Åsa Wengelin, University of Gothenburg

Writing presents significant challenges for many children, particularly for those with reading and writing difficulties such as dyslexia. Beyond persistent spelling problems, these children frequently produce texts of lower overall quality than their peers. The latter is commonly attributed to cognitive bottlenecks during transcription, which place heavy demands on working memory and restrict the resources available for higher-level writing processes. One proposed way to reduce transcription demands is the use of speech-to-text (STT) technology. However, previous research on STT as a writing support for children has yielded mixed results, suggesting that its effectiveness may depend on multiple interacting factors.

Thus, we set out to investigate for whom and under what conditions speech-to-text can function as an effective writing support for children with reading and writing difficulties. We focus on two central explanatory factors: (a) the linguistic strengths and limitations of the speech recognition system, particularly in relation to different lexical and phonological word properties, and (b) individual differences in children’s cognitive and linguistic skills.

58 children aged 10-13 dictated preset sentences and task-oriented texts using speech-to-text, which were compared to keyboarded texts. Analyses addressed (a) the system’s recognition accuracy as a function of words’ linguistic properties, (b) children’s success in dictating these sentences, (c) their success in producing extended texts using speech-to-text, and (d) the extent to which sentence-level dictation performance, together with children’s cognitive and linguistic characteristics, predicts successful text production using speech-to-text. The findings highlight the importance of considering individual profiles and task demands when implementing STT in educational contexts