Early interest in dictation programs for second language (L2) pronunciation learning emerged following rapid advancement of automatic speech recognition (ASR) and increased availability of commercial programs in the 1980s and 1990s (Rabiner & Juang, 2008). Dictation programs, which utilize ASR to provide text of users' speech, were not created for nonnative speakers (Cucchiarini & Strik, 2018), but researchers grew interested in whether transcripts could provide individualized feedback for learners (Coniam, 1999; Derwing, Munro, & Carbonaro, 2000). The usefulness of dictation depends upon the accuracy of the transcript and whether mistranscriptions are due to pronunciation errors. Ideally, a program would recognize speech as humans would, with mistranscriptions resulting from pronunciation errors that also reduce human listener intelligibility (Derwing et al., 2000). Twenty years ago, Coniam (1999) and Derwing et al. (2000) examined the accuracy of a popular dictation program, Dragon Naturally Speaking, for nonnative English speech. At the time, Dragon was a frontrunner in its field and dominated the late 1990s market (Pinola, 2011). Coniam (1999) asked 20 participants (10 first language English, 10 first language Chinese) to read two passages to the program after training it to their pronunciation, finding substantial differences in accuracy between native and nonnative speech. Derwing et al. (2000) asked 30 participants (10 first language English, 10 first language Spanish, and 10 first language Chinese) to dictate 60 sentences to the program, while audio recording. The recordings were played for 41 native-speaking listeners who transcribed and rated the speech samples on accentedness and comprehensibility. Expert raters also marked each sentence for segmental errors. Dragon transcribed less accurately than human listeners, particularly for nonnative speech. While