In my last post, I discussed the question of whether Levelt’s Main Interruption Rule – speech, in a rule, is interrupted immediately after the detection of an error – is still valid considering new empirical findings. My conclusion has been that it is valid at least for phonological errors, that is, for that kind of errors which is relevant for my theory of stuttering, the Attention Allocation Theory. There, it is assumed that stuttering is caused by interruptions of speech flow due to invalid error signals – signals that resemble the feedback of phonological errors.
The idea behind this theory is that stuttering is neither a breakdown of speech control (which would not account for the very distinctive symptoms) nor caused by a special obscure pathomechanism. Instead, stuttering is assumed to be caused by a normal component of speech control – the error detection mechanism – that produces invalid error signals because of distorted, degraded, or incomplete sensory input. In other words: The idea behind my theory is that the mechanism of stuttering and the mechanism of interruption for error repair are one and the same.
In stuttering, the interruption of speech flow (inhibition of a speech motor program, resulting in a block, repetition, or prolongation) occurs against the speaker’s will. If the mechanism of stuttering and the mechanism of interruption for error repair are one and the same mechanism, then interruption for error repair must be an automatic brain response, independent of the speaker’s will. This however seems to be a matter of debate:
Seyfeddinipur, Kita, and Indefrey (2008), for example, begin the Abstract of their article as follows: “When speakers detect a problem in what they are saying, they must decide whether or not to interrupt themselves and repair the problem, and if so, when.” Likewise, Levelt in his book “Speaking. From intention to articulation” writes: “When a speaker detects trouble that is sufficiently alarming according the speaker’s current standards, the decision will be taken to interrupt.” (p. 478)
On the other hand, Levelt (1983) writes: “From an analysis of 959 spontaneous self-repairs it appears that interrupting follows detection promptly, with the exception that correct words tend to be completed.” (p. 41). “Interruption follows detection promptly” does not sound like as if a conscious decision is made, a weighting of costs and benefits of an interruption, which might take some more time. J. D. M. Layer (1972) did not view conscious awareness as necessary for either monitoring or the decision to interrupt speech. He stated. “It is often the case that a speaker makes a slip and corrects it, without either the speaker or the listener being aware that a slip has occurred” (p. 141; cf. Blackmer & Mitton, 1991, p. 176).
I think we should differentiate between several kinds of error. As mentioned in my last post, some authors define the term ‘error’ very broadly. For instance, Seyfeddinipur et al. (2008) rate a fresh start as in “this house had um the entrance was big” (838) as an error, but suchlike changes in formulation or even in conceptualization are normal in spontaneous speech. The speaker, in this case, did not actually make and repair an error, he could continue without fresh start: “this house had a big entrance”. Many of them what is rated as an ‘error’ in psycholinguistic studies is more the manifestation of tentativeness in speech planning, in conceptualization or formulation: The speaker does not actually say something wrong or mispronounce a word, but changes the way of formulation or the intended message.
Looking at speech errors in a narrower sense, namely at production-based errors, we can distinguish two cases: (1) saying something inappropriate, contentually incorrect or ambiguous, which may evoke misunderstanding or irritation, and (2) phonological and grammatical errors, and confused words, which do not necessarily result in misunderstanding (the listener often knows what the speaker has intended to say), which however can appear as if the speaker is confused or does not master the language.
In the first case, the speaker, after having realized the problem, decides whether or not to interrupt him-/herself and make a repair, and if so, when – just as Seyfeddinipur and colleagues put it – and most production-based errors in spontaneous speech are of this kind. In the second case, however, I think that the interruption of speech flow is not the result of the speaker’s decision, but rather an automatic brain response, and even the subsequent repair does not seem to result from a voluntary decision, but rather from a habit; it seems to be a kind of automatized behavior.
Errors of the second sort are, in a sense, violations of language rules – phonological, grammatical, or semantic rules (defined structures or relations), and speakers seem to have a strong need to say things in the correct manner, at least when talking in their native language – even if the listener has quite understood what a ‘cuf of coffee’ means. Similarly, when I, as a stutterer, have a block at the onset of a word, and the listener already knows what I want to say and says the word for me, then I nevertheless have the strong need to say this word myself again in a correct manner without stutter. This redundant behavior does not result from a decision; it is rather like an automatic behavior.
Interestingly, even 3- and 4-year-old children repair their speech errors (Manfra, Tyler, and Winsler, 2016). They do so when talking with others, but also when talking to themselves, and in the latter condition, there were significantly higher numbers of immediate repairs than delayed repairs, whereas in talking with others the percentage of delayed repairs was higher. This again suggests that immediate error repair is an early-learned and automatized behavior, delayed repairs, by contrast, rather result from decisions.
By the way, Manfra and colleagues believe that immediate repairs indicate pre-production monitoring, but I think that’s wrong. Pre-production monitoring makes sense when we are required to avoid mistakes, e.g., in public speaking or in negotiations, but not in soliloquy. The question of whether a pre-articulatory monitoring takes place in spontaneous speech is of some relevance for my theory of stuttering. Read more here in the main text.
But let us go back from repair to interruption: In Section 1.1, I describe how a sensorimotor sequence is learned and how, in the course of learning and automatization, feedback-based control is replaced by feedforward control: In the initial period of learning, each step of a sequence is checked for correctness and completeness before the next step can start. If an error has happened, it can be repaired by repeating the last step only, such that not the entire sequence (in the worst case) must be repeated. In this way, a continuous self-monitoring develops ensuring that the sequence is correctly executed and that errors are immediately repaired.
Self-monitoring is conscious in the early period of learning with the effect that the sequence is not yet fluently executed, because short breaks are needed to evaluate every step. Later, after execution became automatic, self-monitoring runs alongside and is mostly unconscious. It ensures that the next step can start only if the preceding one is correct and complete. It intervenes only if an error was detected, and after a reaction time. Levelt’s Main Interruption Rule, in my view, describe just such an alongside-running, feedback-based self-monitoring in speaking: An automatic and unconscious monitor continuously checks the steps of a speech sequence and interrupts speech flow immediately if a step appears erroneous.
One may argue, that we are quite able to deliberately mispronounce words or make errors in grammar without any internal mechanism interrupting our speech. We can, for example, deliberately say “cracedole” instead of “crocodile” in a sentence without any problem. In a German children’s song (“Drei Chinesen mit nem Kontrabass”) , all vowels are replaced by /a/ in the second strophe, by /e/ in the third, by /i/ in the fourth strophe, and so on. In another German children’s song (“Auf der Mauer, auf der Lauer sitzt ne kleine Wanze”), one phoneme is shaved off the word Wanze (bug) in each following strophe (Wanz, Wan, Wa …); in the last strophe the whole word is omitted, by which the verse becomes syntactically incomplete. How can we do all this without problem when an automatic error-detection-and-interruption mechanism exists in the brain?
We obviously can deliberately override Levelt’s Main Interruption Rule. Volitional behavior overrides automatic control. However, the human ability to speak fluently and widely correct in phonology, grammar, and syntax without thinking of rules, even without explicit knowledge of rules, and to spontaneously produce new sentences possibly never said before – this amazing ability and its acquisition depend on automatic control and automatic error response. Taken together, I cannot provide evidence for my position that interruption of speech after the detection of phonological or grammatical errors is automatic, but (1) there are some findings suggesting that immediate repair is often automatic, and (2) such a mechanism is essential for motor sequence learning in general and particularly for language acquisition: It allows us to learn and to automatize the correct and fluent execution of sensorimotor sequences, among them in speaking.
to the top