The effects of bilingualism on stuttering during late childhood

Abstract

Objectives

To document distinct patterns of language use by bilingual children (use of an alternative language exclusively, LE, or along with English, BIL). To establish how these patterns affect onset of stuttering, school performance and recovery rate relative to monolingual speakers who stutter (MONO).

Design

Clinical referral sample with cases classified by speech-language therapists. Supplementary data obtained from speech recordings, interviews with child and family.

Setting

South-East England, 1999-2007.

Participants

Children aged 8-12 plus who stuttered (monolingual and bilingual) and fluent bilingual controls (FB).

Main outcome measures

Participants’ stuttering history, SATS scores, measures of recovery or persistence of stuttering.

Results

The sample of 317 children had 69 bilinguals (prevalence rate of bilingualism in the stuttering sample was 21.8%). 38 children used a language other than English primarily or exclusively in the home and 36 of these (94.7%) bilinguals who stuttered did so in both their languages. There were fewer LE than BIL stuttering children at time of first referral to clinic (of the bilinguals who stuttered, 15/38, 39.5%, were LE and 23/38, 60.5%, were BIL). The reverse was the case in the fluent control sample (of the bilinguals who did not stutter, 28/38, 73.7%, were LE and 10/38, 26.3%, were BIL). The association between stuttering and bilingual group (LE/BIL) was significant by χ² and this is consistent with a higher chance of stuttering for BIL than LE speakers. For speakers who stuttered, age of stuttering onset for LE and BIL was similar to that reported for MONO groups (4 years 9 months, 4 years 10 months and 4 years 3 months for LE, BIL, MONO respectively) and males were affected in each of these groups to about the same extent (the male/female ratio was 4.1:1, 4.75:1 and 4.43:1 for LE, BIL and MONO respectively). For the 29 children who were old enough to complete the assessments, educational achievement at key stages one and two was not affected by either form of bilingualism relative to the MONO and fluent bilingual groups. For these same children, recovery rate for LE and MONO controls was significantly higher by χ² than for those who were BIL since birth (recovery rate for LE and MONO together was 55%, and for BIL was 25%).

Conclusions

There was an increased chance of stuttering onset for BIL children. The chances of recovery from stuttering were lower for BIL speakers than for LE and MONO speakers.

Introduction

Bilingualism has been regarded as a risk factor for stuttering. ^{1 2} However, there is little information about how a second language affects chances of stuttering onset and of recovery. Consequently, a study was conducted on all referrals for stuttering for children first seen when they were between the ages of eight and ten years who attended clinics that specialized in diagnosis and treatment of stuttering. Children were selected who used a second language in the pre-school years either: a) exclusively (these learned English at school, termed LE) or b) along with English in the home (bilingual from birth, termed BIL). The majority of the children were seen again when they were aged 12 plus. Prevalence rates of stuttering for LE and BIL children as compared with all referrals are reported to see if a disproportionate number of speakers of a language other than English is referred to stuttering clinics. Checks were made to verify that the LE and BIL children were stuttering by comparing age of stuttering onset and gender ratio for these speaker groups against monolingual speakers from the referral sample who stutter (MONO). Early school performance of LE and BIL children who stutter was compared with that of a MONO group and bilingual children who are fluent (the fluent bilingual group, termed FB). The patterns of onset and recovery in the LE, BIL and MONO groups were compared.

Methods

Participants and sub-groups

317 children who stuttered participated. They all: a) started school in the UK at age four or five, b) first presented at clinic when aged between 8 and 10 years, and c) lived in the greater London area. Stuttering was confirmed by a specialist speech-language therapist at the clinic. Reported onset of stuttering usually occurs before age six. The attendance at clinic two or more years later is partly due to time needed to process the children in the health system; these are secondary referrals to the specialist clinic, and devolved budgets to local area health authorities can cause delay. Recordings were taken to estimate the percentage of stuttered syllables and for the stuttering severity instrument measure SSI-3 ³ to be applied (see below for details of how SSI-3 was calculated).

When each child was seen initially, all 317 caregivers were asked whether they used a language other than English exclusively or sometimes in the home. In cases where use of a second language other than English was reported, the caregivers had a further interview. When the caregiver did not speak English, help with translation was obtained from a close friend who spoke the same language as well as English. The interview obtained background details about the child, the family and information about languages used in the home. Details were collected about the biological relation to the child (all were parents of the child they cared for). Gender and age of onset of stuttering were obtained. The child’s Standard Attainment (SATS) educational test scores at ages 7 and 11 (key stages one and two) were obtained for English, mathematics and science. It was also established whether the caregiver spoke English to the child from birth up until when the child went to school, and whether care was given by this person exclusively, or some of the time, up to age 5 years. The caregivers indicated whether the child spoke English when they entered school. Information was obtained about where the child was born and, in cases where the child was an immigrant, the age at which they had entered the UK and the past history of language used in their former country. The language information was used to classify children as LE or BIL.

There were 69 cases where at least one language other than English was spoken in the home. Caregivers of 38 of those 69 children reported that they primarily, or exclusively, used a language other than English in the home and had done so since the birth of their child. This criterion excluded people who opted to speak a language other than English in the home for their child’s educational/social advancement. All 38 families have continued to participate. Nine of the group of 38 have not yet reached 12 years, leaving 29 who were categorized as recovered from stuttering or not.

The 38 selected children were divided into LE and BIL groups. The LE children may be regarded as not bilingual until the age of school onset. The LE group consisted of 15 children (39.5% of the 38 children) who did not speak English when they started school. During this period, English was not used by the caregivers, there was no replay media over which English was heard and if there were siblings, they did not speak English up to the age of school entry.

Twenty-three (60.5%) of the children spoke English in addition to the primary language spoken in the home (BIL group). They were all exposed to English in the home from birth.

A group of monolingual English speaking children who stuttered (MONO) was selected in order to assess the impact stuttering has on epidemiology and education and to see how stuttering severity compared with the experimental groups. The MONO children were selected at random from the entire sample of such children referred to clinic, subject to the restrictions that the group had to match the LE and BIL groups for age and gender. They had similar socio-economic backgrounds and attended similar schools to the LE and BIL groups.

When children in all these groups passed 12 years of age, they and their families were interviewed about their stuttering and school record, and the children were recorded so that a stuttering severity measure could be obtained (to confirm whether or not their stuttering had continued).

A second control group, of fluent children who used a language other than English in the pre-school years, was recruited (fluent bilingual, termed FB). These children were age-matched to those children who stuttered at second attendance, so that SATS scores were available. The children were recruited from schools in the same catchment areas as the clinics and they had similar education and socio-economic backgrounds to the LE and BIL children, and reported no history of speech/hearing problems. They were divided into LE and BIL groups using the same criteria as above. The LE and BIL groups of FB groups were used to compare educational attainment level with those obtained for LE and BIL children who stuttered.

Classification as persistent/recovered

Stuttering was reassessed when the child was seen at age 12 plus, by: 1) the child, 2) the caregiver, and 3) a researcher (the same for all children). All the scales used in these assessments have been normed (see supplementary material). The child questionnaire and the researcher report forms are given as supplementary material (the parent questionnaire was the same as the child questionnaire except that statements were changed from the first to the third person). For the child and caregiver, seven of the 15 questions on Boberg and Kully’s questionnaire ⁴ were employed, and a further question was constructed which combined three more of their questions. For each of the eight questions, the child or caregiver indicated extent of agreement on a five-point scale. Each question was scored 1 to 5, where 1 represented fluent behavior and 5 dysfluent behavior. The scores across all questions were summed, the maximum score being 40. Scores lower than 21 were considered ‘recovered’ and scores greater than 21 were considered ‘persistent’. The scores on these questionnaires were correlated with SSI-3 scores as a validation criterion (part of the norming procedure described in the supplementary materials). This showed that these cutoffs divided speakers at the low end of the moderate scale (stuttering could not be moderate to have recovered and had to be at least moderate to be designated persistent).

The researcher visited each child’s home and recorded an interview that lasted approximately 90 minutes. During his visit, the researcher talked with a caregiver and the child about the speech problem and experience in clinic. He also sought their views about communication style and self-confidence in a range of typical environments. These included home and social gatherings with adults and children in and out of school. Performance and experience in school were assessed in terms of inter-personal relationships with staff and other children (including bullying). General health issues were also examined, including frequent absence from school and childhood illnesses. The researcher subsequently assessed speech fluency, social-conversational skills, and whether the child had a positive self image/confidence about speech, using the recordings and notes taken at the home-visit. Each of the three assessments was scored on a scale of 0 (good) to 3 (poor). The scores for the three factors were summed to give one score of between 0 and 9. A score of 5 or above indicated still stuttering.

To be designated as persistent, the caregiver, child and researcher all had to rate the child as still stuttering. To be designated as recovered, the caregiver, child and researcher all had to designate the child as not stuttering. All cases were unambiguously classified on all three criteria. ^{5 6 7 8} All participants have been followed up for a minimum of 12 months and substantially longer in some cases (mean length of follow up for all speakers is 31.5 months with an sd of 24 months). The recovered participants showed no relapse and none of the participants designated as persistent recovered during this period.

Percentage of syllables stuttered out of the total syllables spoken in a two-minute recording of spontaneous speech taken on the second occasion was used to make a further check on persistence/recovery. All recovered speakers had less than 4% stuttered syllables whereas all persistent speakers had greater than 4% stuttered syllables. Yairi and Ambrose reported that 3% stuttered syllables distinguished most speakers who stutter from fluent controls. ⁹ From this, the recovered speakers can be considered to be close to fluent whereas the persistent speakers cannot.

Stuttering severity assessment (SSI-3)

SSI-3, a standard measure of stuttering severity, was made when the child was first seen and again at age 12 plus. ³ The assessments were always conducted on samples spoken in English, including two cases who did not stutter in English but did in their first language. This was because SSI-3 is not available for these languages and no norms are available. For SSI-3, a monologue, a dialogue and a text, all of which contained at least 200 syllables were recorded using a Sennheiser K6 microphone and a Sony DAT recorder. Associated physical concomitants such as tics and twitches were noted. SSI-3 scores were obtained by qualified personnel.

Statistical analysis

SPSS 11.0 (release 11.5.0; SPSS, Chicago) was used for the descriptive analysis, and for the non-parametric χ² tests. For parametric measures either independent t tests (SATS performance categories) or ANOVA (age of onset, SATS absolute scores and SSI-3 scores) were used for assessing differences between groups. For the parametric measures, 95% confidence intervals were calculated and p < 0.05 was considered to be significant.

Ethical approval

The means by which ethical approval was sought and received was by application to the UCL Committee on the Ethics of Non-NHS Human Research. Project ref 0754/003.

Results

Prevalence of LE and BIL in a sample of speakers who stutter

Prevalence of second language in the sample was 69/317 or 21.8%, which compared to 28.4% of bilingual children in general reported by the London Education Authority. A goodness of fit between the obtained data and the estimates using 28.4% as the expected value for bilingualism was not significant (χ² = 3.702, df = 1, p = .054). Thus there does not appear to be a difference between the number of children who used a minority language in this sample and in London schools in general.

Two (5.3%) of the 38 LE and BIL children stuttered in only one language. They were both BIL children who did not stutter in English. This is in agreement with reports that show that for bilingual speakers who stutter, it is usual to stutter in one language only. ^{10 11}

The number of LE children who stuttered (15) was lower than the number of BIL children who stuttered (23). However, when FB children were partitioned into LE and BIL groups using the same criteria, the reverse was true, with 28 LE and 10 BIL children. The association between LE and BIL and fluency group (children who stuttered/children who were fluent) was significant (χ² = 9.051, df = 1, p = .003). Thus, there was a relatively high proportion of BIL children who stuttered compared to those who were fluent.

Individual profiles and family structure of LE and BIL children who stutter compared to monolingual controls who stutter

The groups of children who stutter were compared on other characteristics to determine whether there were any differences between the LE, BIL and MONO children. Stuttering affects more males than females. ^{8 9} This was true of all three groups of children who stutter with male/female ratios of 4:1, 4.75:1 and 4.43:1 for LE, BIL and MONO groups respectively. A χ² showed that there was no association between these groups and gender, indicating that all three groups had the same gender imbalance towards males.

Stuttering usually starts early in development, but some time after language onset. ^{8 9} Mean age of onset for the LE, BIL and MONO groups was 4 years 9 months, 4 years 10 months and 4 years 3 months respectively. Again this corresponds with estimates obtained on other children in this age range ⁸ though children younger than those examined here report earlier onset. ^{9 12 13} A between-groups ANOVA showed that onset of stuttering did not differ between the three groups. For onset and gender comparisons, LE and BIL speakers have similar patterns to MONO speakers who stutter.

Early school performance of LE and BIL children relative to monolingual children who stuttered and bilingual controls who did not stutter

The caregivers reported SATS scores for English, mathematics and science for the children at key stage one (age seven) and key stage two (age 11). Age seven was before the child attended clinic but after they had had a minimum of two years’ schooling.

SATS performance scores for the age group that are regarded as exceptional, beyond expectation, at the level expected and below expectation are published for each key stage. Children were assigned to one of these categories separately for each subject examined (English, mathematics and language). Although performance levels would be expected to be lower for key stage one than for key stage two, the distributions can, nevertheless, still be compared. This was done for the LE, BIL, MONO and FB LE and FB BIL groups. The distribution of performance levels of the children across these groups was not significantly different by χ² (there was no association between SATS performance category across the five speaker groups p = .865).

Separate ANOVAs were conducted for English, mathematics and science on absolute scores. For each of these analyses there was one between-group factor (speaker group with five levels; LE who stuttered, BIL who stuttered, MONO who stuttered, FB who were LE and FB who were BIL) and one within-groups factor (assessment stage with two levels: key stage one, key stage two). For all three assessment types there was a significant improvement across assessment stages (English, F (1,44) = 160.762, p = <.001; mathematics, F (1,44) = 102.271, p = <.001; science, F (1,44) = 129.175, p = <.001) but no effect of speaker group. The interaction between speaker group and assessment stage was marginally significant for mathematics (F (4,44) =2.647, p = .046). This arose because the FB who were LE scored higher at mathematics at key stage one, but fell back to the same level as the other speaker groups at key stage two.

Severity assessment and recovery rates of LE and BIL children compared to monolingual controls who stutter

SSI-3 severity estimates ³ were examined at first attendance and at age 12 plus. The children were first divided into those who recovered and those who persisted in stuttering at age 12 plus, as severity measures were expected to differ across ages for the two types. The SSI-3 scores were examined in a three-way ANOVA with two between-groups factors (factor 1, speaker groups, has three levels: LE, BIL and MONO; factor 2, recovery type, has two levels: persistent or recovered), and one within-group factor (age with two levels: 8-10 and 12 plus). There was a significant decrease over ages: F (1,44) = 29.119, p< .001. There was also a significant main effect of recovery type, F (1,44) = 40.639, p < .001, but no significant main effect of speaker group, F (2,44) = .186, p = .836. There was also a significant interaction between age and recovery type, F (1,44) = 51.785, p < .001. Figure 1 shows this interaction on SSI-3 scores for persistent and recovered speakers at the two ages. The SSI-3 scores of the persistent groups were about the same across the two ages (scores around 30 points).The SSI-3 scores of the recovered group decreased from about 25 points at the first test age to around 15 at the second test age.

Figure 1.

The SSI-3 scores of the recovered and persistent groups over the two ages.

Stuttering tends to persist in BIL speakers and there is a higher risk of them starting to stutter. LE and MONO were collapsed together because the numbers in the LE group were low when these speakers were divided into persistent and recovered cases, and the patterns in the collapsed groups were similar. There was a significant association between speaker groups (BIL and LE-MONO) and recovery outcome (persistent/recovered) χ² = 4.381, df = 1, p = .036. The BIL group had a 25% recovery rate whereas LE and MONO together had a 55% recovery rate. These and the other statistics are summarized in Table 1.

Table 1.

Summary of statistical test performed. Throughout, “ns” signifies not significant and “sig“ significant

a) Comparisons between LE and BIL speakers
Stutterers	M/F ratio	% stutter in one/both	Age of stuttering	SATS	SSI-3
		language	onset
LE: N = 15	4:1	0%	4y 9m	Same as all peers	Same as other stutterers
BIL: N = 23	4.75:1	5.3%	4y 10m	Same as all peers	Same as other stutterers
	ns	-	ns	ns	ns

b) Comparison between all bilingual stutterers and bilingual non-stutterers
	Prevalence L2 in respective sample
Bilingual stutterers	21.8%
Bilingual non-stutterers	28.4%
	ns

c) Comparisons between stuttering and non-stuttering samples
	Incidence of LE and BIL	Recovery rate of LE and BIL
Stuttering sample	Fewer LE than BIL	Recovery rate of LE > BIL
Non-stuttering	More LE than BIL sample	-
	Sig	Sig

Discussion

The incidence of bilingualism in the clinical sample was 21.8%. This was roughly comparable with reported incidence of bilingualism in the same geographical area (28.4% of pupils in the London Educational authority are bilingual). Stuttering in one language by bilinguals is rare (only 5.3% of the sample of 38 bilingual children stuttered in just one of their languages). ¹⁰ The similar gender ratios (all around 4:1 for LE, BIL and MONO groups) and similar reported age of stuttering onset (around four and a half year for all groups) supported the view that all groups stuttered.

The BIL group was particularly prone to starting to stutter. Thus, at the age at which the children first attended clinic, there were more BIL than LE children whereas the reverse was the case in a control fluent sample. The statistics supported the view that there was a higher chance of BIL starting to stutter as the association between stuttering/not stuttering and BIL/LE was significant. This showed that BIL and LE groups were distributed differently in the stuttering, compared to the non-stuttering, sample (60.5% of the stuttering group were BIL whereas only 26.3% of the non-stuttering group were BIL). The BIL group also had a lower chance of recovery. Recovery rate at 12 plus was higher for LE and MONO combined than for BIL. The statistics supported the position that BIL had a lower chance of recovery as there was a significant association between BIL versus LE-MONO group and recovery outcome. Inspection of the data revealed only 25% of the BIL group recovered whereas 55.5% of the LE-MONO group recovered. Together, these findings suggest that if a child uses a language other than English in the home, deferring the time when they learn English reduces the chance of starting to stutter and aids chances of recovery later in childhood. A final factor of note is that school performance was not affected with respect to whether the child stuttered or not.

What is already known on this topic

Popular conception and early data ^{1 2} suggest an increased risk of stuttering for bilingual speakers, although the data are sparse and this claim has been questioned recently

The Lidcombe treatment programme has been reported to be an effective treatment for young bilingual children who stutter ¹⁴

What this study adds

Bilinguals usually stutter in both their languages (rather than just one). If a minority language alone is used in the home up to age five, the chances of starting to stutter is lower and recovery rate higher than for children who acquire English as well as a minority language during this period

Learning English concurrent or after a minority language does not affect educational attainment at key stages one and two.