🌀 nonlinear sciences

Deviance from a pink noise regime in the temporal organization of semantic relations in psychosis

This study demonstrates that individuals with psychosis exhibit abnormally strong long-range temporal correlations in their semantic speech patterns, deviating from the typical "pink noise" regime toward excessive persistence, which parallels known departures from criticality in brain function associated with the disorder.

Original authors: Paola Moreno Ancalmo, Emre Bora, Rieke Roxanne Mülfarth, Svenja Seuffert, Tilo Kircher, Frederike Stein, Philipp Homan, Wolfram Hinzen

Published 2026-06-23

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Paola Moreno Ancalmo, Emre Bora, Rieke Roxanne Mülfarth, Svenja Seuffert, Tilo Kircher, Frederike Stein, Philipp Homan, Wolfram Hinzen

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

The Big Idea: Listening to the "Rhythm" of Thought

Imagine you are listening to a river. In a healthy, natural river, the water flows with a specific kind of rhythm. It has big waves, small ripples, and everything in between. If you look at the water over a short time, it looks chaotic. But if you zoom out and look at it over a long time, you see a pattern: the small ripples look just like the big waves, just smaller. Scientists call this "pink noise." It's a "Goldilocks" zone—not too random, not too rigid, but perfectly balanced. This balance allows a system (like a brain or a river) to be flexible and adapt quickly.

This study suggests that when people speak, their thoughts usually flow with this same healthy, balanced rhythm. However, in people with psychosis (a condition affecting how the brain processes reality), this rhythm gets stuck.

The Experiment: Measuring the "Flow" of Words

The researchers wanted to see if the way people with psychosis speak is different from how healthy people speak. They didn't just look at what words were used; they looked at how the meaning of those words changed over time.

The Tool: They used a computer program (an AI) that understands the "meaning" of words. Imagine every word has a location on a giant map. If you say "dog" and then "cat," those two words are close together on the map. If you say "dog" and then "spaceship," they are far apart.
The Test: They took recordings of people talking and measured the distance between every word they said. This created a line graph showing how much the meaning of the conversation was jumping around or staying the same.
The Math: They used a special math test (called DFA) to see if that line graph had a healthy, balanced rhythm (pink noise) or if it was broken.

What They Found: The "Stuck" Record

The results showed a clear difference between the groups:

Healthy Controls: Their speech had a balanced rhythm. Their thoughts moved naturally, jumping between ideas but still keeping a flexible connection to the past. It was like a river flowing freely.
People with Psychosis: Their speech showed a different pattern. The researchers found that the meaning of their words was too persistent.

The Analogy:
Imagine a record player.

Healthy speech is like a record playing a song where the needle moves smoothly across the grooves, changing notes and rhythms naturally.
Psychosis speech is like a record where the needle gets stuck in one groove. It keeps repeating the same pattern over and over. The thoughts don't drift away naturally; they get "stuck" in a loop, holding onto the same ideas for too long.

In scientific terms, the patients had "excessive persistence." Their semantic fluctuations (changes in meaning) were too strong and too long-lasting, moving away from the healthy "pink noise" balance toward something more rigid.

Why Does This Matter?

The paper suggests that language is a window into how the brain is working.

The Brain Connection: We know that healthy brains operate on this "pink noise" balance. When the brain is sick (like in psychosis), that balance breaks, and the brain gets "stuck" in rigid patterns.
The Speech Connection: This study shows that the same "stuck" pattern we see in brain signals also shows up in the way people speak. The way they string words together reveals that their internal "rhythm" has shifted away from the healthy sweet spot.

Important Limitations (What the Paper Doesn't Say)

It's not a diagnosis tool yet: The paper does not say doctors should start using this math test to diagnose patients. It is a "proof of concept" showing that the math works.
It's not about specific symptoms: The researchers found that this "stuck rhythm" didn't perfectly match specific symptoms like hearing voices or being depressed. Instead, it seems to be a sign of a deeper, fundamental problem with how the whole system is organized.
It needs more data: The study had to remove some people because they didn't talk long enough. The math needs a lot of words to work, which is hard to get in a short doctor's visit.

Summary

Think of healthy thinking as a jazz musician improvising: it's structured but flexible, with a natural flow that connects the past to the future. In psychosis, the paper suggests the musician gets stuck playing the same few notes over and over. The study proves that by measuring the "distance" between words, we can mathematically detect this "stuck" rhythm, suggesting that the way people with psychosis speak is a direct reflection of a deeper change in how their brains organize time and meaning.

Technical Summary: Deviance from a Pink Noise Regime in the Temporal Organization of Semantic Relations in Psychosis

Problem Statement
Language production is a temporal process unfolding across multiple scales, from words to discourse. Complex systems theory suggests that healthy systems operate near a "critical point," exhibiting scale-free dynamics (pink noise) characterized by long-range temporal correlations. Deviations from this regime—toward uncorrelated white noise or overly persistent brown noise—are established markers of pathology in physiological domains (e.g., gait, heart rate). While criticality signatures have been observed in various levels of language (e.g., Zipf's law, mutual information decay), it remains unclear whether the temporal organization of semantic relations in speech exhibits similar scale-free dynamics and whether deviations from this regime characterize psychosis. Traditional Natural Language Processing (NLP) metrics often fail to capture these long-range temporal dependencies, focusing instead on local transitions or static global metrics. This study addresses the gap by investigating whether speech in psychosis exhibits altered temporal scaling properties in semantic similarity time series.

Methodology
The study analyzed clinical transcripts from three independent datasets: Zürich (Switzerland), Marburg (Germany), and Izmir (Turkey). These datasets included patients with schizophrenia spectrum disorders (SSD), schizotypy groups, major depressive disorder (MDD), and healthy controls (HC).

Data Preprocessing: Audio recordings were converted to text using automatic diarization and manual review. Only continuous speech segments of 100 or more words were retained to ensure stable Detrended Fluctuation Analysis (DFA) estimation.
Semantic Embedding: Word-level semantic embeddings were generated using the Qwen3-Embedding-8b model via the SentenceTransformers library.
Time Series Construction: A continuous time series was created by calculating the cosine similarity between adjacent word embeddings ( $K(X, Y) = \langle X, Y \rangle / (||X|| \cdot ||Y||)$ ).
Detrended Fluctuation Analysis (DFA): DFA was applied to these cosine similarity time series to quantify long-range temporal correlations. The process involved:
- Integrating the mean-subtracted time series.
- Dividing the integrated series into non-overlapping windows of varying lengths ( $n$ ).
- Removing local linear trends within each window.
- Calculating the root mean square fluctuation $F(n)$ for each window size.
- Determining the scaling exponent ( $\alpha$ ) as the slope of $\log F(n)$ regressed on $\log n$ .
- An $\alpha$ of 0.5 indicates white noise (uncorrelated); values approaching or exceeding 1 indicate brown noise (highly persistent).
Statistical Analysis: Participant-level $\alpha$ values were averaged across tasks. Linear regression models (OLS) were used to compare groups (patients vs. controls) within each dataset, adjusting for covariates (number of words, age, sex). Partial Spearman correlations were computed to assess relationships between $\alpha$ and symptom scores (PANSS, SAPS, SANS), controlling for diagnosis.

Key Contributions

Application of DFA to Semantic Space: The study introduces a method to quantify the temporal scaling of semantic relations in discourse by applying DFA to BERT-derived (specifically Qwen) cosine similarity time series.
Cross-Dataset Validation: The findings are validated across three independent datasets involving different languages (Swiss-German, German, Turkish) and diverse clinical populations.
Linking Language to Criticality: The work extends the criticality hypothesis from physiological signals and word frequency distributions (Zipf's law) to the temporal dynamics of semantic similarity in psychosis.

Results

Group Differences: Across all three datasets, patients with SSD exhibited significantly elevated DFA scaling exponents ( $\alpha$ $α$ ) compared to healthy controls.
- Zürich: SSD patients showed significantly higher $\alpha$ than low schizotypy controls ( $\beta = 0.0295, p = 0.001$ ). High schizotypy controls did not differ from low schizotypy controls.
- Izmir: Chronic schizophrenia (SZH) patients showed significantly higher $\alpha$ than healthy controls ( $\beta = 0.0303, p = 0.006$ ). First-episode psychosis (FEP) and schizoaffective (SZA) groups showed trends but did not reach statistical significance against controls.
- Marburg: SSD patients showed significantly higher $\alpha$ than healthy controls ( $\beta = 0.0241, p = 0.039$ ). The MDD group did not differ significantly from controls.
Symptom Correlations: The relationship between $\alpha$ and specific symptom dimensions was inconsistent. In the Zürich and Marburg datasets, correlations were scattered across positive and negative symptoms, with few surviving FDR correction. The Turkish dataset showed no significant associations.
Covariates: The number of words spoken was negatively associated with the scaling exponent across datasets. Age showed a positive association in the Zürich and Marburg datasets.

Significance and Claims
The authors claim that the elevated scaling exponents in psychosis indicate abnormally strong long-range correlations and excessive persistence in semantic fluctuations. Within a complex systems framework, this represents a departure from the near-critical regime (pink noise) characteristic of healthy systems toward a more persistent, "brown noise" regime.

The paper posits that this alteration in the temporal organization of language may serve as a readout of altered neural dynamics. The findings parallel existing literature showing disrupted power-law scaling and increased temporal persistence in the spontaneous neural activity of patients with SSD. The authors suggest that just as reduced neural density has been linked to deviations in word frequency distributions (Zipf's law), deviations in the temporal scaling of semantic similarity may reflect a deeper, primary disturbance in the self-organized criticality of the brain's language system.

The study concludes that while the scaling exponents did not reach the canonical $\alpha = 1$ of pink noise (likely due to noise in vector representations and short time series), the consistent shift toward higher $\alpha$ values across independent samples provides empirical evidence for a systems-level disruption in the temporal organization of meaning in psychosis. The authors caution that the scaling exponent may not index specific symptoms but rather a fundamental breakdown in the system's organizational principles.