![[Adobe Stock]](https://sp-ao.shortpixel.ai/client/to_webp,q_glossy,ret_img,w_750,h_375/https://www.psypost.org/wp-content/uploads/2026/11/conversation-750x375.jpg)
Enjoyable conversations rarely follow a straight line of constant agreement, nor do they jump randomly from one unrelated remark to the next. According to a new study published in Science Advances, more rewarding chats tend to show a stronger version of a mathematical pattern known as “pink noise,” in which speakers align on a topic, drift apart to bring in fresh ideas, and reconnect. This give-and-take was linked to greater enjoyment among both strangers and friends, a deeper sense of connection among strangers, and a higher chance that speed daters agreed to a second date.
One central channel in spoken conversation is semantics, which refers to the meanings of the words people choose. Past studies have often focused on linguistic entrainment, the tendency for conversation partners to gradually match each other’s word choices, speaking styles, vocal pitch, and body movements, linking higher overall similarity to romantic interest and group cohesion.
Treating similarity as a single average or a steady upward climb runs into what the authors describe as a logical limit: if two people stayed aligned on the exact same ideas the entire time, the exchange would likely grow repetitive and dull. Many traditional approaches also model dialogue as the sum of two separate speakers rather than an interactive system that evolves over time.
“When people talk about a particularly enjoyable conversation, they often use metaphors like ‘in sync’ or ‘on the same wavelength,'” first author Katie O’Nell, a doctoral candidate in the Department of Psychological and Brain Sciences at Dartmouth College, told PsyPost. “In the study of human interaction, these metaphors are often taken literally — we think about alignment and synchronization as signatures of good conversation.”
Don’t miss the next big psychology finding
Every morning, PsyPost sends a short briefing on the most interesting new peer-reviewed research in psychology and neuroscience.
“But in the context of words and their meanings, over-synchronization can actually be a bad thing — no one wants to hold a conversation with a parrot,” O’Nell explained. “So our goal for this paper was to develop a mathematical framework for understanding how moving both toward and away from each other in meaning space can help conversationalists feel like their interactions click.”
Working with colleagues including senior author Emily Finn, an assistant professor in the Department of Psychological and Brain Sciences at Dartmouth College who leads the Finn Lab, O’Nell drew on the physics concept of signal noise color to examine how alignment shifts across both brief exchanges and broader themes. Noise color describes how fluctuations in a sequence of measurements are distributed between rapid changes and slower trends.
At one extreme sits white noise, where fast and slow shifts carry equal weight and each moment varies unpredictably; in a chat, speakers might connect on one turn before abruptly pivoting to an unrelated comment on the next. At the opposite extreme sits red noise, also called Brownian drift, where slow trends dominate and speakers either repeat the same point for long stretches or drift apart without reconnecting.
Pink noise occupies a middle ground between random chaos and rigid repetition, following a pattern in which slower shifts carry proportionally more weight than rapid ones while smaller cycles of stability and change nest inside larger arcs. In a pink-noise conversation, speakers would stay aligned long enough to explore a topic, diverge when one person introduces a new angle, and realign as the other picks up that thread.
To measure this pattern, the researchers split conversation transcripts into uninterrupted speaking turns and passed each turn through a computer language program that translates sentence meaning into 768 numerical coordinates. Measuring the distance between adjacent turns on this mathematical map produced a timeline of how close or far apart the two speakers’ meanings were from turn to turn.
The scientists then analyzed how fluctuations along that timeline varied across short and long stretches of turns, yielding a noise score where 0.5 represents white noise and 1.0 represents pure pink noise. Because shorter chats can distort raw scores, the authors scrambled each conversation’s timeline 1,000 times to establish a random baseline and measured how far the real chat rose above it, posting their analysis code and publicly shareable data online.
Their first analysis examined an open dataset of 1,656 online video chats in English between strangers asked to talk for at least 25 minutes as if meeting at a social event, averaging about 320 turns. Out of all 1,656 chats, 94 percent scored above the 0.5 white-noise mark, averaging 0.62.
That average sits closer to white noise than to pure pink noise, but conversations reliably scored above their random baselines. Shuffling the real order of turns weakened the signal, indicating that the pattern depended on how remarks built on earlier ones.
Follow-up tests provided evidence that pink noise reflected joint coordination rather than an individual speaking habit. Isolating a single speaker’s turns shifted the signal toward white noise, and a model combining both partners’ individual scores explained just 1.4 percent of the variation in the pair’s joint score. Among 94 people who completed five or more chats, scores from chats involving the same person were no more alike than scores from chats involving different people, a comparison that was not statistically significant.
Scrambling where speaker handoffs occurred while keeping all words in order also weakened pink noise in joint transcripts, as did rearranging the sequence of steadier and more changeable stretches of conversation. Artificially splicing turns from two unrelated chats inflated pink noise scores because near-zero similarity flattened rapid fluctuations, prompting the authors to limit their coordination claims to real interacting pairs.
Next, the researchers compared pink noise scores against 205 survey questions covering enjoyment, connection, engagement and memory, demographics, self-rated personality traits, and partner trait ratings. Applying strict statistical adjustments for testing many questions without a pre-selected hypothesis, they identified seven outcomes positively tied to pink noise: overall enjoyment, wanting to keep talking, personal self-disclosure, perceived partner disclosure, wanting to stay rather than end early, thinking one’s partner wanted to keep talking, and chatting past 25 minutes.
Across summary scores for the six survey categories, pink noise showed a small but consistent positive relationship with enjoyment and connection, whereas links with memory and engagement, demographics, and personality traits were not statistically significant. Joint pink noise scores predicted enjoyment and connection better than individual speakers’ scores, and the links held after adjusting for basic features like chat length and word counts, as well as for simpler measures of how similar the partners’ remarks were on average and how steadily they converged.
Pink noise also explained variation in enjoyment beyond three mathematical models tracking momentum, path alignment, or shared information when tested one by one, and beyond two of the three for connection. When all three models were combined at once, pink noise was no longer a statistically significant predictor on its own, which the authors suggest is because those separate metrics divide up the fast and slow timescales that pink noise captures in one measure.
Because the stranger dataset involved casual small talk, the team tested whether the findings held in 97 video chats between Black and non-Black college friends, originally collected by one of the co-authors for a separate project. Pairs spoke for an average of 18 minutes about a formative personal experience shared by the Black friend and then completed surveys.
In these friend dialogues, pink noise showed a modest positive relationship with enjoyment, while links with connection and personal traits were not statistically significant. The enjoyment link held after adjusting for average turn length, word counts, and transcription quality, though adding total duration and turn count left the direct link not statistically significant, an outcome the authors suggest reflects friends talking well past the 10-minute minimum when enjoying themselves.
To test a concrete real-world decision rather than survey ratings, the scientists analyzed 31 ten-minute phone dates between strangers from a speed-dating podcast called Blink Date. The 19 pairs who matched by both answering “yes” or “maybe” to a second date showed stronger pink noise than the 12 pairs who did not match, even after accounting for call length and turn count, while unmatched pairs’ difference from random baselines was not statistically significant. The authors describe this finding as preliminary evidence that pink noise predicts partner choice in a natural setting.
In physical skills like walking rhythm, pink noise tends to strengthen with practice, leading the team to track 1,032 episodes across the first 50 installments of 25 free-flowing, amateur two-host podcasts, excluding married couples, scripted shows, and guest episodes. Pink noise grew stronger across those first 50 episodes even after controlling for episode length, turn and word counts, and transcription quality, a trend the authors note is consistent with co-hosts refining their conversational coordination over time.
Finally, the researchers prompted Claude 3.7 Sonnet, a large language model trained on vast amounts of human writing, to script 50 enjoyable and 50 unpleasant 100-turn coffee-shop chats between strangers without mentioning alignment or noise. Human raters scored the enjoyable scripts an average of 4.8 out of 5 compared with 2.4 out of 5 for the unpleasant ones. The enjoyable dialogues displayed stronger pink noise than the unpleasant dialogues, whose difference from random baselines was not statistically significant, which the authors suggest indicates that the model picked up this rhythm implicitly from human text.
O’Nell offered a practical suggestion for everyday conversations. “Don’t be afraid to slow your conversations down — asking follow-up questions and fully exploring a topic before you move on can yield unexpected fun and connection!” she said.
As with all research, there are some caveats to keep in mind. Every human conversation studied was observational rather than experimentally controlled, so the results cannot establish that pink noise causes people to enjoy a chat. Pairs who already click may instead settle into this rhythm naturally. The statistical links were also fairly small, indicating that word-meaning dynamics are just one piece of conversation quality alongside other verbal and nonverbal signals.
Several analyses, including the 31 speed dates and 97 friend chats, also used relatively small samples, and the podcast trend cannot distinguish adapting to one specific partner from building a generalizable social skill.
“This work is, critically, observational: we ran these analyses on existing recordings of conversations,” O’Nell emphasized. “But there are a ton of empirical questions this work opens up!”
“What kinds of interventions move conversations toward pink noise?” she asked. “Do these interventions work differently for people with social deficit disorders? How do different conversational goals (e.g., debates or tutoring sessions) shape the noise structure of a conversation?”
To address those questions, the authors note that future experiments will need to actively guide conversational pacing to test cause and effect, which they suggest could eventually inform conversational software, support tools for people who find social interaction difficult, such as individuals on the autism spectrum, or frameworks for navigating tense discussions.
The study, “Pink noise in dyads’ semantic similarity trajectories predicts conversation enjoyment,” was authored by Kathryn C. O’Nell, Kiara L. Sanchez, and Emily S. Finn.
