Abstract
Inner speech is a silent verbal experience and plays central roles in human consciousness and cognition. Despite impressive studies over the past decades, the neural mechanisms of inner speech remain largely unknown. In this study, we adopted an ecological paradigm called situationally simulated inner speech. Unlike mere imaging speech of words, situationally simulated inner speech involves the dynamic integration of contextual background, episodic and semantic memories, and external events into a coherent structure. We conducted dynamic activation and network analyses on fMRI data, where participants were instructed to engage in inner speech prompted by cue words across 10 different contextual backgrounds. Our seed-based co-activation pattern analyses revealed dynamic involvement of the language network, sensorimotor network, and default mode network in situationally simulated inner speech. Additionally, frame-wise dynamic conditional correlation analysis uncovered four temporal-reoccurring states with distinct functional connectivity patterns among these networks. We proposed a triple network model for deliberate inner speech, including language network for a truncated form of overt speech, sensorimotor network for perceptual simulation and monitoring, and default model network for integration and ‘sense-making’ processing.
Introduction
Most research on inner speech has historically focused on "covert speech"—the simple act of repeating a single word or phoneme in one's head. While these studies have identified core language regions, they fail to capture the richness of our actual internal lives. We do not merely repeat sounds; we plan conversations, rehearse speeches, and narrate our lives within complex situational contexts.
The authors argue that "situationally simulated inner speech" (SSIS) is a distinct cognitive process. It requires more than just language processing; it involves the dynamic integration of external cues with episodic and semantic memories to create a structured narrative. This study aims to identify the neural mechanisms that allow the brain to perform this high-level integration and "sense-making" during deliberate imagination.
Methods
The researchers employed a multi-stage approach to map the neural architecture of inner speech. First, a term-based meta-analysis using NeuroQuery identified core brain regions associated with 14 inner-speech-related terms. This was followed by Seed-based Co-activation Pattern (CAP) analysis using fMRI data to identify large-scale networks that fire in sync with specific "seed" regions, such as the ventral somatosensory areas.
To capture the temporal dynamics of these networks, the team used Dynamic Conditional Correlation (DCC) to track connectivity fluctuations frame-by-frame. Finally, K-means clustering grouped these patterns into four distinct "temporal states." The primary dataset involved 19 healthy participants who were given 10 different background stories (e.g., acting as an astronaut) and asked to mentally incorporate 61 abstract cue words into that specific narrative.
The Triple Network Model
The study proposes that deliberate inner speech relies on the orchestration of three distinct networks:
- Language Network (LAN): Responsible for the "truncated" form of speech, including syntax and word choice.
- Sensorimotor Network (SMN): Provides perceptual simulation and monitoring—the internal "feeling" of speaking and hearing oneself.
- Default Mode Network (DMN): Acts as the "sense-making" hub, integrating the narrative with memory and context.
Notably, the DMN—often considered a "task-negative" network—was found to be an active and essential participant in narrative integration during SSIS.
Key Results
The study pinpointed the bilateral ventral precentral/postcentral gyri (vPreCG/vPoCG) and the middle superior temporal gyrus (STG) as the primary engines of inner speech. CAP analysis confirmed that the LAN, SMN, and DMN consistently co-activate during situational imagination. DCC analysis identified four reoccurring states of functional connectivity, with strong and persistent connections maintained between the inferior frontal gyrus (IFG), STG, and vPreCG across all states. Additionally, substantial activation in the cerebellum (Crus I/II) suggests its role in motor planning and cognitive monitoring of internal speech.
Discussion and Limitations
One intriguing finding was the researchers' inability to decode which specific story a participant was imagining based on brain activity. This suggests that the Triple Network represents the machinery of imagination rather than the specific semantic content of the thoughts.
A major limitation remains the "black box" of imagination; because the speech is entirely internal, there is no behavioral output to verify the vividness or exact content of the narratives. Future work should include subjective ratings of "vividness" and explore whether concrete words, which rely more heavily on the perceptual system, alter these network dynamics. This research provides a foundational framework for understanding how the brain weaves abstract thoughts into the coherent internal monologues that define human cognition.
Summary updated