In speech perception, timing and content are interdependent. For example, in distal rate effects, context speech rate determines the number of words, syllables, and phonemes heard in an unchanging target speech segment. Such results confront psycholinguistic theory with the chicken-and-egg problem of concurrently inferring speech timing and content, and the interrelated issues of narrowing the search space of speech interpretations without bias and optimizing the speed/accuracy tradeoff in online processing. We propose listeners address these issues by managing the timing of speech-related computations. Specifically, we claim: (1) Listeners model speech timing as part of a speaker model; (2) variable-length sequences of morphosyntactic units are the basic increments of speech inference; and (3) listeners adaptively schedule inferential updates and computationally intensive operations according to (4) fluctuations in uncertainty predicted by the speaker model. We illustrate these claims in a mechanistic model – vowel-onset-paced syllable inference – explaining multiple psycholinguistic results, including distal rate effects.