2017 Plan, Scan, Speak, Detect, Stop, Repair

In this paper we develop some predictions on interactional segmental speech errors and their repairs, based on a computational model of self-monitoring proposed by Hartsuiker and Kolk (2001) and modified by Nooteboom and Quené (2017). These predictions were tested in a four-word tongue twister experiment eliciting such errors and their repairs, in word initial and word medial position.

Findings are:

(1) The statistical distributions of error-to-cutoff times, although truncated close to 0 ms, are nearly complete for both initial and medial consonant errors. This implies that against prediction interruption of the speaking process after internal error detection takes more time than speech initiation.

(2) The distributions of cutoff-to-repair times are censored at 0 ms, but cutoff-to-repair times are longer for medial than initial consonants, showing that against prediction repairing takes more time for medial than for initial errors.

(3) Detection rate is much lower for medial than for initial consonants and decreases with position of the misspoken word in the tongue twister sequence. This probably reflects predicted variation in selective attention.