# Dartmouth study links semantic pink noise to conversation enjoyment

- Published: 2026-11-07 05:28 UTC
- Section: Science › Research
- Tags: semantic similarity, pink noise conversation, conversation dynamics, Kathryn ONell, Emily Finn, dialogue analysis
- URL: https://realtech.media/news/rt-bf45d35623a95eeef5af6fcc_dartmouth_study_links/

**Researchers found that specific fractal-like fluctuations in semantic similarity make two-person conversations significantly more enjoyable.**

![Dartmouth study links semantic pink noise to conversation enjoyment](https://realtech.media/media/2026/11/b2e59adbd6022c1f.jpg)
*Vyacheslav Argenberg, Wikimedia Commons, CC BY 4.0*

Dartmouth College researchers have discovered that interpersonal conversations are perceived as more enjoyable when they exhibit mathematical patterns resembling pink noise, according to a study published on November 6, 2026, in Science Advances and reported via EurekAlert [1]. The investigation, led by Kathryn O’Nell and Emily Finn, analyzed how semantic alignments fluctuate between speakers across various time scales.

In this context, pink noise describes shifts in semantic similarity, reflecting patterns in vocabulary choice and speech themes over time [1]. The dynamic involves partners maintaining alignment for an interval, diverging as one speaker introduces novel concepts, and subsequently realigning as the other participant adopts and expands upon the topic [1]. Traditional speech theories often treat dialogue as a strictly linear exchange, failing to account for these multi-scale structural variations.

## Dataset analysis

The team examined two preexisting datasets of two-person dialogues, measuring the presence of 1/f noise alongside subjective ratings of how much the participants enjoyed the interaction [1]. The analysis confirmed that higher levels of semantic pink noise directly correlated with higher reported enjoyment [1]. Furthermore, researchers identified the same structural patterns across alternative conversational formats, including speed-dating encounters and podcasts hosted by two people [1].

## Text generation tests

To observe the phenomenon in simulated dialogues, the researchers prompted an algorithmic conversational script generator to produce both enjoyable and unenjoyable interactions [1]. The system systematically incorporated more pink noise into the scripts requested to be enjoyable [1]. The study authors suggested that this pattern likely reflects structures absorbed from human-authored training data, demonstrating that dynamic alignment and divergence are fundamental to engaging exchanges [1].

## Sources

1. [EurekAlert](https://www.eurekalert.org/news-releases/1150090)

---

Source: RealTech, realtech.media — https://realtech.media/news/rt-bf45d35623a95eeef5af6fcc_dartmouth_study_links/