ENGLISH INTERDENTAL FRICATIVE PRODUCTION IN DUTCH HERITAGE SPEAKERS LIVING IN CANADA

Headnote

Abstract

This study investigates the production of /0/ and /ð/ by three groups of English-speakers in the community of Norwich, Ontario, Canada. English monolinguals, heritage Dutch speakers (early bilinguals), and L1 Dutch/ L2 English speakers (late-learning bilinguals) /0/ and /ð/ production was measured in both naturalistic and reading tasks. Heritage Dutch speakers produce [0] and [ð] at similar rates to Monolingual English speakers, but the two groups exhibit different allophonic realizations, especially when /ð/ is word-initial and /0/ is word-medial. his study suggests that despite their ability to produce [0] and [ð] , Dutch heritage speakers may manipulate the inherently variable English /0/ and /ð/ production to communicate their Dutch cultural identity. his is the first study to examine both heritage Dutch bilinguals in Canada and non-English as a Foreign Language (EFL) classroom Dutch English.

Keywords: Heritage phonology; Bilingualism; L2 speech learning; Socio-phonetics; Interdental fricatives; Dutch; English

1.0 Introduction

This study aims to add to the body of literature on heritage speakers' phonology by examining how heritage Dutch speakers' English [0] and [ð] production may differ from both 'typical' late L2 learners and native speakers of English. In particular, the purpose of this study is to examine both the type and rate of L1-based phonological transfer. Transfer in this case is the way in which Dutch phonology may affect the acquisition, and specifically the production, of English. Although more work has been done on L2 production (e.g. Best and Tyler (2007); Colantoni and Steele (2008); Flege (1995)), there is a growing body of literature on heritage-speaker phonology (Hoffman & Walker, 2010; Chang, Yao, Haynes, & Rhodes, 2011; Rao & Kuder, 2016).

Dutch has no [0] and [ð] in its phonology, and research has shown that even highly proficient speakers often produce these sounds in English in a non-targetlike manner (Schmid, Gilbers, & Nota, 2014; Wester, Gilbers, & Lowie, 2007). he literature on Dutch L1 English learners has mainly focused on L2 speakers who acquired English in an English as a Foreign Language (EFL) context and little is known about learners who learn English outside of an EFL classroom. Moreover, most studies on immigrant Dutch bilinguals have focused on attrition of Dutch, rather than the acquisition of English (c.f., Crezee (2012); Hulsen (2000); Hulsen, de Bot, and Weltens (2002)). he township of Norwich, Ontario, provides an opportunity to study the English of Dutch bilingual immigrants and heritage speakers, as 9.2% of the population reports Dutch as their mother tongue (Statistics Canada, 2012), and approximately 20% of residents are of Dutch descent (Schryer, 1998, p. 75). he research in this study will focus on these two population groups and attempt to answer the following questions:

1. Do heritage speakers of Dutch differ from monolingual English speakers and late-learning L1 Dutch L2 English bilinguals in their realization of [0] and [ð]?

2. Is Dutch heritage speakers' [0] and [ð] production more similar to those of the late-learning L1 Dutch bilinguals or monolingual English speakers?

2.0 Literature Review

2.1Dutch Presence in Canada

Studies of immigrant language use suggest that Dutch immigrants tend to assimilate quickly in Anglophone societies (Harrison, 2000; Hulsen, 2000). As a result, previous studies on Dutch immigrants and their descendants have largely focussed on the attrition of their L1, rather than their acquisition of English. In Canada, 34% of the Dutch immigrant population married someone whose native language is English or French, and only 1% taught Dutch to their children as a first language (Harrison, 2000). Only 25% of Ontarians who speak Dutch as a mother tongue continue to speak Dutch in the home (Schryer, 1998, p. 163). VanDijk notes that while Dutch Calvinists (like many in Norwich ) are more likely to resist language assimilation in comparison to Dutch Catholics or those of other religions, Calvinist rates of Dutch language use are only marginally higher than other groups (2007). he overall high degree of language assimilation may result in Heritage speakers producing [0] and [ð] in a manner more similar to their English monolingual peers than to late-learning Dutch L1 speakers.

2.2 L1 Dutch L2 English Phonology

In general, L1 Dutch/L2 English speakers can attain a high degree of fluency in English. Schmid et al. (2014) found that high-fluency Dutch L2 English speakers did not differ significantly from English speakers on c-test fluency tasks. However, L1 Dutch speakers' speech is still recognizably accented, in both vowel (Schmid et al., 2014) and interdental fricative production (Wester et al., 2007). Wester et al. (2007) looked at the realizations of [0] and [ð] in naturalistic English speech by highly fluent L1 Dutch/L2 English speakers. hey found that [0] and [ð] were most commonly realized as [t] and [d] in word-initial position and as [s] and [z] in word-final position. hese realizations follow a different pattern than those of English L1 [0] acquisition, which tends to favour [f] (Wester et al., 2007). Wester et al. (2007) attribute this to [t] and [d] being more unmarked sounds (they have fewer features that differentiate them from [0] and [ð] in Dutch phonology). L1 Dutch/ L2 English speakers also confuse [0] and [ð] perceptually, commonly confusing [0] for [f] and [s] and [ð] for [d], [z], and [s] (Cutler, Weber, Smits, & Cooper, 2004). hese patterns of [ð] and [0] production and perceptual confusion suggest that we will find some of these patterns of variable [0] and [ð] production in Norwich's late learning L1Dutch/ L2English bilinguals and Dutch heritage speakers.

2.3 Heritage Phonology

Chang et al.,(2011) compared the production of similar (but distinct) Mandarin and English vowels, plosives, and fricatives by four different groups: native Mandarin speakers, heritage Mandarin speakers with low and high Mandarin exposure, and late Mandarin learners/English native speakers. Of the four groups, both heritage speaker groups produced the similar phonemes with stronger between-language contrasts than the late-learning groups. he heritage groups also maintained strong within-language contrasts. Chang et al. (2011) use Flege's Speech Learning Model (SLM) to explain this. According to the SLM, learners acquire the phonology of their second language by creating contrasting categories for sounds, which are highly dissimilar to those in their native language (new sounds), but they tend to merge similar sounds with categories for their native language. his merger is more likely to occur as the age of acquisition increases. hat is, late learners will be more likely to merge similar sounds. Heritage speakers acquire the sounds of both languages simultaneously and at a young age, so they are more likely to create more distinct and accurate between-language contrasts. As a result, they tend to produce native-like sounds for both of their languages, outperforming late-learners.

Hoffman and Walker (2010) investigated the ethnolects of Chinese and Italian heritage speakers in Toronto. heir findings suggest that differences between monolingual's English and heritage speakers' English are due to group identity rather than learning L2 speech patterns from their parents. Gatbonton, Trofimovich, and Segalowitz (2011) found similar ethnic-identity effects for French L1 English speakers' /ð/ production. L2 speakers who viewed French language maintenance as an important political statement produced significantly less [ð] in their English productions, even when accounting for English proficiency. In their wideranging survey of UK immigrants' intergenerational language use, Parameshwaran (2013) found that while the ethnic or heritage language underwent attrition over successive generations, English (majority language) proficiency was generally native-like for all people raised in the UK. Among Heritage groups, a positive attitude towards assimilation correlated with higher levels of English fluency. Calvinist Dutch Heritage speaker communities have a strong sense of Dutch identity, but they generally prefer to speak English (Van Dijk, 1998). his suggests that while the Dutch community in Norwich has largely linguistically assimilated to the English-language majority, they maintain a strong Dutch identity. hese studies suggest that heritage speakers with stronger cultural ties to the Dutch community may exhibit some "Dutch-like" features in their production of English, or may show some transfer from their Dutch for this reason.

As a counterpoint, there are rare examples of ethnic or ethno-religious identity affecting majority language use even in monolingual English speakers. A North American example of this is the Pennsylvania Dutch Anabaptists (Downing, 2015; Huffines, 1980; Parker, 1991). his group is comprised of ethnically German Old-Order Amish and Mennonite populations living not just in Pennsylvania, but also in Ohio and Southern Ontario. Despite the first migrants arriving in the late nineteenth century, these populations have maintained a dialect of heritage German that exists today (Huffines, 1980). While the majority of young people in these communities are monolingual English speakers (Huffines, 1980), these speakers still "exhibit a large influence of the Pennsylvania German dialect in their speech" (Parker, 1991, p. 23). his influence differentiates the English of the Pennsylvania Dutch from the majority English of the United States (Parker, 1991). Consonantal features include /ð/ stopping, the realization of /0/ as [s], and final obstruent devoicing (Downing, 2015; Huffines, 1980). While these features are similar to German L1 English speakers, Pennsylvania Dutch Anabaptists are largely monolingual English speakers, so L2 theories (like the SLM) cannot be used to describe their speech. Instead, Parker (1991) explains that these English productions are a method of labelling themselves as a member of the Anabaptist community, and therefore, excluding themselves from secular English American culture. While the Dutch Calvinist community in Norwich is not as insular as the Anabaptists, Schryer (1998) notes that the Orthodox Calvinist church in Norwich is one of the two most conservative and insular Dutch church sects in Ontario (the other is in Ancaster). his insularity has resulted in strong social networks that may help to explain the survival of Dutch English features in Norwich Township heritage speakers. his means we may find an increased rate of typical Dutch L1 features in heritage English production. In the context of this study, this could be seen as an increased rate of [d] production as an allophone of /ð/, and [t] and [s] as allophones of /0/.

2.4Monolingual production of the English Interdental Fricative

Flege, Munro, and MacKay (1995) measured the word-initial production of English /0/ and /ð/ by Canadian English monolinguals, and Italian L1 English speakers who began acquiring English at different ages. hey found that both the monolinguals and Italian L1 who acquired English at or before age 8 produced [0] and [ð] in approximately 70% of tokens. hose who acquired English later in life produced both sounds at lower rates - speakers who acquired English as adults only produced [0] and [ð] in 20% of tokens. Speakers who acquired English at or before age 8 (both monolinguals and Italian L1) realized /0/ as [t] and /ð/ as [d] in the majority of the remaining tokens (Flege et al., 1995). his study shows us that /0/ and /ð/ production is variable even within monolingual English speakers.

Zhao (2010) investigated /ð/ production and also found that English monolinguals did not produce [ð] as a fricative in 100% of /ð/ target tokens. [ð] was sometimes realized as "stop-like ð", especially after a pause, vowel, liquid, or fricative. "Stop-like ð" is different from [d], having a significantly higher burst spectrum peak and mean spectral frequency, as well as a significantly weaker and flatter spectral burst shape. Roach even goes so far as to deny that /0/ and /ð/ are fricatives, preferring to describe them as "weak dental plosives" (2010, p. 56). he high degree of similarity between monolingual English production of /d/ and /ð/ suggests that these sounds could be merged in L2 English phonologies under Flege's SLM.

/0/ and /ð/ are not present in all English dialects or discriminated by all native English speakers (Munro & Derwing, 2006). /0/ and /ð/ have been found to be supplanted by /d/ and /t/ in Newfoundland by Boberg (2010), in AAVE by Labov (cited in Gatbonton et al. (2011)), New York City by Best and Tyler (2007) and in Belfast by Milroy (cited in Gatbonton et al. (2011)). /0/ has also been frequently supplanted by /f/ in many British (Blevins, 2004; Hanulikova & Weber, 2012), Australian, and New Zealand dialects (Jekiel, 2012) as well as word-finally and inter-vocalically by AAVE speakers (Jekiel, 2012). /0/ is also frequently confused by English monolingual children - the [0] - [f] contrast is one of the few English consonant contrasts that infants struggle to discriminate (Vihman, 1996 in Blevins (2004)). he interdental fricatives are some of the last sounds to be acquired by North American English speaking children (Bernhardt & Stemberger, 2002). In General American English, /0/ is usually acquired around age 6, while /ð/ is acquired around age 7 (Smith, 2007). Prior to that age, it is common for /ð/ to be produced as [d] and for /0/ to be produced as [f] or [s] (Smith, 2007). he minimal contrastive function of /0/ and /ð/ allows these sounds to be produced variably by English speakers of many different dialects.

3.0 Hypotheses

It is predicted that:

1. he three groups will differ and fall along a continuum, with English speakers producing the highest rate of [0] and [ð]. Late bilingual Dutch immigrants will experience some transfer from their L1, and so will produce the lowest rate of [0] and [ð] (as in Wester et al., 2007). Heritage speakers will produce [0] and [ð] at a rate close to, but lower than the English monolingual speakers (following findings by Chang et al. (2011); Hoffman and Walker (2010)).

2. All three groups will exhibit allophonic variation for [0] and [ð] production. However, the patterns of allophonic distribution will differ between the three groups. he Dutch heritage speakers' patterns of [0] and [ð] allophonic production will be similar to those of the Dutch L1 speakers as reported in Wester et al. (2007). Specifically, they will tend to produce [d] and [t] as allophones at a significantly higher rate than the [s] and Ø (zero) allophones which characterize monolingual English speakers.

4.0 Method

4.1Participants

21 participants were recruited for this study. he participants were between age 20-80, with a mean age of 49. he participants were divided into three groups (LI Dutch/ L2 English bilinguals, heritage Dutch bilinguals, English monolinguals) according to their responses to the Dutch and English fluency portions of the proficiency questionnaire task. All of the participants were recruited from the Township of Norwich, Ontario, Canada. Two of the English monolingual group members were not raised in Norwich, but were raised in other Southwestern Ontario communities.. Table 1 shows the demographic data of the participants (see Appendix A for a detailed participant table).

The experiment consisted of three production tasks. First, all participants were required to complete the Language Experience And Proficiency Questionnaire (LEAP-Q) (adapted from Marian, Blumenfeld, and Kaushanskaya (2007)). his survey asks about basic demographic information and relative experience and proficiency with both English and Dutch. he results of the LEAP-Q were used to sort the participants into groups based on the ages at which they learned Dutch and English, and their relative language proficiencies. Participants who preferred Dutch, ranked their Dutch fluency as equal to or greater than their English fluency, and who learned English as adults were put in the 'L1 Dutch/L2 English' group. Participants who spoke both languages but learned English as children, who prefer English, and who rank their Dutch fluency as equal to or less than their English fluency were put in the 'Heritage' group. he 'English' group consists of self-reported English monolinguals. Due to being educated in Canada, a few English participants reported low proficiency in French. A copy of the LEAP-Q is included in Appendix B.

The second task was a naturalistic production task. he participants were audio-recorded as they completed a picture description task. he participants were asked to describe each of the pictures in detail, "as if [they were] describing it to someone who was attempting to draw it", hey viewed each of the 3 pictures as they were presented one-by-one on a computer screen. he task was self-paced, and took participants a mean of 3 minutes to complete. his task was designed to elicit natural production.

The third task was a reading task. All of the participants were audio-recorded as they read aloud a paragraph that contains target /0/ and /ð/ words. he task was self-paced, and took less than 2 minutes for participants to complete. his task ensured that there were at least 5 tokens from 5 unique words demonstrating each position (initial, medial, and final) and each sound (/0/ and /ð/) from each participant.

4.3Stimuli

In the naturalistic production task, one picture contained details designed to elicit 31 words containing /0/ and /ð/ (seeAppendix C). Due to the naturalistic nature of this task, all 31 words were not produced by each participant. More balanced data was recorded from the reading task. he reading task paragraph contained 31 target words for /ð/: 21 word-initially (i.e., though /ðo/), 5 word-medially (i.e., lather /læ ðsj/), and 5 word-finally (i.e., bathe /beð/). It also contained 31 target words for /0/: see Appendix D.

4.4 Data Analysis

The tokens were auditorily transcribed and coded by the English-native author working in Praat for Mac v5.4 (Boersma, 2013). Each token was labeled as either [0], [ð], [t], [d], [s], Ø (zero), or other. A spectrogram produced by Praat was used to confirm segment labeling. Specifically, short duration productions of /ð/ were checked for evidence of "gaps" that would indicate a closure ([d]). A multiple regression analysis was done in Rbrul (Johnson, 2009) to determine which factors significantly affected [0] and [ð] production rates and patterns of allophone realization. he factor variables included in the analysis were: group (English, Heritage, L1 Dutch/L2 English), task (picture description, reading), gender (male, female), and p osition in the word (initial, medial, final). Two continuous variables were also included: age and word frequency. Word frequency was calculated using the logarithmic frequency from the SUBTLEXus database (Brysbaert & New, 2009) as accessed through the English Lexicon Project's website (Balota et al., 2007). Individual speakers were accounted for as a random effect to prevent any one participant from having possible outlier effects.

5.0 Results

5.1 Voiced Interdental Fricative

A mixed-effect variable rule statistical analysis done in Rbrul showed that /ð/ production varied significantly due to group affiliation (p<0.001), position in the word (p<0.001) and task (p<0.05). he log-odds table is shown in Table 2. Gender, age, and word frequency were not significant predictors of /ð/ production. No interactions between variables were found. he results of each task are discussed separately below.

5.1.1 Picture Description Task

The participants produced an average of 45 /ð/ tokens across a total of 38 unique /ð/ words. English monolinguals produced the highest number of tokens (51.8% of all tokens), which was especially due to two speakers who spoke for 8 and 9 minutes and produced over 180 tokens each. he average token count for the English group with these speakers removed remained higher than the other two groups at 66.5 tokens. he full breakdown of both /ð/ and /0/ token counts and the number of unique words used by each group is presented inTable 3.

The most between-group variability was observed at the word-initial position of /ð/. he groups' production was significantly different from one another (p<0.001). /ð/ also had the most numerous tokens (n=893) due to the high proportion of word-initial /ð/ function words in English. English monolinguals produced /ð/ as [ð] in 85% of tokens (n=484). he remaining tokens were realized as Ø (9%), [9] (3%), and [d] (2%). he L1 Dutch/ L2 English group produced [d] at 60% usage, while also producing [ð] in 20% of tokens. he remaining 20% were produced as Ø (n=191). he Heritage group's rate of [ð] production was between the other two groups, with [ð] produced in 62% of tokens (n=218). Similar to the L1 Dutch / L2 English group, the Heritage group produced [d] as the most common allophone of /ð/, producing [d] in 25% of tokens. he remaining productions were more similar to the English group, with 12% of the tokens realized as Ø, and the remaining 1% as [9].

The three groups produced a total of 58 medial /ð/ tokens across 16 unique words. Both the English and Heritage groups produced [ð] very consistently at this position. he English group produced 92.9% tokens as [ð], 3.6% as [9], and 3.6% as Ø (n=28). he Heritage group produced 15 tokens, of which 93.3% were [ð] and 6.7% were [d]. he L1 Dutch/ L2 English group produced 81.3% [d] and 18.8% [ð] (n=16). Only a single /ð/ token was recorded in the word-final position for this task - an English monolingual production of [ð] in "bathe".

5.1.2 Reading Task

The task effect for /ð/ shows a tendency for all three groups to produce more varied realizations of /ð/ in the reading task. Word-initially, the /ð/ reading task results were very similar to those of the picture description task, though overall the groups produced a lower proportion of [ð] tokens. he English group produced the most [ð] at 73%. he remaining realizations were Ø (12%), [9] (12%), and [d] (3%). he Heritage group produced less [ð] (54%) and unlike the English group, favoured [d] (21%). he remaining tokens were very similar to the English group and were split between Ø (15%) and [9] (10%). he L1 Dutch/ L2 English group successfully produced [ð] in only 15% of tokens at this position, instead producing [d] in the majority of tokens (63%). he remaining tokens were Ø (15%), [9] (3%), [t] (1%), and other (2%).

Word-medial /ð/ words showed more variation than in the picture description task. he English and Heritage groups both produced a majority of tokens as [ð]. hese two groups produced a similar pattern of allophonic variation - using [9], [d], and Ø at similar rates, though the Heritage group produced more [d] than the English group. he L1 Dutch/ L2 English group produced [d] in the majority of tokens (58%), but also produced [ð], [9], [d], [s], Ø, and others.

The standard [ð] production was low in all three groups in final position. he English and Heritage groups both produced [ð] in about 45% of tokens. he English group frequently devoiced [ð] in this position, producing [0] in 43% of tokens. English monolinguals also rarely produced [s] (7%) and [d] (3%). he Heritage group also produced [0], but at a lower rate (30%). his group realized /ð/ as [d] at the highest rate of all three groups (26%). he L1 Dutch / L2 English group was again more variable than the other two groups, and produced [0] in the largest number of tokens (42%). he L1 Dutch/ L2 English group produced [ð] in 24% of cases, while also substituting with [d] (18%), [s] (8%), and other sounds (8%).

5.2 Voiceless Interdental Fricative

A mixed effect variable rule statistical analysis done in Rbrul showed that /0/ production varied significantly as a result of group affiliation (p<0.01) and position in the word (p<0.01). he log-odds table is shown in Table 4. Task, gender, age, and word frequency were tested for, but were not significant factors. here were no significant interaction effects. here was not a significant task effect for /0/ production, so the position and group data presented in this section represent the results for both tasks. Overall, all three groups exhibited a tendency to produce /0/ as [0]. his sound was produced with greater accuracy and with less allophonic variation than /ð/.

Word-initially, all three groups produced [0] in the majority of tokens. he Heritage group (n=97) and English group (n=114) both produced [0] very regularly, producing [0] in over 87% of tokens. he L1 Dutch/ L2 English group produced [0] in 53% of tokens (n=155). he most common allophone of /0/ for the Heritage and English groups was [s]. [s] occurred in approximately 6% of both groups' tokens. he L1 Dutch/ L2 English group also produced [s] in 9% of tokens, but preferred to substitute with [t], which was found in 31% of tokens. he Heritage group also produced [t] as a realization of /0/, but more infrequently than the L1 Dutch/ L2 English group (5%). he English group produced both [t] and [ð] extremely rarely, in only 1 token each (<1%).

The three groups also produced [0] in their majority of medial /0/ tokens. he Heritage group (n=42) and English group (n=48) each produced [0] in about 80% of tokens, while the L1 Dutch/ L2 English group (n=57) produced significantly less [0] at a rate of 50%. he next most frequent allophone in the Heritage and English groups was [ð]. While this voicing occurred in 14% of Heritage tokens and 10% of English tokens, in both groups [ð] was confined to only two words: "something" (/sAm0ip/ as [sAmðip] or [sAmðin]) and "healthy" (/hel0i/ as [helði]). he L1 Dutch/ L2 English group produced the voiced variant in only 2 tokens (4%). he remaining productions by the Heritage and English groups were minimal. he English group realized /0/ as Ø (4%), [t] (2%), and [s] 2%, while the Heritage group produced [t] (5%) and [s] (2%). As in other positions, L1 Dutch/ L2 English group was the most highly variable. he L1 Dutch/ L2 English group's most common realization of /0/ was [s] (18%), though they also produced [t] (16%), [d] (5%), Ø (2%), and other (4%).

Word-final /0/ production was very similar to the word-medial position. All three groups produced [0] in the largest proportion of their productions. Both the Heritage group (n=116) and English group (n=144) produced [0] in over 81% of tokens, while the L1 Dutch/ L2 English group (n=135) produced [0] in 42% of tokens. he English group produced a non-[0] variant in only 13% of tokens. he preferred English group variants were [s] (6%) and [ð] (4%). Infrequent English group variants included [t] (2%) and Ø (1%). he Heritage group patterned in a similar way, but preferred [t] more heavily, producing it in 12% of tokens. Other Heritage group variants included [s] (4%), [ð] (2%), and Ø (1%). he L1 Dutch/ L2 English group substituted /0/ most frequently with [s] (28%) and [t] (22%), while also using [d] (3%), Ø (2%), and other sounds (2%) more rarely. A summary of the results for both /ð/ and /0/ is shown inTable 5.

6.0 Discussion

The first hypothesis was that the three groups' production of [0] and [ð] would differ with the L1 Dutch/ L2 English group producing the lowest rate of [0] and [ð], the English monolinguals the highest rate, and the Heritage group producing a relatively high rate that was lower than the English group. his hypothesis was supported. All three groups' productions of /0/ and /ð/ were significantly different from one another. Group affiliation was the strongest contributor to both /0/ and /ð/ production. /ð/ showed a strong group effect with a log-odds range of 3. /0/ was similarly strong with a log-odds range of 2.546. he English group consistently produced [0] and [ð] at rates equal to or higher than the Heritage group. he two groups' productions of /0/ were more similar than those of /ð/, but were still significantly different. he English and Heritage groups' production of /ð/ was most differentiated wordinitially. he English group produced /ð/ on average 26% more than the Heritage group at this position (21.9% mean difference over /ð/ tokens in all positions). /0/ production was most differentiated at both the initial and final positions with a mean 6% difference in production (5.2% mean difference across all tokens). As predicted, the L1 Dutch/ L2 English group produced significantly less [0] and [ð] than both the Heritage or English groups. Overall, the three groups fall along a continuum as predicted. he English and Heritage groups both produced high levels of [0] and [ð], with the English group producing more [0] and [ð]. he L1 Dutch/ L2 English group was highly dissimilar from both of the other groups, and produced significantly less [0] and [ð] than either of the other groups.

The effect of word position was also a significant predictor for the production of both /0/ and /ð/. While significant, the effect was relatively weak with log-odds ranges of 0.778 for /ð/ and 0.688 for /0/. For /ð/, [ð] production was preferred in the word-medial and word-initial positions and dis-preferred in the word-final position. Lower word-final [ð] production is attested in the literature, as utterance-final voiced fricatives are commonly devoiced in English (Hayes, 2009, p. 93). For /0/, [0] production was only preferred word-initially, while both word-medial and word-final position weakly dis-preferred [0].

The weakest factor in the results was the task effect. Task significantly contributed to /ð/ production, with a relatively weak log-odds range of .556. Unusually, [ð] production was more highly dis-preferred in the reading task, which also showed higher variability in /ð/ production than the more naturalistic picture description task. his is a reversal of the more common sociolinguistic pattern of more standardized and regular production during reading tasks. his could be due to the relative difficulty of producing [ð], combined with the high number of /ð/ and /0/ tokens per utterance in the paragraph (mean of 1.8 /ð/ tokens per utterance and 3.6 /ð/ and /0/ tokens per utterance). his gave the paragraph somewhat of a "tongue-twister" effect that may account for the significant increase in /ð/ production variability on the reading task.

All three groups produced [0] at higher rates than [ð], suggesting that /0/ may be less marked or easier to produce than /ð/. his distinction is mostly overlooked in the literature, as the two interdental fricatives are generally treated as a single category (c.f. Flege et al. (1995); Gierut and Storkel (2009); Jekiel (2012)). Where /0/ and /ð/ are differentiated, as in Hancin-Bhatt (1994, in Brannen (2011)) and Cutler et al. (2004), it was found that [0] was more difficult to discriminate than [ð]. he results of Wester et al. (2007) study of Dutch L1 English /0/ and /ð/ production correspond with those of the current study - [0] was produced at higher rates than [ð]. However, Wester et al. did not comment on this differentiation in their study. /0/ being less marked than /ð/ fits with the universal pattern of voiceless obstruents being less marked than their voiced counterparts (Major & Faudree, 1996).

The second hypothesis was that the allophonic realizations of /0/ and /ð/ would differ between the three groups. It was predicted that both the L1 Dutch/ L2 English and the Heritage speakers would realize [t] and [d] at a high rate, while the English group would produce Ø and [s] at a higher rate. his hypothesis was partially supported. he groups did show different allophonic inventories for /0/ and /ð/. he English group's most commonly realized allophones for /ð/ were [0] and Ø. Monolingual English allophones for /0/ were [s] word-initially and finally and [ð] word-medially. he variable voicing of /0/ and /ð/ has been explained as more dependent on position and part of speech rather than a phonemic voicing difference (Smith, 2013). he Smith (2013) findings were supported in this study because, as mentioned in section 5.2, production of /0/ as [ð] was lexically restricted. Heritage speakers generally followed this model of allophone realization, but not at the same rate. he Heritage speakers' patterns of /0/ and /ð/ allophone realization were most dissimilar to English monolinguals at /ð/ in word-initial position, and for /0/ in word-final position. hese positions showed variation more similar to L1 Dutch/ L2 English speakers than to English L1 speakers. At these positions, the Heritage group frequently produced the alveolar stops [d] and [t] in a way which was similar to the L1 Dutch/ L2 English group and very dissimilar to the English speaker group. Word-initial /ð/, was produced as [d] in 23% of Heritage tokens and 62% of L1 Dutch / L2 English tokens, but in only 3% of English tokens. Word-final /0/ was produced as [t] in 12% of Heritage speaker tokens, while the L1 Dutch /L2 English group produced [t] in 26% of tokens and the English group in only 2%. hese two patterns show that in these two positions, the Heritage group's allophonic realization of /ð/ was very unlike the native speakers of English. he Heritage group was not identical to the L1 Dutch/ L2 English group either. he L1 Dutch/ L2 English group's productions were much more highly variable than the other two groups. Late-learning L1 Dutch/ L2 English speakers exhibited greater variability in all positions, and never produced an allophone more consistently than 62% of tokens. he L1 Dutch/ L2 English group was also the only group to produce the "others" category sounds [f] and [ts], and the only group to produce "others" sounds for both /0/ and /ð/. he only other "others" productions were by the Heritage group: 3 /ð/ tokens that were produced as [z].

Except for the cases discussed above (wordinitial /ð/ and word-final /0/), the allophonic production of the English and Heritage groups were similar. he overall similarity of the Heritage and English groups' /0/ and /ð/ allophone production is not surprising, in the context of similar findings of early-acquiring bilinguals such as Chang et al. (2011) and Flege et al. (1995) who found that heritage speaker productions were largely identical to those of the monolingual majority population. However, while similar, the heritage Dutch speakers did maintain significant differences from English monolinguals. his difference was especially clear when comparing the production of word-initial /ð/, and word-final /0/, but it was also present in the overall higher variability of Heritage productions and the more frequent production of [d] and [t] variants. heories of L2 speech acquisition - including the concepts of transfer, lack of contact, and impoverished input - cannot adequately explain the patterns of heritage Dutch speakers' /0/ and /ð/ production. Unlike late-learning Dutch bilinguals, heritage Dutch speakers are integrated into Canadian English society and have a native grasp of English. he LEAP-Q results showed that they generally prefer to use English both at home and at work. his, combined with the Heritage speakers' acquisition of English at a young age, means that a lack of contact with L1 English speakers cannot explain the differences between English monolinguals' and Dutch heritage speakers' /0/ and /ð/ production patterns. Instead, we propose that variable /0/ and /ð/ production by heritage speakers may be promoted by the effect of ethnic or cultural identity.

Ethnocultural identity has been shown to affect target language (e.g., English) production patterns by Hoffman and Walker (2010). Hoffman and Walker found that an increased orientation to one's ethnic community and identification with the ethnic community correlates with ethnolinguistic variation. Dutch heritage speakers living in Norwich strongly identify themselves with their heritage culture and their /0/ and /ð/ allophone inventories may have been affected by this affiliation. In addition, the in-group insularity promoted by the Calvinist church may play a role by promoting linguistic markers that can identify one as a member of that in-group. his finding would mirror those of Parker (1991) and Huffines (1980) in the Anabaptist Pennsylvania Dutch communities. Because the LEAP-Q is not directed towards religious affiliation or other non-linguistic cultural markers, the question of identity will have to be more directly addressed in future work.

7.0 Conclusion

The data presented in this study are a first look at the Dutch-English bilinguals living in Norwich, Ontario and a new insight into heritage speakers' majority language phonology. While heritage speakers generally produce native-like English, their production of /ð/ and /0/ significantly differs from both the majority population of English monolinguals and their parents' generation oflate-learning L1 Dutch bilinguals. Heritage speaker speech production falls along a continuum between the "typical" L1 and L2 speakers of English - heritage speakers share similarities with both groups (especially monolingual English speakers), but they do not fit completely into either category. he difference between heritage and monolingual English phonology may be understood through the ethnic orientation framework developed by Hoffman and Walker (2010). In order to differentiate themselves from their nonDutch speaking peers, heritage speakers use subtle linguistic cues to mark their speech's Dutch origins, while maintaining intelligibility. his is done primarily through word-initial /ð/ and word-final /0/ production. Intelligibility and native accent are maintained through a monolingual-like high percentage of [ð] and [0] productions. Dutch ethnic membership can instead be communicated through the careful manipulation of the intrinsically variable monolingual English /ð/ and /0/ allophone inventory. By replacing a portion of English allophones that are typical of monolinguals with those allophones typical of Dutch L1 speakers, Dutch heritage speakers do not compromise their association with either of their Dutch or Canadian identities.

Footnote

* Graduate of the Masters of Arts in Linguistics program at the University of Western Ontario, and is a doctoral candidate for Library and Information Science at the same institution. Her research interests are in language acquisition, sociolinguistics, and text analytics. Her e-mail address is [email protected]

** Assistant professor in the Department of Modern Languages and Literatures at the University of Western Ontario. She holds a PhD in Hispanic studies from the University of Toronto. Her work has focused on second language speech learning. She is primarily interested in the effects of auditory-orthographic interaction and integration in bilinguals. She has also worked on speech production, processing and dialectal acquisition and change bilinguals. Her e-mail address is yrafat@ uwo.ca

Notes

1. Log-odds reflect the difference from a baseline measure of all three groups combined. Positive values reflect that the variable is used at levels higher than baseline, while negative values are lower than baseline. More information about interpreting log-odds can be found in Johnson (2009).

2. hough the census does not separate "Calvinist" from other Christian denominations, over 30% of Norwich's population identified as "Other Christian" (Statistics Canada, 2013), and Norwich's Netherlands Reformed church (an Orthodox Calvinist denomination) serves approximately 2100 parishioners (Van't Zelfde, 2015).

References

References

Balota, D. A., Yap, M. J., Cortese, M. J., Hutchison, K. A., Kessler, B., Loftis, B., Treiman, R. (2007). he English lexicon project. Behavior Research Methods, 39(3), 445-459.

Bernhardt, B., & Stemberger, J. P. (2002). Intervocalic consonants in the speech of English-speaking Canadian children with phonological disorders. Clinical Linguistics & Phonetics, 16(3), 199-214. doi:10.1080/02699200110112583

Best, C. T., & Tyler, M. D. (2007). Nonnative and secondlanguage speech perception: Commonalities and complementarities. In O.-S. Bohn & M. J. Munro (Eds.), Language experience in second language speech learning: In honor of James Emil Flege (Vol. 17, pp. 1334). Amsterdam, the Netherlands: John Benjamins, B.V.

Blevins, J. (2004). Evolutionary phonology: he emergence of sound patterns. Cambridge;New York;: Cambridge University Press.

Boberg, C. (2010). he English language in Canada: Status, history, and comparative analysis. Cambridge;New York: Cambridge University Press.

Boersma, P. (2013). he use of praat in corpus research Oxford Handbooks Online: Oxford University Press.

Brannen, K. J. (2011). he perception and production of interdental fricatives in second language acquisition. (PhD Dissertation), McGill University, Ann Arbor. Retrieved from digitool.library.mcgill.ca/ thesisfile106279.pdf

Brysbaert, M., & New, B. (2009). Moving beyond kucera and francis: A critical evaluation of current word frequency norms and the introduction of a new and improved word frequency measure for american english. Behavior Research Methods, 41(4), 977-990. doi:10.3758/BRM.41.4.977

Chang, C. B., Yao, Y., Haynes, E. F., & Rhodes, R. (2011). Production of phonetic and phonological contrast by heritage speakers of Mandarin. he Journal of the Acoustical Society of America, 129(6), 3964-3980. doi: 10.1121/1.3569736

Colantoni, L., & Steele, J. (2008). Integrating articulatory constraints into models of second language phonological acquisition. Applied Psycholinguistics, 29(03), 489-534. doi:10.1017/S0142716408080223

Crezee, I. (2012). Language shift and host society attitudes: Dutch migrants who arrived in New Zealand between 1950 and 1965. International Journal of Bilingualism, 16(4), 528-540.

Cutler, A., Weber, A., Smits, R., & Cooper, N. (2004). Patterns of English phoneme confusions by native and non-native listeners. Journal of the Acoustical Society of America, 116(6), 3668-3678. doi:10.1121/1.1810292

Downing, L. (2015). Dutchified English in an Ohio mennonite community. (Master's hesis), he Ohio State University, Ohio.

Edwards, A. (2010). Dutch english: Tolerable, taboo, or about time too? English Today, 26(1), 19-25. doi:10.1017/S0266078409990563

Flege, J., Munro, M. J., & MacKay, I. R. A. (1995). Effects of age of second-language learning on the production of english consonants. Speech Communication, 16(1), 1-26. doi:10.1016/0167-6393(94)00044-B

Flege, J. E. (1995). Second language speech learning: heory, findings, and problems. In W Strange (Ed.), Speech perception and linguistic experience: Issues in cross-language research (pp. 233-277). Timonium, MD: York Press Inc.

Gatbonton, E., Trofimovich, P., & Segalowitz, N. (2011). Ethnic group affiliation and patterns of development of a phonological variable. he Modern Language Journal, 95(2), 188-204.

Gierut, J. A., & Storkel, H. L. (2009). Markedness and the grammar in lexical diffusion of fricatives. Clinical Linguistics & Phonetics, 16(2), 115-134. doi:10.1080/0269920011011287

Gollan, T. H., Starr, J., & Ferreira, V. S. (2015). More than use it or lose it: he number-of-speakers effect on heritage language proficiency. Psychonomic bulletin & review, 22(1), 147-155.

Hancin-Bhatt, B. (1994). Phonological Transfer in Second Language Perception and Production. PhD Dissertation. University of Illinois at UrbanaChampaign. Retreived from: http://hdl.handle. net/2142/21209

Hanulikova, A., & Weber, A. (2012). Sink positive: Linguistic experience with th substitutions influences nonnative word recognition. Attention, Perception, and Psychophysics, 74(3), 613-629. doi:10.3758/s13414011-0259-7

Harrison, B. (2000). Passing on the language: Heritage language diversity in Canada. Canadian Social Trends, 14-19.

Hayes, B. (2009). Introductory phonology. Singapore: Wiley-Blackwell.

Hoffman, M. F., & Walker, J. A. (2010). Ethnolects and the city: Ethnic orientation and linguistic variation in toronto english. Language Variation and Change, 22(1), 37-67. doi:10.1017/S0954394509990238

Huffines, M. L. (1980). English in contact with Pennsylvania German. he German Quarterly, 53(3), 352-366. doi:10.2307/404912

Hulsen, M. E. H. (2000). Language loss and language processing: hree generations of Dutch migrants in New Zealand (Dissertation). Radboud University Nijmegen, Nijmegan. Retrieved from http://repository.ubn.ru.nl/ handle/2066/18901

Hulsen, H., de Bot, K., & Weltens, B. (2002). Between two worlds. Social networks, language shift, and language processing in three generations of Dutch migrants in New Zealand International Journal of the Sociology of Language (Vol. 2002, pp. 27).

Jekiel, M. (2012). he evolution of english dental fricatives: Variation and change. (Master's Master's hesis), he University of Adam Mickiewicz. Retrieved from https://www.academia.edu/7477404/he_evo lution_ of_English_dental_fricatives_variation_and_change

Johnson, D. E. (2009). Getting off the goldvarb standard: Introducing rbrul for mixed-effects variable rule analysis. Language and Linguistics Compass, 3(1), 359383. doi:10.1111/j.1749-818x.2008.00108.x

Major, R. C., & Faudree, M. C. (1996). Markedness universals and the acquisition of voicing contrasts by Korean speakers of English. Studies in Second Language Acquisition, 18(01), 69-90.

Marian, V., Blumenfeld, H. K., & Kaushanskaya, M. (2007). he language experience and proficiency questionnaire (leap-q): Assessing language profiles in bilinguals and multilinguals. Journal of Speech, Language, and Hearing Research, 50(4), 940-967. doi:10.1044/10924388(2007/067)

Munro, M. J., & Derwing, T. M. (2006). he functional load principle in esl pronunciation instruction: An exploratory study. System, 34(4), 520-531. doi:10.1016/j.system.2006.09.004

Parameshwaran, M. (2013). Explaining intergenerational variations in English language acquisition and ethnic language attrition. Ethnic and Racial Studies, 37(1), 2745. doi:10.1080/01419870.2013.827794

Parker, R. (1991). Influences of the Pennsylvania German dialect on the English spoken in" Pennsylvania Dutch country" as a regional identity marker. Term Paper. Hannahs, Napoli.

Rao, R., & Kuder, E. (2016). Research on heritage spanish phonetics and phonology: Pedagogical and curricular implications. Journal of New Approaches in Educational Research, 5(2), 99-113A. doi:http://dx.doi.org/10.7821/ naer.2016.7.171

Roach, P. (2010). English phonetics and phonology: A practical course (4 ed.). Italy: Cambridge University Press.

Schmid, M., Gilbers, S., & Nota, A. (2014). Ultimate attainment in late second language acquisition: Phonetic and grammatical challenges in advanced dutch-english bilingualism. Second Language Research, 30(2), 129-157. doi:10.1177/0267658313505314

Schryer, F. J. (1998). he netherlandic presence in ontario: Pillars, class and Dutch ethnicity. Waterloo, ON: Wilfrid Laurier Univ. Press.

Smith, A. B. (2007). General american english speech acquisition. In S. McLeod (Ed.), he international guide to speech acquisition (pp. 128-147). Clifton Park, NY, USA: homson Delmar Learning.

Smith, B. (2013). An acoustic analysis of voicing in American English dental fricatives. Ohio State University Working Papers in Linguistics, 60, 117-128.

Statistics Canada. (2012). Census subdivision of Norwich, tp (Ontario) - census 2011. (98-310-XWE2011004). Ottawa, Ontario Retrieved from http://www12.statcan. gc.ca/census-recensement/2011/as-sa/fogs-spg/Factscsd-eng.cfm?LANG=Eng&GK=CSD&GC=3532002.

Statistics Canada. (2013). Norwich township, Ontario (code 3532002). (99-004-XWE). Ottawa, Ontario Retrieved from http://www12.statcan.gc.ca/nhs-enm/2011/dppd/prof/details/page.cfm?Lang=E&Geo1=CSD&Cod e1=3532002&Data=Count&SearchText=Norwich&Se archType=Begins&SearchPR=01&A1=All&B1=All&T ABID=1.

Van Dijk, J. (1998). Ethnic persistence among DutchCanadian Catholics and Calvinists. Canadian Ethnic Studies/Etudes Ethniques au Canada, 30(2), 23-49.

Van't Zelfde, A. (2015). Norwich (canada). Norwich (Canada) - Gereformeerde Gemeenten. Retrieved from http://gergeminfo.nl/buitenland/gemeenten-amerikacanada/classis-east/norwich-canada

Wester, F., Gilbers, D., & Lowie, W (2007). Substitution of dental fricatives in English by Dutch l2 speakers. Language Sciences, 29(2-3), 477-491. doi:10.1016/j. langsci.2006.12.029

Zhao, S. Y. (2010). Stop-like modification of the dental fricative /ð/: An acoustic analysis. he Journal of the Acoustical Society of America, 128(4), 2009-2020. doi:doi:http://dx.doi.org/10.1121/1.3478856.

Recebido em: 31/03/2017

Aceito em: 10/07/2017

APPENDIX D

Please review the following paragraph. When you are ready, read it aloud into the microphone.

Ruth sighs and looks at the clock. It is almost three-thirty, and at three-thirty, she will bathe her two dogs: hor and Zeus. he weather had been bad on hursday: it rained heavily and even thundered a bit. Still, Ruth had taken them to the park and thrown the ball for them. he dogs are very muddy, but they both loathe the bathtub. Ruth decides that hor will go first, and she tries to soothe him as she drags him towards the bath. he other dog waits impatiently in the kennel. Ruth takes a deep breath, and pushes hor into the water. hough he was seething and angry a second ago, now hor seems to enjoy the water, and he bites it playfully with his teeth. Ruth takes the shampoo and works up a lather in his fur, getting soap everywhere. She can breathe more easily now. She washes the shampoo out and brushes hor's fur smooth again. "hese dogs are not so difficult - I can do this!", she thinks. When hor is clean, Ruth brushes his teeth too, to keep him healthy. She has special toothpaste and a toothbrush that are made for dogs. hor and Zeus are two lucky dogs! hey are clean, and now their bathing is done.