Quantifying Experience with Accented Speech to Study Monolingual and Bilingual School-Aged Children’s Speech Processing

Abstract

Children around the world often grow up with multiple language varieties and are exposed to regional and second-language accents. This linguistic heterogeneity presents both benefits and challenges for cognitive and language development. Recognizing the importance of input variability in theories of language processing, researchers are now using more nuanced assessments of language experience that go beyond simple ‘monolingual’ versus ‘bilingual’ categories. These assessment methods capture the gradient nature of language exposure and use. This article provides a narrative review of recent research on the role of different accents and languages in children’s environments. It emphasizes the importance of applying gradient assessments of accent variation to both bilingual and monolingual populations. In doing so, a more comprehensive understanding of speech processing in heterogeneous contexts among school-aged children can be achieved.

1. Introduction

Across many parts of the world, it is common for children to grow up hearing linguistically diverse languages and varieties (Grosjean, 2010; De Houwer, 2021). For example, in Calcutta, India, a child may grow up with different standard and regional varieties of Bengali as well as English (Satyanath, 2015). In some parts of Mexico, children learn indigenous languages at home but use Spanish at school (Mulík et al., 2021). In China, children immigrating or migrating from rural areas to urban centers encounter dialect variability (Dong, 2018). In many European countries, including Great Britain (e.g., Jeffries, 2016; McCarthy & Evans, 2019), Sweden (Bodén, 2011), Germany, and Switzerland (e.g., Levy & Hanulíková, 2023), numerous children grow up learning a standard variety alongside a regional variety while also hearing learner varieties, or second-language (L2) accents. This is particularly true in many areas where these varieties are spoken alongside a standard variety (e.g., König et al., 2020; Auer, 2011). Clearly, the type and amount of variation that children experience and are exposed to differ considerably across regions, social contexts, and families. Traditionally, such variation has often been overlooked in psycholinguistic studies on speech processing. The paper aims to discuss recent studies that have investigated the role of such heterogeneous input on speech processing and how researchers measure such input variability to examine children’s spoken language processing. By ‘variability’, we refer to the range of differences in accents as well as in speech styles or language use more generally. It reflects the diversity of speech patterns across different social, regional, and language contexts and includes any differences in how language is produced, understood, or adapted across settings or individuals. An accent refers to the distinctive phonetic characteristics, including segmental and prosodic, of speech that reflect the speaker’s regional origin, language background, language proficiency, or sociocultural context. There is no individual who speaks accent-free; all speakers have an accent. In all cases, accents are dynamic, reflecting a relationship between language identity, social interaction, and language use. The focus of this review will be on first-language (L1) regional accents and second-language (L2) accents. We adopt a narrative approach to provide an overview of the topic, with a focus on summarizing and synthesizing key ideas.

2. Why Does Variability Matter?

The effects of heterogeneous speech input on children’s language development have become a topic of increased scientific interest. Understanding the impact of heterogeneous environments on cognitive and language development helps address challenges related to supporting heritage languages and providing informed advice to parents, teachers, and speech therapists. In many countries across the globe, a significant portion of children grow up in multilingual environments, making this issue particularly relevant. In countries like Germany, at least one in five children under six does not speak German as the predominant language at home (Federal Ministry for Family Affairs1). According to the 2020 Children and Youth Immigration Report, around 35% of children in Germany have an immigration background (Deutsches Jugendinstitut e.V. et al., 2020). There have been public discussions calling for German to be the home language in families with an immigration background (e.g., Staubhaar, 2018) and urging parents to use German regardless of their proficiency or how comfortable they feel using the language. This has caused insecurity among many parents and prompted scientists to oppose such recommendations, because there is no scientific or clinical evidence that acquiring another language requires parents and children to stop using their family language or dialect (Bialystok, 2001; De Houwer, 2021).

The diversity of factors in a child’s environment, including the quantity and quality of the input, shapes language skills (Kidd et al., 2018). Parents and caretakers exhibit large individual differences in language skills, regardless of whether they grew up speaking one or two languages. Similarly, children show individual variation in the number of interlocutors and their respective language varieties. Moreover, many parents who perceive themselves as late learners and who consider their own speech as accented report that their children do not have an accent in their speech (Chambers, 2002). While variability in input might in theory lead to variation in language development, recent studies suggest that what matters most is not the quantity of first- (L1) or second-language (L2) speakers in the home environment but the quality of interaction and the proficiency of any given speaker (Kidd et al., 2018; Unsworth et al., 2019). Similarly, lexical diversity appears to be a more important predictor of language development than input quantity, and lexical diversity is strongly associated with socioeconomic status, particular among North American families (for an overview, see Kidd et al., 2018), though similar patterns have been observed across many regions.

Only a few studies have examined the effects of long-term accent exposure on monolingual and multilingual children’s speech development, including the roles of regional and second-language accents across various languages and cultures. These studies tend to focus on production (e.g., Khattab, 2007; Levy & Hanulíková, 2019) or comprehension of unfamiliar accents (e.g., Leung, 2012; Durrant, 2014; McDonald et al., 2018; Levy et al., 2019). Much of this research examines infants or preschool children and often merely compares monolingual and bilingual groups without an in-depth analysis of the input children receive.

It is astonishing how little we still know about the long-term consequences of diverse accent experience and how it affects both production and perception. The processing of second-language and regional accents is co-modulated by the acoustic and perceptual characteristics of the accents themselves. Research has shown that several factors play a role in this process: familiarity with the accent, listening effort required, and pronunciation distance from the listener’s own variety (Van Engen & Peelle, 2014; Bent, 2018; Hanulíková, 2019, 2021; Levy et al., 2019; Weatherhead et al., 2019; Bent et al., 2021). Listening effort is linked to the processing load experienced when there is a mismatch between the speaker’s pronunciation and the listener’s variety. The greater this mismatch, the more effort is required (Van Engen & Peelle, 2014). To quantify these mismatches, Levy et al. (2019) employed the Levenshtein distance measure, a metric used to assess the similarity between two strings (e.g., a standard variant vs. a regional variant). Their findings suggest that children might struggle more with unfamiliar regional accents of their first language than with second-language accents. Subsequent studies have confirmed and expanded on these effects of pronunciation distance across various first- and second-language varieties (Bent et al., 2021; Hanulíková, 2021, 2024). These findings challenge the common assumption that first-language varieties are inherently more intelligible, cognitively less demanding, and socially more acceptable than second-language speech. Instead, they suggest that the relationship between first- and second-language speech processing is more complex and dependent on the specific acoustic and perceptual features of the accents involved. Therefore, we need a more in-depth analysis of the respective accents and varieties in the input children receive.

Questions regarding the impact of heterogeneous speech input on spoken language processing are connected to theoretical questions about how words are stored in the mental lexicon, whether lexical representations are flexible or unstable, and how they are accessed. Many theories assume that upon hearing a word, listeners map it onto its representation in the mental lexicon (Cutler, 2012; McQueen, 2005), but there is disagreement with respect to the nature of the representations. In a speech community where a certain pronunciation variant (e.g., in a regional accent or colloquial speech style) is highly frequent alongside a standard pronunciation form, both variants might be stored and represented. For example, in Swabian-accented German, the half-open and half-closed unrounded front vowels /ɛː/ and /eː/ are often used interchangeably. Speakers might have representations for both the Standard German pronunciation /ˈleːʁɐ/ and the Swabian variant /ˈlɛːʁɐ/ (Lehrer, ‘teacher’; for more examples, see also Sumner & Samuel, 2009, for standard (rhotic) General American and regional (non-rhotic) NYC-English variants; Bürki et al., 2010, for French variants of the word ‘fenêtre’ with and without schwa; Hanulíková & Weber, 2012 for L2-variants).

Storing two representations of a word (e.g., one standard variant and one regional variant) might lead to greater perceptual flexibility when accessing words and to greater variability in production. Bilinguals have been shown to benefit from their experience with two phonological systems, and children exposed to different accents might develop similar flexibility in speech processing. Compared to monolinguals, bilinguals may demonstrate superior cognitive control and attention in some tasks, which helps them deduce patterns (Bialystok, 2001; van Hell & Poarch, 2014), recognize subtle acoustic cues (Muench, 2011), and even enhance word learning (Kaushanskaya et al., 2014; Menjivar & Akhtar, 2017; Skoruppa, 2019; but see meta-analysis by Lehtonen et al., 2018, showing no clear cognitive advantage for bilinguals). Phonetically diverse input may lead to learning strategies that bilinguals—and possibly bidialectals (children growing up with two regional varieties or typologically related dialects)—can benefit from, especially in situations where they are exposed to phonologically variable speech. For example, when listening to unfamiliar accents, linking phonologically variable speech input to previously stored lexical representations may be easier for children who grow up in heterogeneous speech environments. Frequent exposure to different accents may help in inferring commonalities within phonological categories by drawing on sampled information about a speaker’s pronunciation (Singh et al., 2018; De Bree et al., 2017).

On the other hand, increased phonological variability due to more than one language or accent in a child’s input may lead to less stable lexical representations (Bosch & Ramon-Casas, 2011; Baker & Trofimovich, 2005). The interaction between different phonological systems in a bilingual or bidialectal child may influence speech processing and production such that vowel categories are merged (Gildersleeve-Neumann et al., 2008; Khattab, 2007; Marecka et al., 2015; McCarthy et al., 2014), even in early bilinguals (children who acquire two languages from a very young age, Darcy & Krüger, 2012). Exposure to different pronunciation variants may result in increased competition between lexical representations, leading to processing costs (Clopper, 2014). It remains unclear whether heterogeneous speech input leads to greater flexibility, less stability, or increased phonological variability.

Finally, learning in the context of speech variability may initially be slower and more challenging, but it ultimately leads to better generalization (Raviv et al., 2022) and learning (Perry et al., 2010). This seems to be true for phonological learning (Bradlow & Bent, 2008; Sumner, 2011; Sadakata & McQueen, 2013; Van Heugten & Johnson, 2014; Levy et al., 2019; Hanulíková, 2021, 2024; Rost & Murray, 2010; Lev-Ari, 2018), lexical learning (Barcroft & Sommers, 2005, 2014), as well as for morphosyntactic processing (Engel & Hanulíková, 2020; Hanulíková et al., 2012). Note that some studies fail to find positive effects of variability (e.g., Brekelmans et al., 2022).

Given that bilingual, bidialectal, and monolingual experiences are dynamic in nature, the inclusion of a more nuanced analysis of the respective accents and varieties in the input is crucial for both applied and theoretical contexts. But how do researchers quantify children’s experience?

3. Quantifying Experience with Languages and Accents

One of the many challenges in research on variability is the assessment of the input children are exposed to. For very young children, daylong recordings have been used and provide rich information about the input (Casillas & Cristia, 2019). Alternative approaches, particularly for older children, rely on questionnaires and self-assessments of monolingual and bilingual speakers. Most of these approaches treat bilingualism as a categorical variable (e.g., Darcy & Krüger, 2012; Evans & Lourido, 2019; Kroll & Bialystok, 2013), using criteria such as parental native language (e.g., Bialystok et al., 2010; Buac et al., 2016), fluency in another language (e.g., Lee & Iverson, 2012; Menjivar & Akhtar, 2017), or whether the child’s input contained at least 10% exposure to another language (e.g., Ribot et al., 2018; Skoruppa, 2019). However, such criteria are often arbitrary, and it is difficult to compare studies that use disparate criteria for defining monolinguals and bilinguals (De Bruin, 2019). Studies that focus on second-language acquisition consider diverse variables such as age of acquisition (Birdsong, 2005; Flege et al., 1999; Johnson & Newport, 1989), L2 proficiency, and language dominance (e.g., Bialystok & Feng, 2009; Blumenfeld & Marian, 2007; Hanulíková & Ekström, 2017). A more recent trend is to use the amount of input in one language in order to examine how it influences performance in the other language or in a newly learned second language (Yow & Li, 2015; Buac, 2019; Blom, 2010; Flege & Wayland, 2019; Marian & Hayakawa, 2021; Hanulíková, 2023). Many of these studies test adults and compare different groups of bilinguals (e.g., high/low L2 exposure). In the next section, we take a closer look at how experience tends to be quantified across studies on adults and children.

3.1. Quantifying Language Experience

Measuring the amount of language experience is challenging, and several approaches have been proposed, particularly in the context of adult L2 acquisition (e.g., Flege et al., 1997; Gullifer & Titone, 2019; Lloyd-Smith et al., 2020; Piske et al., 2001; Sulpizio et al., 2020; Hanulíková, 2023). Standardized L2 exposure questionnaires have been used to establish variables such as daily exposure or daily language use in different communicative contexts. For example, Lloyd-Smith et al. (2020) developed a scale for calculating language experience across the lifespan (for adults), investigating whether the amount of Italian use leads to an L2 accent in the L1 (Italian) and in the L2 (German) of adults living in Germany. Their ‘Italian Use Score’ assesses Italian heritage language use across 20 aspects of language use in a detailed questionnaire, with categories such as ‘current Italian use’, ‘time spent in Italy’, and ‘Italian use at home’, and then assigns weights to these scores. Gullifer and Titone (2019) also use a fine-grained scale that characterizes the degree of balance between a bilingual’s two (or more) languages in daily usage. They collected information about participants’ language use in different contexts (e.g., “Please rate the amount of time you use each language at home”) with scores on a 1–7 scale (“no usage at all” up to “usage all the time”) to then assign proportions of usage of the two languages in the different contexts (p. 10). Using principal component analysis, they determined the impact of the individual variables on self-rated L2 accentedness and proficiency. Similarly, Hanulíková (2003) used self-reported daily L1 use of bilinguals to calculate active bilingual usage as a continuum to predict articulatory accuracy in an L3.

Some studies with multilingual children use measures of the amount of input in children’s respective languages based on parental questionnaires (e.g., Tuller, 2015; Uzal et al., 2015; Buac, 2019; Levy & Hanulíková, 2019, 2023). This approach is part of a broader discussion in the psycholinguistic community, where the search for a continuous measurement of bilingualism has gained increasing attention. Marian and Hayakawa (2021) summarize the available tools used to quantify bilingualism and emphasize the need for a collaborative effort to develop a bilingualism score or quotient. They recognize that the many variables influencing bilingualism cannot be simplified to just a few relevant factors and that interactions between variables and between populations must be taken into account. As a common goal for bilingualism research, they urge scientists to present very detailed information on their bilingual and monolingual populations in order to increase comparability, facilitate meta-analyses, and discuss the feasibility of a bilingualism quotient similar to the established intelligence quotient. While the number of studies considering bilingual assessment steadily increases, monolinguals are less likely to be assessed in a similar fashion. Instead, monolinguals are frequently used as a “control” group. The gradient assessment approach is relevant for both bilinguals with varying degrees of exposure to multiple languages, dialects, and accents as well as monolinguals exposed to different accents or dialects and, at times, also to different languages (see Figure 1).

The use of continuous measurement can be extended to neurolinguistic studies as well (Baum & Titone, 2014). Sulpizio et al. (2020) applied functional connectivity to examine the effect of age of acquisition (AoA), proficiency, and current language use on brain plasticity in bilingual adults. They suggest that a continuous variable capturing language use is more informative than a categorical treatment (bilingual vs. monolingual). This is because categorical groups may lead to heterogeneous categories, obscuring the effect of language experience on neural changes. They found that both the executive control network and the language network (important loci of bilingual language processing) were influenced by all three bilingual experience variables. Sulpizio and colleagues propose that bilingualism should be measured using gradient scales (see also DeLuca et al., 2020; and see Baum & Titone, 2014, for a neurolinguistic discussion on the disadvantages of binary monolingual and bilingual groups).

3.2. Continuous Measures of Language Experience

Child researchers mostly continue using categorical measures for language background/experience, which can be a useful and less complex approach for a range of research questions. Surrain and Luk (2019) compared 186 studies on monolingual and bilingual general language outcomes, cognitive measures, and literacy, by considering two variables of self-reported language use: a categorical measure (monolingual vs. bilingual) and a gradient measure of home usage. The majority (147) of these studies reported information on language use. Among these, half of the studies employed a categorical variable (monolingual vs. bilingual), while the other half used a gradient measure such as an estimated percentage of language use per week. However, of the 111 of the studies that focused on child participants, only 36% reported language use on a gradient scale. This is surprising, especially because research on bilingual development has shown that the amount of input children receive in each of their two languages predicts language outcomes in bilingual children (e.g., Cheung et al., 2019; De Cat, 2020).

In order to measure language experience in bilingual children, researchers have recently developed more detailed questionnaires that take into account factors such as the age at which the second language was first acquired, the number of hours per week each language is spoken, and the number of interlocutors for each language in a bilingual’s environment (Buac, 2019; De Bruin, 2019). Similarly, Gutiérrez-Clellen and Kreiter (2003) used variables like years of exposure, amount of home language input, school language input, and amount of reading (all based on parent and teacher questionnaires) to examine grammatical performance during narration in 7- to 8-year-old Spanish-English bilinguals in California. In general, their data from parent and teacher questionnaires were well suited for predicting bilingual proficiency.

With respect to continuous measures of bilingual exposure, the main difference from Buac’s (2019) study is that Gutiérrez-Clellen and Kreiter did not collapse the individual measures into one variable but kept them separate, which could create a multicollinearity problem in statistical models (see also Levy et al., 2019; Levy & Hanulíková, 2023). Although Gutiérrez-Clellen and Kreiter’s study focuses on grammatical performance, the operationalization of language experience is very detailed, including factors such as media use in each language, parents and teachers’ estimation of proficiency, and input. A similar approach is taken by Tuller (2015), who uses measures of quantity and quality of input in the LITMUS-PABIQ questionnaire to diagnose language impairment in bilinguals. A commonly used questionnaire for assessing bilingual language experience and proficiency, the LEAP-Q (Marian et al., 2007), has also been adapted for use with children and translated into 26 languages. It considers language proficiency, age of acquisition, language preference, and, importantly, the amount of language exposure. Exposure to each language is measured on a scale of zero (never) to ten (always) for different contexts such as interacting with family, watching television, or interacting with friends. Parents are also asked to estimate the percentage of current exposure to each language.

Other tools used specifically with child populations and based on parental questionnaires include the Language Exposure Assessment Tool (DeAnda et al., 2016), the Alberta Language and Development Questionnaire (Paradis et al., 2010), and the Bilingual Language Profile (BLP, Gertken et al., 2014). Finally, Levy and colleagues developed parent and school staff questionnaires to assess monolingual and bilingual children’s exposure to diverse languages, accents, and dialects, including the home environment as well as school and leisure time activities (Levy et al., 2019; Levy & Hanulíková, 2019, 2023).

Bedore et al. (2012) present one of the few studies to directly compare the use of a continuous variable based on input measures and a categorical variable (monolingual/bilingual), using a large sample (n = 1029) of Spanish-English bilingual pre-kindergarteners and kindergarteners. They examined the effects of different experience measures on proficiency and language dominance to determine whether these different variables would lead to the same classification into monolingual and bilingual groups as in traditional studies with binary categories. They found that the amount of language use was a good predictor of performance (English morphosyntactic and semantic development). Importantly, their results showed that current use, more than age of acquisition, accounted for much of the variance in language dominance. Some of the children who would be categorized as monolingual English speakers based on interview data were able to perform relatively well in Spanish. They argued that current language use is useful for operationalizing bilingual development (for an overview of how experiential factors relate to bilingual children’s developing language skills, see also Unsworth, 2016; Cattani et al., 2014; Ribot et al., 2018; Cohen, 2016).

Taken together, studies that measure experience on gradient scales, taking into consideration the amount of input in each language, show that although more complex, these measurements provide more detailed predictions of the source of variability in language performance. Most importantly, they show that the amount of current language experience (e.g., measured in hours per week) can be more informative than the traditional binary grouping or age of acquisition measures. We now turn to studies that include assessments of accented speech in a child’s environment.

3.3. Measuring Accent Experience

Child research that focuses on experience with accented speech is more limited compared to studies measuring language experience. Most studies categorize groups of children with regional accent experience as subgroups of monolingual populations or compare bilingual and bidialectal experience (Durrant, 2014; Ross & Melinger, 2017; Lee-James & Washington, 2018). Durrant (2014) defines bidialectals as infants with at least one parent who speaks a dialect that differs from the local area. She finds that these infants perform similarly to bilinguals in mispronunciation detection tasks and show comparable production variability. Durrant classifies these infants as a distinct group within monolinguals and does not consider that bilingual children often also hear different regional varieties or that the amount of input in dialects differs widely across speech communities. In such cases, it might be more viable to use continuous measures of input instead of categorical groups and subgroups. Francot et al. (2017) explicitly investigated the explanatory power of grouping children into monodialectal and bidialectal groups. They tested whether distinct monodialectal and bidalectal groups would emerge based on vocabulary size in Standard Dutch and the Limburg dialect in 5–9-year-olds growing up in bidialectal Limburg. The results showed large interindividual differences in the vocabulary test. Children produced responses that fell along a Dutch-Limburgian continuum, with no child using only one variety. The authors suggested that a classification into monodialectal and bidialectal is not feasible, at least for this group. This result could be explained by means of linguistic change in the Limburg region, such as innovation and dialect leveling, as well as children’s mixed exposure to different language varieties—Standard Dutch at school and dialect at home. Additionally, the experimental setting may have encouraged children to respond in Standard Dutch more often than they would in a more informal setting. Francot et al.’s results suggest that language use cannot be neatly classified into categorical groups. Instead, the findings support the use of continuous measures to capture accent experience.

Another possible limitation of the studies described above is that they often consider only accent exposure in the home language environment. In the majority of studies, bidialectal children are grouped according to their parent’s dialect use, and parents are asked to rate their children’s accent exposure. Accent exposure outside the home, such as at school or in the local community, is not captured by these measures. Importantly, parents may not be aware of the extent to which their schoolchildren are exposed to accented speech outside the home. Several studies have demonstrated the importance of peer groups and environments outside the home for children’s language development and their exposure to accented speech (Floccia et al., 2012; Francot et al., 2017; Tagliamonte & Molfenter, 2007). These studies suggest that the accents that children are exposed to in their speech communities (e.g., at school, during their leisure time, and when interacting with friends) should be accounted for. One possibility is to consider teacher questionnaires as a complement to parental questionnaires (Gutiérrez-Clellen & Kreiter, 2003; Levy et al., 2019).

An alternative approach is a fine-grained method for quantifying children’s accent and language input suggested by Buac (2019). Buac explored word learning from Spanish-English accented speech in bilingual preschoolers and developed a Summary of Exposure to Accent Questionnaire (SEA-Q). Based on this parental questionnaire, she measured exposure to Spanish-accented English using the following variables: the length of bilingualism (child’s age minus AoA in years), the number of L2 English speakers who regularly interact with the child, and the strength of the L2 accent in English in the child’s input. The latter was calculated using the formula “(proportion of time spent using English when interacting with child × strength of accent in English) ÷ (number of days the person interacts with child ÷ 7 days of the week)” (Buac, 2019, p. 67). While this method captures L2 accent experience in a detailed manner, it is limited to exposure to second-language accents. Exposure to regional accents may not have been feasible for the population tested but is important in many speech communities, linguistic landscapes, and geographic regions. In addition, exposure to second-language accents was equated with exposure to Spanish accents, but it remains unclear whether children also heard other non-native accents. Finally, language exposure was operationalized in terms of age of acquisition and the number of L2-accented speakers in the child’s environment, without directly reflecting current exposure to different speakers and accents.

A slightly different approach has been suggested by Levy et al. (2019), see also (Levy & Hanulíková, 2019, 2023; Hanulíková, 2024), using a continuous scale to measure accent and language experience across monolingual and bilingual children. To capture variables such as language, dialect, and accent input (regional and second-language), Levy and colleagues used parental and teacher questionnaires to assess home and school environments (including the school’s afternoon care), applying a largely emic perspective. Parents, teachers, and caretakers at school were asked which languages (e.g., German, Turkish), dialects (Bavarian, Swabian), and accents (French accent in German, Swiss accent) they use when interacting with their child and which languages and accents the child hears from other adults such as grandparents, from siblings, friends, at leisure time activities, and when using media. They were also asked how many hours per week the child spent with these persons and activities. Parents also rated their own pronunciation in German and the pronunciation of other caregivers. Based on this information, Levy and colleagues estimated the percentage of input each child received in Standard German, regionally accented German, L2-accented German, and other languages. All key caregivers who interact with the children in their class or in after-school groups provided self-assessments of their own pronunciation in German and evaluated the children’s accents. These measures were then used to calculate a percentage exposure score for regional and L2 accents as well as for Standard German and another language (see Figure 1).

Figure 1 shows pie charts visualizing the percentage of input in different varieties per child and averaged across monolinguals and bilinguals. It shows that among these primary-school children from southern Germany, there is hardly one idealized monolingual speaker (hearing a standard variety 100% of the time) or an ideal bilingual child (hearing two languages or varieties equally often). The figure shows that all these children vary substantially in their input, with some monolingual children showing exposure to L2 accents and sometimes even to other languages. Similarly, some bilingual children show limited input in their first language and substantial input in regional varieties of German. Admittedly, one of the weaknesses of any questionnaire such as this is its reliance on self-assessments of what it means to speak a standard variety or to speak with an accent and how respondents perceive these concepts. However, the more information we receive from different caregivers and teachers, the more likely we are to approximate the relative heterogeneity of the child’s speech environment.

In their studies, Levy and colleagues used this continuous assessment and demonstrated that speech processing and word learning are not compromised when children are exposed to regional and L2 accents as well as to different languages. For example, using a sentence repetition task, Levy et al. (2019) showed that repetition of utterances with familiar and unfamiliar regional accents was facilitated for children with a large amount of regional accent experience. In a picture-naming task, Levy and Hanulíková (2019) found that experience with variable input led to greater variability in the production of some German vowels. Children who frequently heard different accents produced more variability in vowels than children who experienced less variable input. This variability did not necessarily lead to children being judged as sounding more accented in general (Levy & Hanulíková, 2022). Instead, it was primarily the amount of input in another language that predicted the rating of words and vowels as sounding accented. Finally, a study on word learning (Levy & Hanulíková, 2023) found that primary-school children benefit from experience with regional and second-language varieties when learning new words from different speakers with unfamiliar accents. Taken together, although accent exposure has been shown to lead to greater variability in production, it has also led to greater flexibility in speech perception and word learning. Children exposed to a large amount of variable speech input appear to have benefited from this experience and were able to generalize this experience to new, unfamiliar accents.

Importantly, Levy and colleagues found no differences between monolingual and bilingual children in unfamiliar accent perception or word learning, nor in related research on social preferences (Hanulíková, 2024). The only differences observed were in vowel measurements between monolinguals and bilinguals, which were interpreted, in line with previous studies, as stemming from the interaction between different languages. Moreover, apparent acoustic differences between the vowels of monolinguals and bilinguals were not significant when factors measuring accent experience and language input were considered. These studies further challenge the exclusive use of categorical comparisons and show that input in different accents and languages modulates but does not negatively impact speech processing, production, and learning. Particularly in tasks that involve unfamiliar accents, variable input can at times even improve school-aged children’s perceptual adaptation and learning. Finally, the effects of variability on speech processing appear to apply equally to monolinguals and bilinguals, highlighting similarity rather than differences in underlying processes.

Table A1. Studies that operationalize accent exposure as continuous/scale variables.

Study	Variety	Method(s)	Age (N)	Exposure	Key Result(s)
(Buac, 2019) (Exp. 1)	Spanish- and Korean-accented English	Word learning (eye-tracking)	4–5 yo (50)	Ordinal 5-point scale from ‘no experience’ to ‘daily home exposure’	Experience with an L2 accent enhances learning from a speaker with the same accent but not from a speaker with a different L2 accent.
(Buac, 2019) (Exp. 2–3)	Spanish- and Korean-accented English	Word learning (eye-tracking)	4–5 yo (95)	Length of bilingualism (child’s age minus age of acquisition), number of non-native speakers interacting with the child, strength of L2 accent in English	A high amount of Spanish accent experience reduces word learning from American English native speakers; a higher number of L2-accented speakers is associated with lower American English language skills.
(Levy et al., 2019)	Standard German, second-language- and regional-accented German	Speech intelligibility (sentence repetition)	8–11 yo (33 monolinguals, 27 bilinguals)	Number of hours per week a child spends with Standard German, languages other than German, regional and L2 accents of German; information provided by parents and school-personnel for diverse activities and interactions at home, at school, during leisure time activities, and with media.	More experience with regional accents improved sentence repetition performance in the regional and in the standard accent. More experience with L2 accents did not help in either accent condition.
(Levy & Hanulíková, 2019)	Standard German L2- and regional-accented German	Vowel production (picture naming)	8–11 yo (33 monolinguals, 27 bilinguals)	As in (Levy et al., 2019)	Increased exposure to input variability leads to greater variability in vowel production (measured by Euclidean distances). Bilinguals did not show greater variability compared to monolinguals, but there were some differences in F1 formant values.
(Levy & Hanulíková, 2023)	Standard German, L2- and regional-accented German	Word learning (spot-it-paradigm)	7–11 yo (43 monolinguals, 45 bilinguals)	As in (Levy et al., 2019)	Successful word learning was predicted by the amount of input in regional and L2 accents but not by exposure to other languages (i.e., by bilingualism).
(Poarch et al., 2019)	Standard German and Swabian German	Executive function (Flanker & Simon tasks)	Adults (34)	Daily language usage with family, friends, at work/university, collapsed into one (%) variable, used together with proficiency measurements to create a Swabian dominance score *	Balanced bidialectals perform worse on two executive function tasks than Swabian-dominant bidialectals.
(Porretta et al., 2016)	Chinese-accented English	Cross-modal priming (visual world paradigm)	Adults (96)	Number of weekly interactions with non-native speakers of English on a scale from 0 (Never) to 10 (Daily), converted to a proportion by dividing by 10, multiplied by the percentage of those interactions including speakers with a Chinese accent (range = 0–100)	Accent experience leads to higher activation strength and improves the time course of word recognition in accented speech.
(Porretta & Tucker, 2019)	Chinese-accented English	Pupil dilation	Adults (85)	As in (Porretta et al., 2016)	Accent experience reduces processing effort.
(Porretta et al., 2020)	Chinese-accented English	Predictive processing (visual world paradigm)	Adults (60)	Participant estimation of their total experience interacting with speakers with a Chinese accent as a percentage of their lifetime interactions (range = 0–30)	Accent experience leads to an advantage in predictive processing in accented speech.

* See https://sites.la.utexas.edu/bilingual/scoring-and-interpreting-the-results/ accessed on 15 February 2023, for guidelines on how to compute bilingual/bidialectal dominance scores.

in Appendix A gives an overview of some studies that operationalized long-term accent experience as a continuous variable. Because there are only a few such studies, the table includes research with both child and adult participants. The table shows that attempts to quantify accent experience are relatively recent and include different scales, ranging from self-ratings and ordinal measures of exposure that consider only the number of interactions with accent speakers to the more detailed scales that include measures of speaker number and the degree of accentedness (Buac, 2019). There are only a handful of studies that include continuous scales for both regional and L2 accent experience (see Levy and colleagues).

4. Directions for Future Research

While there has been extensive research on speech perception, production, and word learning in monolinguals and bilinguals, few studies have incorporated measures of the variety of input children receive, especially as a continuous measure in combination with language experience (see De Cat, 2020, and Yow & Li, 2015, for similar approaches). Levy et al.’s (2019) approach advances this area by measuring the amount of input in other languages and regional varieties of German, as well as regional and L2 accents, offering a more fine-grained operationalization compared to categorization into groups (e.g., Durrant, 2014) or levels on a scale (Porretta et al., 2016). However, assessing exposure in primary-school children remains challenging, and there is room for methodological considerations.

One challenge in achieving comparability across studies lies in the distinction between etic and emic perspectives on concepts such as language, standard, dialect, and accent. The etic perspective refers to an external and analytical description of a given concept provided by researchers, focusing on objective and observable aspects. In contrast, the emic perspective describes the internal, experiential view of speakers within a speech community, reflecting lay classification and perception. Many studies that quantify experience with accented speech adopt an etic approach, as they rely on externally defined categories (e.g., standard, dialect, foreign accent, regional accent). However, what may seem like clear distinctions from an etic perspective might not align with the lived experiences and classifications across individuals and diverse social and cultural contexts. Incorporating emic perspectives can thus provide valuable insights into how speakers themselves conceptualize and perceive language variation. Several factors influence these perceptions, rooted in each country’s specific linguistic landscape, the cultural and historical background, distribution and usage patterns of languages, and the relative prestige associated with different varieties (e.g., Hanulíková, 2021; Fiedler et al., 2019). For example, a language variety classified as a dialect in one country may be considered a separate language in another (e.g., in cases like the Chinese “dialects”). Constructing unified questionnaires requires careful consideration of these linguistic landscapes and cultural contexts. At the same time, balancing both etic and emic approaches remains challenging. Future research should aim for greater transparency in how these two perspectives may influence research design and interpretation. Most questionnaires focus on children’s input and do not sufficiently assess the child’s language output and use across diverse social contexts. An alternative approach, as demonstrated by Braun et al. (2021), involves using an app to record children’s naturalistic home interactions between parents and their mono- or bidialectal German-learning toddlers. Another method to capture children’s input in a more detailed way is through naturalistic observation, which has been primarily used with young children (see Frank et al., 2021). However, daylong recordings, as used for infants and young children (Casillas & Cristia, 2019), may be less feasible for school-aged children due to time restrictions and ethical issues related to data privacy. Nonetheless, some studies suggest that patterns of input and output are often strongly correlated in children (Pearson, 2007; also see Unsworth et al., 2019), which might justify prioritizing input measures in some cases. To address this gap, future studies could explore alternative methods such as structured observations, teacher reports, and shorter, targeted recordings, which may offer practical and ethically viable ways to capture children’s language use in different contexts.

Parental and teacher questionnaires are commonly used and can be adequate tools for quantifying input, because they provide information on children’s language exposure at home and outside the family (see Frank et al., 2021, for a recent discussion on methods for examining children’s language learning, with a focus on parent reports). However, they might not provide a complete picture of children’s language environments due to several limitations. Parental questionnaires rely on self-report data, which can be influenced by assumptions about language use rather than direct observations. In addition, parents may not always be aware of the full extent of their child’s language exposure, especially in school or diverse social contexts. For instance, Levy et al. (2019) supplemented parental questionnaires with additional surveys completed by schoolteachers and afternoon-school staff. These surveys revealed discrepancies between parents’ assumptions about Standard German usage during school hours and the actual language use by afternoon teachers, who sometimes spoke L2 or regional accents (based on their self-assessments or researchers’ personal interactions with the teachers). This suggests that these interactions and naturalistic observations provide important contextual information that complements questionnaire data.

To improve quantification methods, researchers could benefit from conducting interviews with both parents and teachers. For example, Rydland and Grøver (2021) used interviews to assess both the quantity and quality of parental input in predicting the vocabulary skills of bilingual children in Norway. They employed trained research assistants fluent in the parents’ first language, which likely yielded more accurate results than questionnaires administered in the majority language, as not all parents may fully understand them. However, Rydland and Grøver relied on a relatively simple scale for assessing language exposure, categorizing it into three levels: mostly Norwegian, both languages, or the other language. Future research might benefit from combining interviews with more fine-grained quantification measures, such as recording the number of hours instead of relying on broad levels.

As suggested by several studies, the operationalization of accent experience could also include the proficiency level of the caretakers who provide the child’s input (Hoff & Shanks, 2020; Core, 2020; Unsworth et al., 2019). These studies demonstrate that the quality of child-directed speech, particularly the proficiency of a speaker, significantly impacts children’s language skills (e.g., vocabulary size; see Hoff & Shanks, 2020). Hoff and Shanks identified differences in the lexical and grammatical properties of maternal child-directed speech between first and second-language speakers. Unsworth et al. (2019) also examined L2 parental speech, correlating parental self-reports with first language speaker’s evaluations of parental proficiency. Their study aimed to determine whether input quality or quantity in Dutch and another language affected bilingual preschoolers’ language proficiency in Dutch (measured via active and passive vocabulary tasks, and via sentence comprehension and word structure subtests of the CELF-2 test). They found that the best predictor of children’s proficiency was not the proportion of first language input but the proficiency of the second-language parents (i.e., language richness and the family constellation). However, their variable ‘family constellation’—which expressed whether both parents or only one parent used the other language—correlated strongly with the amount of input measures. Unsworth et al. (2019) concluded that high-quality input from proficient second-language speakers is more beneficial for children’s language skills than low-quality input from less proficient speakers (see also Place & Hoff, 2016; and see Ramon-Casas et al., 2023).

Another issue to consider is the extent to which the focus should be on a child’s current or past language environment. Focusing solely on current exposure neglects the developmental experience that has led up to that point in time. Interestingly, Levy et al. (2019) found effects of current, but not past, exposure to accented speech heard between 0 and 3 years of age. Similarly, Bedore et al. (2012) found that current use was more informative for children’s language profiles than past use. An argument in favor of assessing only current language use is presented by Flege (2009), who argues that respondents may be more accurate when reporting on their current language use compared to more general questions or past language use. He also proposes a method using smartphone recordings to sample experience (Flege & Wayland, 2019), focusing on current language input and use. Notwithstanding this, it is possible that accent and language experience scales could be improved by incorporating past language use (cf. Lloyd-Smith et al., 2020).

Although the majority of continuous scales for quantifying language input are designed for bilingual populations rather than for measuring accented speech, they can also be applied to accents. The proposal of Marian and Hayakawa (2021) for a bilingualism quotient shows that the problem of quantifying language input poses a significant challenge for many researchers. They suggest increasing the level of specificity when assessing the ‘bilingualism score’ of a given population. For example, one might consider that variables such as exposure might be more meaningful in younger children while competence might be more meaningful in older children, with variables weighed accordingly. While this approach may optimize the fit of the measurements to a specific population, such highly specific rankings risk reducing comparability between studies. Nevertheless, future research on bilingualism and bidialectalism must consider the heterogeneity within these populations and use continuous measurement methods (see, e.g., the Quantifying Bilingual Experience program “Q-BEx”, https://q-bex.org; see also Gullifer et al., 2021).

An improved tool for measuring experience could also include elements from Buac’s (2019) Summary of Exposure to Accent Questionnaire (SEA-Q). Buac’s scale, based on the SEA-Q, includes the length of bilingualism (child’s age minus age of acquisition in years), the number of speakers of English as a second language who regularly interact with the child, and their accent strength. Compared to the Levy et al. (2019) method, it has the advantage of considering previous exposure in the measurement of accent experience and accent strength. However, it lacks coverage of accents other than a specific second-language accent. Again, such coverage needs to be adapted according to the specific geographical region and linguistic community under consideration, considering not only the Global North but also the Global South with its linguistically rich, heterogeneous environments. A combined, improved scale incorporating elements from Lloyd-Smith et al. (2020), Buac (2019), and Levy et al. (2019) might include quantitative measures of past and present accent input, the child’s own productions, and input in different regional and second-language accents from different speakers. Further research is needed to extend the existing methods for operationalizing language and accent experience. Ideally, a scale or quotient specifically designed to quantify accented input (like the bilingualism scale in the Q-BEx project) could be developed and used across different laboratories to increase comparability across studies.

Finally, with the advent of machine learning technologies, future studies might consider shifting to continuous classifications of speakers using machine learning models trained on the speech production of bilinguals and monolinguals (for bilinguals, see Coco et al., 2025). Additionally, research on gradient effects of speech variability could be expanded to include lexical, morpho-syntactic, and sociolinguistic development (as partly performed by Soto-Corominas et al., 2022; for sociolinguistic development, see also Johnson & White, 2020).

In summary, there is converging evidence for the importance of fine-grained measures of language and accent experience, although measuring the extent of accent and language experience remains a methodological challenge. Many studies have shown that operationalizing accent and language input as a continuous variable leads to meaningful predictors of children’s performance across tasks. To advance the field, researchers call for uniform approaches to language experience assessments. These suggestions are well-founded; however, experience variables need to be constructed individually for different accent types and languages, distinguishing between etic and emic perspectives across countries. Uniform approaches, therefore, require collaboration across countries and language varieties to account for the diversity of language communities and the varying lay conceptualizations of languages, accents, and dialects. This approach would provide a more comprehensive and accurate assessment of children’s linguistic environments, contributing to our understanding of language development in diverse linguistic contexts.

5. Conclusions

We began by asking how variation due to diverse languages and accents affects speech processing in childhood. Studies that consider such heterogeneous speech contexts highlight the significant role that experience with different accents and languages plays in shaping speech processing and language learning in primary-school-aged children. These studies demonstrate that exposure to diverse speech inputs fosters greater flexibility in perception and word learning, while also contributing to variability in speech production. Importantly, they underscore the necessity of fine-grained measures of language and accent experience, as well as the value of considering accent and language input as continuous variables. Moving forward, there is a clear need for more research across structurally diverse speech communities, countries, and populations to fully understand the impact of speech variability on language processing and learning in both monolingual and bilingual children. When discussing languages, dialects, and L1/L2 accents from a cross-cultural perspective, it is essential to consider different emic understandings. What might appear like clear distinctions from a researcher’s (etic) perspective may not align with the lived experiences and cultural interpretations of the speakers themselves. This complexity underscores the need for nuanced, context-sensitive approaches in linguistic research. Such research will be invaluable in informing sociopolitical debates and providing evidence-based guidance to parents, educators, and speech therapists on the effects of linguistically diverse environments on children’s language development.