Altered temporal scaling of semantic similarity fluctuations in psychosis

arXiv:2606.23043 · cond-mat.stat-mech, nlin.AO · Submitted 2026-06-22 · Read on arXiv

Listen

Transcript

Introduction to the show: ident: Condensed Matter. The day in condensed matter: materials, phases and the measurements under them.

Kai: We're looking at the paper: "Altered temporal scaling of semantic similarity fluctuations in psychosis".

Mira: Here's the gist of the paper. The paper investigates whether speech in psychosis exhibits alterations in temporal scaling properties within its semantic similarity time series, suggesting a possible deviation from criticality in speech in the disorder <ref:2606.23043#pg20>.

Kai: This research is significant because it extends the framework of scale-free dynamics from neural activity to language, suggesting that altered scaling at the language level may serve as a readout of underlying changes in the dynamics of neural organization <ref:2606.23043#pg18>.

Mira: Temporal Scaling Framework The notion of pink noise refers to ‘scale-invariant’ temporal dynamics, where fluctuations exhibit similar statistical structure across time scales <ref:2606.23043#pg3>.

Kai: Departures from a regime associated with such scale-free organization—toward uncorrelated ‘white’ noise or overly-persistent ‘brown’ noise—have been widely identified as markers of pathology across physiological and cognitive domains <ref:2606.23043#pg19>.

Paper summary: Kai: So the paper, "Altered temporal scaling of semantic similarity fluctuations in psychosis," page one introduces the idea that meaning unfolds over time on multiple timescales, which is something we see in nature with things like Zipf's law describing word frequencies <ref:2606.23043#pg1>.

Mira: They set up a theoretical framework using the criticality hypothesis, suggesting that complex systems function optimally near a critical point, a 'sweet spot' between order and disorder.

Lev: The paper suggests that patterns of self-similarity, characterized by scaling laws where one variable is a nonlinear function of another raised to a power, are ubiquitous in nature and language.

Kai: They use this idea to look for empirical signatures of this critical regime in speech anomalies documented in psychosis.

Mira: The study specifically applies detrended fluctuation analysis, or DFA, to quantify the temporal scaling within BERT-derived continuous cosine-similarity time series that track semantic trajectories.

Lev: These time series are captured from clinical transcripts involving patients and controls across three independent datasets: Zürich, Marburg, and Izmir.

Kai: Across all these datasets, the key finding is that individuals with psychosis showed consistently higher scaling coefficients compared to healthy controls.

Mira: This higher coefficient indicates more persistent long-range temporal correlations in the semantic similarity time series for the patient groups than we see in controls.

Lev: They also noted that there was an adjusted linear regression model predicting these DFA scaling exponents from factors like diagnostic group, number of words, age, and sex that was significant overall with an adjusted R squared of zero point three five two.

Kai: So it's not just a general difference; they found a relationship where the scaling exponent can be predicted by those specific demographic and clinical variables.

Mira: The study also mentioned that symptom correlations were scattered, covering both positive and negative symptom dimensions, which means this scaling alteration might not map cleanly to just one specific symptom.

Conclusion: Kai: Looking at the paper, "Altered temporal scaling of semantic similarity fluctuations in psychosis," page one the authors are pointing out that there is evidence of an alteration in the temporal distribution of semantic similarity values specifically in psychosis <ref:2606.23043#pg1>.

Mira: They suggest this points toward a possible deviation from criticality within speech during these disorders.

Lev: The paper suggests that this DFA scaling exponent might not index any particular symptom dimension directly, but instead it could be capturing a disruption in a broader systems-level principle of organization that then shows up as different symptoms downstream.

Kai: That’s an important distinction because it moves the focus from just correlating scaling with one symptom to understanding a deeper organizational change in the system.

Mira: The paper asks an important question for future research: whether this alteration in language dynamics could actually function as a readout of altered neural dynamics under that same analytical framework.

Lev: The observed scaling coefficient values, the alpha values, they don't approach the one over f scaling characteristic of canonical pink noise, which is alpha equals one <ref:2606.23043#pg1>.

Kai: So they are showing that in this context, psychosis isn't just random noise; it’s structured differently in its temporal organization.

Mira: This work connects the concept of scale-free dynamics from neural activity to language, suggesting that altered scaling at the language level could be a readout of underlying changes in the dynamics of neural organization.

Department of Translation & Language Sciences, Universitat Pompeu Fabra · Department of Neurosciences, Health Sciences Institute, Dokuz Eylul University Izmir · Department of Psychiatry, Dokuz Eylul University Izmir · Melbourne Neuropsychiatry Centre University of Melbourne and Melbourne Health · Department of Psychiatry and Psychotherapy, University of Marburg · Center for Mind Brain and Behavior CMBB Hans-Meerwein-Str 6 Marburg · Department of Adult Psychiatry and Psychotherapy, University of Zurich Zurich · & Neuroscience Center Zurich University of Zurich and ETH Zurich

cond-mat.stat-mech, nlin.AO

Submitted: 2026-06-22

Updated: 2026-10-29

License: http://creativecommons.org/licenses/by/4.0/

Importance score: 87/100

The gist: The paper investigates whether speech in psychosis exhibits alterations in temporal scaling properties within its semantic similarity time series, suggesting a possible deviation from criticality in

Key concepts

Temporal Scaling Framework: This framework looks at how fluctuations change across different time scales. Pink noise represents scale-invariant, random behavior where the statistical structure is the same regardless of how long you look at it. Deviations from this pattern are used to identify pathological states.
Detrended Fluctuation Analysis (DFA): DFA is a mathematical tool used to quantify temporal scaling in data, specifically looking at how fluctuations change over time. It extracts a scaling exponent (α) that measures the strength of long-range correlations in the time series.
Scale-Free Dynamics / Criticality: Scale-free dynamics describe systems where fluctuations exhibit power-law behavior across many scales, often associated with critical states. The study hypothesizes that speech in psychosis deviates from this critical state, suggesting a disruption in the system's organizational principles.

Terminology

Semantic Similarity — 16× in this paper · explained in 2 episodes
Temporal Correlations — 10× in this paper · explained in 2 episodes
Embeddings — 8× in this paper · explained in 2 episodes
Cosine Similarity — 6× in this paper · explained in 3 episodes
Scaling Law — 6× in this paper · explained in 2 episodes
Scaling Laws — 6× in this paper · explained in 5 episodes
Detrended Fluctuation Analysis — 5× in this paper
Detrended Fluctuation Analysis (DFA) — 4× in this paper · explained in 2 episodes
Coherence — 3× in this paper · explained in 2 episodes
Generalizability — 2× in this paper · explained in 6 episodes
Hallucination — 2× in this paper · explained in 4 episodes
Interpretability — 2× in this paper · explained in 15 episodes
Universality — 2× in this paper · explained in 2 episodes
Clustering — 1× in this paper · explained in 2 episodes
Convex Hull — 1× in this paper · explained in 2 episodes
Foundation Model — 1× in this paper · explained in 3 episodes
Foundation Models — 1× in this paper · explained in 8 episodes
IEEE International Symposium on Biomedical Imaging — 1× in this paper
Non-stationarity — 1× in this paper · explained in 2 episodes
Perplexity — 1× in this paper · explained in 3 episodes
Power Law Distribution — 1× in this paper · explained in 2 episodes
Scale-Invariance — 1× in this paper · explained in 2 episodes
Semantic Trajectories — 1× in this paper · explained in 2 episodes
Speaker Diarization — 1× in this paper · explained in 2 episodes
Vector Representations — 1× in this paper · explained in 2 episodes

Summary

The paper investigates whether speech in psychosis exhibits alterations in temporal scaling properties within its semantic similarity time series, suggesting a possible deviation from criticality in speech in the disorder <ref:2606.23043#pg20>. This research is significant because it extends the framework of scale-free dynamics from neural activity to language, suggesting that altered scaling at the language level may serve as a readout of underlying changes in the dynamics of neural organization <ref:2606.23043#pg18>.

Temporal Scaling Framework

The notion of pink noise refers to ‘scale-invariant’ temporal dynamics, where fluctuations exhibit similar statistical structure across time scales <ref:2606.23043#pg3>. Departures from a regime associated with such scale-free organization—toward uncorrelated ‘white’ noise or overly-persistent ‘brown’ noise—have been widely identified as markers of pathology across physiological and cognitive domains <ref:2606.23043#pg19>. The study applies detrended fluctuation analysis (DFA) to quantify temporal scaling in BERT-derived continuous cosine-similarity time series capturing trajectories through semantic space using clinical transcripts from patients and controls across three independent datasets <ref:2606.23043#pg3>. DFA scaling exponents (α) were extracted to characterize the strength of long-range temporal correlations <ref:2606.23043#pg19>. Across all datasets, patients exhibited significantly elevated scaling exponents relative to controls, indicating abnormally strong long-range temporal correlations with excessive persistence in semantic fluctuations <ref:2606.23043#pg19>.

Methodology and Data Sources

Three datasets – here referred to as Zürich (Switzerland), Marburg (Germany), and Izmir (Turkey) – were used in this analysis <ref:2606.23043#pg8>. The Zürich dataset is comprised of recordings from 85 SwissGerman speakers: 20 patients with schizophrenia (SSD), 29 individuals with low schizotypy scores (LS) and 36 individuals with high schizotypy scores (HS) <ref:2606.23043#pg9>. The Marburg dataset is comprised of recordings from 129 German speakers: 43 patients diagnosed with schizophrenia and schizoaffective disorder (SSD), 43 patients diagnosed with major depressive disorder (MDD) (according to SCID-I, DSM-IV-TR), and 43 healthy controls (HC) <ref:2606.23043#pg9>. The Izmir dataset is comprised of 148 native Turkish speakers: 57 patients with chronic schizophrenia (SZH), 17 patients with schizoaffective disorder (SZA) and 36 patients with first episode psychosis (FEP), as well as 38 healthy controls (HC) <ref:2606.23043#pg9>. Word-level semantic embeddings were obtained using Qwen3-Embedding-8b <ref:2606.23043#pg10>. Cosine similarity between adjacent word embeddings, was calculated as follows <ref:2606.23043#pg10>

Group Comparisons and Symptom Links

Individuals with psychosis showed a consistently higher scaling coefficients as compared to healthy controls across all datasets, indicating more persistent long-range temporal correlations in semantic similarity time series <ref:2606.23043#pg12>. In the Zürich dataset, an adjusted linear regression model predicting DFA scaling exponents from diagnostic group, number of words, age, and sex was significant overall (R2 = 0.390, adjusted R2 = 0.352; F(5,79) = 10.11, p. Symptom correlations were scattered in the present study, covering both positive and negative symptom dimensions <ref:2606.23043#pg15>.

Conclusion and Future Directions

The present study provides evidence of an alteration of the temporal distribution of semantic similarity values in psychosis, suggesting a possible deviation from criticality in speech in the disorder <ref:2606.23043#pg20>. The DFA scaling exponent may not index any particular symptom dimension but may instead capture a disruption in a broader systems-level principle of organization which then manifests in different symptoms as downstream consequences <ref:2606.23043#pg18>. An important question for future research is whether the alteration in language dynamics documented here could function as a readout of an altered neural dynamics under the same analytical framework <ref:2606.23043#pg18>. The observed scaling coefficient α values do not approach the 1/f scaling characteristic of canonical pink noise (α = 1) <ref:2606.23043#pg19>.

References

The references listed in the paper are as follows: Kello CT, Brown GDA, Ferrer-i-Cancho R, Holden JG, Linkenkaer-Hansen K, Rhodes T, Orden GCV. 2010 Scaling laws in cognitive sciences <ref:2606.23043#pg20>. O’Byrne J, Jerbi K. 2022 How critical is brain criticality? Trends Neurosci. 45, 820–837 <ref:2606.23043#pg20>. Sánchez-Puig F, Zapata O, Pineda OK, Iñiguez G, Gershenson C. 2023 Heterogeneity extends criticality <ref:2606.23043#pg20>. Altmann EG, Cristadoro G, Esposti MD. 2012 On the origin of long-range correlations in texts <ref:2606.23043#pg20>. Brothers T, Swaab TY, Traxler MJ. 2017 Goals and strategies influence lexical prediction during sentence comprehension <ref:2606.23043#pg20>. Peng CK, Buldyrev SV, Havlin S, Simons M, Stanley HE, Goldberger AL. 1994 Mosaic organization of DNA nucleotides <ref:2606.23043#pg20>. Hengen KB, Shew WL. 2025 Is criticality a unified setpoint of brain function? Neuron 113, 2582-2598.e2 <ref:2606.23043#pg20>. Van Orden GC, Kloos H, Wallot S. 2011 Living in the Pink: Intentionality, Wellbeing, and Complexity <ref:2606.23043#pg20>. Hausdorff JM, Mitchell SL, Firtion R, Peng CK, Cudkowicz ME, Wei JY, Goldberger AL. 1997 Altered fractal dynamics of gait: reduced stride-interval correlations with aging and Huntington’s disease <ref:2606.23043#pg20>. West BJ. 2006 Where Medicine Went Wrong: Rediscovering The Path To Complexity <ref:2606.23043#pg20>. Pan W, Ohashi K, Yamamoto Y, Kwak S. 2007 Power-law temporal autocorrelation of activity reflects severity of parkinsonism <ref:2606.23043#pg20>. Luque J, Luque B, Lacasa L. 2015 Scaling and universality in the human voice <ref:2606.23043#pg20>. Hinzen W, Palaniyappan L. 2024 The ‘L-factor’: Language as a transdiagnostic dimension in psychopathology <ref:2606.23043#pg20>. Figueroa-Barra A, Del Aguila D, Cerda M, Gaspar PA, Terissi LD, Durán M, Valderrama C. 2022 Automatic language analysis identifies and predicts schizophrenia in first-episode of psychosis <ref:2606.23043#pg20>. Parola A, Lin JM, Simonsen A, Bliksted V, Zhou Y, Wang H, Inoue L, Koelkebeck K, Fusaroli R. 2023 Speech disturbances in schizophrenia: Assessing cross-linguistic generalizability of NLP automated measures of coherence <ref:2606.23043#pg20>. Mikolov T, Chen K, Corrado G, Dean J. 2013 Efficient Estimation of Word Representations in Vector Space <ref:2606.23043#pg20>. Arslan B et al. 2024 Automated linguistic analysis in speech samples of Turkish-speaking patients with schizophrenia-spectrum disorders <ref:2606.23043#pg20>. Çabuk T, Sevim N, Mutlu E, Yağcıoğlu AEA, Koç A, Toulopoulou T. 2024 Natural language processing for defining linguistic features in schizophrenia: A sample from Turkish speakers <ref:2606.23043#pg20>. He R, Palominos C, Zhang H, Alonso-Sánchez MF, Palaniyappan L, Hinzen W. 2024 Navigating the semantic space: Unraveling the structure of meaning in psychosis using different computational language models <ref:2606.23043#pg20>. Voppel AE, de Boer JN, Brederoo SG, Schnack HG, Sommer I. 2021 Quantified language connectedness in schizophrenia-spectrum disorders <ref:2606.23043#pg20>. Zilber N, Ciuciu P, Abry P, van Wassenhove V. 2012 Modulation of scale-free properties of brain activity in MEG <ref:2606.23043#pg20>. Bleuler E. 1951 The basic symptoms of schizophrenia <ref:2606.23043#pg20>. Maatz A, Hoff P, Angst J. 2015 Eugen Bleuler’s schizophrenia—a modern perspective <ref:2606.23043#pg20>.

Improvements for AI systems

Improved semantic space representation for language models by incorporating temporal scaling analysis of cosine similarity time series to detect deviations from scale-free dynamics, which can be characterized by a scaling exponent alpha. This allows AI systems to potentially identify linguistic features indicative of pathological states in psychosis, as the study found patients with SSD had consistently higher DFA scaling exponents relative to healthy controls.
Development of a new diagnostic metric for language complexity based on temporal organization rather than just lexical or syntactic structure. This metric would quantify excessive persistence in semantic fluctuations by measuring the DFA scaling exponent, potentially serving as a readout for underlying neural dynamics, as suggested by the paper's conclusion that scaling laws in cosine similarities may serve as readouts of underlying dynamical pathology.
Creation of a system capable of distinguishing between different clinical subtypes based on their temporal linguistic signatures. By utilizing adjusted linear regression models described in the methods, the AI could predict or classify patient groups based on their specific DFA scaling exponents across different datasets (Zürich, Izmir, Marburg).

Abstract

Scale-free temporal organization, in which fluctuations show similar statistical structures across timescales, is a feature of many complex systems considered to support optimal functioning. Alterations of such scaling have been reported in several pathologies, but whether comparable alterations characterize the temporal organization of language remains largely unexplored. We address this question in the domain of psychosis, where language anomalies are pervasively documented. Specifically, we apply detrended fluctuation analysis (DFA) to quantify temporal scaling in Qwen3-derived continuous cosine similarity time series, capturing trajectories through semantic space computed from transcripts of picture descriptions across three independent datasets. DFA scaling exponents α were extracted to characterize the strength of long-range temporal correlations. Across all datasets, patients exhibited significantly elevated α relative to controls, indicating increased temporal persistence in semantic fluctuations. This analysis opens a window into the multi-timescale organization of meaning as it unfolds in discourse and reveals a signature of altered temporal scaling. This alteration parallels known departures from critical scaling in brain dynamics in psychosis, suggesting the exploration of possible links between these two.