What colour is "lavender"?

Find the colour of a word one slider at a time. Each choice changes a single dimension, which is how Gibbs Sampling with People turns human judgments into a sampler.

Practice mode Nothing you do here leaves this browser. We plan to open this as a citizen-science study once our ethics amendment is approved. How taking part will work

Ask someone what colour “lavender” is and you get a word back. Ask them to adjust one thing at a time and you can get the colour itself.

Each trial gives you a single slider that changes one property of the colour, its hue, saturation or lightness, and leaves the other two alone. You pick the value that best fits the word. The next trial moves a different property, starting from your choice. A long run of these easy choices behaves like a Gibbs sampler, an algorithm statisticians use to explore a probability distribution. Here the distribution is the colours a word brings to mind.

Here your 15 choices form one chain of your own. In the study each chain had 30 steps, and each step was made by a different person.

Your palette

    Passing it on

    In Gibbs Sampling with People the chain is passed between people: each trial is done by someone new, starting from the previous person’s colour. The end of the chain is then a colour shared by many minds rather than one. As in the study, the slider sometimes runs backwards, and here it also starts at a random place, so its position gives nothing away. The colour wheel stays grey until your chain is done for the same reason.

    The same trick works wherever people can judge what they see or hear: the paper also used it to find which chords sound most pleasant, which voices sound angry or happy, and, paired with an image-generating network, faces that look trustworthy or intelligent. Associations like these vary across languages and cultures, which is why we want many people to play.

    Where this comes from

    Original study Gibbs sampling with people Harrison, Marjieh, Adolfi, van Rijn, Anglada-Tort, Tchernichovski, Larrouy-Maestri and Jacoby (2020). Advances in Neural Information Processing Systems (NeurIPS) 33. The method this page uses. One slider per trial turns a run of easy judgments into a sampler over colours, chords, voices and faces.

    Related work

    More interactive experiments