A researcher asks someone whether they think in words. The answer is “constantly.” During a later day of random prompts, however, the person reports a worded experience at only some of the sampled moments. Neither answer must be dishonest. One asks for a general impression of one’s mind; the other asks about a particular instant. A third task, silently repeating a word on command, asks for something different again. To measure inner speech, researchers first have to decide which of these experiences and performances their claim concerns.
Measuring inner speech means estimating the occurrence, form, content or functional contribution of language-like experience without audible speech. Inner speech may be deliberately produced, may arise spontaneously, may feel like speaking, may resemble hearing a voice, or may be condensed rather than fully sentence-like. This Methods & Measures entry compares questionnaire, momentary report, behavioral interference and neuroimaging approaches. Each provides evidence under particular conditions; none is a transparent window into every private thought.
Define the target before choosing an instrument
“Do you have an inner voice?” compresses several questions. A participant might mean that they sometimes silently rehearse sentences, hear a familiar voice in imagination, narrate decisions, plan in words, or remember conversations. A scale asking about internal dialogue may not measure the frequency of all silent verbal thought. A laboratory task that requires covert repetition may elicit an intentional action rather than capture a naturally occurring one. A study should specify whether its target is occurrence, phenomenological form, function during a task, or a proposed neural process.
Alderson-Day and Fernyhough’s broad review describes inner speech as a heterogeneous phenomenon with developmental, cognitive, phenomenological and neurobiological aspects. That breadth is a reason to narrow a measurement question, not a license to use the same score for every aspect. One person may often use brief word fragments, another may engage in elaborate internal dialogue, and a third may report few experiences that meet a researcher’s definition. They may all use language competently. The operational definition should leave room for that variation.
Inner speech also borders other categories. Visualizing written words is not necessarily the same as experiencing oneself silently speaking. Recalling the sound of someone else’s speech may be inner hearing. Thinking about a concept without apparent words may still be thought. Descriptive experience research has emphasized that everyday words such as “thinking” and “voice” are not applied consistently across people. A questionnaire can give useful standardized categories, but its categories should not be mistaken for the phenomenon itself.
A measurement claim should identify its unit. A momentary report estimates what a participant describes at sampled instants. A questionnaire may estimate a person’s general tendency to endorse certain experiences. A task contrast estimates how a manipulation changes performance. A brain contrast estimates signal differences between conditions. Moving from one unit to another requires an argument and corroborating evidence. A high questionnaire score is not a count of actual daily episodes; activation during covert repetition is not proof that the participant always thinks verbally.
Questionnaires measure remembered tendencies and descriptions
The Varieties of Inner Speech Questionnaire and its revision ask about qualities such as dialogic exchanges, condensed wording, other people in inner speech, evaluative or critical content, and positive or regulatory content. Alderson-Day and colleagues developed the revised instrument through factor analysis in two samples. Those results support a structured way to ask about varieties of inner speech. They do not establish a census of every private episode, nor do they prove that one factor has the same meaning in every language and population.
Questionnaires are efficient. Many participants can answer the same items, allowing investigators to examine associations with other traits or compare groups. The cost is reliance on memory, interpretation and self-description. A respondent must average across situations, notice what counts as inner speech, and map their experience onto the offered phrases. A person whose speech is fragmentary might reject a question that assumes complete sentences. Another might count deliberate rehearsal but overlook spontaneous verbal fragments. The resulting score is a report under an instrument’s wording.
Frequency language deserves particular care. “I talk to myself in my head” could be endorsed because the experience is memorable or important, even if it occupies few sampled moments. Conversely, a common but unremarkable experience may be underreported retrospectively. The response scale’s anchors, examples and time frame can change the task. A validation study should ask whether people understand key terms similarly, whether factors replicate in the target population, and whether scores show expected relationships without assuming they should perfectly match momentary sampling.
The revised questionnaire includes content with possible emotional significance. Endorsing critical inner speech is not a diagnosis. The association between an item and a symptom scale does not show which caused which. Nor does a low score on a positive subscale show an inability to regulate oneself. Clinical and wellbeing claims require independent, appropriate measures and a design that can test direction and context. A questionnaire can start such work, but should not finish it alone.
Sample moments rather than a general impression
Experience sampling asks participants about what is happening near a prompt, often several times across ordinary days. Alderson-Day and Fernyhough compared a generalized inner-speech questionnaire with reports collected through random smartphone alerts. Agreement varied by the kind of inner speech, and momentary endorsements were generally lower than generalized endorsements. That finding does not identify which method is simply “correct.” It shows that the methods ask different questions and that broad retrospective reports cannot automatically stand in for moment-level incidence.
Prompt design matters. If an alert asks “Are you talking to yourself?” it may guide a respondent toward one interpretation. If it asks for a free description, responses may be richer but harder to code consistently. A prompt can interrupt the very experience it seeks to record. Participants may be unable to answer while driving or speaking with someone, making observed samples unrepresentative of their day. Researchers should report alert timing, response rates, delays between alert and response, and how missing prompts were handled.
Descriptive Experience Sampling adds training and a later interview to examine specific sampled moments in depth. Hurlburt and colleagues’ work argues that careful questioning can clarify whether someone was innerly speaking, innerly hearing, seeing an image or experiencing something else. The method’s strength is fidelity to particular moments and attention to ambiguity. Its costs include trained interviewers, substantial participant time, and smaller samples. Even a careful interview remains a reconstruction after the prompt, influenced by the conversation and by the participant’s ability to describe an elusive event.
Momentary sampling also does not produce an unconditional personal percentage. The proportion of prompts at which inner speech is reported depends on waking hours sampled, activity mix, compliance and classification rules. A week dominated by solitary writing may differ from one dominated by conversation. A participant may begin noticing inner speech more because the study repeatedly asks about it. A good report supplies the sampling frame and separates within-person variation from between-person differences.
Behavioral interference tests a functional role
Sometimes the question is not “How often do you experience inner speech?” but “Did a silent verbal process contribute to this task?” Researchers can ask a participant to repeat an irrelevant syllable aloud while doing a memory or reasoning task. This articulatory-suppression manipulation occupies speech-related processes. If it selectively impairs performance compared with a well-matched control, a verbal route becomes more plausible. The inference concerns the task under those conditions, not the person’s entire inner life.
Alderson-Day and Fernyhough’s review surveys such behavioral evidence alongside developmental and neural work. It describes how suppression can affect tasks in which covert verbal rehearsal or labeling would be useful. But the manipulation is not a surgical switch that turns off every kind of inner speech. Speaking aloud adds motor demands, changes attention and may mask auditory information. A participant may adopt a different strategy rather than lose an experience entirely. A researcher should compare suppression with a control task that matches general effort as closely as possible and state which alternative explanations remain.
Task materials matter. Remembering a list of words invites verbal rehearsal; judging a complex visual pattern may not. A null suppression effect on one visual task does not prove that the person lacks inner speech. A positive effect may reflect verbal working-memory coding without a vivid subjective “voice.” The construct of phonological rehearsal and the phenomenology of inner speaking overlap in some circumstances but are not identical. Researchers should resist using a behavioral cost as a direct frequency estimate.
Instructions can create the very process being measured. Asking participants to silently name pictures or rehearse words makes covert speech likely and useful for a controlled experiment. It does not tell us how often they would have used words spontaneously. Conversely, telling participants not to use words can increase monitoring for them. Behavioral tasks are strongest when they specify a mechanistic prediction and include control conditions that address effort, motor interference and task difficulty.
Neural evidence needs a phenomenological anchor
Neuroimaging often contrasts a condition in which participants are told to silently repeat or generate words with a comparison condition. Such studies can identify networks associated with task-elicited covert verbal activity. They do not automatically identify the neural signature of all spontaneous inner speech. The scanner environment, instructions and timing can change what the person does. Activity in a language-related area is compatible with several processes, including preparation, comprehension, memory and attention.
Hurlburt and colleagues paired trained experience sampling with fMRI in a small study. Their comparison of intentionally elicited and spontaneously sampled inner speaking found reason to question whether the two are neurally interchangeable. The study is unusually informative about the measurement problem because it links reported experience to scanner moments, but it involved five participants. It should not be treated as a population map of every inner-speech subtype. Its contribution is a demonstration that task validity and ecological validity need separate examination.
Temporal alignment is difficult. An fMRI signal changes slowly relative to a brief private thought. A prompt asks for a report after a moment, and a participant may have moved between experiences. Motion, auditory scanner noise and the act of preparing to answer can affect the signal. Even when a neural pattern predicts a report above chance, classification performance depends on the training sample, task and chance baseline. A brain decoder is not a mind-reading instrument merely because it distinguishes two experimental conditions.
Neural evidence is most useful when it tests a clear prediction across methods. If a questionnaire describes a subtype, momentary sampling confirms instances of that subtype, and an independently specified neural contrast differs at those moments, the triangulation is stronger than any one measure. Disagreement can also be informative: a task may evoke speech-like motor planning while the participant reports little speech experience. The methods should constrain one another rather than be forced into agreement.
Construct validity is a network of checks
Before data collection, investigators can write a sentence of the form: “This score is intended to estimate X, over time frame Y, in population Z.” For a questionnaire, X may be self-reported tendency toward dialogic inner speech. For sampling, it may be incidence of reported inner speaking at randomly signaled moments. For suppression, it may be the contribution of a verbal process to a specific task. For imaging, it may be task-related signal under covert-speech instructions. The measurement plan should not call all four simply “amount of inner speech.”
Convergent evidence should be expected where theories predict overlap, not demanded everywhere. A person who endorses frequent internal dialogue might report more dialogic moments, but differences in recall and sampling context will leave room for imperfect correlation. A suppression effect in a verbal memory task might relate to ability or strategy rather than to the vividness of inner experience. Divergent evidence matters too: a measure of inner speech should not merely recapitulate verbal ability, general talkativeness or distress.
Language and culture affect item interpretation. An inner monologue may be described with different metaphors; a translation of “voice” may suggest auditory perception or something more abstract. Cross-cultural use of a questionnaire should check its factor structure and item meaning rather than assume the original validation carries over. Children and people with communication differences may need adapted prompts, yet adaptation can change the construct. The sample, language and administration method belong in the results.
Reliability has several meanings. A questionnaire can show stable relative scores while a person’s momentary inner experience naturally varies. A momentary measure can be reliable at classifying a well-described episode without producing a stable person-level rank. An interference task can be sensitive to an experimental manipulation yet noisy for ranking individuals. The required form of reliability depends on whether the study claims a trait, a state, a task mechanism or a group difference.
Do not make absence into impairment
Nedergaard and Lupyan examined people who reported little or no inner speech and tested behavioral consequences in several tasks. Their study is valuable because it links a self-description to behavior rather than assuming that the description alone settles function. It does not justify a universal claim that people without a familiar inner voice lack thought, self-control or language. Subsequent debate about how to establish the absence of a private phenomenon reinforces the need to separate report, sampled experience and task performance.
A person’s report of rare inner speech may reflect genuinely different phenomenology, different definitions, an unfamiliarity with introspective questions, or a mixture. Researchers should not erase that report, but they should specify what was measured. Similarly, a high rate of critical inner speech may warrant careful study without becoming a moral judgment. The aim is to understand forms and functions of private language, not to place minds on a simple good-to-bad scale.
Inner speech may serve planning, memory, rehearsal or self-regulation in some settings. Other processes can serve those functions too. A task effect under suppression demonstrates a local contribution, not that the function always requires words. A neural pattern during covert naming demonstrates activity during that task, not that all reflective thinking consists of speech. The most useful result is often a conditional statement about when a verbal route is used and what changes when it is disrupted.
What the evidence does not show
The cited methods do not provide a single definitive count of how many words a person says internally each day. Questionnaire endorsements can differ from sampled moments, sampled moments depend on prompts and contexts, and interviews remain interpretations of reported experience. A behavioral interference effect does not reveal the exact subjective quality of thought. A language-related brain signal does not prove inner speech occurred without a suitable experiential and task anchor.
Studies of task-elicited covert speech do not automatically generalize to spontaneous inner speaking. The small fMRI experience-sampling study establishes a validity question, not a final neural taxonomy. Neither a low questionnaire score nor a null suppression effect proves that someone never thinks in words. Neither a high score nor a positive task effect establishes a psychological diagnosis or a superior style of thought.
The Institute’s measurement distinctions and design checklist are editorial synthesis of the verified studies below, not a new validated instrument. Their purpose is to keep each conclusion at the level its method can support.
Return to the question
“Do you think in words?” is a useful opening question, but the researcher must decide what would count as an answer. A general scale can describe a person’s remembered tendencies; a sampled prompt can locate a particular episode; an interference task can test whether a verbal route helped performance; and imaging can constrain a neural account when linked carefully to experience. They become stronger together when their differences are reported honestly.
Measuring inner speech well does not require pretending private experience is simple. It requires a precise construct, a method matched to that construct, and conclusions that preserve the distance between a report, a behavior and a brain signal. That precision makes room for the many ways people use words, images and other forms of thought in ordinary life.
Research references
- Alderson-Day, B., & Fernyhough, C. (2015). Inner speech: Development, cognitive functions, phenomenology, and neurobiology. Psychological Bulletin, 141(5), 931–965. View source
- Alderson-Day, B., Mitrenga, K., Wilkinson, S., McCarthy-Jones, S., & Fernyhough, C. (2018). The varieties of inner speech questionnaire revised: Replicating and refining links between inner speech and psychopathology. Consciousness and Cognition, 65, 48–58. View source
- Alderson-Day, B., & Fernyhough, C. (2015). Relations among questionnaire and experience sampling measures of inner speech: A smartphone app study. Frontiers in Psychology, 6, 517. View source
- Hurlburt, R. T., Alderson-Day, B., Kühn, S., & Fernyhough, C. (2016). Exploring the ecological validity of thinking on demand: Neural correlates of elicited vs. spontaneously occurring inner speech. PLOS ONE, 11(2), e0147932. View source
- Heavey, C. L., & Hurlburt, R. T. (2008). The phenomena of inner experience. Consciousness and Cognition, 17(3), 798–810. View source
- Nedergaard, J. S. K., & Lupyan, G. (2024). Not everybody has an inner voice: Behavioral consequences of anendophasia. Psychological Science, 35(7), 780–797. View source