Socio-psychological salience and categorisation accuracy of speaker place of origin

There exists a dearth of research investigating how listeners use their knowledge of variation in their L2 to categorise speaker provenance from stimulus speech. The present study, employing a free classification measure, examined 191 Thai university students' categorisations of the geographical origin of nine speakers of English. Analysis demonstrated participants were generally able to distinguish between native and non-native English speech more broadly, and this distinction was found to be the primary perceptual dimension underlying speaker provenance categorisations. With regards to more fine-grained classifications, recognition rates for Thai, UK, US and Indian English speakers were substantially higher when compared to[5_TD$DIFF] Vietnamese and Australian English speakers, indicating the social[6_TD$DIFF]-psychological salience of the speech forms, rather than geographical proximity, was key in determining categorisation accuracy. Analysis of misidentification patterns showed a tendency for the Thai students to conflate Asian English speech forms, despite substantial phonological and phonetic differences between the English spoken in different Asian nations. Participant comments also indicated segmental features were largely responsible for (mis)categorisations. Consistent with current speaker evaluation[7_TD$DIFF] theories, the findings point to speaker categorisation as an initial processing stage, leading to the activation of[8_TD$DIFF] stereotypes about and attitudes towards the speakers' perceived[9_TD$DIFF] social and ethnic group membership. CrownCopyright © 2019 Published by Elsevier B.V. This is an open access article under the CCBY license (http://creativecommons.org/ licenses/by/4.0/). [10_TD$DIFF]

language varieties. This is especially the case because Thailand currently receives a very high number of overseas tourists, from a vast assortment of different countries and who most often communicate with the local population in English, has recently rapidly internationalised its Higher Education system and -by offering of a range of university courses taught exclusively in English -been rewarded with exponential growth in English-speaking students from (mainly) Asian countries attending its universities, and is a founding member of ASEAN (The Association of Southeast Asian Nations) where, from 2015 onwards, English has been employed as the sole official working language (see [ 1 4 _ T D $ D I F F ] McKenzie [ 1 5 _ T D $ D I F F ] et al. 2016; Snodin and Young, 2015).
Many of the prior studies examining categorisation rates for speaker place of origin more broadly have employed either perceptual dialectology map-tasks (e.g., Preston, 1999) or forced-choice categorisation tasks (e.g., Kristiansen et al., 2018;Prikhodkine, 2018), where listeners are generally required to select from a pre-determined list of countries, regions or varieties. However, the utilisation of free classification measures is of particular theoretical and methodological value since it permits listeners to create their own perceptual labels ; [ 1 1 _ T D $ D I F F ] McKenzie, 2015a). This, in turn, eliminates the potentially confounding effect of researcher-determined categorisations and helps to ascertain the depth to which listeners can classify speech into specific language groups, regions, countries or cities (e.g., the extent to which a speaker may be accurately identified to be from the North America, the United States or New York). The data collected from free classification instruments also lends itself to the analysis of patterns of miscategorisation, as well as correct categorisation, which, for example, can provide valuable information about the extent to which listeners may conflate different forms of English spoken in Asia, Europe or North America, regardless of the level of linguistic similarity between the varieties under consideration. Analysis of (mis)identification patterns can thus increase understanding of (the depth of) listeners' perceptual representations of different speech varieties, help uncover wider language ideologies amongst communities of listeners as well as provide knowledge regarding the precise linguistic features responsible for their categorisations ([ 1 1 _ T D $ D I F F ] McKenzie, 2015a).
To extend prior variety recognition research conducted in Asian contexts, the present study examines how accurately and consistently Thai university students use their knowledge of variation within spoken English to categorise the geographical origin of speakers of nine forms of L1 and L2 English. Based on the findings of prior equivalent studies, the following hypotheses were developed: Hypothesis 1. Thai students will be able to differentiate between native and non-native speakers of English.
Hypothesis 2. The geographical origin of the Thai English speaker will achieve the highest rate of categorisation accuracy.
Hypothesis 3. The forms of English which are most geographically proximate and socio-psychologically salient will achieve the highest rates of categorisation accuracy.
In addition, through the use of a free classification measure, the study also aims to investigate Thai university students' patterns of categorisation and miscategorisation to gain a greater insight into the ideological frameworks surrounding English language diversity in Thailand. In-line with prior research findings the following prediction was made: Hypothesis 4. Thai students will frequently conflate the geographical origin of speakers of different forms of Asian English speech.
Finally, since existing research has indicated that listeners principally attend to segmental features of the speech stimuli to help identify speaker provenance, by requesting open-ended comments regarding the linguistic features responsible for participant categorisations, the following hypothesis was tested: Hypothesis 5. Thai students will most frequently identify phonological features to explain their categorisations of the geographical origin of speakers.

Participants
Data was collected from 191 undergraduate students attending three universities in Thailand: two well-known institutions located in Chiang Mai (n = 63, n = 38), Thailand's second city, and a high-ranking university in Bangkok (n = 90). At the time of the data collection, all participants had studied English for a minimum of 15 years, were continuing to study the language as a main subject or in conjunction with another discipline, and had attained at least upperintermediate English proficiency. The age range of the sample was between 19 and 27 (mean = 21.3, SD = 1.2). Reflecting the gender imbalance within the student population attending universities in Thailand more broadly ( , there were considerably more female (n = 146) than male participants (n = 43), with two individuals identifying their gender as neither male nor female. All reported their nationality as Thai and their L1 as Thai.

Speech stimulus
The stimulus consisted of recordings of nine specific forms of English speech produced by different female speakers, i.e., a verbal-guise design (see Garrett, 2010 McKenzie and Gilmore, 2017). During the pilot stage of the study, from a larger database of digital recordings made by the researchers, eight equivalent Thai undergraduate university students selected each of the nine speech samples as most prototypical (i.e., representative) (Dragojevic et al., 2017;Hogg and Reid, 2006) of each of the English speech forms in question. More specifically, for both the L1 and L2 speech samples selected, prototypicality for the particular English speech form were determined by at least four pilot study participants. It is also notable, that prototypicality did not seem to be influenced by participants' perceptions of the fluency of the speakers in question (see Dragojevic et al., 2017 for contrary evidence).
To control for passage content effects and to be able to present relatively lengthy spontaneous speech samples, all speakers were recorded giving directions on a winding road on the same fictitious map, from an imaginary start position to a castle, with a range of other locations/buildings en-route ([ 1 1 _ T D $ D I F F ] McKenzie, 2010) (see also below). In this way, it was possible to verify the speech contained no potentially confounding personal information regarding speaker provenance, educational background or nationality. In order to ensure naturalistic, spontaneous speech samples, map-task recordings of the nine forms of English were chosen as stimulus -instead of speech recordings employing a prescribed read-aloud text, which would likely feature incidences of spelling pronunciations, greater pausing at syntactic boundaries and unnaturally modulated stress patterns --and thus also capturing authentic and substantial phonetic, phonological, lexical, morpho-syntactic and pragmatic variation between the different speech forms (see Blaauw, 1994 2008c;Van Bezooijen and Gooskens, 1999). Whilst it was not feasible to develop a map-task which could elicit every linguistic feature prototypical of all nine English speech forms employed, particular locations and buildings were incorporated into the design of the map specific to elicit differences between the speakers' use of a number of phonological variables. These included differences in the articulation of postvocalic /r/ (church), H-dropping (hospital) and consonant clusters (bridge) as well as (lack of) phonemic opposition between /l/ and /r/ (lake), /w/ and /v/ (volcano) and /ɖʒ/[ 1 9 _ T D $ D I F F ] and /ʒ/ (bridge). As detailed in Fig. 1, the employment of relatively lengthy speech samples also allowed for tokens of suprasegmental, lexical, grammatical and pragmatic features, prototypical of the forms of English presented for categorisation, to feature in the speech stimuli.
Recordings of three native speakers of English were included in the study. Two of these varieties, Mid-West (Standard) US English and Scottish Standard English, were selected specifically to represent standard varieties of English spoken in the US and the UK respectively ([ 1 1 _ T D $ D I F F ] McKenzie, 2010) as well as to allow for meaningful comparison with the results of equivalent variety recognition research conducted amongst university students in Japan ([ 1 1 _ T D $ D I F F ] McKenzie, 2008a). In light of its relative geographical proximity to Australia to Thailand, a recording of General Australian English was also incorporated.
The stimulus also contained a number of forms of English spoken in Asia. More specifically, following [ 1 1 _ T D $ D I F F ] McKenzie et al. (2016), to complement recent language attitude research conducted amongst Thai university students, recordings of speakers of Thai English, Japanese English, Chinese English and Indian English were presented. A sample of Vietnamese English was included to provide a further example of English spoken in South-East Asia. Similarly, given the historical influence of France in Indochina more widely, albeit less so in Thailand itself, a sample of French English speech was included for categorisation. At the time of recording, all six L2 English speakers had achieved at least a Masters' level qualification, taught in English at a UK university, had attained an advanced level of English language proficiency and, as such, all were extremely fluent speakers of the language. In this way, the speech samples were controlled for potential differences in English language proficiency. All nine speakers reported no history of speech or hearing impairment. The speaker characteristics, including a description of the linguistic features prototypical of each of the forms of English in question, as observed by the researchers from the speech recordings, are detailed in Fig. 1.

Procedure
The data collection was undertaken, during regular class times, at the students' respective universities by two L1 users of Thai (Bangkok) and one L1 English speaker form the UK (Chiang Mai). In an attempt to assist the Thai students' awareness of world geography, as well as to highlight the global nature of English language use, participants were first given a map of the world, marked with national boundaries only. Subsequently, on presentation of each of the speech samples, participants were requested to write responses on a separate questionnaire to two questions (with Thai [ ( F i g . _ 1 ) T D $ F I G ]  Aitken, 1984;MacFarlane and Stuart-Smith, 2012;McKenzie, 2015a;Osburne, 1996;Peters and Burridge, 2012). translations of potentially problematic lexis provided if requested): (i) what is the speaker's native language; and (ii) which country do you think the speaker comes from?
To help determine the reasons behind, and especially the linguistic features responsible for, speaker provenance categorisations, participants were also requested to provide open-ended comments in response to the following: (iii) how did you make this decision; and (iv) identify any specific sounds, words or grammar of this speech. To control for any potential order effects, for each data collection session, the sequence in which the speech samples were played was randomised into one of four blocks. Following established procedure for a within-subjects experiment, all participants listened to each of the samples. During each data collection session, the speech stimulus was played through a high-quality audio system in the classrooms utilised. Participants heard each speech sample once only and completed the tasks individually.

Categorisation of speaker provenance
The first stage of analysis involved the calculation of participant recognition rates for the nine speakers as L1 or L2 English users. Table 1 reveals that whilst there were differences between classification rates, with the speakers from India, Vietnam and Japan most accurately identified as L2 English users, categorisation accuracy was generally high overall (i.e., between 64.4% and 97.9% accuracy). Consistent with Hypothesis 1, this result indicates, when presented with speech stimulus, Thai listeners are highly attuned to differences between native and non-native English speech. Notably, the categorisation of the Thai English speech as L2 English was comparatively low (70.7%). This finding contrasts with the results of similar research undertaken amongst Japanese university students ([ 1 1 _ T D $ D I F F ] McKenzie, 2008a), who were found to classify the English speech of a highly proficient Japanese national as 'non-native' with over 93% accuracy.
More fine-grained analysis of participant categorisations of speaker place of origin was subsequently undertaken. When compared to the correct identification rates for the speech samples as L1 and L2 English, Table 2 demonstrates there was greater disparity between accurate recognition rates for speaker country of origin. A markedly different ranking order for correct categorisations of L1--L2 English (Table 1) and speaker origin (Table 2) was also found. For instance, whilst the speaker from Vietnam achieved the second highest hit rate for correct L1--L2 English categorisations (96.9%),  despite the geographical proximity of Vietnam to Thailand, the geographical provenance of the speaker was least accurately identified (9.4%).

Patterns of categorisation and miscategorisation
In order to gain a deeper understanding of the ways in which the Thai university students employ their awareness of English language variation to differentiate between and identify the provenance of the nine speakers of English, as well as to uncover any specific patterns of categorisation and miscategorisation for each speaker, further fine-grained analysis of the large quantity of the categorisation data, constituting over 1800 responses, was undertaken (Tables 3--[ 2 0 _ T D $ D I F F ] 5). In addition, to identify the specific reasons underlying participant classifications for each of the nine speakers --with a secondary objective of identifying any overlapping patterns of responses between groups of speakers --thematic analysis was conducted on the participants' open-ended written comments for each of the speakers individually. For this reason, representative comments, quoted verbatim, illustrative of the main themes uncovered for each of the speakers are also presented below (participant number, age and gender in parentheses)[ 2 1 _ T D $ D I F F ] Table 3.

Thai English speaker
The highest percentage of categorisation accuracy, supporting the prediction made in Hypothesis 2, was afforded to the speaker from Thailand (60.2%), followed by the Japanese (58.1%) and the Indian speaker (52.9%). Participants who correctly categorised the geographical origin of the Thai speaker, perhaps unsurprisingly, often highlighted their familiarity with and, for some, the clarity of Thai English speech.  Listeners who were unable to accurately categorise the geographical origin of the Thai speaker (60.2%) frequently misidentified her as an L1 English speaker (28.1%), and most especially from the UK (15.7%) or the US (9.4%). Comments regarding the clarity of the Thai English speech again featured prominently amongst these listeners. The above comments indicate that for these Thai participants, both Thai English and L1 English speech may be associated with clarity, presumably, in the case of Thai English, as a result of regular face-to-face interaction with (other) Thai English speakers or, in the case of US English in particular, through frequent exposure to materials employed in English language classrooms or in the English language media. This finding may help explain the results of a recent language attitude study conducted amongst Thai university students where it was found --on both competence and warmth traits --Thai English was evaluated similarly to US and UK English and much more positively than other Asian English speech forms.

Japanese English speaker
A comparatively high percentage of participants (58.1%) were also able to categorise accurately the provenance of the Japanese speaker. A number of the Thai students remarked upon their relatively high levels of prior exposure to Japanese English, principally through the prevalence of Japanese films and J-Pop (Japanese popular music) where, in terms of the latter in particular, singers frequently employ a hybrid of Japanese and Japanese-accented English ([ 1 1 _ T D $ D I F F ] McKenzie, 2008b; Stanlaw, 2004). Given the large numbers of English-speaking Japanese workers and students resident in Bangkok, including those located in the 'Japan-town' areas of Thong Lo and Ekamai, large numbers of more short-term Japanese visitors, as well as the availability of cheap low-cost flights between Thailand's capital city and Japan, some Bangkok participants also mentioned their personal experiences of communicating face-to-face in English with Japanese nationals.
(022F-20) I think about Japanese people speak in the movie (161F-20) I listen to Japanese music (164F-21) I watch Japanese TV programs and they speak in English (028M-21) I used to have Japanese roommates, strong accent (041F-21) I have very good Japanese friends (047F-23) My father is Japanese (189F-22) I listen from tourist Several listeners who accurately categorised the speaker as Japanese also pinpointed distinctive phonological, prosodic and discourse features of her spoken English. These were often seemingly related to the influence of the lack of consonant clusters in the Japanese language sound system, and the associated tendency for L1 Japanese speakers to apply additional vowels to syllable final consonant segments --i.e., vowel epenthesis --when speaking in English (also known as Katakana-English, [ 1 1 _ T D $ D I F F ] McKenzie, 2008b). Other participants highlighted the lack of phonemic distinction between /r/ and /l/, the realisation of /s/ as [ʃ] and /t/ as [tʃ] in word initial position, e.g., 'see' and 'til', the insertion of Japanese fillers --especially aso and eto --as well as a higher pitch, reflecting politeness in the speech of Japanese females. A much larger percentage of listeners were able to accurately classify the Japanese English speech as L2 English (96.3%) and of those participants who could not categorise the specific geographical origin of the Japanese speaker, most perceived her to be from other countries in East Asia, South-East Asia or Asia more broadly (35.7%).

Indian English speaker
More than half of the Thai listeners accurately categorised the provenance of the Indian English speaker (52.9%) and some remarked upon the speech negatively. Several also commented upon their familiarity with Indian English through classroom language study with English teachers from India or by interacting face-to-face with members of the longestablished Indian communities which exist in the major Thai cities, especially in Bangkok (Peleggi, 2007), as well as increasing numbers of tourists from India. Again, the overwhelming majority of participants who were unable to accurately pinpoint the origin of the Indian English speaker correctly classified her to be an L2 speaker of the language (97.9%), most often from other areas within the Indian subcontinent, Turkey or the Middle East. Interestingly, analysis of participant comments indicated that those phonological features most frequently remarked upon in relation to correct categorisations of speaker provenance were also often mentioned with regard to incorrect classifications. This was particularly the case when the speaker was miscategorised as an L1 Turkish or Arabic speaker, languages where the fricatives /ð/ and /?/ are also generally absent from the sound inventory, and thus frequently realised as variants of /t/ or /d/ in spoken English (Sailaja, 2009) Table 4.

Mid-West US English speaker
The provenance of the speaker of standard US English was accurately categorised most frequently of the three L1 English speech forms (46.6%). This finding may be explained by the major economic, political and cultural influence, from the Indochina War onwards, which the United States has held within Thai society. Such influence has resulted in the widespread use of US English speech within the English language media in Thailand as well as the selection of (standard) Many participants who failed to recognise the place of origin of the US English speaker, instead, identified her as a user of another form of L1 English (28.1%), most particularly miscategorising her to be from the UK/England (16.2%) or Australia (11.4%). Whilst most of these listeners did not provide specific linguistic features, the findings point to recognition of the speech at a broader level.

Scottish Standard English speaker
A lower proportion of listeners accurately categorised the provenance of the Scottish English speaker (38.2%). This was the case despite the acceptance of both 'Scotland', 'the UK' or 'Great Britain' as successful categorisations. Some participants who correctly categorised the provenance of the SSE speaker focused on differences between the English spoken in Scotland and other countries in the UK and/or in the United States. A number highlighted the distinctive rhoticity of Scottish English or specific Scottish lexis, e.g., 'wee' and 'yer'.  Analysis indicated a comparatively low proportion of listeners accurately classified the Scottish English speech as L1 English (64.4%). Of those who did, many miscategorised the speaker to be from the United States (16.2%), Canada (4.7%), Australia (4.7%) or New Zealand (0.5%). Again, several participants mentioned the rhoticity of the speech. A sizeable proportion miscategorised the speaker to be from Europe, and France, Italy or Germany more specifically. This result is broadly similar to the findings of equivalent research conducted amongst Japanese university students ([ 1 1 _ T D $ D I F F ] McKenzie, 2008a), where a high percentage also miscategorised Scottish English speech as 'European'. In the present study, most participants who miscategorised the Scottish English speaker did not provide any linguistic features responsible for their incorrect classifications, in contrast with many of those participants who were able to classify accurately the speaker's provenance, again mirroring the findings amongst Japanese university students.

General Australian English speaker
Despite the relative geographical proximity of Australia, the proportion of Thai students who accurately categorised the Australian English speaker was the lowest of the three L1 English forms presented (16.8%). This finding is consistent with the results of a study conducted by Kristiansen et al. (2018), involving listeners from eight European countries, where accurate recognition rates for an Australian English speaker were also much lower when compared to rates for US English or Scottish English speakers. Furthermore, in the case of the present study, the vast majority of participants were unable to identify particular linguistic features responsible for their correct categorisations and, instead, often commented upon perceived differences between General Australian English speech and the English spoken in the UK and the US. However, several listeners mentioned the distinctive use of a high rising tone in declarative clauses by many Australians (Cruttenden, 2014), including in the General Australian English speech sample employed in the present study.
(110M-21) She has English as her native language but she's not from main areas (095F-22) She speaks very clear but not America (090F-23) Have friends from Australia (114M-22) The high pitch of the voice at the end of the words in mountain, airport, castle Participants who did not recognise the provenance of the Australian English speaker were, nonetheless, generally able to categorise her as a L1 speaker of English (82.2%). Although the majority of listeners were again unable to pinpoint any linguistic features, it may be the case that the large percentage who misidentified the speaker to be from 'England/the UK' (38.2%) were influenced by the phonological closeness between General Australian English and the English spoken in South East England (Cruttenden, 2014;Melchers and Shaw, 2011)[ 2 4 _ T D $ D I F F ] Table 5.

French English speaker
Analysis of the comments provided by the relatively low proportion of Thai listeners who were able to categorise accurately the provenance of the French speaker (25.1%) revealed a tendency to focus upon segmental features. These Whilst there was a general absence of comments relating to any specific linguistic features responsible for misidentifications, many participants who were not able to categorise the speaker as French recognised her to be from 'Europe' and most especially as an L1 speaker of a different Romance language. Several participants also incorrectly categorised the speaker to be from South-East Asia, perhaps reflecting the historic influence of French in countries neighbouring Thailand.
(075M-23) She pronounce like some language in Europe and say '''turn on the right[ 2 2 _ T D $ D I F F ] '' so not native (Italy) (149F-21) She sounded like a westerner but not a native speaker (Italy) (013M-22) good English but slight sound difference from original English accent (Portugal) (013M-22) She has an accent like my friends from Vietnam

Chinese English speaker
The task of categorising the origin of the Chinese speaker was also somewhat problematic for the Thai participants: perhaps surprising given the long-established (Thai-speaking) Chinese community dispersed throughout Thailand and the ever-increasing number of Chinese visitors and Chinese nationals studying at Thai universities (

sound is not normal
A sizeable percentage (10.5%) did not attempt to categorise the speaker's origin, suggesting a particularly high degree of confusion amongst these listeners. Most participants who miscategorised the speaker's provenance were again able to classify her as an L2 user of English (70.2%), and often from other nations in East Asia (20.4%) or South-East Asia (13.6%), thus providing evidence to support the prediction (Hypothesis 4) that Thai listeners often conflate the

Vietnamese English speaker
Despite the existence of a physical border between Thailand and Vietnam, the Thai students demonstrated most difficulty identifying the origin of the Vietnamese speaker (9.4%). Some listeners who were able to categorise accurately the speech remarked upon specific linguistic features, especially the deletion of consonants in word final position, particularly /s/ and /t/, as well as the overall monotone quality of the speech. Several other listeners seemed to categorise the Vietnamese speaker through a process of elimination of other forms of (Asian) English.
(069M-22) don't have t in airport (098M-23) no /s/ or /t/ at end of words (093M-23) monotone sound (183F-21) Vietnamese pronounce in the shorten way (011M-22) Sounds like she is from Asia, but not India, Japan, China (094F-23) Her accent sounds like an Asian speaking English The high proportion of listeners who failed to recognise the speaker's place of origin (90.6%), nevertheless, were generally able to categorise her as an L2 English speaker (96.9%) and, frequently, to be from another Asian country (81.2%). A number of participants were critical of the English speech and, again, several detailed specific phonological features responsible for their (mis)categorisations. A large proportion classified the speaker's origin as 'China', suggesting recognition of the influence of an L1 tonal language on the speaker's English.

General discussion and conclusion
The present study, employing a free classification task, sought to determine how accurately and consistently Thai users of English could map the place of origin of speakers of nine different forms of L1 and L2 English. Analysis of participant responses from the introductory 'native language' question revealed relatively high levels of categorisation accuracy of the speech as L1 or L2 English, with particularly high recognition rates for the speakers from India, Vietnam and Japan as L2 English users. As predicted (Hypothesis 1), the high degree of correct classifications thus demonstrates that the Thai listeners' awareness of the differences between native and non-native English speech, whether at conscious or unconscious levels, is both perceptually robust and paramount in the categorisation process. In turn, the analysis points to a strategy of initially differentiating between 'native' and 'non-native' English speech, before attempting more precise categorisations regarding the geographical origin of each speaker. This result --the first to be conducted amongst Thai/ South-East Asian listeners --thus replicates the findings of prior equivalent research, investigating the categorisations of English speech stimulus, involving listeners from the US Bent and Holt, 2017) (Kristiansen et al., 2018) and South Korea/Spain .
Moreover, in the case of the present study, the qualitative comments indicated a propensity for the Thai participants to classify the Japanese, Vietnamese, Chinese, Indian and, to a lesser extent, French English speech as forms of L2 English precisely because they were perceived to be unclear and/or incorrect. In contrast, in the case of the categorisations of the US, Scottish and Australian English speech as L1 English and, notably, the Thai speech as L2 English, the Thai students frequently highlighted the general fluency, clarity and correctness of the speakers' English.
The above comments indicate that when Thai listeners were requested to identify the geographical origin of the speakers of English, the categorisation process also frequently led to the indexing of existing stereotypes regarding the competence of each speaker and, in turn, of the speech itself. This finding offers support to speaker evaluation theories within language attitude research, where it is believed (social) categorisation processes play a central role and, upon the presentation of speech stimuli, the act of classifying a speaker is thought to activate evaluative reactions regarding their social and ethnic group membership (see Dragojevic et al., 2018).
With regards to more fine-grained classifications, substantial differences were found between Thai university students' ability to categorise the place of origin, to varying degrees of specificity, of the nine speakers of English. Given Thai students are likely to have received more exposure to Thai English than other forms of English, especially within English language classes at school and university, it is perhaps unsurprising, and consistent with expectations (Hypothesis 2), the geographical origin of the Thai speaker was accurately identified most frequently. As detailed above, comments regarding the Thai English speech also tended to be broadly positive. Participants most frequently remarked upon the intelligibility and/or clarity of the speech, amongst both listeners who identified, and those who failed to recognise, the speaker's place of origin. Such comments suggest favourable evaluations of, and thus a high degree of linguistic security regarding, the form of English Thai students are themselves likely to speak (see also [ 1 1 _ T D $ D I F F ] McKenzie et al[ 1 6 _ T D $ D I F F ] . 2016). Furthermore, since participant comments regarding the Mid-West US English and Scottish Standard English speech also tended to be favourable, the broadly positive comments afforded to the Thai English speech may help explain the relative frequency with which the Thai speaker was categorised incorrectly to be from the US or the UK.
In support of the prediction made in Hypothesis 3, the Thai students also frequently miscategorised the English speech samples provided by the speakers from Japan, China and Vietnam to be from other countries in East Asia or South-East Asia. This result is especially noteworthy since the English spoken within, and between, these broad areas is phonologically and phonetically very different. For example, largely as a result of L1 influence, there exists substantial variation between the forms of English spoken in different Asian nations in terms of vowel length and the employment of syllable-timed or stress-timed rhythms ( Deterding, 2013). In the case of the present study, whilst the possibility exists that the tendency for listeners to conflate English speakers from different Asian nations is influenced, at least to some extent, by a relative lack of geographical knowledge, the results are of importance since they indicate that (Thai) L2 English users, as well as L1 English users in the US McGowan, 2016) and the UK ([ 1 1 _ T D $ D I F F ] McKenzie, 2015a), frequently group Asian English speakers together. That is to say, Thai listeners, whether below or above the level of individual awareness, may also utilise broader, undifferentiated stereotypes regarding (speakers of) 'Asian English', conceptualised as single entity, to categorise (or not) speaker place of origin solely from speech stimulus.
In the specific case of the English speech samples provided by the Chinese, Vietnamese and, to a lesser extent, Japanese speakers, analysis of participant comments provides evidence that phonological and phonetic features found in all three samples may be particularly responsible for the frequent miscategorisation of the three speakers to come from other countries in East Asia or South-East Asia. More specifically, participants most frequently mentioned a lack of opposition between /l/ and /r/, a tendency to delete consonants in word final position and/or perceived errors in word stress and intonation more broadly.
As discussed above, differences were also found between categorisation accuracy rates for the regional provenance of the US, Scottish and Australian speakers. Despite the relative geographical proximity of Australia to Thailand, as well high rates of correct categorisations of the Australian speech as L1 English, listeners generally found the place of origin of the Australian speaker difficult to pinpoint. A considerably lower proportion of participants were able to identify the speaker of Vietnamese English. Thus, since the most localised speakers were not always the most frequently categorised, the geographical proximity of the place of origin of the speakers to Thailand did not seem to be the most important factor in the recognition process. Instead, the analysis points to the relative social-psychological proximity of the English speech forms presented to the Thai listeners as the strongest determinant of categorisation accuracy. Indeed, when the categorisation rates for the eight (non-Thai) English speech samples were compared, the Japanese, Indian, US and UK speakers were the most accurately identified. Hence, Hypothesis 4, that geographical proximity and socio-psychological salience of the English speech forms would both be important determinants of speaker categorisation accuracy rates, was only partially supported. This finding is important since it extends prior research involving native speaker categorisations and, to the best of our knowledge, the present investigation constitutes the first speaker categorisation study to demonstrate amongst L2 users (of English) that the social-psychological salience of the speech forms is also paramount in determining their levels of accurate identification. Analysis of participant comments also indicated, for the Thai listeners, greater levels of awareness of those forms of English was attained largely through an accumulation of high levels of exposure within the Thai broadcast media, face-to-face contact with speakers and, in the case of US English, through English learning materials employed in English language classrooms at schools and universities in Thailand.