Why Don’t We Speak Faster? Speech Rate, Surprisal, and the Mind’s Bandwidth

Human language evolves towards efficiency so that the maximum amount of meaning can be squeezed into the minimum amount of time and effort. The latter is covered by Zipf’s principle of least effort. The time half of the efficiency equation has not received enough attention. An obvious way to improve efficiency would be to simply speak faster. Anyone who has ever listened to a horse race or a house auction understands that, physically, humans can speak incredibly quickly, and when they do, meaning can be inferred from rapid speech; otherwise, winners would never get called and houses would never sell. Ask yourself: why don’t you speak faster than you do? The obvious answer is that it is unusual. Speech is a social tool, presumably, speech communities unconsciously decide on an appropriate rate. Within languages, people are reasonably good at sticking to that rate. People, not real estate agents.

Across languages, however, there is quite a bit of variability in speech rate. Japanese and English sit at opposite ends of this spectrum. Japanese is a fast language, and English is a slow language. Japanese has a relatively small inventory of speech sounds and strict rules that govern how those sounds can go together. The upshot of this is that Japanese and English converge in how much information they express over time. Information is calculated using Claude Shannon’s surprisal, a simple log transformation of the percent chance of an event, in the case of Coupé et al. (2019), the chance of a given combination of speech sounds in a syllable. So English, with its larger and freer sound system, has many more potential syllable combinations and the percent chance of any specific combination is much lower than that of Japanese. They revealed that all 17 typologically diverse languages converge on around 39 bits of information per second.

This shared limitation suggests that it is our shared cognitive machinery that sets the ceiling for speech rate. The brain is not a passive receiver of speech; it is constantly predicting what comes next. It seems that the distance between what is expected and what is experienced is an important factor in speech perception. This is exactly what Claude Shannon intuited when he defined information as the inverse of predictability: a wholly expected signal carries zero information: meaning lives in the surprise. Seen this way, a syllable is not “information” because of its acoustic complexity, but because of how unlikely it is, given everything the listener’s brain had already anticipated from context, grammar, and prior sound. Japanese packs less surprisal into each syllable because its sound system is so constrained that outcomes are easier to guess; English packs more into each syllable because its freer inventory keeps the listener genuinely uncertain until the sound arrives. Speakers of both languages, then, aren’t really trading in phonemes per second, they’re trading in prediction errors per second, and 39 bits/second may simply be the rate at which a human brain can generate expectations, compare them to incoming sound, and update its model before the next signal arrives. Speech rate, in other words, isn’t tuned to the mouth. It’s tuned to the mind’s bandwidth for being surprised.

However, as noted earlier, humans can infer meaning from rapid speech and that contradicts the idea of a hard ceiling. So, what are the limiting factors? An explanation might exist in the research that looks at the effects of increased speech rate. Rapid or artificially time-compressed speech severely impact memory and processing effort. In a randomized trial, normal-speed speech produced a 27-percentage-point advantage in cued recall over time-compressed speech (Massoud & El Khoury-Malhame, 2026), and pupillometry research suggests that listeners expend more cognitive effort processing rapid speech (Müller et al., 2019). Perhaps the most interesting evidence comes from a study that utilised the Hungry Donkey task. In this task, you help a hungry donkey collect apples by choosing among four doors. Two doors are riskier, giving big rewards but bigger losses, resulting in net loss over time. Two doors give smaller rewards but are net positive. Nobody tells you which is which, so you learn from feedback, and good performance means shifting toward the safer doors over time. The study (Buelow et al., 2020) used speech rate as the experimental condition and found that listening to rapid speech led to making riskier choices. In the context of the horse race caller and the real estate agent, their rapid speech possibly takes a more sinister motivation: to get you to gamble or bid beyond your budget.

These findings raise a question for future research. Rapid speech raises the rate at which surprisal reaches the listener, so its effects on memory, effort, and risk-taking may partly reflect a surge in unpredictable input rather than speed itself. However, my own research shows that while high surprisal words are more difficult to process, they are also better recalled in memory recognition experiments (Kilpatrick & Bundgaard-Nielsen, 2025), which cuts against the memory deficits of rapid speech. This suggests that speech rate and surprisal may not act on memory and decision-making through the same mechanism, and that the risk-taking effect may be better explained by processing load or time pressure than by surprisal itself. Whether the two variables interact could be tested directly by manipulating rate and surprisal independently within a task like the Hungry Donkey. Such work would clarify whether the 39 bits/s convergence marks a genuine cognitive limit, or simply the point past which speech becomes too costly for listeners to bear.

Buelow, M. T., Hupp, J. M., Porter, B. L., & Coleman, C. E. (2020). The effect of prosody on decision making: Speech rate influences speed and quality of decisions. Current Psychology, 39(6), 2129–2139.

Coupé, C., Oh, Y. M., Dediu, D., & Pellegrino, F. (2019). Different languages, similar encoding efficiency: Comparable information rates across the human communicative niche. Science Advances, 5(9), eaaw2594. https://doi.org/10.1126/sciadv.aaw2594

Kilpatrick, A., & Bundgaard-Nielsen, R. L. (2025). Exploring the dynamics of Shannon’s information and iconicity in language processing and lexeme evolution. PLOS ONE, 20(4), e0321294.

Massoud, C., & El Khoury-Malhame, M. (2026). Normal speed versus time-compressed speech in long-term memory retention: A randomized, within subjects experimental design. BMC Psychology, 14(1), 684.

Müller, J. A., Wendt, D., Kollmeier, B., Debener, S., & Brand, T. (2019). Effect of speech rate on neural tracking of speech. Frontiers in Psychology, 10, 449.