The impact of self-monitoring on language production: Errors, latencies, and disfluencies Robert J. Hartsuiker Language production research traditionally exploits patterns of speech errors, speech onset latencies and (to a much smaller extent) disfluencies in order to investigate how the language production system works. Current accounts of language production also postulate a self-monitoring system that inspects overt and internal speech and interrupts and repairs when a problem is detected. Accepting such accounts has an important implication however: The end-product (namely patterns of speech errors, response latencies, and disfluencies) is a function of both the speech production and the self-monitoring systems, and the challenge is to disentagle their contributions. For example, if one type of speech error occurs more frequently than another type, this may be due to the language production system being differentially error-prone, but it may also be due to the monitoring system editing out one type of error more often than another type. This paper therefore discusses the potential role of self-monitoring in shaping speech. The first part deals with the lexical bias effect, a prototypical example of a speech error pattern that has been attributed by some to speech production mechanisms (i.e., feedback) and by others to an editing process. On the basis of new data, a case will be made for the claim that both mechanisms play a role. The second part discusses response time effects in Stroop-like paradigms that, again, have been attributed by some to feedback and by others to (the comprehension component of) self-monitoring. The third part discusses the hypothesis that disfluencies result from monitoring and covert self-repair. This hypothesis will be compared to the alternative hypothesis that disfluencies 'signal' delays in planning. However, the signal hypothesis also needs to assume a role for the monitor: You cannot respond to an upcoming problem without a monitoring system having detected the problem. On the basis of the discussion, I conclude that self-monitoring plays an important role in shaping speech, and that is essential to better understand how this system works. As a first step in that direction, I will make two proposals: (1) the divide between monitoring accounts and feedback accounts is not really a divide: The monitor exploits feedback, precisely for the purpose of maintaining accuracy; (2) because the monitor is sensitive to patterns of activation, it can inspect the process (e.g., is there fierce competition at a given level, and hence a delay) and respond to problems in the process.