Examining the effects of inner speech on uncertainty resolution in visual object recognition

Inner speech – the internal production of language akin to silent self-directed talk – influences diverse cognitive functions including planning, problem-solving and emotional regulation (Alderson-Day & Fernyhough, 2015; Fernyhough & Borghi, 2023). The Linguistic Active Inference Theory (Yao, 2025) proposes that inner speech reflects the brain’s use of linguistic predictions to revise prior expectations and resolve uncertainty. By deploying linguistic categorical priors, inner speech modulates the precision of perceptual hypotheses, enabling rapid commitment to categorical interpretations under ambiguous conditions. However, evidence for language’s influence on perception stems primarily from studies using external verbal cues rather than self-generated inner speech. The present study addresses this gap by testing whether self-generated inner speech reduces visual uncertainty through prior precision modulation that alters decision-making strategies during object recognition. Substantial evidence demonstrates that language modulates visual perception. Verbal labels facilitate object recognition from early infancy: 7- and 12-month-old infants show enhanced object recognition and memory when objects are named (LaTourrette et al., 2023), whilst 3-year-olds and adults perceive object size and shape differently depending on naming conventions (Landau & Leyton, 1999). In adults, hearing verbal labels before viewing ambiguous Mooney images aids recognition and improves visual discrimination, with preparatory changes in alpha-band oscillations reflecting perceptual readiness (Samaha et al., 2018). Language appears to influence perception through modulation of early sensory representations: prior knowledge about unfamiliar objects enhances visual brain responses at approximately 120 ms and facilitates recognition of particularly blurred images (Abdel Rahman & Sommer, 2008). This early P1 component in extrastriate visual cortex responds to prior conceptual knowledge when recognising both unfamiliar visual objects and their written names (Rabovsky et al., 2012). Importantly, hearing words enhances P1 responses to familiar animals and artefacts more effectively than nonverbal sounds, indicating that linguistic categorical cues influence sensory systems more powerfully than associated sensory experiences (Boutonnet & Lupyan, 2015). Beyond simple detection, language improves visual search efficiency. When participants searched for the numeral 2 amongst 5s, hearing “find the two” or “ignore fives” immediately beforehand improved response times and search efficiency (Lupyan, 2007). Visual search performance improves when novel stimuli are categorised as familiar and improves further when targets or distractors are verbally labelled, demonstrating that familiarity, category membership and linguistic labels enhance visual processing beyond perceptual similarity (Lupyan, 2007, 2008). These effects extend to conscious perception: hearing a word biases detection during binocular continuous flash suppression, rendering otherwise imperceptible stimuli detectable (Lupyan & Ward, 2013), whilst objects associated with verbally described functions show enhanced detection during the attentional blink paradigm, accompanied by increased P1 activity (Weller et al., 2019). Critically, these studies employed externally provided verbal cues. Whether self-generated inner speech produces comparable effects remains unexplored. Theoretical accounts propose that inner speech is recruited specifically under conditions of uncertainty or ambiguity, when individuals engage internal reflection to search for stimulus meaning and guide interpretation (Borghi & Fernyhough, 2022). LAIT specifies this mechanism further: under uncertainty, inner speech generates linguistically structured categorical priors that modulate the precision of perceptual predictions. High-precision priors enable rapid commitment to categorical hypotheses based on partial evidence, manifesting as earlier decisions with greater temporal variability (reflecting the confidence to guess before evidence accumulates). Conversely, articulatory suppression blocks linguistic prediction generation, forcing conservative accumulation of sensory evidence and producing slower but more uniform decision patterns (lower temporal variability). Consistent with this proposal, participants identify concrete versus abstract concepts more slowly under articulatory suppression than under control conditions or nonverbal interference (squeezing a softball), suggesting inner speech aids interpretation when stimulus meaning is uncertain (Fini et al., 2022). Neuroimaging reveals heightened activity in left-lateralised language networks during open-ended abstract thinking – regions consistently linked to inner speech (Berkovich-Ohana et al., 2020). The present study explicitly tests whether self-generated inner speech modulates decision patterns during visual object recognition by examining both central tendency (mean decision time) and distributional properties (decision variability). Participants will identify everyday objects presented in videos that progressively unblur over 10 seconds via Gaussian blur reduction. This dynamic unblurring creates a continuous gradient of uncertainty levels, enabling identification of recognition thresholds whilst capturing decision timing – advantages over paradigms applying fixed uncertainty. Participants will complete this identification task under two counterbalanced conditions: articulatory suppression (continuously verbalising the word ‘aluminium’ to occupy the articulatory motor system) versus foot tapping (a nonverbal motor task controlling for secondary task cognitive load whilst preserving speech processes). Comparing conditions reveals whether inner speech supports object identification under uncertainty. We predict that articulatory suppression will reduce decision variability (sigma) by eliminating early guessing enabled by linguistic priors, forcing participants to wait for conclusive sensory evidence. The effect should prove strongest in individuals who habitually rely on verbal thinking, as linguistic prediction opportunities are most relevant for this population. Established individual differences in verbal thinking styles will be measured using the Internal Representations Questionnaire (Roebuck & Lupyan, 2020). We hypothesise that only individuals who rely predominantly on language to generate predictions will show reduced decision variability under articulatory suppression, whilst less verbal thinkers will show minimal condition differences (cf. Nedergaard & Lupyan, 2024). This individual-differences approach recognises heterogeneity in inner speech reliance and tests the specificity of inner speech’s role in resolving visual uncertainty. The study’s purpose is twofold: (1) to test explicitly whether self-generated inner speech modulates decision patterns through linguistic categorical predictions that enable rapid commitment to hypotheses; (2) to determine whether effects are moderated by individual thinking styles. Expected outcomes include reduced decision variability (lower sigma) and slower decision speeds (higher mu) during articulatory suppression compared to foot tapping in high verbal thinkers, reflecting a reduction of premature guessing strategies. Time-course analysis will reveal whether distributional differences concentrate during periods of high uncertainty (early trial time when objects are heavily blurred), confirming that inner speech specifically facilitates rapid categorical commitment when sensory evidence is ambiguous. Such findings would provide direct evidence that inner speech resolves perceptual uncertainty through internally generated linguistic predictions that modulate decision strategies, extending laboratory findings of external verbal cues in fixed uncertainty conditions to naturalistic self-generated language processes in dynamic uncertainty conditions.