Trajectories through semantic spaces in schizophrenia and the relationship to ripple bursts

Matthew M. Nour, Daniel McNamee, Yunzhe Liu, Raymond J. DolanProceedings of the National Academy of Sciences 2023 · DOI

TL;DR

This study uses natural language processing and computational modeling to show that individuals with schizophrenia exhibit less semantically guided word selection during verbal fluency tasks. This behavioral deficit is linked to reduced hippocampal ripple activity, suggesting a disruption in the neural mechanisms underlying cognitive mapping and associative memory in the disorder.

What Problem Does This Paper Address?

Schizophrenia is characterized by profound disruptions in thought and cognition, often manifesting as "formal thought disorder" and conceptual disorganization. These symptoms are hypothesized to stem from disorganization in how individuals represent and process relationships between concepts – essentially, a breakdown in their internal "cognitive maps." While the hippocampus is known to be crucial for forming and utilizing these cognitive maps, and abnormalities in hippocampal activity (specifically, "replay" and "ripples") are implicated in schizophrenia, the direct link between these neural signatures and the behavioral manifestations of conceptual disorganization has remained unclear. This paper aims to bridge this gap by computationally assessing semantic processing during a verbal fluency task and relating it to neural measures of hippocampal function.

Methods

The researchers employed a multi-faceted approach combining behavioral tasks, computational modeling, and neuroimaging:

  • Verbal Fluency Tasks: Participants completed category-based ("animals") and letter-based ("p" words) verbal fluency tasks. This allowed for the assessment of word selection strategies under different cognitive demands (semantic vs. orthographic).
  • Natural Language Processing (NLP) for Semantic Analysis: A pre-trained word embedding model (fastText) was used to quantify the semantic similarity (cosine similarity) between words generated by participants. This allowed for the creation of "semantic spaces" where words are positioned based on their meaning.
  • Computational Modeling of Word Selection: A computational model was developed to simulate the process of word selection during verbal fluency. This model treated word selection as a "local search" within the semantic space, with parameters capturing the influence of semantic and orthographic associations. The model allowed for the quantification of "goal-induced semantic modulation" (Δω), representing how effectively participants adjusted their semantic search based on task demands.
  • Magnetoencephalography (MEG): Participants underwent MEG scans during a resting-state period following a separate sequence-learning task. This allowed for the measurement of neural replay and, crucially, "ripple power" – high-frequency oscillations in the hippocampus known to be involved in memory consolidation and cognitive mapping.

Datasets

The study utilized a dataset of 52 participants: 26 individuals diagnosed with schizophrenia (26 PScz, 13 of whom were unmedicated) and 26 age-, gender-, and IQ-matched healthy controls. The MEG data was collected from a subset of these participants during a previously conducted sequence-learning task. No new benchmark datasets were introduced.

Key Insights and Ideas

  • Semantic "Cognitive Maps" are Disrupted in Schizophrenia: The core idea is that our ability to navigate and retrieve information relies on internal, structured representations of the world (cognitive maps). In schizophrenia, these maps appear to be disorganized, leading to difficulties in coherent thought and speech.
  • Verbal Fluency as a Window into Semantic Navigation: The researchers propose that verbal fluency tasks can be viewed as a form of "mental navigation" through a semantic space. The way individuals select words reveals how constrained their search is by semantic relationships.
  • Hippocampal Ripples as a Neural Correlate of Semantic Navigation: The study hypothesizes that hippocampal replay and ripples, known to be involved in memory consolidation and cognitive map stabilization, are also critical for guiding semantic search.
  • Task-Specific Semantic Modulation: The ability to flexibly shift between semantic and orthographic processing based on task goals is a key aspect of healthy cognition. This "goal-induced semantic modulation" is shown to be impaired in schizophrenia.

Key Results

  • Reduced Semantic Guidance in Schizophrenia: Participants with schizophrenia showed significantly lower semantic similarity between consecutively generated words in the category fluency task compared to controls. This effect was not observed in the letter fluency task, highlighting the task-specific nature of the deficit.
  • Impaired Goal-Induced Semantic Modulation: The computational model revealed that individuals with schizophrenia were less able to adjust their reliance on semantic information based on task demands (i.e., lower Δω). This impairment was linked to negative psychotic symptoms, particularly those related to abstract thinking and spontaneity.
  • Correlation with Hippocampal Ripple Power: Crucially, the degree of goal-induced semantic modulation (Δω) positively correlated with the strength of hippocampal ripple power measured via MEG. This relationship held for both controls and individuals with schizophrenia, suggesting that ripple activity is a shared neural substrate for this cognitive process.
  • No Correlation with Neural Replay: Interestingly, Δω did not correlate with measures of neural replay, suggesting that ripple activity, rather than replay itself, might be more directly linked to the flexible, context-dependent use of semantic information.

Limitations and Future Work

The authors acknowledge several limitations:

  • Pre-trained Word Embeddings: The study assumes that the pre-trained word embedding model adequately represents the semantic knowledge of individuals with schizophrenia, which may not always be the case.
  • Model Simplifications: The computational model does not account for all potential cognitive processes involved in verbal fluency, such as working memory or self-monitoring.
  • Correlational Nature: The findings are correlational, meaning they demonstrate associations rather than direct causation.
  • Time of Measurement: Behavioral and neural measures were acquired at different times, which could introduce variability.
  • Sample Size: The study involved a relatively small sample size.

Future work could involve:

  • Concurrent Measurement: Combining neuroimaging and verbal fluency tasks simultaneously to capture real-time neural dynamics during semantic search.
  • Investigating Other NLP Models: Exploring different NLP models to assess the robustness of the findings.
  • Intervention Studies: Developing and testing interventions aimed at improving semantic processing and hippocampal function in schizophrenia.
  • Exploring Other Cognitive Domains: Extending this approach to other cognitive tasks that rely on structured representations, such as reasoning or planning.

Summary updated

Abstract

Human cognition is underpinned by structured internal representations that encode relationships between entities in the world (cognitive maps). Clinical features of schizophrenia-from thought disorder to delusions-are proposed to reflect disorganization in such conceptual representations. Schizophrenia is also linked to abnormalities in neural processes that support cognitive map representations, including hippocampal replay and high-frequency ripple oscillations. Here, we report a computational assay of semantically guided conceptual sampling and exploit this to test a hypothesis that people with schizophrenia (PScz) exhibit abnormalities in semantically guided cognition that relate to hippocampal replay and ripples. Fifty-two participants [26 PScz (13 unmedicated) and 26 age-, gender-, and intelligence quotient (IQ)-matched nonclinical controls] completed a category- and letter-verbal fluency task, followed by a magnetoencephalography (MEG) scan involving a separate sequence-learning task. We used a pretrained word embedding model of semantic similarity, coupled to a computational model of word selection, to quantify the degree to which each participant's verbal behavior was guided by semantic similarity. Using MEG, we indexed neural replay and ripple power in a post-task rest session. Across all participants, word selection was strongly influenced by semantic similarity. The strength of this influence showed sensitivity to task demands (category > letter fluency) and predicted performance. In line with our hypothesis, the influence of semantic similarity on behavior was reduced in schizophrenia relative to controls, predicted negative psychotic symptoms, and correlated with an MEG signature of hippocampal ripple power (but not replay). The findings bridge a gap between phenomenological and neurocomputational accounts of schizophrenia.
BibTeX
@inproceedings{nour2023trajectories,
  author = {Matthew M. Nour and Daniel McNamee and Yunzhe Liu and Raymond J. Dolan},
  title = {{Trajectories through semantic spaces in schizophrenia and the relationship to ripple bursts}},
  booktitle = {Proceedings of the National Academy of Sciences},
  year = {2023},
  doi = {10.1073/pnas.2305290120},
}
Figure