UBC Undergraduate Research

Do biases affect speech perception? Russell, Jamie 2014-04

Your browser doesn't seem to have a PDF viewer, please download the PDF to view this item.

Notice for Google Chrome users:
If you are having trouble viewing or searching the PDF with Google Chrome, please download it here instead.

Item Metadata


52966-Russell_Jamie_LING_449_2014.pdf [ 336.15kB ]
JSON: 52966-1.0077992.json
JSON-LD: 52966-1.0077992-ld.json
RDF/XML (Pretty): 52966-1.0077992-rdf.xml
RDF/JSON: 52966-1.0077992-rdf.json
Turtle: 52966-1.0077992-turtle.txt
N-Triples: 52966-1.0077992-rdf-ntriples.txt
Original Record: 52966-1.0077992-source.json
Full Text

Full Text

 DO BIASES AFFECT SPEECH PERCEPTION?  Jamie Russell   Undergraduate Honours Thesis LING 449 Bachelor of Arts (Speech Sciences, Hons.)  April 2014  Supervisor: Dr. Molly Babel Department of Linguistics University of British Columbia (Vancouver)                  © Jamie Russell, 2014 	   2 Table of Contents   1. Introduction and Background                                                                                                       3 2. Experiment                                                                                                                                      7 2.1 Stimuli                                                                                                                                        7 2.1.1 Auditory Stimuli                                              8 2.1.2 Visual Stimuli (Photographs)                                                                                        9 2.1.3 IAT Stimuli                                                                                                                   9 2.1.4 Explicit Measure of Ethnic Bias                                                                                  10 2.2 Participants                                                                                                                              10 2.3 Procedure                                                                                                                                 10 2.3.1 Speech Perception in Noise                                                                                         11 2.3.2 Accentedness Rating                                                                                                   11 2.3.3 Implicit Measure of Ethnic Bias: Implicit Association Task                                      12 2.3.4 Explicit Measure of Ethnic Bias                                                                                 13 2.3.5 Social Network Self-Assessment                                                                                14 3. Data Analysis and Results                                                                                                            14 3.1 Sentence transcription                                                                                                              14 3.2 Accentedness Ratings                                                                                                               15 3.3 Implicit Measure of Ethnic Bias: Implicit Association Task                                                    16 4. Discussion                                                                                                                                      16 4.1 Speech Perception in Noise and Accentedness Ratings                                                           17 4.2 IAT Results                  18 4.3 Limitations                              19 4.4 Future Research                 20 5. Conclusions                              21 Acknowledgements                              22 References                                                                                                                                                  23 Appendix A: BKB Sentence lists per talker                                                                                           26 Appendix B: Statements from the Explicit Measure of Ethnic Bias                                                    29     	  	  	  	   3 1. Introduction and Background  As listeners, we are able to extract a host of cues from speech, including speaker size, gender, and age (c.f. Belin, Fecteau and Bedard, 2004). From early on, this information can be used to make a variety of judgements about a speaker, such as friendliness and intelligence (Kinzler and Dejesus, 2012), as well as attractiveness (Babel, McGuire and King, 2014). Indeed, research has highlighted how listeners use speech information to evaluate and categorize individuals (c.f. Drager, 2010). While such listener-related judgments are triggered in the course of perceptual processing, it is of practical and theoretical importance to explore the opposite -- whether our assumptions affect our perception of speech.   Niedzielski (1999:63) posited that just as listeners use auditory and visual information to “create or calibrate the phonological space of speakers”, they also use social information. Her results show that holding stereotypes about other language varieties can affect listeners’ perception of such varieties, and that these beliefs can have the same affect on the perception of one’s own variety. For example, in her study, Detroiters believed that the model talker (also from Detroit) exhibited the raised diphthongs typical of Canadian Raising when the label “Canadian” was given to the voice. Yet when that label was switched to “Michigan”, respondents selected vowels more associated with Standard American English (SAE), even though they were hearing the same Detroit talker. These results suggest that (a) listeners' stereotypes about regional accents affect their percepts and (b) listeners can have inaccurate stereotypes about their own and others' speech varieties.   Categorization into groups is one way that humans impose organization on the world; doing so may allow for more efficient processing of our environments via weighing the relative social importance of the stimuli. Babel, McGuire and Russell (submitted) present evidence from a voice memory task which suggests that listeners pay more attention to members of their own in-group, even if that group is arbitrarily assigned (i.e. based on a personality test/condition). When participants in the study were faced with a surprise test block and asked to identify voices they had previously heard, memory was greatest for voices that belonged to their in-group; the authors suggest that listeners may have therefore paid selective attention to these voices. If we are better at remembering individuals who are similar to ourselves and even find their voices more attractive (Babel, McGuire and King, 2014), we predict that listeners will also be better at attending to in-group speech in noisy environments.   To this end, Yi, Phelps, Smiljanic and Chandrasekaran (2013) found that monolingual native American English speakers were better at identifying keywords in noise when the target speaker was also a native 	   4 speaker of American English as compared to a Korean-accented speaker of American English. An independent group of listeners also rated the native speakers of English as less accented than the Korean-accented speakers. Yi and colleagues assessed the benefit of audiovisual (AV) presentation for these two groups of speakers. While participants benefited from AV integration for both, the benefit was less for the non-native speakers. Indeed, in the AV trials, accentedness ratings decreased for native speakers while they increased for non-native speakers. Yi and colleagues (2013) suggest that this could be due to two factors: first, the speaker-related effect of non-native speech patterns, which show articulatory differences, and second, the listener-related effect of visual bias, which may enhance or exaggerate the expectation of non-nativeness, leading to an AV processing cost. For example, a study by Devos and Banaji (2005) shows that Asian faces are less likely to be rated as English-speaking Americans than White faces when tested with an Implicit Association Task (IAT; Greenwald et al., 1998). Using the same measure, Yi et al. (2013) found that participants who scored higher on a “Caucasian + American” and “Asian + Foreign” IAT demonstrated increased difficulty perceiving non-native speech in the AV condition. This suggests that not only do visual cues provide speaker information, but that this information can activate biases that affect our speech processing.  How far can we push the effect of listener expectations on speech processing? The study by Yi et al. (2013) used populations differing in the nativeness of their English, but do listeners impose non-nativeness on phenotypically Asian but native Canadian English speakers as well? While the current study is methodologically similar to that of Yi and colleagues (2013), some key differences allow us to focus on how visual information can affect perception instead of the combination of AV integration and listener-related effects. Because Yi et al. (2013) presented a video in their trials, it is difficult to determine whether these effects are (1) related to the integration of non-native visual articulatory cues; (2) due to listener attentiveness when accessing visual-phonetic information; or (3) caused by mismatches between expected and observed speech signals. In an attempt to eliminate issues with AV integration with non-native speech and to explore whether social information alone can trigger a reduction in intelligibility, the current study presents half of the trials in the sentence transcription and accentedness ratings tasks with a static visual face prime.  Another difference between our studies lies in the selection of speakers; all twelve self-identified as native speakers of Canadian English and were born and raised in Richmond, British Columbia. They represented two different racial groups: White and Asian. Before beginning the experiment, listeners were informed that the speakers were from Richmond. This was done in an attempt to prime any expectations about habitants of the city, which at 60% has the highest immigrant population of any city in the Lower 	  	  	  	   5 Mainland (Metro Vancouver, 2011). This is a vast increase from 1996, when only 16.1% of immigrants in the Greater Vancouver Regional District (GVRD) settled in Richmond (GVRD Strategic Planning Department, 1997). In this time, the percentage of Richmondites of English, Irish, French and Canadian ethnic origins has decreased a total of 47%, while those of Chinese origin increased 21% (Policy Planning Division of Richmond, 2006). As of 2011, the most frequently reported ethnic origin in Richmond is Chinese (48.5%) whereas English and Canadian account for 11% and 8% respectively (Statistics Canada, 2011).   While amongst the immigrant population in Richmond, 35.6% speak English and/or French at home, the three most common non-official languages include Cantonese (23.6%), Mandarin (15.2%) and Chinese (14.9%), for a total of 53.6% (Statistics Canada, 2011). While only 3% of the population indicated that they spoke two languages as their mother tongue (ie. were bilingual), 86% of citizens reported that they spoke English at work-- this indicates that citizens are becoming proficient in English, even if they don't identify it as a first language (Policy Planning Division of Richmond, 2011). The present study investigates whether our awareness of these changing demographics is keeping pace. By employing a design that includes an explicit measure of bias (a questionnaire regarding stereotypes about the two groups) and an implicit measure (an IAT), we are able to explore whether expectations and stereotypes about race and language affect our perception of speech. For example, this may be the case if participants with a stronger positive “White Canadian” bias (as indicated by the IAT) have more difficulty perceiving Asian-Canadian speech during the visual face-prime trials of sentence transcription, and similarly rate Asian-Canadians as more accented when presented with their voice-photo combination. In this way, instead of listener biases contributing to a reduced efficiency of audiovisual integration for non-native speech, as is posited in the Yi et al. (2013) study, they may instead result from differential social weighting of linguistic variants during speech perception instead (e.g. Sumner et al., 2014). While models of speech perception that incorporate the role of social information are complex, it is possible that such information can trigger pre-existing biases and thus change the amount of attention paid to stimuli (the same general idea posited Babel, McGuire and Russell (submitted) discussed above). For example, if one enters a conversation with the expectation that they will have difficulty understanding their interlocutor, does this assumption have a top-down effect on speech perception, effectively causing or increasing perceptual difficulty?   A study by McGowan (2012) employed a matched guise paradigm wherein listeners completed a speech perception in noise task during which they were presented with either an Asian, Caucasian or Neutral (a basic silhouette) visual prime along with both high and low predictability sentences spoken by a Chinese 	   6 native speaker of Mandarin. McGowan (2012) suggests that the presence of the congruous face and voice pairing (i.e., Asian face + Mandarin speaker) led to higher accuracy during the transcription task; in this way, information as to the broad social identity of the speaker (the stimuli were not real face + voice matches) led to improved performance. Conversely, when listeners were presented with incongruous pairings (e.g. Caucasian face + Mandarin-accented English), performance was lower. While McGowan’s study (2012) was based on Rubin (1992) and Lippi-Green (1997), his findings do not support their hypotheses that when influenced by racial bias listeners a) pay reduced attention or b) expend less energy on decoding speech (part of ‘the communicative burden’). Instead, he posits that such socioindexical information may be best described as having a clarifying or reinforcing role during perception, and therefore is most useful for an “almost accessible signal” versus one that is more degraded. McGowan (2012) also divided his listeners into experienced and inexperienced listeners; those who had exposure to Mandarin speakers, and those who didn’t. His results show that congruent face-primes can also lead to improved performance in the inexperienced group, which he suggests could be due to stereotypes held about Chinese-accented English and its phonetic realizations. In this case, McGowan indicates that stereotypes are having a positive effect on intelligibility, instead of hindering it. This ‘stereotype-as-benefit’ could stem in part from the integration of social and linguistic information during speech perception. Sumner, Kim, King and McGowan (2014) recently proposed a dual-route model of speech perception that, unlike abstractionist or frequency-based theories, allows for simultaneous mapping of these two types of information. They therefore posit a role for meaningful social variation (e.g. speaker attributes, contextual information, social group) in the differential weighting of acoustic information during encoding at the lexical representation level. This process could in turn affect the accessibility and activation of these forms during subsequent perception. Undoubtedly, social information could include or be reinforced by visual cues signalling speaker gender, age, ethnic origin and so on (c.f. Yi et al., 2013; McGowan, 2012). If this is the case, experiences with different groups and their language varieties could be encoded as variants. Returning to the idea of categorization, it is possible that this type of social information could manifest as listener expectations that are called upon when perceiving novel speech at a later time.  In order to investigate how stereotypes or expectations are affecting our perception of speech, positively or negatively, it is necessary to explore whether or not they are held by our participants or not, and to what extent. As discussed in Devos and Banaji (2005), recent research in social cognition has demonstrated that attitudes and beliefs about different social groups exist at “two distinct levels”: one consciously-held and controllable, and the other quite automatic and less conscious. The present study employs a pair of measures, one implicit (the IAT) and the other explicit (stereotypes questionnaire), not 	  	  	  	   7 only to investigate the correlations between ethnic bias and perceived intelligibility of speech, but also to compare how self-reported, conscious beliefs differ from those held subconsciously.   An explicit measure of ethnic bias allows insight into conscious beliefs and attitudes, even though these might not align with the results on the IAT. Examining results from an explicit measure can highlight the socially-constructed filter with which we view the world. For example, while participants in a series of studies by Devos and Banaji (2005) expressed that qualities such as egalitarianism are important to being American, their IAT results consistently showed that White Americans were held to be more “American” than African Americans or Asian Americans. The dominant Western discourse on cultural and ethnic differences is built upon the belief that as humans all groups are equal and should be treated with a “colour-blind” approach (Todd, 2014a). If such a belief extends to our assumptions about the way others speak, it could mean that we expect everyone to speak the same way, either within or across groups. This could in turn affect how we perceive others' speech.  The following experiment uses these two measures (an implicit association task (IAT) and an explicit measure of ethnic bias) with a combination of two other tasks: speech perception in noise (SPIN) and accentedness rating to investigate the relationships between the bias and intelligibility. A social network self-assessment and a background questionnaire also provide further information as to linguistic experience and habits. Analyses investigate performance in both the sentence transcription and accentedness ratings tasks, comparing the audio-only and audio-face prime trials, as well as interactions between these results and the explicit and implicit measures of ethnic bias. This preliminary study only looks at the results of the SPIN task, the accentedness ratings and the results of the IAT; data from the explicit questionnaire, listener ethnicity, linguistic background and social network remain to be investigated. We predict, however, that these factors will play a role, as preliminary results from a study by Wong (in progress) has shown that while listeners are above chance at identifying the self-identified ethnic identity of Chinese, East Indian, and White speakers, listeners who have larger White networks are best at identifying White voices while listeners who have larger Chinese networks are equally good at identifying Chinese and White voices. If this is the case, listeners in the present study may show a benefit during the SPIN task with different social network experience (see McGowan (2012) for discussion on experienced and inexperienced groups).  2. Experiment  2.1 Stimuli 	   8  2.1.1 Auditory Stimuli  Twelve speakers (6=female) recorded a randomized list of 120 sentences drawn from the Bamford, Kowal and Bench (BKB) sentence lists (Bench and Bamford, 1979; see Appendix A). Each target sentence was presented twice using E-Prime 2.0 software (Psychology Software Tools, Pittsburgh, PA) and recorded in stereo at a sampling rate of 22,050Hz using a USB PreAMP and an AKG C520 head-mounted microphone. All recordings were made in a sound-attenuated booth in the Speech In Context lab at the University of British Columbia.   All speakers were born and raised in Richmond, British Columbia and identified as native speakers of Canadian English. Six of the speakers self-identified as ethnically White, and six as Asian. Within each ethnic group there was an even number of males (3) and females (3). They ranged in age from 17 to 25 years old (M = 21.5 years). All speakers except one spoke English at home, instead speaking exclusively Cantonese with their parents. Two other speakers indicated that they spoke both English and Cantonese at home, and only Cantonese with their parents. Regardless of ethnic background, all speakers spoke English at school, with friends, and at work, although one also spoke Mandarin in their workplace.  Five of the six Asian Canadians reported their parents first languages as Cantonese (N=3) or Chinese (N=2). The Cantonese-speaking parents were all born in Hong Kong, whereas the two Chinese-speaking parents were born and raised in Canada (Alberta and BC). One speaker in the Asian Canadian group had parents born in BC who spoke exclusively English.  The speakers in the White Canadian group all had parents who spoke English as their first language, although two had one parent each who was raised bilingually (English and Greek; English and Dutch). All parents were born in BC with the exception of one born in Amsterdam, NE, who moved to BC at age 2.   Elicited sentences were saved as individual sound files and converted to mono. Silence was trimmed from the ends of the files, and they were peak-amplitude normalized. A subset of sentences in this format were used in the accentedness rating task. For the sentence transcription task, however, all sound files were embedded in pink noise at a signal-to-noise ratio (SNR) of -3 dB with a 500 ms buffer of noise at the beginning and end of each file. Three different levels of SNR were piloted (+1 dB, +3 dB and -3 dB) and from this it was determined that +1 and +3 were too easy. 	  	  	  	   9  Sentence stimuli was programmed into the experiment using E-Prime 2.0 software (Psychology Software Tools, Pittsburgh, PA).  2.1.2 Visual Stimuli (Photographs)  All twelve speakers were photographed in a well-lit corner of the Speech In Context lab against a white background. This was done using an Olympus FE-220 digital camera. Speakers were asked to make a “neutral” face to prevent projecting any emotions, which could affect social judgments about the speaker. The photographs were manipulated to black and white using Picasa picture-editing software and cropped to 580 x 850 pixels.   2.1.3 IAT Stimuli  An Implicit Association Task (IAT; Greenwald et al., 1998) was constructed using a subset of the 100 most common last names in Vancouver, BC, as reported by The Vancouver Sun (2007). A total of forty Asian and Caucasian-sounding names were selected and assessed by an independent group of raters (N=21) as being White, Asian, or other. Sixteen “White” last names and fifteen “Asian” last names received unanimous agreement. To create equal groups of fifteen, one more “White” last name was randomly discarded.   The IAT also included 8 semantically “pleasant” words and 8 “unpleasant” words, which were taken from a previous experiment (Babel, 2009). These can also be seen in Table 1.  White Surname Asian Surname Pleasant Unpleasant Smith Wong sunshine bad Brown Chan happy awful Johnson Chen good disease Anderson Li vacation trouble Jones Wang gift pain Taylor Leung paradise failure 	   10 Williams Liu holiday poison Campbell Wu heaven disaster Macdonald Ho   Miller Chow   Thompson Chang   Stewart Cheung   Robinson Zhang   Moore Yang   Clark Huang   Table 1: Stimuli used to construct the Implicit Association Task.  2.1.4 Explicit Measure of Ethnic Bias  The Stereotypes Questionnaire was constructed to gauge consciously-held beliefs and attitudes about White Canadians and Asian Canadians. Each item was designed to reflect pervasive stereotypes and were balanced such that respondents holding the target stereotype would be expected to respond with either a “Strongly Disagree” or “Strongly Agree”. Eight of each type were presented. Examples include “Asian Canadians are involved in more car accidents than White Canadians” and “It is appropriate for signage in Richmond to be in Asian languages”.   Another seven questions were presented in the same manner, although these probed general beliefs about stereotypes and diversity. They included statements such as “Stereotypes have some truth to them” and “Do you think that a having a racially diverse campus is important?”   See Appendix B for the complete list of statements.  2.2 Participants  Ten self-identified native speakers of English between the ages of 19 and 31 were recruited from the University of British Columbia community. All received monetary compensation ($10 CAN) for their time, and none reported any speech, hearing or language disorders.   2.3 Procedure 	  	  	  	   11  Participants were seated at individual PC workstations in a sound-attenuated booth in the Speech In Context lab for the duration of the task. All listeners wore AKG K240 headphones, and logged responses using either the keyboard or a serial response box, depending on the task. They were casually informed that all the speakers they would hear were from Richmond, BC and were given instructions for each task verbally. The experiment took between 35-45 minutes to complete.  2.3.1 Speech Perception in Noise  The first part of the experiment involved transcribing sentences embedded in noise; subjects were warned prior to beginning that this was the case and that it was a difficult task. Each participant heard 120 sentences total, with ten produced by each of the twelve speakers. No sentences were repeated within the task. Each set of ten sentences came from a single BKB list (Bench and Bamford, 1979; see Appendix A). Half of the sentences (N=5) were presented with a black and white photograph of the speaker, while the other half were presented with a set of three fixation crosses (+++) positioned in the same place. This visual prime was always presented 2000 ms before the audio began to encourage participants to look at the computer screen instead of the keyboard. As soon as the sentence began to play, a box appeared on the screen for listeners to type in. Each audio file began and ended with 500 ms of pink noise, and the presentation order of trials was randomized. Pressing the Enter key moved listeners on to the next sentence. Three breaks were programmed to appear every 40 trials.  2.3.2 Accentedness Rating  Following sentence transcription, participants were asked to rate the accentedness of each of the speakers. Two sentences were randomly selected from each speaker's stimuli list; one was presented with the visual face prime (the photograph of the speaker), and the other with three visual fixation crosses, as in the sentence transcription task. These sentences were presented in the clear.   Participants were asked to rate the accentedness of the speaker using a Likert scale from 1 to 9. A choice of 1 indicated “No Foreign Accent” and 9 signified “Very Strong Foreign Accent” following Yi et al. (2013) and Smiljanic and Bradlow (2011). This scale was provided at the bottom of each screen while the audio files played. Responses were entered using the number pad on the keyboard.  	   12 2.3.3 Implicit Measure of Ethnic Bias: Implicit Association Task  Before beginning the experiment, participants were verbally instructed to complete the IAT as quickly and accurately as possible, as response time would be used in analysis. They were also asked to pay close attention to the categories, as they changed sides throughout the task. These same instructions were also presented on the computer monitor prior to the presentation of the IAT. The IAT involves five blocks of speeded categorization, each of which is described in turn below.  Participants responded using a button box, which was positioned to the right of the keyboard. The left-most button (1) was used to categorize items belonging to the left category on the screen; similarly, the right-most button (5) was used to categorize items belonging to the right category. Immediately after categorizing an item, participants were given feedback as to the speed of their response and its accuracy. Correct responses were indicated in blue, whereas incorrect responses were displayed in red. If no response was detected after 3 seconds, the screen flashed white and indicated that no response was detected. It then continued on to the next trial.  The first IAT block is referred to as target-concept discrimination. For this block, the labels Asian and White were presented in either the left or right top corners of the computer monitor. The surnames to be categorized (e.g. Wong, Smith) were then randomly presented in the center of the screen; participants were asked to categorize these last names as Asian or White as quickly as possible using the button box. As described above, a feedback screen was immediately presented following each response; this is the same for all subsequent blocks.  The second block involves an associated attribute discrimination. The left and right categories thus change from Asian and White to pleasant and unpleasant. This time, attribute words with either pleasant or unpleasant connotations (e.g. disease, vacation) are randomly presented in the middle of the screen for categorization.   The third block is a combined test; both Asian and White as well as pleasant and unpleasant labels are presented in the upper left and right corners of the screen. The trials to be categorized thus include both the surnames from Block 1 and attribute words from Block 2. All trials were presented in random order. Last names were presented with the standard capitalization of the first letter, and attribute words were presented in all lowercase.  	  	  	  	   13 The fourth block was the same as the first target-concept block, except the categories Asian and White changed sides of the screen. For example, if participants were presented with Asian on the left side of the screen and White on the right in Block 1, in this block they would reverse so that White would now appear on the left and Asian on right. The categorization task itself remained the same as it was in Block 1.  The fifth and final block was another combined test block. As in Block 4, the target-concepts (Asian and White) were reversed, but the attribute labels pleasant and unpleasant remained on the same sides as they were in Block 3. Therefore if Asian was first paired with pleasant on the left side of the screen and White with unpleasant, in Block 5 White would appear on the left with pleasant and Asian would appear on the right with unpleasant. Within each of the four conditions of this experiment, participants were counterbalanced so that half (N=5) were first presented with Asian and pleasant and the other half were first presented with White and pleasant. For the purposes of this honours thesis, only one condition has been run.  2.3.4 Explicit Measure of Ethnic Bias  Participants were next asked to indicate how strongly they agreed with a series of randomized statements. These statements were organized into two blocks. The first block included sixteen statements specifically investigating common stereotypes about or issues affecting Asian Canadians and White Canadians (e.g. Asian Canadians are involved in more car accidents than White Canadians; It is appropriate for signage in Richmond to be in Asian languages; see Appendix B for the complete list). Subjects were instructed to bring to mind individuals born in Canada when deciding how much they agreed with each statement (as in Devos and Banaji, 2005). This was done in an attempt to differentiate between recent immigrants and Canadian-born citizens. Participants were asked to respond as honestly as possible, and assured that all responses were anonymous.  Each statement was presented in the top centre of the computer monitor. The rating scale was presented at the bottom. This Likert scale ranged from 1 (Strongly Disagree) to 7 (Strongly Agree) and included response options of Disagree (2), Somewhat Disagree (3), Neither Agree nor Disagree (4), Somewhat Agree (5), and Agree (6). A small box appeared in the centre of the screen for participants to respond in. Answers were logged using the number pad on the keyboard, and the next question was automatically presented. Half of the statements were designed so if a participant held a particular stereotype they would be predicted to respond with Strongly Agree; the other half would be expected to elicit a Strongly 	   14 Disagree.  The second block was intended to explore general beliefs about stereotypes and diversity. These questions were presented and responded to in the same way as the first block.  These data have yet to be analyzed and will not be discussed further.  2.3.5 Social Network Self-Assessment  Participants were asked to describe the predominant ethnic composition of their social group. There were four options for answers: 1 (Most of my friends, family and coworkers are White Canadian), 2 (Most of my friends, family and coworkers are Asian Canadian), 3 (My friends, family and coworkers are an equal mix of both groups) or 4 (Most of my friends, family and coworkers are members of a different ethnic group).   These data have yet to be analyzed and will not be discussed further.  3. Data Analysis and Results  3.1 Sentence transcription  Intelligibility scores were calculated by averaging the number of words correctly identified during the speech perception in noise task (McGowan, 2012; Bradlow and Bent, 2008). Following McGowan (2012), words transcribed with obvious spelling mistakes were corrected and punctuation was normalized. As some of the data collected did not show spaces that were inserted between words, spaces were inserted where word boundaries obviously indicated how they should have been parsed in an orthographic typing stream.   Scoring was done using a script that automatically calculated the number of correctly typed words per listener, per sentence and per talker. This data was then analyzed in R (R Development Core Team, 2013) using a mixed effects linear regression model with talker ethnicity and condition as independent variables. There were random intercepts for listener with random slopes for both talker ethnicity and condition. The model returned a significant intercept [B = 0.37, SE = 0.02, t = 15.59] and a main effect of Condition [B = 	  	  	  	   15 -0.11, SE = 0.02, t = -4.52]. The interaction between Ethnicity and Condition was just beyond significant [B = 0.07, SE = 0.03, t =1.9]. These results are shown in Figure 1. The static image of the talkers had negative consequences on speech intelligibility compared to the audio-only condition. The near-significant interaction between Condition and Ethnicity illustrates that the presence of the talker face had a larger negative effect for the Asian Canadian talkers.   Figure 1. Interaction between Intelligibility (proportion words correct) by Condition and Ethnicity.  3.2 Accentedness Ratings  Listeners' accentedness ratings for each voice in each condition were entered into a linear mixed effects model using the same fixed effects structure as that for the sentence transcription task. The model returned with a significant intercept [B = 2.9, SE = 0.35, t = 8.4] and an effect of talker ethnicity [B = -0.67, SE = 0.29, t = -2.3]. Asian Canadian speakers were rated as slightly more accented (M = 2.9, SD = 	   16 1.8) than White Canadian speakers (M = 2.4, SD = 1.8).   We assessed the relationship between intelligibility and accentedness via a simple correlation. Talkers who were less intelligible were rated as more accented [t (238) = -2.13, p < 0.05, cor = -0.14].   3.3 Implicit Measure of Ethnic Bias: Implicit Association Task  The IAT was scored using the improved scoring algorithm designed by Greenwald, Nosek and Banaji (2003). The final IAT scores are centred around 0 and run on a continuum between -1 to 1. This means that scores along the scale indicate varying degrees of positive bias towards either ethnic group. Scores of -1 indicate a positive bias towards White Canadians, whereas scores of 1 indicate a positive bias towards Asian Canadians.  We assessed whether IAT scores predicted Intelligibility scores in a simple linear regression model with Talker Ethnicity as a predictor. There was an effect of IAT [B = 0.08, SE = 0.03, t=2.5, p < 0.05] which indicated that a positive bias towards White Canadians reduced the intelligibility of both White and Asian Canadian talkers.   We similarly examined whether IAT scores predict the Accentedness Ratings in a simple linear regression model with Talker Ethnicity as a predictor. Again, we found a significant effect of IAT [B = -0.76, SE = 0.34, t = -2.7, p < 0.5]; a positive bias towards White Canadians led to an increase in accentedness ratings for all voices in the task.   4. Discussion  Do listeners have more difficulty understanding the speech of Asian Canadians born-and-raised in the Lower Mainland because they assume their speech will be accented? We investigated this question using speech from Asian Canadians and White Canadians, and presented listeners with a series of tasks, including speech perception in noise, accentedness ratings, an IAT, a questionnaire and a social network self-assessment. Importantly, where audio stimuli was used, half of the sentences were presented with a static visual prime and half without; this manipulation allowed us to investigate the role of socio-visual information in both intelligibility tasks and accentedness ratings. While only ten participants have been run in the experiment thus far, the preliminary results from the SPIN task, accentedness ratings and IAT suggest that the role of socioindexical information in speech perception is complex. 	  	  	  	   17  4.1 Speech Perception in Noise and Accentedness Ratings  We predicted that the presence of socioindexical information in the Face+Audio condition would affect intelligibility of speech of both groups, but that it would be more detrimental to Asian Canadian faces. This turned out to be the case: while there was a decrease in the perceived intelligibility for voices of both ethnicities in the Audio+Face condition, listeners found Asian Canadian voices less intelligible than White Canadian voices. Although part of this decline could be ascribed to inexperience processing speech paired with static faces, this does not explain why Asian Canadian voices are more affected. We propose two possible explanations. First, it could be that a mismatch between expected and observed phonetic patterns occurs, and this leads to a decrease in intelligibility for the Asian Canadian voices. Alternatively, the phenotypically-Asian Canadian face prime could lead to a decrease in attention or effort on the part of the listener. At this time it is impossible to attribute our results to one or the other hypothesis, although our findings suggest that they could account for, at least in part, the results found in Yi et al. (2013). Because our experimental design eliminated audiovisual stimuli, our findings suggest that social information (versus audiovisual processing experience) can play a crucial role in perception; this hypothesis is in line with current theoretical work (e.g. Sumner et al., 2014). It will be important to undertake other research to explore this further.  Both groups of talkers were rated on the low-to-mid end of the accentedness scale. This said, regardless of condition, Asian Canadian speakers were rated as slightly more accented than White Canadian speakers. This was found regardless of condition, suggesting that listeners are picking up on something in the speech signal. For example, there may be suprasegmental information causing listeners to assign judgements of higher accentedness. If such cues exist that make Asian Canadian speech recognizable, this could signal the existence of a variety of Canadian English similar to African American English or Chicano English in the US. It is also possible that there are differences in the phonetic content of some of the Asian Canadian speakers as opposed to the White Canadian speakers. Three of the six Asian Canadians spoke Cantonese with their parents at home, and their parents were all born in Hong Kong. Two more Asian Canadian speakers also had parents whose first language was a dialect of Yue Chinese (closely related to Cantonese), but both spoke English at home with their parents and rated their overall proficiency of Mandarin as ‘poor’. Further, their parents were born in Canada. While there are numerous factors that influence a child’s first language, whether they are bilingual or not, and their knowledge of their heritage language, it is possible that those who are second generation Canadians (individuals born in 	   18 Canada with at least one parent born outside of Canada; see Statistics Canada, 2014) and have experience with Chinese may show differences (e.g. as a result of a broader phonological inventory), even if only hearing ambient language.  But why does the presence of a face have a greater effect on intelligibility than it does on an accentedness rating? One possibility lies with the type of task. Transcribing sentences embedded in noise is more of an offline task than rating how accented a voice sounds with a single keystroke (more of an online task). In the SPIN task, listeners had to extract the sentence from noise, and were presumably listening for something that made sense semantically, as is predictable with most phrases we encounter (unless we are linguistics students!). Further, participants had to type out what they heard, which required more attention. Conversely, the sentences used in the accentedness ratings task were presented in the clear, and listeners were only asked to indicate how accented they believed the speech to be. They were also only rating two sentences per speaker (one Audio-only and one Audio+Face), for a total of 24 sentences, versus 120 sentences in the SPIN task. Running more participants will allow us to further investigate the role of condition on both intelligibility and accentedness ratings, as ten participants only allows for general comments regarding data trends.  The negative correlation between accentedness ratings and intelligibility showed a logical pattern: voices that were rated as more accented were also found to be less intelligible, and those rated as less accented were more intelligible. While the negative trend holds for both ethnic groups, there is great variability, especially on the voices that were rated as being more accented. This is likely because there are fewer data points on this side of the scale: the majority of speakers were rated as having “No Foreign Accent”. The existence of such a correlation suggests that regardless of task type, accentedness and intelligibility affect each other.  4.2 IAT Results  The IAT results are puzzling and will benefit immensely from a larger sample size. Eight of the ten participants run thus far exhibit some degree of a positive bias towards White Canadians, meaning that they were faster and more accurate at pairing stereotypically White last names with pleasant attributes, slower at pairing stereotypically Asian last names with the same, faster at stereotypically Asian surnames with unpleasant words, and slower at stereotypically White last names and unpleasant attributes. The other two participants showed some degree of positive bias towards Asian Canadians. We interpret the trend of a bias towards White Canadians as indicative of the more general Western cultural bias in 	  	  	  	   19 Canada, which can also be seen in the US (see Devos and Banaji, 2005). Both countries have histories of large waves of European settlement and struggles between present and immigrated minority groups. It is important to note, however, that IAT results are not representative of racist tendencies or beliefs; instead, they indicate dominant cultural stereotypes.   While we expected to find correlations between pro-White Canadian IAT scores and lower intelligibility and higher accentedness ratings for Asian Canadian voices, the preliminary results are not that clear. We found a positive correlation between intelligibility and IAT scores such that listeners with stronger pro-White Canadian biases found all talkers, regardless of ethnicity, to be less intelligible. They also rated voices as more accented, regardless of group. This is compared to the participants with pro-Asian Canadian biases, who found speakers to be more intelligible independent of group, and rated all voices as less accented. Therefore, instead of finding differential treatment of ethnic groups when a participant held an implicit pro-White or pro-Asian Canadian bias, we found more general differences in how they perceived both ethnic groups, regardless of task. More participants will allow us to both pick apart these findings and allow us to see for certain whether with the changing demographics of Vancouver are leading to changes in implicit IAT beliefs. A larger sample will also let us to investigate the ecological validity of our results.   Perhaps there are different definitions of “accentedness” at play between the two groups. For example, listeners with pro-Asian Canadian IAT scores may have rated all speakers as less accented because they have a broader definition of what it means to be accented. On the other hand, listeners with positive biases towards White Canadians may have a much narrower definition that leads to higher ratings overall. The “broadness” or “narrowness” could be as a result of language experience, which could also lead to differences in perceived intelligibility of speech. While all listeners have lived in the Lower Mainland for at least 2 years, a few have varied living histories and therefore presumably have exposure to a variety of global Englishes or other languages. Future work will look at participants’ IAT scores to see whether linguistic experience or living history correlate.   4.3 Limitations  This study is limited in its statistical power due to the small sample size. An additional thirty listeners will allow for more confidence in findings, as well as more robust effect sizes. Additionally, the first ten participants were selected with minimal criteria: all self-identified as native speakers of English and had 	   20 no speech, language or hearing disorders. Screening for linguistic experience and living history may also allow for increased ecological validity and generalizability to the Lower Mainland population.  It is possible that some of the sentences were more intelligible in noise than others, which could be due in part to intensity modulation, which may have led to a skewed SNR once embedded in noise. For example, if a listener finds a sentence intelligible from the beginning, especially if it begins with an amplitude peak, this could facilitate their understanding of the rest of the sentence. In the future, pre-testing the sentences could allow us to remove those that are highly intelligible. One of the White Canadian speakers also had a few British-like vowels; therefore it is impossible to know whether listeners rated his voice as “accented” because they picked up on this. Further, the meaning of “accented” was specifically not defined, so it is impossible to know what variety of English listeners were using as their baseline when rating accentedness.   Some of the sentences were also recorded in a list-like manner, where speakers used high rise terminal intonation at the end of the sentences. While sentences were selected so that they featured minimal rising intonation, it is possible that this affected how intelligible sentences were in noise. Other things, such as rate of speech or especially breathy or creaky voices could have also become less intelligible in noise. It is also possible that the task was simply too difficult.  Lastly, as one of the participants in the study noted, it is possible that the visual stimuli also signaled cultural information through the type of clothes speakers were wearing in the photographs. Future studies could control for this by having all speakers wear an identical shirt or colour.  4.4 Future Research  The continuation of this study will allow us to further explore interactions between explicit and implicitly held beliefs; it will also allow us to take social network and background information into account when understanding intelligibility and accentedness ratings. Once complete, future study is needed to investigate the replicability of such findings with other ethnic groups in other locations; this will allow us to see whether stereotypes and general expectations regarding linguistic background affect processing in the same way elsewhere. It would also be interesting to complete the same experiment using audiovisual stimuli during the speech perception in noise and accentedness ratings tasks, to see whether the static face stimuli was detrimental because it was unnatural or unexpected. This said, the results would likely be different in some way due to an expected AV boost. 	  	  	  	   21  5. Conclusions  Our judgements about speech can have far-reaching consequences, especially in witness testimony, speech assessments, interviews, and education. Purnell et al. (1999) demonstrate that not only are nonstandard dialects identifiable from simple utterances like hello, but that this information can lead to discrimination. The testimony of witness Rachel Jeantel during the Florida v. Zimmerman case in 2012 was scrutinized by the press and public for being in African American English – a variety of English often misinterpreted as unintelligent by the media and public. While historical and cultural factors play into our implicit and explicit beliefs about groups, no language variety is “better” than another; attitudes and assumptions are imposed by society and culture. It is important to investigate role of social information in speech perception as it occurs in everyday communication.   The preliminary results from this study suggest that socioindexical information about speakers’ ethnic backgrounds affects how intelligible listeners find them. This said, its role in accentedness ratings is more complex. There is a correlation between intelligibility and accentedness ratings such that listeners who were rated as less accented were found to be more intelligible in the speech perception in noise task, and vice versa. We found that participants with positive biases towards White Canadians found voices of both ethnicities to be less intelligible and more accented overall, and those with a positive bias towards Asian Canadians found voices of both groups to be more intelligible and less accented. We believe that a larger sample size will clarify the trends presented in these preliminary results.  Canada promotes and celebrates diversity. Once complete, the present research will contribute to our knowledge of stereotypes, misunderstandings and implicit (or explicit) biases towards different ethnic groups. In this way, it will also examine whether judgements or stereotypes about the types of speech associated with ethnic groups can be corroborated or dismissed by data (i.e. does a variety of Asian Canadian English exist that is readily identifiable by local listeners?) or if our stereotypes about groups affect how intelligible we find them, as preliminary results suggest. A better understanding of the role of socioindexical information in speech perception can contribute to the creation of tools and workshops that facilitate open communication and an inclusive community. In a multiethnic and multilingual environment, especially one with a growing Asian population such as Vancouver (Todd, 2014b), such research becomes increasingly important and relevant.   	   22 Acknowledgements  Thank-you notes go to: Dr. Molly Babel for the guidance, support & excitement. The Russell family for articles, endless cups of coffee, and general love & support. Martin Oberg for the noise & analysis help. Tess Fairburn, Alannah Turner, David Kurbis, Jen Abel, Yulia Olefirenko, Diane Hui, Douglas Todd, SpeeCon, the audience at MURC 2014 – thanks for the laughter, help, discussions & for sparking ideas. Phoebe Wong: we did it! Thanks for being my thesis buddy. Lastly, a huge THANK YOU to the speakers who lent their voices & faces to this project, and to the listeners who participated.                           	  	  	  	   23 References   "Top 100 surnames in the Lower Mainland." The Vancouver Sun. 3 Nov. 2007: n. pag. canada.com. Retrieved from http://www.canada.com/vancouversun/news/weekendreview/story.html?id=32537115-d2c9-4c4d-8442-4406b5577300/  Babel, M. (2009). Phonetic and Social Selectivity in Phonetic Accommodation. Unpublished PhD Dissertation. University of California, Berkeley.  Babel, M., McGuire, G. & King, J. (2014). Towards a More Nuanced View of Vocal Attractiveness. PLoS ONE 9(2).   Babel, M., McGuire, G. & Russell, J. (submitted). Listeners exhibit improved memory for ingroup voice.   Belin, P., Fecteau, S., & Bedard, C. (2004). Thinking the voice: neural correlates of voice perception. Trends in Cognitive Science, 8(3), 129-135.  Bench, J., & Bamford, J.M. (1979). Speech-Hearing Tests and the Spoken Language of Hearing-Impaired Children. London and New York: Academic Press.   Boersma, P., & Weenink, D. (2013). Praat: doing phonetics by computer [Computer program]. Version 5.3.42, retrieved from http://www.praat.org/  Bradlow, A. R., & Bent, T. (2008). Perceptual adaptation to non-native speech. Cognition, 106(2), 707 -729.  Devos, T. & Banaji, M. (2005). American = White? Journal of Personality and Social Psychology, 88(3), 447-466.   Drager, K. (2010). Sociophonetic Variation in Speech Perception. Language and Linguistics Compass, 4(7), 473–480.  Greenwald, A., Nosek, B. & Banaji, M. (2003).Understanding and using the Implicit Association Test: An improved scoring algorithm. Journal of Personality and Social Psychology 85, 197–216.  Greenwald, A., McGhee, D. & Schwartz, J. (1998). Measuring Individual Differences in Implicit Cognition: The Implicit Association Test. Journal of Personality and Soclal Psychology, 74(6), 1464-1480.  	   24 GVRD Strategic Planning Department. (1997). Greater Vancouver’s Population Growth is Fueled by International Immigration (Summary of 1991-1996 Census Data). Retrieved from http://www.metrovancouver.org/about/publications/Publications/Census1996-Immigration.pdf  Kinzler, K. & DeJesus, J. (2013) Northern = smart and Southern = nice: The development of accent attitudes in the United States. The Quarterly Journal of Experimental Psychology, 66(6), 1146-1158.   McGowan, K. (2012). Socioindexical expectation and speech perception in noise: experienced and inexperienced listeners. Rice University.  Metro Vancouver. (2011). 2011 National Household Survey - Bulletin #6 : Immigration and Cultural Diversity (Summary of Statistics Canada Release). Retrieved from http://www.metrovancouver.org/about/publications/Publications/2011CensusNo6-ImmigrationCulturaDiverstiyl.pdf  Policy Planning Department of Richmond. (2006). Ethnicity Hot Facts (Summary of 1996, 2001, 2006 Census Information). Retrieved from http://richmond.ca/__shared/assets/2006_Ethnicity20987.pdf  Policy Planning Department of Richmond. (2011). Languages Hot Facts (Summary of 2006 and 2011 Census Information). Retrieved from http://richmond.ca/__shared/assets/Languages6251.pdf  Purnell, T., Idsardi, W. & Baugh, J. (1999). Perceptual and Phonetic Experiments on American English Dialect Identification. Journal of Social Psychology, 18(1), 10-30.  R Development Core Team (2013). R: A language and environment for statistical computing. R   Foundation for Statistical Computing, Vienna, Austria. ISBN 3-900051-07-0, URL   http://www.R-project.org/.  Smiljanic, R. & Bradlow, A. R. (2011). Bidirectional clear speech perception benefit for native and high-proficiency non-native talkers and listeners: Intelligibility and accentedness. Journal of the acoustical Society of America 130 (6), 4020-4031.  Statistics Canada. (2014). Generation status: Canadian-born children of immigrants. Retrieved from http://www12.statcan.gc.ca/nhs-enm/2011/as-sa/99-010-x/99-010-x2011003_2-eng.cfm  Statistics Canada. (2011). NHS Focus on Geography Series – Richmond: Immigration and Ethnocultural Diversity. Retrieved from http://www12.statcan.gc.ca/nhs-enm/2011/as-sa/fogs-spg/Pages/FOG.cfm?lang=E&level=4&GeoCode=5915015  Sumner, M., Kyung Kim, S., King, E. & McGowan, K. B. (2014). The socially weighted encoding of spoken words: a dual-route approach to speech perception. Frontiers in Psychology, (4)01015.  	  	  	  	   25 Todd, D. (2014, March 7). To avoid stereotyping, forget being ‘colour blind’. The Vancouver Sun. Retrieved from http://www.vancouversun.com/opinion/columnists/Douglas+Todd+avoid+stereotyping+forget+being+colour/9593805/story.html  Todd, D. (2014, April 1).  The most 'Asian' city outside Asia. The Vancouver Sun. Retrieved from http://www.vancouversun.com/opinion/columnists/Douglas+Todd+Metro+Vancouver+most+Asian+city+outside/9674092/story.html  Wong, P. (in prep). Perceptual identification of talker ethnicity in Vancouver. Undergraduate Honours Thesis, University of British Columbia.                           	   26 Appendix A BKB Sentence lists per talker  Speaker BKB List number Sentence  WM1  1  Children like strawberries.   She cut with her knife.   The bag bumps on the ground.   The boy did a handstand.   The car engine is running.   The clown had a funny face.   The green tomatoes are small.   The house had nine rooms.   They are buying some bread.   They are looking at the clock. WM2 20 A cat jumped off the fence.   He paid his bill.   The baby has blue eyes.   The boy got into trouble.   The football hit the goalpost.   The girl caught a cold.   The oven is too hot.   The raincoat is hanging up.   They are going out.   They sat on a wooden bench. WM3 5 A boy fell from the window.   He is washing his face.   She used her spoon.   Somebody took the money.   The bath towel was wet.   The cook cut some onions.   The kitchen sink is empty.   The light went out.   The park is near the road.   They are running past the house. WF1 3 The boy is running away.   The children are walking home.   The dog played with a stick.   The glass bowl broke.   The kettle is quite hot.   The lady wore a coat.   The man cleaned his shoes.   The room is getting cold.   They say some silly things.   She drinks from her cup WF2 4 A boy ran down the path.   He found his brother.   Lemons grow on trees.   She spoke to her son. 	  	  	  	   27   Some animals sleep on straw.   The house had a nice garden.   The machine was quite noisy.   The old man worries.   The wife helped her husband.   They are crossing the street. WF3 7 She found her purse.   The bus went early.   The children dropped the bag.   The dog came back.   The father forgot the bread.   The floor looked clean.   The girl has a picture book.   The orange was quite sweet.   The young people are dancing.   They had two empty bottles. AM1 14 An old woman was at home.   He dropped his money.   She is helping her friend.   The angry man shouted.   The big fish got away.   The dog sleeps in a basket.   The girl plays with the baby.   The kitchen window was clean.   They are drinking tea.   They broke all the eggs. AM2 8 A friend came for lunch.   Police are clearing the road.   She writes to her brother.   The ball broke the window.   The boy forgot his book.   The bus stopped suddenly.   The family bought a house.   The pond water is dirty.   They are shopping for cheese.   They heard a funny noise. AM3 9 A letter fell on the mat.   He listens to his father.   She stood near her window.   The book tells a story.   The five men are working.   The lady packed her bag.   The shoes were very dirty.   The table has three legs.   The young boy left home.   They are climbing the tree. AF1 13 A man told the police.   He is cleaning his car.   Potatoes grow in the ground.   She argued with her sister. 	   28   The big dog was dangerous.   The fruit came in a box.   The husband brings some flowers.   The mouse found the cheese.   They are playing in the park.   They waited for one hour. AF2 16 A lady buys some butter.   He closed his eyes.   She’s paying for her bread.   Some men shave in the morning.   The children wave at the train.   The driver lost his way.   The policeman found a dog.   The raincoat is very wet.   They called an ambulance.   They stared at the picture. AF3 2 He cut his finger.   She is taking her coat.   Snow falls at Christmas.   The ball went into the goal.   The boy knew the game.   The ladder is near the door.   The old gloves are dirty.   The police chased the car.   The thin dog was hungry.   They had a lovely day.  WM = White Canadian Male WF = White Canadian Female AM = Asian Canadian Male AF = Asian Canadian Female            	  	  	  	   29 Appendix B Statements from the Explicit Measure of Ethnic Bias  Stereotypes and Issues Questionnaire  Expected responses if one holds the stereotype Strongly Disagree (1) Strongly Agree (7)  Asian Canadians speak English well.   Everyone in Canada should be fluent in English.  It is appropriate for signage in Richmond to be in Asian languages.  It is true that Asians Canadians are better at math than White Canadians.   The Canadian government should do more in terms of translation for Canadian speakers of Asian languages.  Asian Canadians are involved in more car accidents than White Canadians.  Canada should compensate Asian Canadians for racial discrimination in the past.  Asian Canadian females are less confrontational than White Canadian females.  White Canadian males are less masculine than White Canadian males.   White Canadian students experience less family pressure regarding academic performance than Asian Canadians.  Asian Canadians are more “Canadian” than White Canadians.  Asian Canadians are more likely to excel in business versus trade professions.  White Canadian parents are more strict with their children than Asian Canadian parents.  Asian Canadians are less athletic than White Canadians.  White Canadians haven’t experienced as much racial discrimination in Canada as Asian Canadians.  Asian Canadians excel at classical piano and violin.  General Questions Stereotypes have some truth to them. Stereotypes are perpetuated by misunderstandings. All cultures have negative and positive stereotypes. Stereotypes affect the way we behave when meeting someone new. Stereotypes provide insight into the way groups perceive and judge others. All ethnicities should be treated equally, regardless of existing stereotypes.  


Citation Scheme:


Citations by CSL (citeproc-js)

Usage Statistics



Customize your widget with the following options, then copy and paste the code below into the HTML of your page to embed this item in your website.
                            <div id="ubcOpenCollectionsWidgetDisplay">
                            <script id="ubcOpenCollectionsWidget"
                            async >
IIIF logo Our image viewer uses the IIIF 2.0 standard. To load this item in other compatible viewers, use this url:


Related Items