6 Coordination, Prediction, and Repair

How comprehension and production interact

Comprehension turns incoming speech or text into an interpretation, while production turns an intended message into words and, in speech, articulatory movements. These processes have distinct input and output demands, but they can draw on overlapping linguistic knowledge and meaning representations.

In conversation, participants rapidly switch between listening and speaking. What a person understands can shape the response they plan, and the response can give the other person information about whether a message was understood.

Prediction during comprehension

Comprehension is incremental: listeners and readers interpret language as it arrives rather than waiting for a sentence to end. They use context, linguistic knowledge, and experience to form probabilistic predictions about what may come next. Predictions can concern meaning, grammatical structure, or a likely word form, but they are expectations rather than certainties.

For example, when someone asks, “Would you like tea or…,” a listener may use the context to anticipate coffee. That expectation can help the listener process the next word quickly, while leaving them able to respond if the speaker says juice instead. The listener can revise the interpretation using the incoming sound and context.

Prediction is useful when input is rapid, ambiguous, or noisy. At a dinner table, “Please pass the…” makes salt or pepper more likely than passport. An unexpected word prompts the listener to update the interpretation.

Production and prediction

One proposal, often called prediction by production, is that comprehenders recruit some processes used to produce language to anticipate an upcoming utterance. This may help explain how people prepare for likely words and coordinate smoothly in dialogue.

The evidence does not establish that all comprehension prediction depends on production. Prediction can vary with context and task, and how much production-specific processing it involves remains debated.

Monitoring speech

Speakers monitor their own utterances for problems, including an incorrect word, a grammatical mistake, or a message that does not fit their intention. Monitoring can happen before an utterance is spoken, by checking an internal representation, and after speech begins, by listening to one’s own voice. How these routes contribute and interact with general systems for detecting and correcting actions is debated.

For example, a speaker who starts to say, “Turn left at the—” may notice that the intended direction is right, stop, and correct the utterance. Monitoring can affect the words produced as well as the timing and fluency of speech. Not every error is noticed or repaired: detection and correction take time, and speakers may continue despite a mismatch.

Repair in conversation

Repair addresses trouble in speaking, hearing, or understanding. A speaker may initiate and complete a correction: “Meet me on Tuesday—sorry, Wednesday.” A listener can also signal a problem, prompting the speaker to clarify:

“Meet me at the usual place.”

“Which place?”

“The café by the station.”

Conversation research finds that repair is organized in ways that often give the original speaker an opportunity to correct their own trouble. It is more than fixing a mistaken word: it is a coordination process that helps participants restore a sufficiently shared understanding. A listener’s question or puzzled response can alert the speaker to a mismatch, and the clarification can update both participants’ interpretations.

Coordination without perfect communication

Incremental interpretation, prediction, response planning, and feedback link comprehension and production during conversation. Predicting likely input can make comprehension more efficient, although the extent to which prediction relies on production mechanisms remains unsettled.

Speakers monitor their speech, and both speakers and listeners can initiate repair. Together, prediction, monitoring, and repair support coordinated communication without requiring either participant to be error-free.