Speech-language pathologists (SLP) make impressionistic speech intelligibility judgments of spontaneous speech samples (e.g., conversational speech) as part of their assessments of children for speech sound disorder. However, since judgments of intelligibility are established with an auditory-perceptual assessment of spontaneous speech by a clinician, they are susceptible to a variety of sources of error and biasing effects of speaker characteristics (Kent, 1996). This is particularly an issue with bilingual children who are disproportionately misidentified as having speech and language impairments (Artiles et al., 2002; Cycyk et al., 2022; Sullivan, 2011). When trained and untrained listeners were provided with visual information (i.e., physical appearance) about a child speaker’s race, it impacted their assessment of speech production accuracy (Evans et al., 2018). So if a listener expects a child’s speech to sound a particular way based on their physical appearance, they anticipate and make judgments based on those expectations. In particular, listeners’ beliefs about a speaker’s ethnic and/or linguistic background influence the way they perceive speech by either inhibiting or facilitating accurate perception (Kutlu et al., 2022; Melguy & Johnson, 2021; Vaughn, 2019).
Given this background, we wondered if bilingual SLPs would rate the intelligibility of bilingual speakers differently than monolingual SLPs would. To answer this research question, we collected perceiver data using video and audio-only recordings of three child speaker groups which included a monolingual English control group, a bilingual Spanish-English group, and a bilingual Mam-English group. Note that Mam is a Mayan language spoken in Central America that is not related to Spanish and thus has different sounds and language structures. Additionally, all monolingual-English children were White (given the demographic of the recruitment area) and the Spanish- and Mam-English bilingual children were Hispanic. The goal of this research was to improve the multicultural sensitivity of SLPs to reduce the number of bilingual children who are misidentified as having speech and language impairments. By understanding what influences SLPs when they are making speech intelligibility judgments of spontaneous speech, graduate-level training and assessment procedures can be modified so they are less biased.
Overall, seeing a speaker results in higher intelligibility than hearing speech alone because of lipreading cues. Intelligibility ratings of conversational speech samples provided by school-age bilingual children in two presentation modalities (i.e., audio-only and audio-visual [AV]) were examined to determine whether there were other impacts (e.g., linguistic bias) of seeing the children. Forty speech-language pathologists (SLP) rated the intelligibility of conversational speech samples provided by 24 typically developing school-aged children in three groups (eight in each): bilingual Spanish-English, bilingual Mam-English (Mam is an Indigenous, Mayan language spoken in Guatemala), and monolingual English (control). SLPs heard audio-only and AV recordings, with individual two-minute speech samples in a randomized order, in an online experiment. Data were analyzed using a linear mixed-effects model to determine if there were effects of presentation modality or speaker group on the intelligibility ratings. No statistical differences were found between intelligibility ratings from the audio-only and AV presentation modalities for any speaker group. The Mam-English speaker group had significantly lower intelligibility than the Spanish-English and control groups.
SLPs rated conversational speech samples of monolingual English and bilingual child speakers as having similar intelligibility regardless of whether they heard audio-only or AV recordings. Given the advantage of lipreading, the lack of differences between AV and audio-only ratings indicates a linguistic bias. Clinically, when SLPs use intelligibility ratings in their assessments of bilingual children’s speech for speech sound disorders, it is vital to consider the influences of lipreading, language exposure, and linguistic bias.
Coauthor: Allison Haskill
This study examines differences between two language sample types, conversation and narrative, to determine if there is an optimal type for assessing bilingual school-age children’s syntactic complexity and lexical diversity. Conversation and narrative language samples were elicited from 24 typically developing school-age children in three groups who participated in the study: monolingual English, bilingual Spanish-English, bilingual Mam-English. Two measures of syntactic complexity and two measures of lexical diversity were calculated. Preliminary results indicate that bilingual children’s mean length of utterance varies systematically with language sample type, as does number of different words.
Results demonstrate differences in morphosyntactic and semantic measures between narrative and conversational language samples elicited from bilingual children. The optimal type of language sample elicited by a speech-language pathologist may be best determined by the suspected area of difficulty for each specific child.
Coauthor: Melissa Redford
This study examined global intelligibility estimates (GIE) of school-age children’s speech by speech-language pathologists (SLPs). To understand GIE, we compared it to other ratings of intelligibility and comprehensibility and determined whether children with speech sound disorder (SSD) were identified. We also examined the influence of children’s language skills and non-native accents. Spontaneous and experimentally controlled speech was elicited from 30 children in two age and four language groups: English-speaking monolinguals with SSD; typically developing (TD) English-speaking monolinguals, and TD English-speaking multilinguals with either Spanish or Mam as their first language. In Experiment 1, spontaneous speech was evaluated by 42 SLPs, who provided GIE and language skill ratings (LSR). SLPs heard experimentally controlled speech to generate laboratory-based intelligibility scores (LIS) and comprehensibility ratings (CR). Experiment 2 investigated the impacts of independent language measures and accentedness ratings on perceptual measures. Experiment 1 showed effects of child age and language group on all measures, but CR distinguished children with SSD from TD children better than LIS and GIE. LSR and all measures of intelligibility/comprehensibility were correlated. Perceived accentedness accounted for significant variance in GIE, LIS, and CR above and beyond language skill and its correlates (Experiment 2).
The findings suggest that perceptual assessments are more sensitive to speech differences when the samples on which they are based are controlled for language factors. Clinicians should be careful when using intelligibility ratings to diagnose SSD because our inability to look past non-native accents and language skills may contribute to intelligibility ratings.