Voices in the head: A population-level map of inner-speech modes and agency
PsyArXiv Preprints June 25, 2026 Peer reviewed DOI: osf:2s7a5_v1 via PsyArXiv
Summary
A large online survey of 5,289 respondents mapped how seven types of inner speech co-occur across three scenarios. The 21-dimensional rating space is continuous, not categorical, organized into a text family and a voice pair, with own voice and visual imagery as standalone axes. The principal opposition is between own voice and visual imagery, refining dual-coding theory: the phonological half of verbal code opposes imagery, while the orthographic half is nearly orthogonal. Own voice forms an axis independent of other-voice, and the two carry opposite-signed agency associations. Free-text descriptions from 1,903 respondents corroborate this voice-led organization, providing a reference map for idiographic and clinical studies.
Study at a glance
| Design | cross-sectional survey |
|---|---|
| Sample size | 5,289 |
| Population | online survey respondents |
| Key finding | The structure of inner speech is a continuous mixture organized into a text family and a voice pair, with own voice versus visual imagery as the principal opposition. |
Abstract
Inner-speech research documents striking individual diversity through descriptive phenomenology, self-report questionnaires, and extreme phenotypes, yet prior studies typically sample a few hundred respondents, and no design has mapped how the major modes co-occur or trade off at population scale. We addressed this gap with a video-based online survey (clean N = 5,289): respondents rated seven inner-speech types across three scenarios, alongside agency and sensory-modality self-reports; to our knowledge, it is the largest survey to map this co-occurrence. The study is exploratory: we chart the space's shape rather than test effects, and that shape is continuous and voice-led. The 21-dimensional rating space is a continuous mixture rather than discrete categories, organized into a text family and a voice pair with own voice and visual imagery as standalone axes; reproducible profiles emerge only in the lower-dimensional sensory-modality subspace. The principal opposition is own voice versus visual imagery, not text versus imagery, refining dual-coding theory: only the phonological half of the verbal code opposes imagery, while the orthographic half is near-orthogonal. Own voice also forms an axis independent of the other-voice pair, and the two carry opposite-signed agency associations that are small, as expected for a single item, but consistent in sign. Because the agency and voice items share self/other wording, we treat continuity with sub-clinical voice-hearing as a hypothesis. Unsupervised topic modeling of 1,903 free-text descriptions further corroborates this voice-led organization. These coordinates provide a reference map for idiographic and clinical studies of inner-voice experience.