The current study is set to investigate two debates. First, it has been controversial whether Mandarin tone representation is independent from segments in phonological encoding (e.g., Wan & Jaegar, 1998, 2003; Wan, 2007b) or dependently like lexical stress (e.g., Chen, 1999; Chen, Chen, & Dell, 2002). Secondly, previous studies had different views at whether inner speech is as feature-specified as overt speech (e.g., Corley, Brocklehurst, and Moat, 2011; Oppenheim & Dell, 2008, 2010, 2012), in which the core is the existence of phonemic similarity effect in inner slips and its size. Therefore, we employed a self-report-error tonguetwister paradigm (adapted from Brocklehurst & Corley, 2011) in inner and overt speech, comparing native Taiwanese Mandarin speakers’ tonal and segmental errors in the following Match/Mismatch (Tone × Syllable) conditions: (1) tone × syllable match; (2) tone mismatch; (3) syllable mismatch; (4) tone × syllable mismatch.

Although we preliminarily confirmed tonal error exists both in overt and inner speech, its amount in inner speech (only 7 of 197 slips) was rare, which limited our inferential analysis. Secondly, in the mixed-effect regression models, we found significant effect from syllable match/mismatch in predicting overt speech errors (i.e, segmental and tonal) and inner segmental errors, but not in tone match/ mismatch effect on error likelihoods. This suggested that tone conditions are not as influential as segments to speech errors, which might support Chen’s theses. Also, the presence of syllable match/mismatch effect in inner and overt segmental errors proved that phonemic similarity effect is shared in inner and overt Mandarin speech, implying that Mandarin inner speech is as feature-specified as overt speech, and this might reject some findings from Indo-European languages that inner speech is impoverished at feature level (e.g., Oppenheim & Dell, 2008, 2010, 2012). Details about the reason were discussed.