Dictation automatic speech recognition for second language pronunciation assessment: Focus on age-related bias

Dictation automatic speech recognition (ASR) technology offers significant advantages in second language (L2) pronunciation assessment, a claim supported by research showing strong correlations between ASR and human-rater scores. However, ASR systems may introduce biases linked to age-related physiological changes that affect performance. Despite studies on age bias, none address dictation systems in assessment contexts. This study investigates age bias in Google Voice Typing (GVT), Microsoft-Transcribe (MS-T), and Apple Siri for English L2 pronunciation assessment. We analyzed 1,000 university test responses spanning five L1 groups (French, Spanish, Persian, Arabic, Chinese) and three age groups (<30, 30–44, >44). Recordings were processed through GVT, MS-T, and Siri to generate word accuracy scores. Regression analyses revealed the target ASR systems favored test takers ≤29 years (p =.001). These findings highlight dictation ASR-related age biases and underscore the need for fairness in automated assessments.