-War World, 168–193. London: Routledge. Google Scholar
Staum Casasanto, Laura. 2010. What do listeners know about sociolinguistic variation? Penn Working Papers in Linguistics 15(2). 40–49. Google Scholar
Strand, Elizabeth A. & Keith Johnson. 1996. Gradient and visual speaker normalization in the perception of fricatives. In Dafydd Gibbon (ed.), Natural language processing and speech technology, 14–26. Berlin Mouton de Gruyter. Google Scholar
Sumner, Meghan & Reiko Kataoka. 2013. Effects of indexical variation on spoken word recognition. Journal of Acoustical Society of America Express Letters 134. 485–491. Google Scholar
Sumner, Meghan, Seung Kyung Kim, Ed King & Kevin McGowan. 2015. The socially weighted encoding of spoken words: A dual-route approach to speech perception. Frontiers in Psychology 4(1015). 1–13. Web of ScienceGoogle Scholar
Walker, Abby. 2014. Crossing oceans with voices and ears: Second dialect acquisition and topic-based shifting in production and perception. Columbus, OH: Ohio State University dissertation. Google Scholar
Wells, John C. 1982. Accents of English I: An introduction. Cambridge & New York: Cambridge University Press. Google Scholar
Yu, Alan, Carissa Abrego-Collier & Morgan Sonderegger. 2013. Phonetic imitation from an individual-difference perspective: Subjective attitude, personality and “autistic” traits. PLoS ONE 8(9). e74746. Web of ScienceCrossrefGoogle Scholar
A Appendix: Experimental stimuli
Each sentence was presented with the target word final (omitting the material in parentheses), and with the target word medial (including the parenthetic material).
I wanted to go horse riding with my friend but she didn’t have a spare bit
(so I had to stay at home)
I couldn’t be bothered cooking so I went and got fish
(but it took forever since they were so busy)
During the colder months John and I both wanted to stay fit
(so we decided to join a yoga class)
Celine Dion’s new song “love me forever” was a massive hit
(on the Mongolian top of the pops show)
In the end I had to ask my wife where on earth I had left my tool kit (’cause I’m notoriously bad at losing them)
I was going to put the heater on but then I realised that the fire had already been lit
(so I didn’t need to worry after all)
The elephant took a wrong turn and fell into a pit (constructed by nomadic ivory poachers)
My friend and I eventually want to work on a cruise ship (but we need to do some training first)
It was during the test period and so I couldn’t find anywhere to sit (in any of the university libraries)
I charmed her with my good looks, smiling blue eyes and striking wit (all a vain attempt to impress her mother)
This was the first time that Johnny had ever picked up a cricket bat
(and he was so nervous about it he was sure he was going to faint)
At 4 year’s old, there was no way that Warren was old enough to have his own cat
(so his mum promised him one for his tenth birthday instead)
Michael certainly had a tendency to find some very funky hats
(in various op shops, some of which turned out to be worth a lot of money)
We spent all afternoon anxiously looking for Matt (but he was nowhere to be found)
It was going to be uncomfortable wherever we sat (so we just chose to take the next seat that we came across)
Very quickly, I realised that this was actually a horribly uncomfortable bed
(and that I needed to go back and have it exchanged)
The morning after, I honestly couldn’t remember making the bet
(so I just paid him just in case it was legitimate)
The government should really do something about student debt (so that young people are actually encouraged to go to university)
(but he was nowhere to be found)
Some people like to think that Mr Tarantino is sick in the head (but I prefer to believe that he is a talented producer)
For as long as I could recall I had always wanted a pet
(but my mother wouldn’t allow me to have one)
B Appendix: Australian Attitude Index Questionnaire (AAI)
Participants in the NZ1 iteration were asked to read the following sentences and for each one, provide a rating indicating how much they agreed with it. The scale provided was:
1 strongly disagree, 2 disagree, 3 agree, 4 strongly disagree.
These values were placed into the Principal Components Analysis, and the labels used in
Figure 5 are shown in parentheses. For the labels ending in P, the value provided by participants was subtracted from 1, so that higher values aligned with positive attitudes.
Australians and New Zealanders are very similar. (similar)
I find it annoying when people get New Zealanders and Australians confused. (annoyingP)
Australians and New Zealanders agree on the important issue. (agree)
Australia has a good human rights record. (rights)
In most sports, the team I most want New Zealand to beat is Australia. (sportsP)
Most stereotypes about Australians are false. (stereotypes)
I would enjoy living in Australia. (living)
Australia is a place I’d like to go for a holiday. (holiday)
New Zealanders and Australians have very similar accents. (accents)
In most sports, if New Zealand is not playing, then I tend to support Australia. (sports2)
Throughout this paper, we use Wells’ (1982) lexical set labels to refer to different vowel classes.