Stuttering or not? Analysis of language exposure effects on fluency assessment Seetal Ahluwalia Drs. Nan Ratner, Yasmeen Faroqi-Shah, and José Ortiz Department of Hearing and Speech Sciences, University of Maryland Undergraduate Honors Thesis May 3, 2024 1 Abstract Language exposure is hypothesized to impact bilingual speakers’ levels of typical and stuttering-like disfluency. The current study examined the relationship between English language exposure before school age and bilingual children’s speech fluency during an English task. The sample included 33 Spanish-English bilingual children from the English-MiamiBiling corpus at CHILDES. Participants were asked to narrate Mayer’s (1969) wordless picture book, Frog, Where Are You? Children who spoke only Spanish in the home were referred to as “MonoSpanHome” while children who spoke English and Spanish at home were referred to as “BilingHome.” It was hypothesized that children in the “MonoSpanHome” group would be more disfluent than their “BilingHome” peers. It was found that the “MonoSpanHome” participants had an increased number of typical disfluencies than their “BilingHome” counterparts. However, the number of stuttering-like disfluencies and total overall disfluency was similar between the two groups. Keywords: Stuttering, Bilingualism, Narrative Task Background Fluency and Disfluency There are numerous connotations of the word “fluency.” In the context of speech, it refers to a speaker’s ease, effort, and proficiency (Logan, 2015). Observable features of speech include continuity, smoothness, and rate. Continuity of speech is defined as the connected flow of components of spoken messages (Logan, 2015). Smoothness is synonymous with the rhythm or prosody of one’s speech. Finally, rate refers to the speed of speech. These factors, along with the effort a speaker expends in delivering a message, are considered in determining one’s fluency (ASHA, 2024). 2 Disfluency is not equivalent to stuttering, as there are numerous factors that contribute to disfluent speech. For example, conversation demands and environmental factors increase a speaker’s likelihood of being disfluent (ASHA, 2024). Every speaker experiences disfluency at some point or another; in fact, many people are disfluent when speaking publicly, yet are not diagnosed as people who stutter (PWS). Disfluent speech is characterized by hesitations, use of filler words, and repetitions of words or phrases. In order to differentiate between disfluency and stuttering, one must carefully analyze the type and frequency of a speaker’s disfluencies, as PWS exhibit different disfluencies than people who do not stutter (PWNS). Typical and Stuttering-like disfluencies When speaking, individuals pause, revise, and repeat parts of their message; these aspects of conversational speech are exhibited by PWS and PWNS alike. These brief interruptions in speech are disfluencies (Arslan et al., 2023). Typical disfluencies (TDs) include filled and unfilled pauses, word and phrase revisions, and phrase repetitions (Bloodstein et al., 2021). Stuttering-like disfluencies (SLDs) differ from TDs in that they include unintentional syllable repetitions, prolongations, and blocks (Coleman, 2013). Blocks are characterized by a disruption of air flow, where the speaker pauses and is unable to produce sounds (ASHA, 2024). The frequency and type of disfluencies differ between PWS and PWNS. In their 1999 study, Ambrose and Yairi compared children who did and did not stutter. PWS produced nearly three times as many disfluencies per 100 syllables than PWNS. Additionally, they found that PWNS mostly produce interjections and revisions. While they may exhibit sound prolongations, partial word repetitions, and other SLDs, PWNS produced these disfluencies at a much lower rate than PWS. These trends alone are not sufficient as a basis of stuttering diagnosis, an individual’s perceived affect is a crucial factor in distinguishing disfluency from stuttering. PWS 3 differ from PWNS in their feelings, interactions, or thoughts about their stuttering; their affect, behavior, and cognition, respectively (Bloodstein et al., 2021). Bilingualism and Fluency Language Knowledge and Disfluency Though there is insufficient evidence to substantiate the claim that bilingualism is a risk factor for stuttering, there is research that supports a link between the two with numerous studies on the impact of bilingualism on fluency. The relationship between speaking more than one language and being disfluent was studied by Woumans et al. (2021) and Rojas et al. (2023). Woumans and colleagues (2021) studied twenty-eight bilingual PWS to understand the influence of language knowledge on speech disfluency. Similarly, Rojas and colleagues (2023) aimed to examine differences in speech disfluency. They studied 106 Spanish-English bilingual children with typical development. The studies’ differed in sample size and selection; Rojas et al. participants were specifically Spanish-English bilinguals, whereas Woumans et al. were either Dutch-French or Dutch-English bilinguals. Both studies concluded that bilinguals produce more disfluencies in their less dominant language. This difference in production of disfluencies indicates a relationship between language knowledge and speech fluency. Specifically, Rojas and colleagues (2023) found that English dominant bilinguals produce fewer TDs and SLDs in their English compared to Spanish, whereas Spanish dominant bilinguals produced fewer SLDs in Spanish than English. The current study aims to further investigate the relationship between language knowledge on speech disfluency in bilingual children by analyzing the effects of English exposure before school entry, age 5, on Spanish-English bilingual children’s speech fluency for an English narrative task. Linguistically Responsive Assessments 4 Assessment tools and guidelines normed on monolingual English speakers are commonly used in the field. The use of these tools on bilingual speakers can lead to false-positive identification of stuttering for typically fluent speakers (Byrd et al., 2020). A potential consequence of this includes utilization of inappropriate therapy strategies. For example, a bilingual speaker incorrectly identified as a child who stutters (CWS) would receive stuttering therapy, while they would benefit from English for Speakers of Other Languages (ESOL). Misdiagnosis of Disfluency as Stuttering Stuttering is a complex, neurodevelopmental, fluency disorder. It is not identified solely by the presence of disfluencies; rather it is marked by affective, behavioral, and cognitive components, or “the ABCs” of stuttering (Bloodstein et al., 2021). The more relevant piece of the condition is not what is observed by the listener, but what is experienced by the speaker (Panzarino et al., 2024). Most PWS experience affective reactions to stuttering; these reactions include frustration, guilt and shame, and fear and anxiety (Bloodstein et al., 2021). Panzarino et al. (2024), note that even behavioral aspects of stuttering, as well as physical tension, differentiate it from typical disfluency. The muscle tension experienced by PWS, specifically adults, is reportedly prevalent in the jaw, front of mouth and throat, and the chest and abdomen (Bloodstein et al., 2021). Secondary behaviors are also commonly exhibited by PWS. Examples of these mannerisms include visible tension in one’s face (e.g., frowning), head jerking, and excessive eye blinking (Bloodstein et al., 2021). Cognitive factors in stuttering refer to avoidance, often a result of anticipation of stuttering, through remaining silent, circumlocution, and having someone speak for the PWS (Bloodstein et al., 2021; Panzarino et al., 2024). Speech pathologists can assess stuttering in multiple languages, regardless of the extent of their knowledge of the language (Byrd et al., 2020). Basing stuttering diagnosis solely on 5 frequency of disfluency can lead to increased misdiagnosis of multilingual speakers. The likelihood of false-positive diagnoses for bilingual children increases when they do not have “native-like” knowledge, common in young children who are acquiring either their first or additional language (Byrd et al., 2020, p. 2). Additionally, two key factors that contribute to increased misdiagnosis of bilingual speakers are insufficient understanding of SLDs in languages other than English and use of assessment materials normed on monolingual English speakers. There is limited credible research on the relationships between bilingualism and fluency and bilingualism and stuttering. In fact, one study (Travis et al., 1937) that is often used as evidence for the claim that bilingualism is a risk factor for stuttering has been shown to be seriously flawed. Gahl (2020) investigates the claims, methodology, and data analysis of Travis and colleagues. According to Gahl (2020), two primary concerns were that the prevalence rates reported in Travis et al. were internally inconsistent and that statistical significance is not reliable. Multiple children who stutter from the monolingual group were excluded from the prevalence estimate which created a discrepancy in the reported rates between groups. Additionally, prevalence of stuttering was lower among trilinguals than bilinguals and monolinguals. A significant consequence of the claims by Travis et al. (1937) is that parents are erroneously advised that raising their child to be bilingual will increase the risk of stuttering, leading to heritage language loss. In order to address this, a commitment to accurate research on bilingualism and stuttering is needed. Introduction Summary of Research Question and Hypothesis We used the CHILDES English-MiamiBiling corpus to investigate how exposure to English before school entry affects fluency for an English task. Participants were asked to tell a 6 story in English. “MonoSpanHome” children were expected to be more disfluent than “BilingHome” children because of their presumed lower English language knowledge, as they were monolingual Spanish speakers until school entry. Typical and stuttering-like-disfluencies were hypothesized to be elevated amongst “MonoSpanHome” participants, as they were asked to complete an English based task. If this hypothesis is supported, we will have supported the findings of Woumans et al. (2021), which concluded that speaking in a non-dominant language increases both typical and stuttering-like disfluencies in bilingual persons who stutter. We hypothesized that non-stuttering children who spoke only Spanish in the home, or did not have English exposure before school age, would be more disfluent when narrating in English than their counterparts who spoke both English and Spanish, or had English exposure before school age. Methods Data for this study came from Pearson et al. (2002). They studied the differences between monolinguals’ and bilinguals’ narrative task performance, as well as the impact of dual language instruction on bilinguals’ narrative task performance in both languages. Furthermore, Pearson and colleagues analyzed SES and school type on narrative task performance. The sample includes 240 participants who narrated Frog, Where Are You; bilinguals told the story in English and Spanish on different days. Pearson and colleagues (2002) hypothesized that there would be an unequal observed difference of narrative ability in English for monolingual and bilingual speakers. Additionally, they expected that dual instruction would positively impact bilingual speakers’ performance on a narrative task. Lastly, a predictive relationship was expected on exhibited narrative abilities between languages. The cross-sectional design of the study required that conclusions remain tentative. A major conclusion was that time spent learning Spanish did 7 not inhibit progress in English. The current study utilized Pearson and colleagues’ data from 2002 out of convenience. Participants We analyzed data from 33 participants from the English-MiamiBiling corpus of the Child Language Data Exchange System (CHILDES) at TalkBank.org (Pearson et al., 2002). Participants were asked to narrate the wordless story book, Frog, Where Are You? by Mayer (1969). There were twelve males and twenty-one females who ranged between seven and ten years old at time of recording, with an average age of 8.85 years. There were twenty-one participants from middle socioeconomic status (SES) families and twelve participants from low SES families. Age, sex, and SES were not matched between groups. All children were English-Spanish bilinguals. However, some children had exposure to English at home before school entry, while others were monolingual Spanish speakers until school entry. There were seventeen participants in the “MonoSpanHome” group and sixteen participants in the “BilingHome” group. All participants were considered to be typically-developing and typically fluent when enrolled in the study. Participants were second graders who were between seven to eight years old and fifth graders who were between ten to eleven years old. Differences between participants included sex, school type, socioeconomic status (SES), and grade. School types included an English immersion program for Hispanic students, a dual or “two-way” bilingual program for Hispanic students with 50% Spanish and 50% English instruction, a regular monolingual English classroom for non-Hispanic students, and a monolingual English classroom with primarily Hispanic students. Participants were from middle and low SES backgrounds. 8 Data Collection and Analysis Bulleting and Updating Coding of the Language Samples All transcripts were linked, reviewed, and coded for fluency by the author. The original orthographic representations of participants’ speech were used as a starting point. After downloading all applicable transcripts and audio samples from the CHILDES English-MiamiBiling corpus, the corresponding pairs of files were linked. Audios and transcripts were linked on CLAN through the bulleting process. This allowed for utterances to be separated into short clips, allowing for more manageable segments for coding. Before beginning the coding process, all transcripts were checked for misspellings through the Mor command on CLAN. To begin coding, the “Walker Controller” window was opened. The settings for “Walk length” and “Loop number” were adjusted to 40000 msec., and 3, respectively. This allowed for individual bullets to be replayed thrice in their entirety with the click of one button, F6. Each utterance was heard at least three times, difficult or lengthy ones were replayed as needed. Transcripts were coded for fluency according to FluencyBank conventions. Typical and stuttering-like disfluencies were denoted by their corresponding code. To be considered complete, transcripts were run through the Check and Mor commands on CLAN. Any errors were troubleshot and fixed. This assured that the transcripts’ contents were aligned with CLAN’s conventions. 9 Table 1 Coding Stuttering and Other Fluency Behaviors Note: This table is from A Clinician’s Complete Guide to CLAN and PRAAT (Ratner et al., 2023). FluCalc Upon completion of coding and reviewing of all 33 transcripts, FluCalc was utilized to compute fluency-related behaviors. FluCalc is a fluency computation program that allows for efficient analysis and comparison of fluency behaviors within an individual or collection of samples. This program created an Excel spreadsheet where it reported numerous fluency and grammatical analyses along with participants’ identification number, age, sex, school type, SES, and grade. Examples of provided analyses include number of utterances, words, and syllables. Additionally, FluCalc provided the number and percentage of all TDs and SLDs. 10 Results In total, thirty-three audio samples were applicable for analysis. The applicable samples from the English-MiamiBiling corpus consisted of 1, 216 utterances and 9, 881 words. To be deemed applicable, transcripts needed to have accessible, matching audio files with minimal background noise. Of the 269 available English transcripts, only thirty-seven had audio samples. Thirty-three out of the thirty-seven available files were used, with the remaining four being excluded because of mismatch between the audio files and written transcripts or the participants’ speech being indiscernible due to excessive background noise. Table 2 demonstrates the difference of speech disfluencies observed in the samples amongst groups. At a glance, pauses, a typical disfluency, had the largest percentage in both groups. Conversely, broken words were not common and did not occur in the “MonoSpanHome” group at all. Table 2 Total Percentage of Syllables of Typical and Stuttering-Like-Disfluencies per Group “MonoSpanHome” “BilingHome” Prolongation 28.51 20.36 Broken Word 0 0.28 Blocking 1.86 2.75 Repeated Segment 0.64 0.92 Word Repetition 25.3 19.8 Phrase Repetition 12.65 11.26 Word Revision 10 6.46 Phrase Revision 13.82 12.67 Phonological Fragment 6.44 7.4 11 Pause 133.29 100.92 Filled Pause 17 4.68 Figure 1 depicts the higher frequency of typical disfluencies in the “MonoSpanHome” group (M = 11.37, SD = 3.09) compared to the “BilingHome” (M = 8.96, SD = 2.62), t = -2.40, p = 0.022. There was a -2.40 mean difference between groups for typical disfluencies. Children who did not have English exposure before school age were significantly more typically disfluent than their counterparts who did speak English in the home. The two outliers in the “MonoSpanHome” group are represented by red dots in Figure 1. One participant was far more typically disfluent (20.08%) and one was far less typically disfluent (4.74%) than others in the group. The participants with the highest percent of pauses, a typical disfluency, were from the Spanish until school entry group, with 17.67% pauses found in their audio sample. Additionally, the number of filled pauses was highest among a participant from the “MonoSpanHome” group. This participant had 6.20% filled pauses in their audio sample. Two participants from the “MonoSpanHome” were the most typically disfluent, with 15.89% and 20.08% TDs in their samples. Figure 1 Percent of Typical Disfluencies for Children from Monolingual Spanish Homes and Bilingual English-Spanish Homes 12 The percent of stuttering-like disfluencies between groups were not significantly different between groups as shown in Figure 2. The “MonoSpanHome” had an average SLD of 3.31 with a standard deviation of 3.22, whereas the “BilingHome” had an average SLD of 2.81 and a standard deviation of 2.13, t = -0.53, p = 0.60. Similar to Figure 1, there is an outlier depicted by a red dot in Figure 2 the “MonoSpanHome” group where one participant experienced significantly more stuttering-like disfluencies (15.12%) than others in the group. There is less variability among the percent of SLDs in the “MonoSpanHome” group, with upper and lower limits much closer apart than those of the “BilingHome” group. The participant with the highest counts of prolongations, a stuttering-like disfluency, was from the “MonoSpanHome” group with 7.75% prolongations over total number of syllables in the sample. A “MonoSpanHome” participant had the highest percent of stuttering-like disfluencies with 15.52% SLDs over total number of syllables in the sample. Figure 2 Percent of Stuttering-Like Disfluencies for Children from Monolingual Spanish Homes and Bilingual English-Spanish Homes 13 The total percentage of TDs and SLDs is not significantly different between “MonoSpanHome” and “BilingHome.” Figure 3 depicts the slightly higher frequency of total disfluencies in the “MonoSpanHome” group (M = 14.68, SD = 5.27) compared to the “BilingHome” (M = 11.77, SD = 3.57), t = -1.84, p = 0.075. There were two outliers in the Spanish only until school entry group whose percent of total disfluencies were above others in the group. These participants had total disfluencies of 31.01% and 22.89% over total number of syllables in the sample. Figure 3 Total Percent of Stuttering-Like Disfluencies and Typical Disfluencies for Children from Monolingual Spanish Homes and Bilingual English-Spanish Homes 14 The hypothesis that children who only spoke Spanish until school entry, “MonoSpanHome,” would have higher rates of disfluency was supported for typical disfluencies, but not stuttering like disfluencies. Discussion Children who had Spanish exposure until school entry were more typically disfluent when narrating a story in English. This is supported by the increased percent of typical disfluencies in the “MonoSpanHome” group. The percentage of total disfluency, TDs and SLDs, were not significantly different between groups, indicating that children with English exposure before school age were as disfluent as their counterparts without English exposure. Woumans et al. (2021) found that bilingual PWS are more disfluent in their non-dominant language. This is relevant to our findings, as participants from the “MonoSpanHome” group produced numerous disfluencies during the assigned task in English. While the current study only compared the two groups’ English responses, it is expected that analysis of the Spanish audio samples would yield results that support those of Woumans et al. (2021). 15 Given that the findings of the current study do not indicate stark differences between the rates of disfluency between the two groups, the increased likelihood of hypothesized misdiagnosis of bilingual children as PWS is concerning. Misdiagnosis may be a result of a lack of understanding of stuttering. The total percent of TDs and SLDs in the “BilingHome” group being similar to the “MonoSpanHome” group indicates that children are disfluent regardless of language knowledge. Sixteen participants had weighted SLD (WSLD) scores above 4.0 which is highly suggestive of clinical diagnosis of stuttering in young children (Ambrose & Yairi, 1999). Eight participants from both groups had WSLDs above 4.0. That being said, it is unlikely that all 16 of these children can be diagnosed as people who stutter. In fact, it is not possible to confidently claim that any of the participants stuttered, as tension and stuttered quality of speech were not observed. WSLD alone cannot be used to determine whether the children are PWS; rather, the childrens’ affect and behavior must be taken into consideration when diagnosing children as people who stutter. Limitations Limitations of the current study include the absence of inter-coder reliability as the author was the sole coder of all samples. Similarly, differentiation of short, medium, and long pauses as (.), (..), and (...) was subjective. Moving forward, it would be beneficial to have these files coded by another researcher to determine whether the fluency codes used by the author were reliable. In the future, coders should be blinded to participants’ groups to limit any implicit biases. Additionally, in order to truly replicate the findings of Woumans et al. (2021), participants’ English audio sample should be compared to their Spanish audio sample. This would allow for a clearer understanding of whether a specific group is more disfluent in their less dominant language. Lastly, population size needs to be increased to be able to truly address the question of 16 bilingualism as a risk factor for stuttering. The current study cannot fully answer the question, as it is not known if any participants were PWS. Recommendations for Speech Pathologists To better serve bilingual populations, speech pathologists must thoroughly understand the differences between stuttering and disfluency before diagnosing anyone as a PWS. When misdiagnosed, bilingual PWNS receive inaccurate forms of therapy. For example, a Spanish-English bilingual PWNS may likely benefit from English as a Second Language (ESOL) instead of stuttering management therapy. One clear distinguishing factor that speech pathologists must understand is that PWS tend to be visibly frustrated, anxious, or upset when they are more disfluent. Additionally, speech pathologists should encourage parents to use their native language with their children. They should not feel pressured to use the dominant language, as it has not been proven that bilingualism is a risk factor for stuttering. 17 References Ambrose, N. G., & Yairi, E. (1999). Normative disfluency data for early childhood stuttering. Journal of Speech, Language, and Hearing Research, 42(4), 895–909. https://doi.org/10.1044/jslhr.4204.895 American Speech-Language-Hearing Association. (2024). Fluency disorders. American Speech-Language-Hearing Association. https://www.asha.org/practice-portal/clinical-topics/fluency-disorders Arslan, B., Aktan-Erciyes, A., & Göksun, T. (2023). Multimodal language in bilingual and monolingual children: Gesture production and speech disfluency. Bilingualism: Language and Cognition, 26(5), 971–983. doi:10.1017/S1366728923000196 Bloodstein, O., Ratner, N. B., & Brundage, S. B. (2021). A handbook on stuttering (7th ed.). Plural Publishing, Inc. Byrd, C., Werle, D., Coalson, G. A., & Eggers, K. (2020). Use of monolingual English guidelines to assess stuttering in bilingual speakers: Journal of Monolingual and Bilingual Speech, 2(1), 1–23. https://doi.org/10.1558/jmbs.12733 Coleman, C. (2013, September 26). How can you tell if childhood stuttering is the real deal? . ASHAWire. https://shorturl.at/avyAF. Gahl, S. (2020). Bilingualism as a purported risk factor for stuttering: A close look at a seminal study (Travis et al., 1937). Journal of Speech, Language, and Hearing Research, 63(11), 3680–3684. https://doi.org/10.1044/2020_jslhr-20-00364 18 Logan, K. J. (2015). Fluency Disorders. Plural Publishing, Inc. (San Diego, CA). Mayer, M. (1969). Frog, where are you? New York, Dial Books for Young Readers. Panzarino, R., Vanryckeghem, M., Bedwell, J. S., & Wendt, O. (2024). Relationships between psychological distress and affective, behavioral, and cognitive experiences of stuttering. American Journal of Speech-Language Pathology, 33(1), 262–278. https://doi.org/10.1044/2023_ajslp-22-00347 Pearson, Barbara Z. (2002). Narrative competence among monolingual and bilingual school children in Miami. In D. K. Oller and R. E. Eilers (Eds.), Language and literacy in bilingual children (pp. 135-174). Clevedon, UK: Multilingual Matters. Ratner, N. B., Brundage, S. B., & Fromm, D. (2023). A clinician’s complete guide to Clan and Praat. TalkBank. https://talkbank.org/manuals/Clin-CLAN.pdf Rojas, R., Irani, F., Gusewski, S., & Camacho, N. (2023). A cross-sectional investigation of disfluencies in typically developing Spanish-English bilingual children. Journal of Fluency Disorders, 77, 1–10. https://doi.org/10.1016/j.jfludis.2023.105988 Travis, L. E., Johnson, W., & Shover, J. (1937). The relation of bilingualism to stuttering. Journal of Speech Disorders, 2(3), 185–189. https://doi.org/10.1044/jshd.0203.185 Watson, J. B., Byrd, C. T., & Carlo, E. J. (2011). Effects of length, complexity, and grammatical correctness on stuttering in Spanish-speaking preschool children. American Journal of Speech-Language Pathology, 20(3), 209–220. https://doi.org/10.1044/1058- 0360(2011/10-0019) 19 Woumans, E., Van der Linden, L., Hartsuiker, R., Duyck, W., Moerenhout, C., de Partz, M. P., Pistono, A., De Letter, M., & Szmalec, A. (2021). Speech fluency in bilinguals who stutter: Language proficiency and attentional demands as mediating factors. Journal of Fluency Disorders, 69, 105850. https://doi.org/10.1016/j.jfludis.2021.105850