Speech Monitoring During Self-Comprehension
Abstract
Spontaneous speech is full of false starts, rephrasing, and speech errors. Speakers’ ability to detect errors and other problems in their own speech is commonly explained by monitoring mechanisms that rely on pre-articulatory representations, production states, or derivative predictions. We develop an alternative, cue-integrative account in which verbal self-monitoring is a natural consequence of what we have previously termed self-comprehension: speakers’ integration of contextual and sensory information to understand their own unfolding speech without requiring that meaning or intention be fully specified before articulation. Two sources of anticipatory information are central to the account, neither requiring explicit predictive or forward-model representations. First, we apply to self-comprehension the recently proposed idea of multi-scale pattern completion during speech comprehension, which allows partial input to activate continuations at multiple linguistic levels, generating contextually constrained anticipation. Second, anticipatory coarticulation can provide auditory and somatosensory information about upcoming speech before the relevant segment is produced, challenging the assumption that sensory feedback is too slow for online self-comprehension. We propose three routes to error detection: divergence between pattern-completion-based and coarticulatory anticipations; divergence between pattern-completion-based anticipation and post-articulatory sensory feedback; and post-articulatory self-comprehension in the absence of sufficiently informative anticipation. These routes provide a framework for explaining error detection across the observed range of error-interruption latencies, including very rapid interruptions. This framework also provides alternative interpretations of error-detection findings under noise-masked auditory feedback and of cortical attenuation during speaking without requiring a separate pre-articulatory monitoring channel, explicit predictions, or prediction-error computations.
Related
- Speech Tracking in fMRI Reveals a Distinct Response to Sentence Boundaries During Comprehension
- Older Adults Show Preserved Neural Signatures of Audiovisual Integration but Altered Links to Speech Comprehension
- The Listening Effort Profile of Eye Movements: Easy, Difficult, and Impossible Speech Comprehension
- Development and Evaluation of a Realistic Audio-Visual Multi-Talker Speech Comprehension Test
- Electrophysiological Indices of Hierarchical Speech Processing Differentially Reflect the Comprehension of Speech in Noise
BibTeX
@misc{lind2026speech,
title = {{Speech Monitoring During Self-Comprehension}},
author = {Andreas Lind and Petter Johansson and Lars Hall},
howpublished = {PsyArXiv},
year = {2026},
month = oct,
doi = {10.31234/osf.io/2edgj_v1},
url = {https://osf.io/preprints/psyarxiv/2edgj_v1/},
note = {CC BY 4.0},
}
RIS
TY - UNPB
TI - Speech Monitoring During Self-Comprehension
AU - Andreas Lind
AU - Petter Johansson
AU - Lars Hall
PY - 2026
DA - 2026/10/23
PB - PsyArXiv
DO - 10.31234/osf.io/2edgj_v1
UR - https://osf.io/preprints/psyarxiv/2edgj_v1/
AB - Spontaneous speech is full of false starts, rephrasing, and speech errors. Speakers’ ability to detect errors and other problems in their own speech is commonly explained by monitoring mechanisms that rely on pre-articulatory representations, production states, or derivative predictions. We develop an alternative, cue-integrative account in which verbal self-monitoring is a natural consequence of what we have previously termed self-comprehension: speakers’ integration of contextual and sensory information to understand their own unfolding speech without requiring that meaning or intention be fully specified before articulation. Two sources of anticipatory information are central to the account, neither requiring explicit predictive or forward-model representations. First, we apply to self-comprehension the recently proposed idea of multi-scale pattern completion during speech comprehension, which allows partial input to activate continuations at multiple linguistic levels, generating contextually constrained anticipation. Second, anticipatory coarticulation can provide auditory and somatosensory information about upcoming speech before the relevant segment is produced, challenging the assumption that sensory feedback is too slow for online self-comprehension. We propose three routes to error detection: divergence between pattern-completion-based and coarticulatory anticipations; divergence between pattern-completion-based anticipation and post-articulatory sensory feedback; and post-articulatory self-comprehension in the absence of sufficiently informative anticipation. These routes provide a framework for explaining error detection across the observed range of error-interruption latencies, including very rapid interruptions. This framework also provides alternative interpretations of error-detection findings under noise-masked auditory feedback and of cortical attenuation during speaking without requiring a separate pre-articulatory monitoring channel, explicit predictions, or prediction-error computations.
ER -