Foreign Accent in Pre- and Primary School Heritage Bilinguals

: Previous research has shown that the two languages of early bilingual children can inﬂuence each other, depending on the linguistic property, while adult bilinguals predominantly show inﬂuence from the majority language to the minority (heritage) language. While this observed shift in inﬂuence patterns is probably related to a shift in dominance between early childhood and adulthood, there is little data documenting it. Our study investigates the perceived global accent in the two languages of German-Russian bilingual children in Germany, comparing 4–6-year-old (preschool) children and 7–9-year-old (primary school) children. The results indicate that in German the older children sound less accented than the younger children, while the opposite is true for Russian. This suggests that the primary school years are a critical period for heritage language maintenance.


Introduction
Heritage speakers (HSs) are typically characterized as bilinguals growing up with a minority language, passed on by one or both parents, along with the dominant language of the society. Even though heritage speakers acquire two languages during early childhood, the societal (majority) language (ML) typically becomes their dominant language throughout the lifespan (herein, we will use the term "dominance" in terms of proficiency). The term heritage language has traditionally been used in the context of immigrant languages, indigenous languages and colonial languages. In this study, we focus on an immigrant minority language, specifically Russian in Germany, and on children in the second generation. Child (heritage) bilinguals are at the crossroads between first (L1) and second language (L2) learners. They are pre-critical period, L1 acquirers, who develop their phonological representations and their articulatory and acoustic-perceptual skills in tandem. At the same time, they bear resemblance with late L2 learners because they represent cases of language contact within an individual.
Pronunciation is one of the most puzzling phenomena in HSs. With regards to proficiency in their heritage language, HSs are generally said to possess native-like levels of pronunciation and fluency (e.g., Montrul 2008, p. 163;Polinsky and Scontras 2020, p. 8;Rothman 2009, p. 157). Indeed, research has shown that HSs have relatively "authentic" pronunciation, differing in this respect from late L2 learners (Au et al. 2008;Kupisch et al. 2014;Flores and Rato 2016;Kupisch et al. 2020;Chang et al. 2008;Saadah 2011). This is not surprising, since many phonetic and phonological properties are amongst the earliest acquired properties of language and should therefore be relatively resistant to attrition and resilient to cross-linguistic influence (CLI) (Montrul 2008, p. 71;Polinsky and Scontras 2020, p. 8).
In reality, however, HSs are often deemed foreign-sounding in their heritage language, and native speaker raters can easily detect accent features of the speakers' dominant language (Kupisch et al. 2014;Lloyd-Smith et al. 2020;Kupisch et al. 2020). Benmamoun German has only 24 consonants and, unlike in Russian, voiceless stops are aspirated, but a comparatively larger vowel inventory with 15 short and long monophthongs and four diphthongs (Hall 2000). German is also subject to substantial regional variation. The varieties relevant to our study are the Alemannic variety spoken in Konstanz and surroundings and the variety in Berlin, which, despite some idiosyncrasies, resembles Standard German more closely. For example, as a result of regional variation, speakers of the Alemannic variety might produce/z/as voiceless [s] (Beckman et al. 2009) and palatalize/s/ (phonological process/s/ → [ ]/_[+stop]) (Auer 1990). The variety spoken in Berlin is characterizable as an urban dialect, which means that there are some dialect features (e.g., production of the front vowels [ I ] and [ε] as rounded, i.e., as [ Y ] and [oe]; Wiese and Freywald 2017), but speakers tend to avoid them. Germans can typically identify speakers from other regions of Germany, even though they cannot always determine the exact region. The varieties of Russian are comparatively more homogeneous. There are a few prominent accent features (e.g., fricativization of /g/ in Southern Russian), but variation is much less pronounced than in German. The data in the present study are representative of the Central dialect group, which is characterized by a variety of vowel reduction patterns in contrast to the Northern dialect group where the reduction is weak or lacking (e.g., producing / ∂ / vs. /o/ in unstressed positions) (Kasatkin 1989).
Our study compares the accents of Russian-German bilingual preschool and primary school children. We are particularly interested in potential differences between the languages as well as between different age groups (before and after starting primary school). Our results show that, when speaking German, the younger children are perceived as foreign-sounding more often than the older children, while when speaking Russian, it is the older children who tend to be perceived as more foreign-sounding. This suggests that the phonological systems of bilingual primary school children are still malleable and that it is during these years that a shift from sounding accented in the ML to sounding accented in the HL may take place, and could potentially be prevented. Our paper is structured as follows: In the next section, we provide some background on previous research on phonological development in bilingual children, and findings on the perceived accents of early bilingual adults. In Section 3, we formulate our research questions and hypotheses. Section 4 presents the results, which are discussed in Section 5. We conclude with some notes on the societal relevance of our findings.

Background
The acquisition of phonology is a gradual process, which starts very early but continues into the preschool years. For example, German-learning children produce the consonants [m b d t n] 90% correct between 1;6-1;11, while [j] and [ inventories. The phonological system of Russian contains 42 phonemes, with only five vowels (Avanesov 1956;Halle 1959;Panov 1967;Bondarko 1998), and is characterized by a large variety of positional vowel and consonant alternations. Vowel alternations are constrained by stress, which can fall on any syllable and can move. Russian further distinguishes between soft (palatalized) and hard consonants, and voiceless stops are pronounced without aspiration. German has only 24 consonants and, unlike in Russian, voiceless stops are aspirated, but a comparatively larger vowel inventory with 15 short and long monophthongs and four diphthongs (Hall 2000). German is also subject to substantial regional variation. The varieties relevant to our study are the Alemannic variety spoken in Konstanz and surroundings and the variety in Berlin, which, despite some idiosyncrasies, resembles Standard German more closely. For example, as a result of regional variation, speakers of the Alemannic variety might produce/z/as voiceless [s] (Beckman et al. 2009) and palatalize/s/ (phonological process/s/ → [ʃ]/_[+stop]) (Auer 1990). The variety spoken in Berlin is characterizable as an urban dialect, which means that there are some dialect features (e.g., production of the front vowels [ɪ] and [ɛ] as rounded, i.e., as [ʏ] and [oe]; Wiese and Freywald 2017), but speakers tend to avoid them. Germans can typically identify speakers from other regions of Germany, even though they cannot always determine the exact region. The varieties of Russian are comparatively more homogeneous. There are a few prominent accent features (e.g., fricativization of /g/ in Southern Russian), but variation is much less pronounced than in German. The data in the present study are representative of the Central dialect group, which is characterized by a variety of vowel reduction patterns in contrast to the Northern dialect group where the reduction is weak or lacking (e.g., producing /ə/ vs. /o/ in unstressed positions) (Kasatkin 1989).
Our study compares the accents of Russian-German bilingual preschool and primary school children. We are particularly interested in potential differences between the languages as well as between different age groups (before and after starting primary school). Our results show that, when speaking German, the younger children are perceived as foreign-sounding more often than the older children, while when speaking Russian, it is the older children who tend to be perceived as more foreign-sounding. This suggests that the phonological systems of bilingual primary school children are still malleable and that it is during these years that a shift from sounding accented in the ML to sounding accented in the HL may take place, and could potentially be prevented. Our paper is structured as follows: In the next section, we provide some background on previous research on phonological development in bilingual children, and findings on the perceived accents of early bilingual adults. In Section 3, we formulate our research questions and hypotheses. Section 4 presents the results, which are discussed in Section 5. We conclude with some notes on the societal relevance of our findings.

Background
The acquisition of phonology is a gradual process, which starts very early but continues into the preschool years. For example, German-learning children produce the consonants [m b d t n] 90% correct between 1;6-1;11, while [j] and [ŋ] are only acquired between 3;0 and 3;5, and the sibilants [s] and [z] and the affricate [ts] are phonetically unstable up to the age of 5;11 (Fox 2007). Similarly, the acquisition of Russian phonology is characterized by early acquisition of the vowel [a] and on the late acquisition of the vowel [ ɨ ] (e.g., Shvachkin 1948;Gvozdev 1961). The speech of young Russian children is characterized by an overall softness of consonants. According to Zharkova (2005), palatalized consonants are the most frequent substitutes for other consonant phonemes at an early age. In contrast, stressed syllables are shown to be acquired very early, and Russian-speaking children make few mistakes in choosing which syllable to stress, but they can omit unstressed syllables before the age of 3. The fact that the acquisition of phonology, although it starts very early, continues throughout childhood compromises the idea that phonology is generally early acquired. It is likely that the late(r) acquired ] are only acquired between 3;0 and 3;5, and the sibilants [s] and [z] and the affricate [ts] are phonetically unstable up to the age of 5;11 (Fox 2007). Similarly, the acquisition of Russian phonology is characterized by early acquisition of the vowel [a] and on the late acquisition of the vowel [ inventories. The phono vowels (Avanesov 195 by a large variety of pos constrained by stress, w tinguishes between sof nounced without aspir voiceless stops are aspi and long monophthong stantial regional variati spoken in Konstanz an idiosyncrasies, resembl gional variation, speak (Beckman et al. 2009(Beckman et al. ) 1990). The variety spok that there are some dia rounded, i.e., as [ʏ] and Germans can typically they cannot always de tively more homogene tion of /g/ in Southern R The data in the presen characterized by a varie group where the reduc positions) (Kasatkin 19 Our study compa mary school children. W languages as well as b school). Our results sh ceived as foreign-soun Russian, it is the older c suggests that the phono leable and that it is dur sounding accented in t paper is structured as f vious research on phon perceived accents of ear tions and hypotheses. We conclude with som

Background
The acquisition of tinues into the preschoo sonants [m b d t n] 90 between 3;0 and 3;5, an unstable up to the age o is characterized by ear vowel [ ɨ ] (e.g., Shvach is characterized by an o atalized consonants are an early age. In contra Russian-speaking child they can omit unstress phonology, although it the idea that phonolog ] (e.g., Shvachkin 1948;Gvozdev 1961). The speech of young Russian children is characterized by an overall softness of consonants. According to Zharkova (2005), palatalized consonants are the most frequent substitutes for other consonant phonemes at an early age. In contrast, stressed syllables are shown to be acquired very early, and Russian-speaking children make few mistakes in choosing which syllable to stress, but they can omit unstressed syllables before the age of 3. The fact that the acquisition of phonology, although it starts very early, continues throughout childhood compromises the idea that phonology is generally early acquired. It is likely that the late(r) acquired properties are particularly vulnerable to CLI in bilingual children, in particular in combination with (typically) massive exposure to the ML in the context of entering primary school, if not earlier, when some phonological properties are not yet stable.

Cross-Linguistic Influence in the Phonologies of HSs during Early Childhood
When it comes to the bilingual acquisition of phonology, most research during the past three decades has focused on whether the speech of bilingual children differs in qualitative and quantitative ways from that of monolingual children. Systematic differences would suggest that there is interaction between the two linguistic systems of the bilingual, i.e., CLI. Most of the relevant research in this area has focused on young bilinguals who are either simultaneous bilinguals (with exposure to both languages from birth) or early sequential bilinguals (with exposure to one language from birth and sequential exposure to the other before the age of three). Traditionally, there were two research trends. The first trend represents studies trying to show that bilingual children start out acquiring both of their languages with one unified speech system (Vogel 1975;Volterra and Taeschner 1978). Under the second approach, bilingual children are assumed to build up two systems from the beginning, although these two systems may interact (Paradis and Genesee 1996).
More recent studies within phonology-oriented research, based on in-depth analyses of production data, have tried to model cross-linguistic interaction (see, e.g., Lleó and Cortés 2013;Kehoe 2015;Lleó 2018). These detailed linguistic studies suggest that the acquisition of phonology may be accelerated or delayed depending on the specific structures under investigation. For example, findings on the same group of German-Spanish bilingual children revealed different patterns of interaction: acceleration of codas in Spanish, deceleration of the vowel length distinction in German, and transfer of long lag voicing into Spanish (Kehoe 2015). The use of the term "transfer" in this approach refers specifically to the presence of a non-native sound or structure in one of the bilingual's languages, which comes from its presence in the other language, thus differing from its more traditional use, where it is synonymous with cross-linguistic interaction. It can be considered a qualitative difference to monolinguals. By contrast, acceleration and deceleration refer to quantitative differences between monolinguals and bilinguals, which implies that bilinguals undergo qualitatively similar processes as monolinguals but they do so earlier/faster or later/more slowly. In the following, the three types are illustrated on the basis of studies on German-Spanish children in Germany (see Kehoe 2015 for an excellent overview with more examples).
A clear example of acceleration has been provided by Lleó et al. (2003), who showed that German-Spanish bilinguals (HSs of Spanish) produced syllable-final codas in Spanish earlier than Spanish monolinguals, arguably due to experience with more complex codas in German. However, with regard to the same phenomenon in comparable language combinations other authors found deceleration or no influence (e.g., Gildersleeve-Neumann et al. 2008;Almeida et al. 2012;Ezeizabarrena and Alegria 2015), suggesting that language external factors also play a role. An example of deceleration comes from a study by Kehoe (2002) on the acquisition of vowel length in German-Spanish bilinguals. The German vowel system is more complex than the Spanish one and Kehoe (2002) findings suggest indeed that bilingual children experienced difficulty acquiring the more marked system of German vowels. Interestingly, German was the ML in this study, which means that at least at an early age, CLI can also affect the majority language. Moreover, the children did not differ from monolinguals in their acquisition of the five-vowel system of Spanish, their HL, which implies that in some cases, the HL is not affected by influence even if there are differences between the ML and the HL. An example of transfer was provided by Kehoe et al. (2004), who have shown that one of the four German-Spanish children they investigated transferred long lag VOT from German to Spanish. Both German and Spanish have the voiceless stops /p, t, k/, but the phonetic basis underlying the voicing distinction is different in the two languages, as they are realized with long lag in German and with short lag in Spanish. Since transfer of long lag was also found in a study on English-Japanese bilinguals with Japanese as their HL (Johnson and Wilson 2002), it is possible that transfer only affects the HL, while quantitative effects (acceleration and deceleration) can be bidirectional. Finally, note that evidence for language interaction can still be found at primary school ages. For example, Lleó (2018) investigated assimilation of Languages 2021, 6, 96 5 of 19 place in coda nasals in the Spanish productions of German-Spanish bilingual children in Germany. Comparisons to monolinguals and bilinguals growing up in Spain showed that in the productions of the bilinguals living in Germany the phenomenon was still unstable at the age of 7;0. In summary, research on young bilingual children has provided plenty of evidence for language interaction. The influence can be quantitative or qualitative, it can affect the HL and the ML, and it does not necessarily cease when children enter primary school.
There are very few accent rating studies with children, looking at accent globally. "Global" means that there is no focus on any particular phonetic or phonological feature, because an accent can result from a combination of different segmental or suprasegmental features. Accent rating studies with children are particularly challenging because, due to language development, children's "accents" are different from those of adults for an extended period of time, yet without being foreign. To date, only three accent rating studies have been carried out with children (Asher and García 1969;Snow and Hoefnagel-Höhle 1977;Wrembel et al. 2019): The former two explored accentedness in bilingual children's ML, and the latter in their HL Polish. Asher and García (1969) analyzed the accentedness of Spanish-English bilinguals aged 7-19 years compared to English monolinguals in their ML English in the US. Ratings were based on four English sentences rated by high school students. Findings show that all bilinguals were rated as accented; however, with an early age of onset upon arrival and with increasing length of exposure the amount of children rated as near-native increased. Snow and Hoefnagel-Höhle (1977) conducted two studies on accentedness to compare the abilities of children and adults (L1 English) to pronounce words in a new foreign language (Dutch): In study 1, three raters judged 136 English-speaking children and adults. In study 2, one rater judged 47 children and adults. The authors found that the pronunciation of the older participants was initially more target-like than that of the younger ones, and that after about a year of exposure this trend started to reverse. Wrembel et al. (2019) combined a detailed phonetic analysis with accent ratings investigating the speech of Polish-English bilinguals in their HL Polish. The rating study was based on data from a sentence repetition task and complemented by data from a narrative task. The bilinguals were preschool and primary school children (mean age 5.79) who were exposed to English (the ML) before the age of three and had at least one Polish-speaking parent. The raters for this study were teachers and teacher trainees, who assessed the degree of accentedness on a 7-point Likert scale. The Polish-English bilingual children were perceived to be significantly more accented than monolingual Polish children, and the amount and quality of the HL Polish input was found to be the main predictor. The relationship in global accentedness between the children's two languages (HL and ML) during childhood has not been explored, so far. An open question concerns the adequacy of raters for accent rating studies with bilingual children. There has been some discussion in the context of late second language learners on whether it is crucial to work with experienced (i.e., phonetically trained) raters, but with no clear advantage pointing to either of the two types (see Jesney 2004). Therefore, in a study on early bilinguals, Kupisch et al. (2014) engaged 50% of each type, finding no difference between those groups. An additional challenge pertaining to child speech is the fact that children's sound systems are still in the process of developing and that it may be difficult to distinguish "developmental" accents from "foreign" accents. This would justify prioritizing phonetically untrained raters, as long as they are familiar with child language, provided raters with the combined expertise (phonetics and child speech) are unavailable.

Global Accent in Adult Early Bilinguals
A foreign accent has been defined as any perceived divergence from a native speaker, resulting not from the presence of regional varieties but from the influence of another language (Derwing and Munro 2009). HSs are native speakers by definition, but they may nevertheless be perceived as foreign-sounding. Often, a foreign accent results from a number of features, including segmental and suprasegmental properties as well as phonological rules, which is why relevant studies have often used the term "global" ("global foreign accent"). At the same time, a single sound that is pronounced in a divergent fashion may be sufficient to contribute to a perceived foreignness.
While many accent rating studies have focused on late L2 learners (see Jesney 2004 for an excellent overview), only a few studies have assessed the perceived accents of early bilinguals and have included ratings for both languages. Kupisch et al. (2014) compared HSs when speaking their HL to (i) monolingual speakers of the same language, (ii) late L2 learners and (iii) bilinguals with the same language combination but an inverse dominance relation between ML and HL. For example, HSs of Italian in Germany were compared to (i) monolingual controls from Italy, (ii) L2 speakers of Italian and (iii) HSs of German in Italy speaking Italian as their ML. In addition, there was a (mirror) experiment for German, where the same two early bilingual groups were compared to monolinguals and L2ers. Crucially, all heritage bilinguals in this study had been exposed to both of their languages from birth so that potential effects of Age of Onset (AoO) could be excluded when comparing the ML and HL. The results showed monolingual-like ratings for HSs when speaking their ML and advantages over late L2ers when speaking their HL. Nevertheless, HSs were more often than not deemed foreign speakers of their native language. The study also included experiments with French HSs in Germany and German HSs from France with similar outcomes. Finally, individual speakers were compared in their two languages, and the comparison indicated that a native accent in the ML did not coincide with a foreign accent in the HL, or vice versa, as there were also speakers who were perceived to be native in their two languages.
In a follow-up study with HSs of Italian in Germany (Lloyd-Smith et al. 2020), HSs with sequential exposure to German were studied. The results confirmed those reported above, i.e., HSs performed on a par with monolingual speakers when speaking their ML, while in their HL they outperformed late L2ers but were nevertheless often perceived to have a foreign accent. It was also shown that a later AoO in German neither had a negative effect on the speakers' accents when speaking German, nor a positive effect on their accent when speaking Italian. The latter might have been expected because an earlier AoO in German implies less time for the HL to develop independently. However, speakers who had reported using Italian more were perceived to sound more native-like when speaking Italian, which parallels Wrembel et al. (2019) study with Polish-speaking children. This suggests that language use is at least as important a factor for HL development and maintenance, as AoO is (said to be) an important factor for the development of the ML. Another interesting aspect comparing this study with Kupisch et al. (2014) was that the rate of perceived foreignness was lower: 49% in this study as compared to 70% in the Italian HSs of the previous study. This could be related to the different geographical settings, as Lloyd-Smith et al. (2020) study was situated in the South where density of the Italian populations is higher than in the region where Kupisch et al. (2014) had tested their Italian speakers. It could also be related to the higher number of sequential bilinguals in the latter study, i.e., speakers with a later AoO in German who spoke only Italian in the home. In another study with HSs of Turkish in Germany , these findings were confirmed once more, except that in this study the HSs were often perceived to be mildly accented when speaking their ML. The authors argued that what the raters have perceived might have been features of a variety of German, "Kiezdeutsch" (see Wiese 2012), that is often spoken by young immigrants and native Germans in cities with larger linguistic minorities.
In summary, accent rating studies with HSs during adulthood show that these speakers are often, though not always, perceived to be foreign speakers of their HL but more rarely of their ML. Amount of language use seems to be a crucial factor for the development of a native-sounding accent in the HL, while it is somewhat controversial whether an early AoO in the ML is necessary for developing a native-sounding accent in the ML, as long as exposure happens during the preschool years. Other factors, such as the density of the population, which comes with more opportunities to use the language, could play an additional role.

Research Questions and Predictions
Based on what we know from previous studies, we have formulated the following questions and predictions (P): RQ1. How often are monolingual children perceived to have a foreign accent? RQ2. Are bilingual children perceived to have a foreign accent more often than monolingual children when speaking the majority language? RQ3. Are younger bilingual children perceived to have a foreign accent more often than older children when speaking the majority language? RQ4. Are older bilingual children perceived to have a foreign accent more often than younger children when speaking the heritage language? P1. Monolingual children will occasionally be rated as accented, but the incidence of perceived foreignness will fade in the older children. This hypothesis is motivated by the fact that monolingual children develop their phonological systems gradually and that some phonological phenomena are late acquired. Given individual variation, not all adultlisteners might be familiar with all accent features, even if they are used to hearing child speech and sensitized prior to the experiment. P2. When speaking German, the bilingual children will be perceived as more accented than the monolingual children due to cross-linguistic influence from Russian. This hypothesis is motivated by the literature on young bilingual children having shown that the two languages of bilingual children can mutually influence each other. Of course, a later age of onset in German (in the case of children who speak only Russian at home) and schooling in German might play an additional role. P3. Perceived accentedness in German, the majority language, will decrease over time. The motivation for this is twofold. First, children's speech becomes more adult-like with increasing age (see P1). Second, rating studies of early bilinguals during adulthood have shown that adult early bilinguals are rarely considered to be foreign-sounding. P4. Perceived accent in Russian, the children's heritage language, will increase in the older bilingual children. The motivation for this prediction is again twofold. Research on child bilinguals has shown that the two languages of bilingual children show crosslinguistic influence. However, there is no indication that this influence is unidirectional from the majority language in the heritage language or vice versa. During adulthood, by contrast, heritage speakers tend to be accented in their heritage language and not in their majority language (Kupisch et al. 2014;Lloyd-Smith et al. 2020). Thus, a shift from bidirectional to (more) unidirectional influence must take place during their development in later childhood, catalyzed by a change in exposure patterns. We assume that this period coincides with entry to primary school where exposure and use of the majority language steadily increases.

Child Participants and Preparation of Materials
We recorded German-Russian simultaneous and sequential bilingual children when narrating picture-based stories in German and in Russian, using the German and Russian versions of the Multilingual Assessment Instrument for Narratives (MAIN) task (Gagarina et al. 2012) 2 . Narratives of 12 children aged 4-6 years (henceforth Group I, mean age 5;4), and of 12 children aged 7-9 years (henceforth Group II, mean age 8;4) were selected based on sound quality and availability of recordings of the same child in both languages. The reason for choosing these ages was as follows: In Germany, children start school at the age of 6 years. Thus, children aged 4-6 will have had exposure to German but presumably not as much as with the beginning of school. At the age of 6, exposure to German is supposed to increase substantially in both quantity and quality, because alphabetization starts as well. Moreover, the educational setting might trigger changes in the children's identity (self-concept) and attitudes.
All children were born and raised in Germany (either in the Konstanz or the Berlin area 3 ). Most of the children (19/24) were growing up in families in which both parents were Russian-speaking and in which only Russian was spoken. These children could be characterized as early sequential bilinguals, since their parents generally reported that their children had their first (intensive) experience with German around the age of three years when entering kindergarten. It needs to be acknowledged, however, that, since the children were growing up in Germany, they had most likely had some exposure to German before, so that the reported AoO of three years can only be approximate. The remaining children (5/24) grew up in families in which one parent was Russian and the other German, and in which both languages were spoken. These children could be characterized as early simultaneous bilinguals, as they had been exposed to both languages from birth with their parents generally following the one parent-one language strategy. Since the information on AoO is more problematic, as mentioned above, we prefer to refer to the groups in terms of family type ("Russian" vs. "mixed") rather than AoO. Although we have not collected detailed information on input quantity and quality, it seems reasonable to say that the children from Russian families have been exposed to Russian more substantially.
Sound files of 30 s (±10%) per child in each language were extracted from the MAIN narratives (based on two of the MAIN stories). The length of the samples varied slightly in order to avoid interruptions within a sentence. In some cases, edits were required due to the lack of uninterrupted 30-s speech samples. The quality of the files was double-checked by native speakers of each language.
We further included control data from 12 monolingual Russian and 12 monolingual German age-matched children. Fewer control samples were included to avoid a bias towards more unaccented samples (we assumed that not all bilinguals would sound accented). The mean ages of Group I and II for the Russian controls were 4;8 and 8;6 respectively. The mean ages of Group I and II for the German controls were 5;1 and 8;1 respectively. The control data were included to ensure that the raters were able to identify a monolingual accent in German or Russian respectively (see Table 1 for an overview).

Rating Task and Raters
The sound files were judged by native speakers of Russian (Russian experiment) and German (German experiment) who lived in Russia and Germany, respectively. The main inclusion criterion for the raters was being familiar with child speech by spending time with 4-9-year-old children on a regular basis. Most of the participating raters (n = 36) had children, grandchildren 4 or younger siblings in this age range. This choice of raters had the disadvantage that they were phonetically untrained (unlike in Wrembel et al. 2019), but it came with the advantage that they know what children at comparable ages sound like. We opted for this choice because there is mixed evidence concerning the need of phonetically trained raters, while it was crucial that the raters would be sensitive to developmental aspects of child speech.
The sound files were randomized and presented to the raters by means of a PowerPoint Presentation. The raters' task was to determine for each sample whether the child has "no accent", a "weak foreign accent" or a "strong foreign accent". 5 The rating task consisted of two parts: preschoolers were rated in the first part, school children in the second part. Before the experiment, the raters were familiarized with the procedure and sensitized to the fact that they would listen to children. The raters were instructed to listen carefully to the children's short narratives, and to determine whether the specific child had a (weak/strong) foreign accent or not. The raters could listen to each recording twice. We explained that a foreign accent corresponds to a pronunciation in German or Russian respectively that is influenced by characteristics of another language. In addition, the raters were asked to ignore dialectal features and grammatical errors, and they were reminded of the fact that child speech differs from adult speech as young children often use shorter sentences and have a smaller vocabulary. The German raters participated in the experiment via internet in a Zoom meeting (due to the pandemic). The sound was checked carefully before starting the experiment.
Twenty-two Russian raters (16 females) with a mean age of 44 years (age range 18-69) participated in the Russian experiment. Most of them were mothers and grandmothers (age range 31-69) and one was an 18-year-old sister of a preschool boy. All raters were born and have lived most of their lives in Ivanovo, a city in Central Russia. They all spoke standard Russian without particular dialectal features. Our monolingual controls were from the same city. Twenty-one German raters (13 females) with a mean age of 44 years (age range 22-72) participated in the German experiment. The majority of the raters were parents of preschool and/or school children (n = 14), while others were primary school teachers (n = 3), grandparents (n = 2) and babysitters (n = 2) of 4-9-year-old children. At the time of testing, most of the raters (n = 14) were living in the Berlin area; the remainder came from other areas in Germany including Western and Southern Germany, often coinciding with the area in which they had grown up.

Results
In Figure 1, we plotted the distribution of foreign accent ratings ("weak accent", "strong accent" or "no accent") in Russian (left) and German (right) by the family type of the children: heritage mixed (family in which German and Russian are spoken), heritage Russian (family in which both parents speak Russian), monolingual Russian family in Russia or monolingual German family in Germany. In order to calculate the mean accentedness of speakers (see Figures 2 and 3), accent ratings were converted into numeric scores, where "no foreign accent" corresponded to 0, "weak foreign accent" corresponded to 1, and "strong foreign accent" corresponded to 2. The measure of internal consistency reliability of the ratings in each language, Cronbach's alpha (ltm package, Rizopoulos 2006), was 0.899 for Russian and 0.703 for German data. numeric scores, where "no foreign accent" corresponded to 0, "weak foreign accent" corresponded to 1, and "strong foreign accent" corresponded to 2. The measure of internal consistency reliability of the ratings in each language, Cronbach's alpha (ltm package, Rizopoulos 2006), was 0.899 for Russian and 0.703 for German data.  numeric scores, where "no foreign accent" corresponded to 0, "weak foreign accent" corresponded to 1, and "strong foreign accent" corresponded to 2. The measure of internal consistency reliability of the ratings in each language, Cronbach's alpha (ltm package, Rizopoulos 2006), was 0.899 for Russian and 0.703 for German data.  As expected, monolingual Russian and German children were most often rated as unaccented (86% and 83% respectively), although they were perceived as having a weak foreign accent, respectively, 12% and 16% of the time. Bilingual children from families with two Russian-speaking parents were rated as unaccented 50% of the time in Russian, but only 20% of the time in German. In this type of family, accent strength in Russian was mostly perceived to be "weak" (34% of the time) and less often as "strong" (17% of the time), while in German, the accent was perceived to be weak 44% of the time and strong 36% of the time. Bilingual children with one German-speaking parent were perceived as unaccented 40% of the time when speaking Russian, and only 19% of the time when speaking German. In this family type accents tended to be perceived as strong in Russian and as weak in German: 42% of the cases in Russian and 21% in German were rated as strongly accented; 18% of the cases in Russian and 60% of the cases in German were rated as weakly accented.
The effect of family type and age group on the perceived accentedness of the children was investigated with an ordinal mixed-effects logistic model in R (Ordinal package, Christensen and Christensen 2021), where accent score (ordinal variable taking values 0, 1 or 2) was predicted from family type, age group and their interaction, and "rater" was included as a random intercept term. For Russian, the model indicates a significant effect of family type, and a significant interaction between family type and age group, suggesting that the relation between perceived accentedness and age depends on the family type (Tables 2 and 3). For German, the interaction term was not included due to the incomplete information problem (Field et al. 2012, p. 322), resulting from the absence of "strong accent" ratings in German monolingual school children. The results are presented in Tables 4 and 5. For German, both the family type and age group are significantly associated with perceived accentedness. As expected, monolingual Russian and German children were most often rated as unaccented (86% and 83% respectively), although they were perceived as having a weak foreign accent, respectively, 12% and 16% of the time. Bilingual children from families with two Russian-speaking parents were rated as unaccented 50% of the time in Russian, but only 20% of the time in German. In this type of family, accent strength in Russian was mostly perceived to be "weak" (34% of the time) and less often as "strong" (17% of the time), while in German, the accent was perceived to be weak 44% of the time and strong 36% of the time. Bilingual children with one German-speaking parent were perceived as unaccented 40% of the time when speaking Russian, and only 19% of the time when speaking German. In this family type accents tended to be perceived as strong in Russian and as weak in German: 42% of the cases in Russian and 21% in German were rated as strongly accented; 18% of the cases in Russian and 60% of the cases in German were rated as weakly accented.
The effect of family type and age group on the perceived accentedness of the children was investigated with an ordinal mixed-effects logistic model in R (Ordinal package, Christensen and Christensen 2021), where accent score (ordinal variable taking values 0, 1 or 2) was predicted from family type, age group and their interaction, and "rater" was included as a random intercept term. For Russian, the model indicates a significant effect of family type, and a significant interaction between family type and age group, suggesting that the relation between perceived accentedness and age depends on the family type (Tables 2 and 3). For German, the interaction term was not included due to the incomplete information problem (Field et al. 2012, p. 322), resulting from the absence of "strong accent" ratings in German monolingual school children. The results are presented in Tables 4 and 5. For German, both the family type and age group are significantly associated with perceived accentedness. We conducted post hoc pairwise comparisons (Lenth 2016) in order to compare mean accentedness scores between different participant groups (see Appendix A for detailed results). The data are illustrated in Figure 2, which plots mean accent ratings by age group in the different populations (error bars indicate a 95% confidence interval). The analysis shows that for Russian, both family type and age significantly affect the perceived accentedness of the children. Thus, monolingual Russian preschoolers were rated as significantly less accented than their peers from bilingual families with two Russianspeaking parents (z ratio = −4.907, p < 0.0001), who, in turn, are rated as less accented than children coming from mixed heritage families (z ratio = −3.408, p = 0.0086). The same step-wise distribution is also evident for school-age children, where monolingual Russian children are perceived to be less accented than bilingual children from families where both parents speak Russian (z ratio = −7.696, p < 0.0001), who again are rated as less accented than children from families with one Russian-and one German-speaking parent (z ratio = −3.310, p = 0.012).
For Russian, the effect of age is evident in that preschoolers are perceived as less accented than school children. The effect holds for the bilingual children from Russian bilingual families (z ratio = −4.844, p < 0.0001), and for the children coming from mixed bilingual families, it is marginally significant (z ratio = −2.780, p = 0.06). At the same time, no difference in accentedness scores is found between monolingual preschoolers and schoolage children. For German, we find an overall effect of family type, such that monolingual German children are perceived as less accented than bilingual children with one Russianand one German-speaking parent (z ratio = −11.068, p < 0.0001), who, in turn, are rated as less accented than children with two Russian-speaking parents (z ratio = −2.414, p < 0.04).
The effect of age is significant overall, such that school-age children are perceived as less accented than preschoolers (z ratio = −4.340, p < 0.0001). Figure 3 plots, for each bilingual child, the mean accent score in German against the mean accent score in Russian. A Pearson's test did not show a significant correlation between the two scores (r = 0.16, p = 0.45). The visual inspection of the scatter plot indicates that there are children who are perceived as not having an accent in any language and other children who are perceived as foreign-accented in one of their languages, while very few children are perceived as strongly accented in both of their languages.

Perceived Accent in Monolingual Children
We predicted (P1) that monolingual children will occasionally be rated as accented but that the incidence of perceived foreignness will fade in the older children. This prediction was motivated by the fact that monolingual children develop their phonological systems gradually and, given individual variation, even listeners sensitized to hearing child speech may sometimes mistake a native accent for a foreign accent.
The monolingual children in our study were perceived to be (weakly) accented 12% (Russian) and 16% (German) of the time, respectively. Whether this small percentage of misclassifications is due to the raters' lack of familiarity with the dialectal features in the speech of individual children, or with difficulties distinguishing the idiosyncrasies of child speech from foreign accents, is hard to tell. The inter-rater reliability in our Russian study suggests that the ratings are reliable. However, they are somewhat less reliable for German, where dialectal variation is more typical, both in the language per se and also in our sample. Therefore, although we did not find that raters from specific areas (e.g., Berlin) systematically misclassified particular children (e.g., from Southern Germany), an overall effect of dialectal variation might be at play. Moreover, in previous studies with monolingual raters for various target languages, the highest rate of "misclassifications" (20%) was found in a study on German (Kupisch et al. 2014). 6 Future studies might want to avoid such effects by including homogeneous speaker samples, and raters who are trained phonetically, familiar with the dialects of the target system and with developmental features of child speech. However, such studies would come at the cost of ecological validity, because in real life situations the two target systems of bilinguals typically do show regional variation, and those people who judge their speech, even if unconsciously, rarely ever happen to be linguists.

Perceived Accent in the Majority Language of Bilingual Children
Our second prediction (P2) was that, when speaking German, the bilingual children would be perceived as being more accented than monolingual children due to CLI from Russian. This prediction is motivated by the literature on young bilingual children showing cross-linguistic interaction, i.e., influence into both the minority and the majority languages. Moreover, although AoO effects are not always visible in HSs during adulthood (see Stangen et al. 2015), they may (still) be visible during childhood. Given that the youngest children in our study were only four years old and partially from monolingual Russian families, their exposure to German had, in the most extreme cases, only started 1-2 years before the data were collected. The results show that, when speaking German, the bilingual children tended to be perceived as accented (79%), compared to only 16% of the time in monolingual children, although in both groups the accents were mostly perceived to be mild. Strong accents were the exception and occurred more often in children from exclusively Russian speaking families with a later AoO in German. Thus, both monolingual and bilingual children are sometimes perceived as accented, but the incidence of perceived foreignness is higher in the latter population. This indicates that developmental features, "foreign" features and, potentially, dialectal features are not always easy to tease apart.
We further assumed (P3) that perceived accentedness in German would decrease over time, since rating studies of adult early bilinguals had shown that adult early bilinguals are rarely ever deemed foreign-accented. It was indeed the case that the older children were perceived to be less accented than the younger children, and this trend occurred in both family types. While we need to be cautious in not over-interpreting our data, these results could be taken to suggest that speaking (only) the HL in the family does not harm the development of the ML, at least not as far as accentedness is concerned. Of course, strictly speaking, our study does not allow us to conclude that a foreign accent in the ML decreases over time, because we have not looked at the same children longitudinally. While we can only speculate that by the time these children are adults, they will sound just like monolinguals, previous studies with adult bilinguals (e.g., Kupisch et al. 2014;Lloyd-Smith et al. 2020) suggest that this is likely.

Perceived Accent in the Minority Language of Bilingual Children
We finally predicted (P4) that the incidence of perceived accents in Russian, the children's HL, would increase in the older bilingual children. To begin with, we found that bilingual children from Russian families or families in which both German and Russian were spoken were perceived to be accented more often (50% and 60% of the time, respectively) than monolingual Russian children (12%). However, they were less likely to be perceived as accented when speaking Russian than when speaking German. Their foreign accents, if perceived, were more likely to be strong when they came from families with one German parent (42%) than when coming from Russian-speaking families (17%). Since the children were exposed to Russian from birth, their perceived accents must be due to the influence of German, which is arguably stronger if German is spoken in the family. These findings are unexpected under the assumption that early exposure is sufficient to attain a native-like accent, but they are expected given previous findings of CLI in early bilingual children. We further observed that the likelihood of a perceived foreign accent was higher for school children than for preschoolers. This was also expected given the high incidence of perceived foreignness in adult HSs in previous studies (e.g., Kupisch et al. 2014Kupisch et al. , 2020Lloyd-Smith et al. 2020), as well as Wrembel et al. (2019) finding that Polish-English bilinguals in the UK are perceived to be more accented than their monolingual peers from Poland. However, it is somewhat counterintuitive that older children, despite having had more language experience with the target language than younger children, do not move closer to the target but further away, at least in terms of pronunciation.
Finally, although one might have expected an inverse correlation in accentedness in the children's two languages, such that children who sound native in German would sound foreign in Russian and vice versa, no such correlation was found. There are children who sound like monolinguals in both languages and children who are perceived as foreignaccented in one of their languages. The number of children being perceived as strongly accented in both languages is small. This is in line with previous studies on adult HSs in which both of the bilinguals' languages were compared.

Conclusions and Relevance
In summary, German-Russian children were deemed foreign-sounding significantly more often than monolingual children. This was the case in both languages. There was a difference between the majority language and the heritage language, such that a perceived accent was more likely in German. When speaking Russian, children from German-Russian families were more likely to be perceived as foreign than bilingual children with two Russian-speaking parents, while the opposite was found for German. The incidence of a perceived foreign accent decreased from younger (preschool) to older (primary school) children in German, while increasing for Russian. The latter may be caused by a gradual change in the relative exposure to both languages, as the beginning of school comes with more German input and qualitatively different input, including formal and written registers. This change in linguistic experience may be accompanied by changes in the HS's attitudes, caused by the prevalent ideologies in the educational settings of in countries with only one national language (e.g., Hornberger and Wang 2008;Valdéz et al. 2008). Germany represents such a setting.
The observed increase in accent over time in Russian, and its decrease in German must be seen with caution. The first and foremost reason is that we did not investigate individual children longitudinally. Although unlikely, it is possible that we have included school children who happened to be generally more accented in Russian, and preschoolers who happened to be generally more accented in German. Second, our study leaves open which properties exactly are subject to CLI and the related question what exactly the raters perceived to be "different" from monolinguals. Anecdotally, the raters have commented on the pronunciation of /r/, vowel quality and rhythm in German, vowel quality, lack of palatalization and stress in Russian. However, we cannot exclude that the raters also based their judgment on morpho-syntactic properties, even if we explicitly told them not to do so. This might have been the case especially for the younger children when speaking German, as they sometimes omitted articles and produced gender and case errors. Such properties are developmental, i.e., also found in monolingual German children, but in monolingual acquisition they typically occur at an earlier age. Lastly, there was a lot of individual variation, suggesting that factors beyond family type and age were at play. From a methodological perspective, these might include the small sample size as well as a higher degree of misclassifications (compared to previous studies with adults) due to less familiarity with dialectal or developmental features.
Finally, one could question the relevance of such a study, asking why it is important that bilinguals sound "monolingual-like", given that being bilingual is part of their identity and having an accent is part of being bilingual, just as it is typical to speak a standard language with a regional dialect if one comes from a region in which a dialect is (still) spoken. In other words, do we, by studying bilingual children's accents, promote the idea of a "monolingual habitus" (Gogolin 2008)? It is, of course, not our intention to promote a "monolingual habitus", nor do we subscribe to the idea that there is only one "correct" way of speaking a language. On the other hand, it is a fact that a foreign accent is a feature that can reveal a person's origin and identity. If there are positive associations with this origin, this is beneficial. However, an accent may also have negative associations. As shown in a large-scale study by Gärtig et al. (2010), Germans show positive attitudes towards French-, Italian-, English-or Spanish-sounding foreign accents, but not necessarily for certain other accents, and a third of the respondents mentioned comprehension difficulties in conversations with migrants. Such attitudes can affect people's well-being. One example is that young people with Turkish citizenship sometimes report that people in Turkey identify them as coming from Germany, while in Germany they count as "the Turkish", which could make them feel like strangers in both countries ). As to the majority language, potential employers may unfairly discriminate against applicants with non-native sounding accents due to stereotyping or cultural biases (Munro 2003;Hosoda and Stone-Romero 2008). From a linguistic perspective, it will be crucial to investigate in future studies whether differences in phonological production are mirrored in perception. Perception differences might have implications for the acquisition of morpho-syntactic properties, e.g., gender and number, which are typically marked by word-final sounds (Colantoni et al. 2020). If these are perceived differently, they may also be produced differently. Taking these points together, the choice about whether or not to consciously develop one's accents, e.g., by increasing use and/or exposure, should be made by the speakers themselves, and a first step towards this is to gain knowledge about when a foreign accent emerges.