Artificial intelligence and communication disorders: Enhancing access to language assessment for multilingual children
Background: Multilingual children are under-identified for communication disorders compared to monolinguals, due to monolingual-centric assessment practices and limited home-language tools. A shift has begun to identify inclusive strategies, including narratives and non-word repetition tasks. However, a usage barrier is the limited number of bilingual speech-language pathologists (SLPs). Artificial Intelligence (AI)-based translation is emerging; however, it is based on English-centric adult models, which may influence relevance to multilingual children.
Aim: We examine the extent to which AI-based translation can facilitate home-language assessment scoring and identify multilinguals at risk for communication disorders.
Method/approach: Ethics approval was obtained (Pro00151738). After consent, caregivers completed the C-QUEB and ALDeQ language-background questionnaires. We used Zoom to assess 50 multilingual children on home-language narrative storytelling (Edmonton Narrative Norms Instrument [ENNI] and Global Tales) and non-word repetition (Quasi Universal Non-Word Repetition Task). We scored children based on three home-English language translation/scoring conditions: i: AI text translation (via DeepL and Google Translate), ii: translation by a team member, and iii: direct home-language scoring by a bilingual team member.
Results/findings: Our ANOVA indicates significant differences between direct bilingual scoring and AI scoring (F = 3.4; p < 0.01). We found scoring differences based on the type of AI translation, language distance, and linguistic content.
Conclusion: Our findings support English monolingual SLPs, educators, and psychologists in conducting home-language assessment scoring. Early access to language assessment positively influences a child’s long-term language and overall development. We promote diversity in child development and AI research by generating data from underrepresented languages.
Implications for children and families: If you or your child speaks more than one language, our research ensures you can access language assessments in your home language, so you don’t miss important language activities and support.
Implications for practitioners: If you work with multilingual children, our research can help you with home-language assessment scoring and identifying communication disorders, even when you don’t speak the child’s language.
Funding: University of Alberta.
Keywords: artificial intelligence, multilingualism, children, early language assessment, communication disorders
This presentation relates to the following United Nations Sustainable Development Goals: