Synaesthesia: The People Who Taste Words and See Sound

Synaesthesia is involuntary and automatic. The experience arrives without effort or intention, and the synaesthete cannot suppress it. That is the first line of the definition and it already does real work, because it separates the phenomenon from a preference or a habit of thought. You do not decide that the letter A is red, and you cannot decide to stop.

The experiences are percept-like rather than merely thought. Synaesthetes report seeing, tasting or feeling the concurrent sensation, though not always projected outward into the world: for many it happens in the mind's eye rather than out on the page. They are also usually unidirectional. The inducer triggers the concurrent, so the grapheme brings the color, but the color does not bring back the grapheme.

There is no single synaesthesia. Over 60 types have been documented, and essentially any pairing of sensory or cognitive dimensions can in principle produce one. Grapheme-color synaesthesia, in which letters and numerals carry colors, is the most studied variant by a wide margin. Day-color and month-color forms are common, as are spatial forms in which a sequence occupies a fixed arrangement in space.

The documentation is older than the modern research program. In 1881 Francis Galton published a paper called Visualised Numerals in the Journal of the Anthropological Institute, collecting and printing diagrams from his own subjects showing how each of them mentally saw a sequence of numbers arranged in a fixed spatial form. It is one of the first scientific documentations of what is now called number-form or spatial-sequence synaesthesia, and it arrived more than a century before any of the imaging work below.

The gold-standard diagnostic is consistency over time, and it is a genuinely good test because it is hard to fake and simple to score. Give a synaesthete a list of letters and ask for the color of each. Then do it again after a long delay. Baron-Cohen, Harrison, Goldstein and Wyke reported in 1993 that a synaesthete's pairings hold above 90 percent consistency even when years separate the two sittings, with a measured 92.3 percent one-year retest figure across their nine subjects. Control subjects, deliberately asked to memorize an arbitrary set of pairings and then retested, manage roughly 30 to 40 percent, and they were retested after only a few weeks. The gap is not subtle. It is the difference between a perception and an association.

Which is why the flat dismissal fails. The claim that synaesthesia is just imagination, just metaphor, or just a learned association is not merely disputed, it is contradicted. Three independent lines converge on the same answer: behavioral consistency across time, interference effects of the Stroop type, in which one automatic response measurably obstructs another, and neuroimaging. Our own research file tiers the dismissal at Tier 4, and that is where we carry it.

The old figure was about 1 in 2,000, roughly 0.05 percent. It comes from Baron-Cohen, Burt, Smith-Laittan, Harrison and Bolton in 1996, who estimated prevalence from people responding to a magazine advertisement, and who also reported a female-to-male ratio of about 6 to 1. Both numbers carry the signature of the method that produced them. If you find your subjects by asking who will come forward, what you learn about first is who comes forward.

The modern figure comes from Simner and colleagues in 2006, and it is the study most often quoted and most often described wrongly, including by us. The paper was built specifically to avoid self-referral: a large, non-self-referred sample drawn from the general public at a public science event, adults and children, tested directly with an objective color-matching task. What it measured directly was grapheme-color synaesthesia, at about 1.1 percent. The widely quoted headline of at least 4.4 percent is a composite lower-bound estimate for synaesthesia of any type, assembled by combining that one measured rate with published rates for other variants drawn from the wider literature. Our own research file describes the study as a screening of 500 university students, which is close to the one thing it was designed not to be, and we correct that here. One point of genuine remaining uncertainty, named rather than smoothed over: secondary sources disagree on the sample size, most describing roughly 1,190 participants and at least one giving 500, and this could not be resolved against the paper's own methods text.

For grapheme-color synaesthesia specifically, the working estimate is 1 to 2 percent. That is the figure to reach for if you need one, and it comes with a label attached: it describes a single variant, the most studied one, and not synaesthesia as a whole.

The early female predominance has largely dissolved. It was reported in the first wave of studies, but it may reflect ascertainment bias, since self-referral skews toward the people who volunteer to be studied, disproportionately women in those samples. When Simner and colleagues looked at a sample that had not selected itself, prevalence came out roughly equal between the sexes.

Ramachandran and Hubbard proposed in 2001 that grapheme-color synaesthesia arises from structural cross-activation between two neighboring pieces of cortex: the grapheme-recognition area in the posterior fusiform gyrus and the color-processing area, V4 and V8. The proposal's strength is its simplicity. Those two regions are anatomically adjacent, and adjacency is exactly where an extra connection would be least surprising.

The imaging supports it. In fMRI, grapheme-color synaesthetes shown black-and-white letters activate V4, the color area, and non-synaesthetes shown the same letters do not (Hubbard and colleagues, 2005). Diffusion tensor imaging finds increased white-matter connectivity between the grapheme and color areas in synaesthetes (Rouw and Scholte, 2007). One is a functional signature and one is a structural signature, both of them measurements taken in a scanner rather than a model drawn on an outline of a brain, and both point at the wiring the model predicts.

Where would the extra connections come from? Maurer and Mondloch proposed in 2005 that normal development involves the selective pruning of cross-modal connections that are more widely present at birth, and that adult cross-activation may be what incomplete pruning leaves behind. Note carefully what this claim is: a proposed developmental mechanism for the adult wiring. The stronger claim that every newborn is actively synaesthetic is a different proposition, tiered lower, and it appears further down.

The alternative is disinhibited feedback, proposed by Grossenbacher and Lovelace in 2001. On this account nothing extra is wired in. Higher cortical areas constantly send feedback signals down to lower sensory areas, that feedback is normally inhibited below the threshold of awareness, and in synaesthetes the inhibition is reduced so the signal becomes consciously expressed. Its advantage is coverage. It accounts for the variants that link brain areas nowhere near each other, such as lexical-gustatory synaesthesia, where a word produces a taste. Anatomical adjacency cannot explain that one, because there is none.

Synaesthesia runs in families. Early pedigree work by Baron-Cohen and colleagues in 1996 suggested autosomal dominant inheritance with incomplete penetrance, which is to say the pattern can be passed on and still not appear in a given family member.

Then the genome looked back. Asher and colleagues ran a genome-wide linkage study in 2009 on families with auditory-visual synaesthesia and identified suggestive loci on chromosomes 2q24, 5q33, 6p12 and 12p12. No single synaesthesia gene has been identified. The condition appears polygenic, and different forms may involve different genetic variants entirely, which sits comfortably with the picture the mechanism section leaves behind: not one thing, several.

Multiple studies report an association between synaesthesia and artistic or creative ability. The paper cited for it in the sources below is Rothen and Meier's Higher Prevalence of Synaesthesia in Art Students, from 2010, and it is worth being exact about which half of the usual sentence that paper actually carries. What it reports is a prevalence gap: grapheme-color synaesthesia at roughly 7 percent among art students against roughly 2 percent in a non-art-student control group. Our own research file additionally credits Rothen and Meier with the separate finding that synaesthetes score higher on measures of visual imagery and creativity, and with the statement that synaesthesia is over-represented among artists, musicians and writers. Those are different claims from the one the titled paper carries, we could not tie them to a specific publication in this pass, and we name the join rather than let one link stand for all of it. What none of this establishes is the direction of causality or the cognitive mechanism connecting the two, both of which our own file records as debated.

The famous names need taking apart. Our own research file lists, in a single flat sentence, the composers Liszt, Messiaen and Scriabin, the artists Kandinsky and Hockney, and the physicist Feynman. Checked individually, those six cases are nowhere near equal in evidentiary weight. Two are strongly documented in the subject's own words. One is actively contested in the scholarly literature. One rests on a secondhand anecdote. One was not independently verified in our own checking. And one is a memoir passage from a man who was never tested.

That passage is worth precisely what it is worth. It is a genuine, directly attributable, first-person description from a named individual with no reason to invent it. It is also a description, not a measurement. Feynman was never a subject in any of the studies this article cites, and nobody ever ran the retest on him. Testimony of this kind is evidence that a person reports an experience. It is not, on its own, evidence about what is happening in a brain, and the entire architecture of the consistency test exists because that gap is real.

One more flattering story deserves the same restraint. Mirror-touch synaesthesia is the variant most often tied to empathy research, and the tie is genuinely unresolved. Some published work finds mirror-touch synaesthetes scoring higher on empathy measures, with a hypothesized relation to mirror-neuron systems in motor cortex. Other published work finds no significant relationship between mirror-touch synaesthesia and empathy at all. Our own research file names the variant and makes no empathy claim about it whatsoever, and until the literature agrees with itself, neither will we.

The strong version of the pruning story says that all neonates are synaesthetic, living in a blooming, buzzing confusion of cross-modal activation that development gradually refines away. It is a suggestive idea and it fits neatly onto the mechanism above. It is also very difficult to test directly, because the subjects cannot yet tell you anything, and that is exactly why it sits at Tier 3 while the pruning mechanism it extends sits at Tier 1. Fit is not evidence.

Psychedelic drugs, including LSD, psilocybin and mescaline, commonly produce temporary synaesthetic experiences: hearing colors, seeing sounds. The proposed connection is disinhibition, with serotonin 2A receptor activation reducing cortical inhibition in something like the way the disinhibited-feedback model describes for developmental synaesthesia. Whether the drug-induced and the developmental forms actually share identical neural mechanisms is unconfirmed by our own file's account, and the resemblance between two descriptions is not evidence that one process underlies both.

The claim that everyone is secretly synaesthetic is overstated, and it is overstated in an interesting way, because the thing it points at is real. Cross-modal correspondences do exist across the general population. High pitch associates with bright and with small. The bouba/kiki effect, in which invented words are consistently matched to particular shapes, is robust. But these are statistical tendencies in judgment, not the involuntary, percept-like, temporally consistent experiences that define synaesthesia. The difference is the whole content of the machinery this article opened with: a tendency is not a percept, and a preference tested once is not a pairing that survives a decade.

The file's framing is that synaesthesia matters to consciousness studies because it demonstrates that subjective conscious experience, qualia, can differ dramatically between individuals perceiving the same physical stimulus: the word Tuesday triggering a specific shade of blue for one person and no color at all for another. That framing is worth examining rather than swallowing. What synaesthesia establishes, uncontested, is that different people have measurably different, consistent, involuntary cross-modal perceptual responses to the same stimulus. Whether that difference amounts to proof that qualia differ, in the loaded philosophical sense of that word, is a further step, and it is not a step this file argues for. It belongs to the qualia debate proper, where the terms are fought over rather than assumed, and that debate has its own article in this wing, Qualia: The Redness of Red and the Limits of Science, which carries the fight between realism and illusionism in full. We point there rather than re-run it here.

The second framing is that synaesthesia is a natural experiment in cross-modal binding, a case of the brain's normal integration process reaching features it usually keeps apart. That is our file's own description and it is a fair one, but it is worth being exact about what stands behind it. Our companion research file on the binding problem, K_2_08, defines cross-modal binding generally, as the integration of information across the senses, with the McGurk effect and the rubber-hand illusion as its standard examples, and it does not mention synaesthesia once: a full-text search of that document returns zero matches for the word. That is a fact about our own shelf and not a survey of the wider literature, which we have not made here. The bridge between the two files is a reasonable inference, and it is ours, not a finding imported from somewhere stronger.