You are mid-sentence, explaining something you know well, and then it happens: “I was going to, um…” A tiny stall. Most people treat these moments as verbal hiccups, signs of nervousness or a word stuck on the tip of the tongue. But a convergence of neuroscience research now points to a different explanation. Those “ums” and “uhs” are not empty noise. They are byproducts of the brain’s command-and-control systems firing in real time.
And a team at Baycrest Centre for Geriatric Care in Toronto thinks that tracking those small vocal stumbles could eventually help doctors spot the earliest signs of cognitive decline, potentially before a patient or their family notices any memory trouble at all.
The brain signature behind every “um”
The most direct evidence comes from a 2020 study published in Scientific Reports that used electrocorticography, a technique that places electrode grids directly on the surface of the brain, to record cortical activity while participants described pictures aloud. When speakers produced an “uh” or “um,” the researchers detected a distinct shift in high-gamma neural activity, a frequency band associated with active cognitive processing. Crucially, that shift occurred in regions tied to speech planning and executive oversight, not just the motor areas that move the tongue and lips. The finding demonstrated that these brief fillers carry a distinct cortical fingerprint, one that looks nothing like the signal produced by ordinary content words.
Behavioral experiments tell a consistent story. A dual-task study published in 2001 in the Journal of Speech, Language, and Hearing Research asked participants to narrate stories while simultaneously performing a secondary attention task. Now more than two decades old, the experiment remains influential because its results have held up: as attentional load increased, the rate and type of filled pauses shifted in predictable, measurable ways. The results showed that fillers draw from the same limited pool of cognitive resources that powers attention and planning, and the authors quantified how added task demands altered the balance between silent hesitations and spoken “ums.”
Even earlier experimental work, published in Language and Speech in 1963, examined how speakers used pauses while solving addition problems out loud. That study found that filled pauses were not scattered randomly. Speakers placed them at moments when internal computation competed with the need to keep talking, and the fillers clustered at points of peak processing load as the sums grew harder. Decades later, the finding still holds up: hesitation sounds mark active management of competing demands, not a lapse in competence.
Taken together, these three studies, spanning direct brain recording, controlled behavioral experiments, and cognitive-load tasks, converge on the same conclusion. Filled pauses are outputs of executive processing. They mark moments when the brain is juggling competing demands and uses a tiny vocal placeholder to buy time while higher-level systems catch up.
From lab insight to clinical possibility
The newest application of this idea comes from researchers at Baycrest Centre for Geriatric Care. Their work, published in early 2025 in the Journal of Speech, Language, and Hearing Research (DOI: 10.1044/2025_JSLHR-24-00268) and distributed through ScienceDaily using Baycrest materials, links pauses and fillers to executive function in the specific context of early dementia detection. The team used automated tools to quantify how often older adults hesitated, restarted, or filled gaps while speaking, then related those patterns to standard neuropsychological tests of attention and planning.
The premise is straightforward: if filled pauses reliably track executive-function load in healthy speakers, then unusual changes in those patterns might signal the earliest stages of cognitive decline. Current frontline screening tools, such as the Montreal Cognitive Assessment (MoCA) or clock-drawing tests, require a clinical visit and a structured testing session. A speech-based approach could, in theory, work from a recorded conversation or even a phone call, lowering the barrier to screening considerably.
That potential is real, but so are the caveats.
Why the science is not yet settled
Several gaps stand between these findings and a tool a doctor could confidently use. The electrocorticography study recorded brain activity under tightly controlled conditions, with participants describing specific pictures in a lab. Whether the same high-gamma patterns hold during free-flowing conversation, where topic, emotion, and social pressure all shift unpredictably, has not been confirmed. Electrocorticography also requires surgical electrode placement, so it cannot serve as a screening method itself. Its value is foundational: it establishes that fillers have a measurable neural signature tied to executive systems.
The Baycrest team’s clinical claims, while promising, remain early-stage. The full methods and participant demographics of the early 2025 JSLHR paper are behind a paywall, making independent evaluation of sample size, demographic balance, and effect magnitude difficult for outside reviewers. Without those details, the strength of the link between filler patterns and clinical cognitive scores cannot be fully assessed, and it is unclear how well the model would generalize to populations beyond the one studied.
Context poses another challenge. Research published in Language and Speech examining disfluency rates in natural conversation has documented that the frequency of “ums” and “uhs” varies significantly by age, gender, conversational role, topic complexity, and the relationship between speakers. A person discussing an unfamiliar subject with a stranger will produce more fillers than someone chatting with a close friend, regardless of cognitive health. Any screening algorithm that ignores these social and situational drivers risks generating false alarms or, worse, missing genuine decline in people whose baseline disfluency rate is naturally low.
Individual language habits add another layer of complexity. Some speakers favor “uh” over “um.” Others rely on silent pauses or floor-holding phrases like “you know” or “like.” These stylistic choices can remain stable for decades and may be shaped by region, education, or bilingual status. Distinguishing a lifelong speech pattern from a subtle, clinically meaningful shift will require longitudinal data collected over months or years, not a single recording at one point in time.
No published study has yet cross-referenced the social-context disfluency data with the executive-function measures from the Baycrest work. Until researchers calibrate their models against these known sources of variation, the clinical utility of filler-based screening remains an open question. Robust validation will need diverse samples, repeated measurements, and careful adjustment for conversational setting.
What your “ums” actually tell you about your brain
The strongest evidence in this body of research comes from direct measurement. The electrocorticography study recorded neural activity at millisecond resolution and showed a clear cortical distinction between fillers and regular speech. The dual-task experiment manipulated cognitive load and observed corresponding changes in disfluency output. The arithmetic study demonstrated strategic placement of pauses under measurable cognitive pressure. All three were published in peer-reviewed journals with explicit methods linking hesitation sounds to executive control.
The Baycrest findings sit one step further from direct proof. The press materials describe a meaningful link between speech disfluencies and executive function in a dementia-detection context, but the full dataset and analytic pipeline have not been independently replicated. The claim is grounded in plausible neuroscience, and the underlying journal article carries a DOI, yet the clinical application is best understood as a hypothesis under active testing rather than a validated diagnostic tool.
For now, the safest conclusion is this: the next time you catch yourself saying “um” in the middle of a thought, your brain is not failing you. It is doing something sophisticated, holding the line while its planning systems work through a bottleneck. Whether clinicians can eventually harness that signal to catch cognitive decline earlier than current methods allow is a question that will take larger studies, longer follow-ups, and a much deeper understanding of how conversation shapes the sounds we make when we are thinking out loud.
More from Morning Overview
*This article was researched with the help of AI, with human editors creating the final content.
