Investigating Low Literacy with Speech-to-Text Technology: Exploring Effects of Word Frequency and Age of Acquisition on Visual Word Recognition

Abstract

Low literacy remains a pressing issue, with over 3,1 million Dutch adults facing challenges in reading, writing, and math. These difficulties often go unrecognized due to stigma and limited identification tools. The present study investigates whether a speech-to-text-based visual word recognition task can offer a more accessible method for assessing word processing in low-literate individuals. Specifically, it examines how word frequency and age of acquisition (AoA) influence task performance. Forty-seven adults with low literacy levels completed a simulated evaluation of speech-to-text software. Participants repeated auditory word stimuli and indicated whether the automatic transcription matched their spoken word. Accuracy and reaction time were analyzed using (generalized) linear mixed models ((G)LMM). Results showed that both AoA and word frequency significantly affected reaction time: earlier-acquired and more frequent words were processed faster. However, only word frequency significantly influenced accuracy, with more frequent words recognized less accurately. Contrary to expectations, AoA did not significantly predict accuracy. Literacy level predicted reaction time, but no interaction effects were found between literacy and lexical variables. Exploratory analyses indicate that digital skills and reading frequency are also associated with task performance. This study confirms the potential of combining speech technology with psycholinguistic tasks to study low literacy in a less invasive way. It also highlights the need for stimuli that better align with individual linguistic experiences.