Development of phonological constancy

From five months, infants have been shown to demonstrate a preference for a familiar, local accent over unfamiliar, nonlocal accents (Butler, Floccia, Goslin, & Panneton, Reference Butler, Floccia, Goslin and Panneton2011). Studies of infants’ word learning have found that accent differences initially prevent the recognition of familiar words when they are heard in an unfamiliar accent (cf., Best & Kitamura, Reference Best, Kitamura and Burton2012; Schmale, Cristià, Seidl, & Johnson, Reference Schmale, Cristià, Seidl and Johnson2010). These studies have found that by 12–19 months, infants can abstract across different and unfamiliar accents in order to understand familiar words.

The ability to comprehend familiar words in an unfamiliar accent reflects children's development toward understanding the principle of ‘phonological constancy’ (Best, Tyler, Gooding, Orlando, & Quann, Reference Best, Tyler, Gooding, Orlando and Quann2009), whereby the phonology of a word is kept intact despite variations in its phonetic realization. Best et al. (Reference Best, Tyler, Gooding, Orlando and Quann2009) found that 19-month-olds but not 15-month-olds accepted nonnative pronunciations of familiar words and advocate a ‘perceptual attunement’ account to explain this development. They suggest that young children first develop dialect-specific phonetic patterns through their perception of the articulatory gestures of their native language/dialect. By 19 months, children have learned the phonological distinctiveness of the phonemes in their native language/dialect and have acquired the corresponding skill of phonological constancy; they are able to detect the phonetic variations they encounter as belonging to the same phoneme, by comparing the variations in terms of their articulatory gestures. The perceptual attunement account draws on the ‘perceptual assimilation model’ (PAM) (Best, Reference Best and Strange1995), originally developed as an explanation of how adult listeners deal with nonnative phonemes in cross-language speech perception. Therefore, PAM is able to explain and link together processes of speech perception throughout childhood and adulthood, as well as accounting for perceptual development of both native and nonnative contrasts.

Best et al.’s (Reference Best, Tyler, Gooding, Orlando and Quann2009) account focuses on children's linguistic input and their developing comparisons between incoming sounds. As such, it is compatible with a usage-based explanation such as exemplar theory, in which both probabilistic methods of learning and the creation of higher-level abstractions are proposed. The advantage of such an account for the current study is that the explanation incorporates a description of both the storage and accessing of a combination of phonological and social information (see ‘Theoretical account’ below). Such an account is therefore at least partly based on the individual child's exposure.

Previous research provides conflicting results regarding the role of infants’ previous exposure to accent varieties once they have reached the stage of phonological constancy. Prior exposure to an unfamiliar accent under laboratory conditions was found by Schmale, Cristia, and Seidl (Reference Schmale, Cristia and Seidl2012) to help 24-month-old infants’ understanding and processing of the accent, but this was contradicted by van Heugten and Johnson's (Reference van Heugten and Johnson2015) study of 28-month-old infants.

Similarly inconsistent results are found in studies of children's immersive experience of an accent. Floccia, Delle Luche, Durrant, Butler, and Goslin (Reference Floccia, Delle Luche, Durrant, Butler and Goslin2012) found that 20-month-old children brought up in a rhotic community were quicker at recognizing words pronounced in a rhotic form, regardless of whether they were “mono-accentual” (with two rhotic parents) or “bi-accentual” (with at least one nonrhotic parent). They interpret this as suggesting that a child's phonological representations are conditioned by their community rather than by their parents. However, van der Feest and Johnson's (Reference van der Feest and Johnson2015) study of 24-month-old Dutch children found that children with a “Mixed Input” (nonlocal parents) were able to detect mispronunciations by a speaker with their parents’ accent, whereas children with “Uniform Input” (local parents) ignored these mispronunciations because they were not familiar with the accent. Therefore, rather than ignoring the input from their parents (as Floccia et al.’s [Reference Floccia, Delle Luche, Durrant, Butler and Goslin2012] study suggests), the children with Mixed Input were able to utilize the mixed evidence from their linguistic input to decide the relevance of phonological contrasts that they heard.

Overall, the results of studies with infants demonstrate a development in children's sensitivities to accent variability. Infant studies are based on speaker discrimination and/or word learning which reflects the infants’ familiarity with the individual or accent that they are hearing. Their findings are, therefore, a preliminary step in understanding more about how speakers with the same accent can become categorized together conceptually as children mature. The current research investigates the emergence and development of such categorization among children at a key point in their sociolinguistic development.

Children's sociolinguistic development

Studies uncovering the development of sociolinguistic skills in the preschool years have found that children from the age of two acquire accent-specific phonological variation in their production (Foulkes, Docherty, & Watt, Reference Foulkes, Docherty and Watt1999; Roberts, Reference Roberts1997; Roberts & Labov, Reference Roberts and Labov1995). Small-scale sociolinguistic patterns relating to gender have also been discovered, such as girls’ higher rates of (-t, d) deletion in Roberts’ (Reference Roberts1997) study of preschool children in Philadelphia, and gender-specific variation of /p t k/ in preschool Tyneside English (Foulkes, Docherty, & Watt, Reference Foulkes, Docherty and Watt2005) and in primary school Australian English (Tait & Tabain, Reference Tait and Tabain2016). Style-shifting of certain variables has been evidenced in the speech of children from the age of three (Smith et al., Reference Smith, Durham and Fortune2007, Reference Smith, Durham and Richards2013) and has been found to develop as they get older (Kerswill & Williams, Reference Kerswill and Williams2002). The probability and rates of such style-shifting vary according to the variable itself, as well as according to the levels of style-shifting that children are exposed to in their input (Roberts, Reference Roberts1997; Smith, Durham, & Fortune, Reference Smith, Durham and Fortune2009; Smith et al., Reference Smith, Durham and Fortune2007, Reference Smith, Durham and Richards2013).

A few studies have investigated young children's perceptual awareness of regional accents after infancy, from the age of five. However, their differing methodologies and assumptions deliver conflicting conclusions. Studies by Floccia, Butler, Girard, and Goslin (Reference Floccia, Butler, Girard and Goslin2009) and Wagner, Clopper, and Pate (Reference Wagner, Clopper and Pate2014) found that children under the age of seven were not able to group speakers according to their regional accent. In these studies, the children only heard two example sentences in each accent before they were then asked to categorize ‘aliens from elsewhere’ (Floccia et al., Reference Floccia, Butler, Girard and Goslin2009) or different color puppets (Wagner et al., Reference Wagner, Clopper and Pate2014) into two groups based on these examples. Therefore, the tasks may simply have been too difficult as they required the children to have a very good working memory. Additionally, Floccia et al.’s [Reference Floccia, Butler, Girard and Goslin2009] task assumed that the children would know what an alien is and understand the link between where speakers are from and how they speak. Such metalinguistic awareness is a difficult and abstract skill that is being developed but is not fully mastered at such a young age, as Beck (Reference Beck2014) found in a task investigating children's explicit awareness of the link between accent and regional origin. Beck directly addressed this question by asking 5-7-year-olds to listen to two speakers with different regional accents and answer the question, “Can you guess why these two people talk differently?” She found that only 32% of the children gave the correct answer, and 42% declined to answer at all, indicating that most of the children had not made a metalinguistic connection between regional accent and geography.

Beck (Reference Beck2014) also investigated regional accent awareness among 5-7-year-olds by running an experiment with a simpler design than the grouping tasks discussed above. In her ABX discrimination task, children heard single words which were chosen to reflect the vowel quality differences between two different accents. They heard three speakers pronounce each word and were asked which two speakers sounded most alike. Beck found that the children in her study were able to discriminate between a familiar, local (Philadelphian) and a nonfamiliar, nonlocal (General Southern) regional accent, with an average of 64% correct answers. Beck's study therefore presents some evidence that children from five years are able to match speakers based on regionally distributed pronunciations. However, her experiment was limited to testing children's ability to discriminate between sounds and match them accordingly.

The current research goes beyond investigating sound-matching within words by testing their interpretation of variation at the level of the phoneme across different words (Level 2 of the experiment) as well as across different phonemes (Level 3 of the experiment). Additionally, by using a categorization task rather than an ABX discrimination task (see ‘Methodology’ section), the results of the current study can be said to indicate more reliably that the children are using the differences to group speakers rather than simply matching sounds. Furthermore, a driving question for the current research is whether children of an even younger, preschool age show an emerging awareness of accent; therefore, the age of focus for the current study is on 3-4-year-olds.

The current study also investigates the children's performance across the age range, as well as considering gender-based differences in the results. These factors are included as a rough approximation of the potential impact of maturational factors, such as increased processing speed that develops with maturation of the brain (Murphy, Reference Murphy2004). Infant girls have been found to mature physically at a faster rate than boys (Bornstein, Hahn, & Haynes, Reference Bornstein, Hahn and Haynes2004) and, therefore, they may have an early advantage in language processing tasks due to earlier brain maturation. More pertinent to the current study's focus on sociolinguistic development, any gender-related differences are likely to represent children's early gendered socialization. As mentioned above, research has found differences in the production patterns of young children in line with gendered norms, indicating that this particular aspect of their language socialization starts early. Furthermore, research shows that girls and boys can receive different linguistic input from their parents. In a community in Newcastle, Foulkes et al. (Reference Foulkes, Docherty and Watt2005) found that the child-directed speech (CDS) produced by mothers to their sons contained a higher proportion of nonstandard variants compared to CDS produced by mothers to their daughters. The authors proposed that “mothers are tuning their phonological performance in line with their child's developing gender identity” (Foulkes et al., Reference Foulkes, Docherty and Watt2005:198). It is also possible that this form of children's linguistic socialization affects their overall perception of linguistic variation, and, therefore, gender is an important consideration in the current study.

A theoretical account

A theoretical account of the stages in the phonological/sociolinguistic acquisition process is needed to explain how perceptual awareness progresses throughout childhood and beyond. Many usage-based accounts of the cognitive processes involved in language acquisition now advocate exemplar theory (ExT) as the best way to explain the storing of both linguistic and social information (cf., Pierrehumbert, Reference Pierrehumbert2003). Exemplar theory is a theory of memory and categorization originally developed in psychology. ExT proposes that we store detailed episodic traces in memory and that these memory traces affect how we process and interpret our future experiences.

At the heart of an exemplar model of memory is the idea that individually encountered stimuli are stored with details of the individual encounter. In speech encounters, this can range from the phonetic detail of the pronunciation made by the individual at the time, to social detail such as aspects of the speaker's accent or social background. When similar stimuli are then encountered in speech processing at a later point, these episodic traces, and the details stored alongside them, are accessed together.

In describing how children learn socially structured variation alongside their phonology, Docherty, Foulkes, Tillotson, and Watt (Reference Docherty, Foulkes, Tillotson, Watt, Goldstein, Whalen and Best2006) credited an exemplar approach with being able to account for the connection between language and its social context, and, in particular, how phonetic properties can be aligned with social referents, such as particular speakers, particular genders, and particular accents. They compared data of child-directed speech in Newcastle from Foulkes et al. (Reference Foulkes, Docherty and Watt2005) with adult-to-adult speech in the same community. Findings indicated that some of the community patterns of adult-to-adult speech were emphasized in child-directed speech. For example, in adult-to-adult speech, women were found to produce a higher proportion of (word-medial) standard /t/ compared to men. In CDS, both women and men increased their use of standard /t/, but there was still a much larger proportion of standard /t/ among women. They also found evidence that the children's own productions were reproducing the fine-grained phonetic variability to which they were exposed. As Docherty et al. (Reference Docherty, Foulkes, Tillotson, Watt, Goldstein, Whalen and Best2006) described, in line with the patterns of their local accent and their gender, children specifically associate phonetic variability with certain kinds of speaker. “Thus exemplar models may offer a plausible means of accounting for the learning and emergence of features of socially structured variation alongside other systematic aspects of sound patterning” (Docherty et al., Reference Docherty, Foulkes, Tillotson, Watt, Goldstein, Whalen and Best2006:414). From this sociolinguistic perspective, children are learning socially structured variation alongside their phonology, and both kinds of patterning lead to stored abstractions across their encounters.

While the focus in ExT is on the role of individually stored episodic traces and their detail, there is a consensus view among many proponents of the theory that some level of abstraction is also an important part of the process (Docherty & Foulkes, Reference Docherty and Foulkes2014:46). Cutler, Eisner, McQueen, and Norris (Reference Cutler, Eisner, McQueen and Norris2010) suggested that only a hybrid model of speech processing, which includes a role for both abstractions and episodes, can account for evidence showing that listeners adjust their interpretation of phonemes after limited exposure to deviant realizations by individuals (such as found by Norris, McQueen, & Cutler [Reference Norris, McQueen and Cutler2003] and McQueen, Cutler, & Norris [Reference McQueen, Cutler and Norris2006]). In this case, stored individual encounters, detailing the phonetic realizations of the speakers, contribute to a changed distinction that listeners make on the phonological level. Similarly, the build-up of encounters that listeners have with speech exemplars indexing social information about speakers can explain the development of categories pertaining to social-indexical distinctions, such as those based on speakers’ regional accents.

Foulkes (Reference Foulkes2010) hypothesized that we cognitively categorize speakers based on our accumulation of individual speakers’ exemplars and, as a result, that differences between individual speakers form the basis for the development of these speaker categories. For example, it is likely that fairly early in life, children are exposed to individual speakers who are easily categorizable “in a relatively neat tripartite structure” (Foulkes, Reference Foulkes2010:20) as ‘adult males,’ ‘adult females,’ and ‘children.’ Due to large differences in pitch and formant frequencies, exemplars from these individuals are grouped together in this three-way distinction. Less tangible groupings, such as those based on accent, are likely to develop later, through the accumulation of more exposure to speakers with these accents. Therefore, an individual's experience of individual speakers with different accents (as explored in the current study), is central to such a model.

The phonological variables

This study investigates preschool children's awareness of accent features indicative of the distinction between speakers from the north and south of England, using the phonological variables in the bath, strut, and face lexical sets (Wells, Reference Wells1982a). In York, North Yorkshire, where the research took place, the local accent includes pronunciations of bath, strut, and face, which are prototypical of the Central North (a region defined by Hughes, Trudgill, & Watt [Reference Hughes, Trudgill and Watt2012]).

The Yorkshire accent extends to cover the accent of speakers from the other county subdivisions of Yorkshire (West and East) that the city of York itself borders (cf., Haddican, Foulkes, Hughes, & Richards, Reference Haddican, Foulkes, Hughes and Richards2013; Tagliamonte & Roeder, Reference Tagliamonte and Roeder2009). The bath vowel (and its realization as [a] in the North or [ɑ:] in the South) and the strut vowel (and its realization as [ʊ] in the North or [ʌ] in the South), are described as among the most conspicuous accent features in differentiating a Northern accent from a Southern one (cf., Hughes et al., Reference Hughes, Trudgill and Watt2012; Wells, Reference Wells1982a/Reference Wellsb). The differences in pronunciation of the strut and bath vowels are commonly seen as linguistic stereotypes (in Labov's [Reference Labov1972] terms) of the north/south of England as they are often overtly commented on by lay listeners as characterizing speakers from these two broad geographic regions. The realization of the face vowel differentiates speakers from the Central North who use [e:], and those from both the Midlands and the south of England who use [eɪ]. The monophthongal variant [e:] is a ‘mainstream’ Northern variant (Watt, Reference Watt2002), however, it is often not used by middle-class speakers in these areas, who use the more typically Southern/SSBE (Standard Southern British English) diphthongal pronunciation [eɪ]. The ongoing rise in the use of the diphthongal variant in Central North regions such as York is a change in progress, as reported by Haddican et al. (Reference Haddican, Foulkes, Hughes and Richards2013). Together, therefore, the bath, strut, and faceFootnote ¹ vowels form a distinction between the vowels used in a local Yorkshire accent (situated in the Central North) and those used in a SSBE accent (pronunciations typical of the southeastof England).

Participants and background information

Twenty preschool children (ten 3-year-olds, ten 4-year-olds; twelve females, mean age 3;10; eight males, mean age 3;11) took part in the experiment. These children were all attending one of two different nurseries in York; nine children aged 3;1 to 4;6 from one nursery and eleven children aged 3;2 to 4;7 from another nursery.

The children's parents were asked to provide regional background information. Eighteen of the twenty children were born in York; one child moved from Germany to York aged five months, and another child moved from London and had been living in York for 17 months. For the purposes of the statistical analysis, the children were split into two groups according to whether they had at least one Yorkshire parent (ten children, mean age 3;10) or no Yorkshire parents (nine children, mean age 3;11, missing information from one child), with the region of Yorkshire defined as set out above. This distinction was made in line with second dialect studies in which children are usually classified as bidialectal if they move to a new town from elsewhere and have two nonlocal parents (e.g., Chambers, Reference Chambers1992; Payne, Reference Payne and Labov1980; Tagliamonte & Molfenter, Reference Tagliamonte and Molfenter2007; Trudgill, Reference Trudgill1981). Although most of the children in the current experiment were born in the local region, the ‘outsider’ status of their parents represented their exposure to nonlocal varieties at home. Those children with no Yorkshire parents had parents from a range of regions throughout the UKFootnote ², such as Hampshire, London, the West Midlands, Northern Ireland, Tyne & Wear, and Aberdeenshire (see Appendix, Table A1 for a full list).

Experiment design: Stimuli

Sentence-length stimuli were constructed with the target word (i.e., the word featuring the accent difference) at the end of each sentence in order to draw the children's attention to the target vowel. The rest of the words in each sentence were chosen carefully in order to avoid including any words with accent differences corresponding to diagnostic features of the north and south (as defined by Wells, Reference Wells1982a). For example, the sentence shown in (1) features the bath vowel in the word ‘basket.’

1. (1) This is my basket

In each iteration of the experimental procedure (see below), children were presented with two ‘reference sentences,’ each spoken by a different cartoon image. They were then asked to match a set of ‘grouping sentences,’ (again each linked to a separate cartoon image), to one or other of the reference sentences. The reference sentences in each set were worded the same as each other, so that the only difference between them was the target vowel pronunciation. For example, there is only a bath vowel ([a]/[ɑ:]) distinction in (2) and (3).

1. (2) This is my b[a]sket

2. (3) This is my b[ɑ:]sket

The grouping sentences were designed with vowel pronunciation differences in line with three different levels of the experiment (referred to as levels 1–3), in order to test different aspects of the children's accent awareness. In line with the findings from infant phonological development described above, the children were presumed to have reached the stage of phonological constancy. It was therefore assumed that the children would have no trouble comprehending the words when they were pronounced either in a Yorkshire or a SSBE accent. The first level in the experiment aimed to explicitly test whether the children were able to pick up these lower-order phonetic patterns by testing their ability to group speakers based on different pronunciations of the same word, as shown in (4).

1. (4) Reference sentence: ‘This is my basket’

   Grouping sentence: ‘Put me in a basket’

   Accent difference: ‘b[a]sket’ versus ‘b[ɑ:]sket.’

This first level was therefore testing whether, despite having reached a level of phonological constancy, children were still able to interpret the phonetic variation between these sounds to the extent that they would be able to use them as grouping criteria. The second level was a higher-level test of the extent to which children were able to use both their knowledge of abstraction and variation, as this task asked the children to group speakers based on different pronunciations of the same phoneme but in different words, as shown in (5).

1. (5) Reference sentence: ‘We need to walk on the path’

   Grouping sentence: ‘I want to walk on the grass’

   Accent difference: ‘p[a]th’/’p[ɑ:]th’ vs. ‘gr[a]ss’/‘gr[ɑ:]ss’

The second level was therefore testing whether the children could hear these phonetic differences across different words, relying on their awareness of phonological constancy across words, as well as their ability to interpret phonetic variation within the phonemes.

The third level tested children's more abstract knowledge, as the task asked the children to group speakers across different phonemes, as shown in (6).

1. (6) Reference sentence: ‘What did you break?’

   Grouping sentence: ‘It was a glass’

   Accent differences: ‘br[e:]k’/‘br[eɪ]k'vs.‘gl[a]ss’/’gl[ɑ:]ss’

This third level was therefore more explicitly testing the social indexical knowledge developing among the children in relation to their awareness of speakers belonging to abstract regional categories (Yorkshire versus SSBE). Overall, these different levels aimed to track the stages of development: from children achieving phonological constancy to being able to use their wider abstract reasoning to link phonetic variation to higher-order differences relating to regional accent groups.

Table 1 presents a summary of the different levels, with example words used in the experiment.

Table 1. Experiment design with examples

Stimuli recordings

Stimuli for the experiment were recorded from one bidialectal speaker, a 25-year-old female, who was able to switch between two different accents and produce vowel pronunciations typical of both Yorkshire and the South East of England. Using the same speaker helped to ensure that the children would focus on the phonological accent variation of the speaker guises during the experiment, rather than making decisions based on other characteristics of individual speakers’ voices.

The speaker was recorded reading a list of sentences, using a Zoom H4n recorder which was set to record at a 16 bit 44.1kHz sampling rate. The speaker was asked to read the set of sentence stimuli, first with a SSBE pronunciation and then with a Yorkshire pronunciation of the target word. In order to keep the pronunciation of the rest of the stimulus consistent, she was instructed to read the rest of the sentence naturally, in her normal accent (standard and not regionally distinctive). This meant that the focus of the point of comparison would be on the end word itself, as this was the only one that differed. Furthermore, as the speaker's prosody was another variable with the potential to indicate her regional background, this was kept controlled to some extent by keeping stress placement consistently on the final word in the sentence.

Experimental procedure

The children took part in the experiment individually, either in a quiet corner of the nursery or at the child's home with their parent(s) present. The experiment was presented on a laptop computer, and the children listened to the audio stimuli through headphones.

The experiment was designed to be run in Microsoft PowerPoint as a slideshow, with pictures and sound clips and each slide consisting of a different grouping task. In order to keep the experiment short, there were ten tasks in all, each consisting of five trials. Each task was presented as a different screen (see Figure 1), and each trial involved matching a cartoon image to its group on the basis of a stimulus sentence. Tasks 1–3 (the first 15 trials) consisted of Level 1 sentences, tasks 4–7 (the next 20 trials) of Level 2 sentences, tasks 8–9 (the next 10 trials) of Level 3 sentences. Task 10 consisted of a mixture of Levels 2 and 3, with 3 of the grouping sentences matching the phoneme of the reference sentence and two grouping sentences containing a different phoneme. Therefore, in total the children carried out 15 trials at Level 1, 23 trials at Level 2, and 12 trials at Level 3. This uneven number of trials per level is accounted for in the statistical analysis of the results, which considers the results of each level separately.

Figure 1. Screen shot of first ‘teddy bear’ grouping task.

The first three grouping tasks (for the Level 1 sentences) involved cartoon bears. In each of these tasks, two mother bears were displayed at the bottom of the screen. They were made distinguishable by having different color patches on their fur (see Figure 1). Each mother bear was linked to a sound clip of one of the reference sentences and next to each mother bear was a picture of the subject of the sentence, for example a basket in the case of the first sentence. At the top of the screen, pictures of five identical baby bears were displayed, each linked to an audio clip of one of the grouping sentences.

The experiment was presented as a game for the children to play; they were asked to group the ‘lost’ baby bears with their mother bears. The experimenterFootnote ³ controlled the playing of the audio files, clicking on each of the characters that rocked from side to side while the corresponding sound clip played. Each mother bear was heard first, and then each baby bear was heard, after which the child was asked to indicate the mother bear they belonged to by pointing at the screen. The experimenter then dragged the picture of the baby bear over to the mummy bear that the child had indicated. For the sake of consistency, in each task, three of the baby bears were linked to the Yorkshire sound clip and two of the bears were linked to the SSBE sound clipFootnote ⁴. The baby bears were arranged in a random order each time, to prevent children forming a pattern-based decision. Similarly, while the mother bear on the left of the screen was consistently played first, the accents of the mother bears were played in a pseudo random order (either Yorkshire or SSBE first). A trial was logged as ‘correct’ if the baby bear was grouped with the same-accented mother bear.

To keep the task varied and interesting, the grouping tasks for Levels 2 and 3 used pictures of cartoon mothers and their daughters instead of teddy bears. The grouping tasks were primarily the same, with two mothers (distinguished by different color dresses) displayed at the bottom of the screen and five daughters arranged randomly at the top of the screen.

Children who did not want to continue and those who failed to understand the task did not take part in the second part of the experiment based on the stimuli from Levels 2 and 3 (featuring the mothers and daughters). Out of the 20 child participants from the first part of the experiment, 15 went on to do the second part of the experiment: six 3-year-olds (five females, one male) and nine 4-year-olds (five females, four males). Overall, the mean age for the ten females was 3;11 and for the five males was 4;1. There were five children who had no Yorkshire parents (mean age of 4;2) and nine children who had 1+ Yorkshire parents, (mean age 3;10, missing information for one participant).