Content uploaded by Carlos Velasco
Author content
All content in this area was uploaded by Carlos Velasco on Oct 13, 2023
Content may be subject to copyright.
Food Quality and Preference 112 (2023) 105009
Available online 11 October 2023
0950-3293/© 2023 The Author(s). Published by Elsevier Ltd. This is an open access article under the CC BY license (http://creativecommons.org/licenses/by/4.0/).
The taste of colours
Carlos Velasco
a
,
*
, Francisco Barbosa Escobar
b
,
c
, Charles Spence
d
, Juan Sebastian Olier
e
a
Centre for Multisensory Marketing, Department of Marketing, BI Norwegian Business School, Oslo, Norway
b
Department of Food Science, Faculty of Science, University of Copenhagen, Frederiksberg, Denmark
c
Department of Food Science, Faculty of Technical Sciences, Aarhus University, Aarhus, Denmark
d
Crossmodal Research Laboratory, Department of Experimental Psychology, University of Oxford, Oxford, UK
e
Department of Cognitive Sciences and Articial Intelligence, Tilburg University, Tilburg, The Netherlands
ARTICLE INFO
Keywords:
Colour
Taste
Crossmodal correspondences
Statistical regularities
Semantic congruence
ABSTRACT
A multitude of crossmodal correspondences have now been documented between taste (gustation) and visual
features (such as hue). In the present study, new analytical methods are used to investigate taste-colour corre-
spondences in a more ne-grained manner while also investigating potential underlying mechanisms. In
Experiment 1, image processing analysis is used to evaluate whether searching online for visual images asso-
ciated with specic taste words (e.g., bitter, sweet) generates outcomes with colour proportions similar to those
that have been documented in the literature on taste–colour correspondences. Colour–taste matching tasks
incorporating a much wider colour space than tested in previous studies, were assessed in Experiments 2 and 3.
Experiments 3 and 4 assessed the extent to which the statistical regularities of the environment, as captured by
food object categories, might help to explain the aforementioned correspondences and to what extent the cor-
respondences are present in online content associated with specic tastes, respectively. Experiment 5 evaluated
the role of statistical regularities in underpinning colour-taste correspondences related to the stage of ripening of
fruit. Overall, the ndings revealed consistent associations between specic colours and tastes, in a more
nuanced manner than demonstrated in previous studies, while at the same time also showing that both food
object categories and the stage of fruit ripening signicantly inuenced colour and taste perceptions. This, in
turn, suggests that people might base these correspondences on both the foods present in their environments, as
well as the natural changes that they undergo as they ripen. The results are discussed in light of the different
accounts that have been suggested to explain colour-taste correspondences.
1. Introduction
Vision is critical when it comes to setting our food and drink ex-
pectations and thereafter our avour perception (e.g., Delwiche, 2012;
Hurling & Shepherd, 2003; Piqueras-Fiszman & Spence, 2015; Sun and
Gauthier, in press). Indeed, it has been suggested that vision and ret-
ronasal olfaction may provide some of the most important cues as far as
setting people’s expectations associated with the likely taste of foods and
beverages (Hutchings, 1977; Stevenson, 2009). Perhaps unsurprisingly,
the research that has been published to date has documented a number
of associations between visual features and tastes (see Spence, 2023, for
a review). Some of these associations fall within the literature on
crossmodal correspondences, that is, the sometimes surprising, associ-
ations that have repeatedly been documented between features present
in (or assessed via) different senses (Spence, 2011, 2022; Walker, 2016).
A variety of crossmodal correspondences between visually-presented
features including shape curvature and symmetry (Turoman et al.,
2018), textures (Barbosa Escobar et al., 2022), and colours (Saluja &
Stevenson, 2018), on the one hand, and taste and/or taste words on the
other, have been documented to date. For instance, people typically tend
to associate sweetness with roundness and pinkish-red hues and other
tastes such as bitterness and sourness with shape angularity and yellow/
green hues (e.g., Velasco et al., 2015, 2016).
In the present study, we focus on taste–colour associations. As such,
it is important to clarify rst what taste and colour are. Taste refers to
the basic gustatory sensations experienced in the oral cavity, especially
the tongue, with the basic qualities of taste commonly assumed to refer
to sweet, sour, salty, bitter, and increasingly also umami (e.g., Avery
et al., 2020). People often confuse taste with avour. Technically
speaking, though, avour relates to the interaction between taste,
* Corresponding author.
E-mail address: carlos.velasco@bi.no (C. Velasco).
Contents lists available at ScienceDirect
Food Quality and Preference
journal homepage: www.elsevier.com/locate/foodqual
https://doi.org/10.1016/j.foodqual.2023.105009
Received 31 July 2023; Received in revised form 3 October 2023; Accepted 3 October 2023
Food Quality and Preference 112 (2023) 105009
2
olfaction, and possibly also trigeminal sensations that form a unitary
percept (Auvray & Spence, 2008; Spence, Smith, & Auvray, 2015). In
fact, all of the senses contribute to the perception of avour, either in
terms of helping to set people’s avour predictions, and/or in terms of
contributing directly to avour perception (Spence, 2017; Stevenson,
2009, 2014). Colour perception is intricately linked to the human visual
system’s sensitivity to electromagnetic radiation within the visible
spectrum. This sensitivity is contingent upon the specic wavelength of
light, which, in turn, determines the appearance of objects and light
sources through three fundamental perceptual dimensions: hue, satu-
ration, and lightness. Hue corresponds to the pure spectrum of colours,
often equated with colour names such as red, green, and blue. Saturation
relates to the perceived degree of whiteness in a colour, while lightness
pertains to the intensity of light seemingly reected from the object.
These perceptual dimensions collectively contribute to the comprehen-
sive understanding of colour and its visual representation (Taveras-Cruz
et al., 2022).
Over recent decades, a number of studies have explored the associ-
ations between tastes and colour hue (e.g., Koch & Koch, 2003; Marks,
1978; O’Mahony, 1983; Spence et al., 2015). Broadly-speaking, the
different studies reveal a general tendency for people to associate
sweetness with red and pink, sour with green-yellow, salt with blue and
white, and bitter with darker colours such as black and brown (e.g.,
Tomasik-Krotki & Strojny, 2008; Velasco et al., 2016; Wan et al., 2014;
see also Spence & Levitan, 2021, for a review).
Whilst the research documents the existence of a number of such
crossmodal correspondences between tastes and colour hues and a range
of potential explanations have been suggested, there are two major
limitations that the present research is designed to address. First, many
of the studies involve situations in which the participants were forced to
choose from a small set of colour options. Second, it is still not fully clear
why it is that people associate seemingly-unrelated visual and gustatory
features in the consensual ways that the research shows that they do.
The research outlined here involves new analytical approaches (e.g.,
image processing analysis of colour in relation to taste searches), a more
comprehensive colour space in the matching tasks, as well as a new
experiment designed to assess how source object changes relate to both
colours and tastes. Comparable analytical methods have been used in
previous research both in terms of image processing analysis (Motoki,
Takahashi, & Spence, 2021) and ne-grained colour scales (Huang et al.,
2019, 2020). Notably, though, when it comes to image processing
analysis, in contrast to previous research that analysed all the pixels in
an image, our study concentrated exclusively on the hue, saturation, and
value of the most salient pixels, thus ignoring interference from colours in
other parts of the background of the image. We also used features
extracted with deep learning models to compare the semantic content of
images of specic objects and those associated with tastes. In addition,
whilst we use a 1536 ×384 px rectangle with HSV values (16384 colour
points), some of the previous studies have focused on the default
Microsoft Ofce 127-colour palette. Lastly, previous research using a
broader set of colour alternatives, has focused on foods and their a-
vours (e.g., tea, coffee), with taste ratings associated with them just
being one aspect of the studies, whilst the present research focuses
exclusively on taste.
Spence and Levitan (2021) recently outlined several potential
mechanisms that might lie behind the aforementioned correspondences.
They describe four possibilities: the rst suggests that colour-taste cor-
respondences may be based on the crossmodal similarity of the
component unisensory stimuli (though see Di Stefano & Spence, 2023;
Spence & Di Stefano, 2022). The second mechanism indicates that
people might be sensitive to the statistical regularities of the environ-
ment and, through a process of associative learning, come to internalize
those mappings that occur between tastes and colours in our environ-
ments (e.g., Maga, 1974; cf. Parise, Knorre, & Ernst, 2014). Spence and
Levitan clarify, though, that this mechanism leads to the question of
whether colour–taste correspondences may be mediated by a specic
source object. In other words, do people sometimes choose to associate
colours and tastes because, when asked about these mappings, they use a
benchmark object as a reference point (i.e., such as associating yellow
and green with sour because they bring to mind citrus fruits, such as
lemons and limes, at least if one happens to live in Europe)? Providing
some insights into this matter, Speed and her colleagues (2023) recently
conducted a study designed to assess whether odour-colour associations
were mediated by concurrent verbalization. Their studies revealed that,
even though colour associations are somehow related to semantic fac-
tors (e.g., odour naming accuracy), they are not based on odour labels.
Finally, Spence and Levitan (2021) point to a potential emotional
mediation account of at least some taste-colour correspondences. It is
worth noting that this kind of explanation has also been used to explain
many other correspondences, such as those between taste and shape or
scent and colour (i.e., where a statistical or source object account has yet
to be forthcoming; Schifferstein & Tanudjaja, 2004; see Spence, 2020,
for a review). The idea here is that the participants in experimental
studies might choose to associate tastes and colours on the basis of the
affective tone of the individual sensory stimuli. It is important to bear in
mind here that the different mechanisms of crossmodal correspondences
need not be mutually exclusive but might be complementary when it
comes to explaining them (Spence, 2011).
In the present research, the focus is on trying to provide insights into
the specic colour spectrum associated with each taste and the
explanatory power of the statistical account of taste–colour correspon-
dences. Thereafter, the aim was to assess the extent to which people use
source objects, and the transformations that they undergo, to help guide
the colour–taste associations. To that end, ve studies were conducted
that were designed to replicate and extend previous ndings on col-
our–taste correspondences.
In Experiment 1, image processing analysis was used to evaluate
whether visual images coming from an online search using specic taste
words as search terms leads to image outcomes with most salient areas
or objects containing hues that resemble the taste-colour correspon-
dences that have been documented. In Experiment 2, a colour–taste
matching task was conducted with a wider colour space, relative to
previous studies in order to replicate the ndings of Experiment 1. In
Experiment 3, Experiment 2 was replicated and extended by adding a
question to inquire what came to mind when the participants were
performing the matching task in order to assess whether specic source
objects guided the associations. In Experiment 4, the aim was to capi-
talize on the results of Experiment 3 and conduct a new search using the
methods developed in Experiment 1. The objective was to evaluate the
similarity between images associated with tastes and specic food cat-
egories and determine whether the associations documented in Experi-
ment 3 could also be found in images obtained from a web search
engine. Finally, in Experiment 5, people’s colour and taste associations
with fruits at different stages of the ripening process were evaluated.
This relates to what are presumed to be correlated changes in both
colour and taste, as well as aroma, avour, and possibly also texture.
2. Experiment 1: Image processing
2.1. Method, procedure, and analyses
This experiment was designed to determine the distribution of col-
ours in the most salient areas of images associated with specic taste
words and retrieved online through a search engine. To that end, the
retrieved images were processed by extracting the HSV (Hue, Satura-
tion, Value) representations of the pixels’ colours in each image’s most
salient areas as described below.
Image acquisition: Images associated with specic tastes were
retrieved using the Google Cloud Platform and the custom search API.
The API is used, as opposed to a manual search, because of easiness in
downloading the material, and the possibility of specifying the location
and language of the results, regardless of where the search was
C. Velasco et al.
Food Quality and Preference 112 (2023) 105009
3
performed. The geolocation was set to the United States of America (US),
and the search language to English. The search terms used were: bitter,
salty, sour, sweet, and umami, followed by the word food. The API
returns the URL to the most relevant images associated with a given
search term, as determined by the engine. The rst 50 results from the
image search were downloaded for further processing to determine the
colour distributions in the salient areas. The results used were capped at
50 since additional results would include repeated or unrelated images
as they become less relevant or associations are less direct.
Saliency estimation: Images can contain different areas in the fore-
ground and background; however, we were interested in those parts of
the image containing the most relevant components, or objects. As such,
only some elements in each image are likely suitable for determining the
distribution of colours that are associated with a specic taste. The Deep
Learning model presented in Kroner et al. (2020) was used to estimate
the most salient areas of an image and then consider only those pixels.
The model was trained on the SALICON dataset (Jiang et al., 2015). This
model takes a raw image as its input and generates a mask indicating a
saliency per pixel (See Appendix A for examples of downloaded images
and their estimated saliency maps). The saliency was normalised be-
tween 0 and 1, and only those pixels with a saliency greater than 0.9
were considered further. Some of the images contained faces or pack-
aging; however, in the images containing faces, the saliency maps
highlighted the food and not only the faces. Furthermore, these images
represented less than 4% of all the images. Similarly, there were images
presenting packaging, but these were less than 2% and, in most cases,
the colours featured were in line with the associated food. So, the effect
of face and packaging pixels was assumed to be minor, and images were
kept as they still represent associations with the visual representation of
tastes online.
Colour distributions: All the pixels in the most salient areas were
considered per picture. A colour distribution per image was calculated,
and then all distributions per image were averaged to form a distribution
per taste. Distributions were generated as two-dimensional histograms
for hues and values, in a format that allowed for comparison with the
results of Experiments 2 and 3. An average distribution over all images
was calculated per taste by adding all the individual image distributions
and forming a normalised histogram.
Palette: The results of Experiments 1, 2, and 3 are superimposed on
the colour palette used for Experiments 2 and 3. The colour palette
consisted of a 1536 ×384 px rectangle varying in HSV (see Fig. 1). In
this palette’s horizontal axis, all hues values in HSV ranging from 0 to
255 are arranged. In the vertical axis, saturation and value vary simul-
taneously. Value decreases linearly from 255 on the top of the palette to
35 on the bottom. Saturation increases linearly in the rst third of the
axis from 35 to 254 and is maintained at 254 afterwards. That allows for
a wide range of value combinations and saturation levels for each hue.
Moreover, the top part of the palette contains lighter colours, with
darker ones on the bottom, while always maintaining a hue (as the
saturation never drops to 0). This colour palette alternative allows us to
quantify variations in hue and brightness. In addition, given that we are
using coordinates to represent colour variation, they can be used to es-
timate a give spatial location (mean) with some measure of variability
(standard deviation), in our case the average colour region and its width.
2.2. Results and discussion
The results of this Experiment 1 are visualised in Fig. 2 (Experiment
1). They show that estimates of salient areas of those images that are
associated with sweetness have predominantly red hues, notably darker
shades. That is mainly due to images depicting desserts, pastries, and
chocolates. There is a high concentration of yellow and orange hues and
darker shades in those images that were associated with the word ‘salty’.
There are some elements of blue and lighter reds. These results emerge
from images depicting snacks as examples of salty food and using blue
elements in the image compositions.
For sour, salient areas mainly depict dark oranges, yellows, and reds.
Closer inspection to the images reveals that these colours appear to come
from citrus fruits such as lemons, oranges, and grapefruits. Some images
also show cocktails, sour candy. In the images associated with bitterness,
the range of hues is broader and includes green, yellow, orange, and red.
Green is associated with bitter vegetables and fruits, while red comes
from some fruits and drinks. Lastly, umami images depict brighter reds
and dark browns, which may be linked to the presence of this taste in
tomatoes, meats and fermented foods (cf. Ikeda, 2002).
Overall, these results appear to indicate that, although certain image
colours match the searched tastes following the documented taste–-
colour correspondences (e.g., sweet and red), others did not (e.g.,
bitterness and green, yellow, orange and red) (Saluja & Stevenson,
2018). Notably, however, previous studies typically include only a
selected number of colours. For this reason, in Experiment 2, we move
on to evaluate how taste–colour correspondences occur in tasks that
include a larger colour spectrum to choose from than previous studies.
3. Experiment 2: Matching task
3.1. Participants
A total of 245 native English speakers from the US took part in the
study. Nevertheless, the data from one participant with duplicate an-
swers were removed from the analyses. The nal sample consisted of
244 participants (121 males, 119 females, 4 other) aged 18–50 years (M
=31.99 years, SD =7.84). The participants were recruited from Prolic
(https://www.prolic.com/) and were compensated with GBP 0.45. All
of the experiments involving human participants reported here com-
plied with the World Medical Association’s Declaration of Helsinki.
Before starting each experiment, all of the participants provided their
informed consent to take part.
3.2. Apparatus and materials
Unlike previous studies, which have typically used a narrow set of
colours to evaluate people’s colour-taste associations, we used a colour
palette to obtain highly granular data. The colour palette consisted of a
1536 ×384 px rectangle with HSV values as described in Experiment 1
(see Fig. 1).
Fig. 1. Colour space used in the tasks of Experiments 2 and 3. Horizontal plane (x-coordinates) reect variations in hue, and vertical plane (y-coordinates) reect
variations in brightness.
C. Velasco et al.
Food Quality and Preference 112 (2023) 105009
4
3.3. Design and procedure
The experiment was programmed and conducted in Gorilla
(https://gorilla.sc/), and only laptops or desktops could be used. In the
experiment, participants were tasked with selecting the colour that they
most strongly associated with each of the ve basic tastes (sweet, salty,
sour, bitter, and umami), presented one at a time, by clicking on the
colour palette. We use taste words in the present research. They have
been shown to give rise to similar patterns of results as when actual
tastants are used (Saluja & Stevenson, 2018).
The participants were rst introduced to the experiment and asked to
provide their informed consent. Next, they were presented with detailed
instructions. Subsequently, they were asked to calibrate their screen
with Gorilla’s built-in calibration method using a credit card. Subse-
quently, to familiarise themselves with the procedure, they completed a
practice trial of the task, in which they had to select the colour they most
strongly associated with fatty taste. Afterwards, the participants
completed the experimental trials.
3.4. Analyses
First, two nonparametric tests for repeated measures data were
performed with taste as a within-participants factor and x- (hue) and y-
coordinates (brightness) as dependent variables, respectively. The Wald
test for each nonparametric test were estimated using the {nparLD} R
Package. Signicant main effects were further analysed using the Wil-
coxon Test through the {stats} R package. After that, the results were
visualised via a 2D kernel estimation density plot based on all partici-
pants’ colours selected for each taste and displayed the results with
contours. To do this, the geom_density_2d function from the {ggplot2} R
Package was used.
3.5. Results and discussion
A signicant main effect of taste was observed for both x-
coordinates, W(4) =145.79, p <.001, and y-coordinates, W(4) =
200.97, p <.001, suggesting differences in both hue and brightness as a
function of taste. The x-coordinates were signicantly higher for
sweetness than for the other tastes (p <.001). They were also higher for
saltiness (p =.038) and umami (p =.007) than for sourness. No other
differences were observed (ps >0.999). In terms of the y-coordinates,
bitterness and umami were signicantly higher than saltiness, sourness,
and sweetness (ps <0.001), no difference was observed between
bitterness and umami (p =.110), and sourness was signicantly higher
than sweetness and saltiness (ps <0.001). Saltiness and sweetness did
not differ signicantly (p >.999).
In Table 1, the relative treatment effects (RTE) are presented,
providing an indicator of the size of the effect. As indicated by Mar-
molejo-Ramos et al. (2013), the RTE denotes the likelihood that a
randomly selected observation from a given subset of data is greater
than a randomly selected observation from the entire dataset, and its
value falls within the range of 0 to 1. The largest effect observed for the
x-coordinates (hue) is for sweetness, whereas the largest effects docu-
mented for the y-coordinates (brightness) were for bitterness and
umami.
The results of the 2D density plots showed that participants tended to
associate different basic tastes with specic areas of the colour palette
(see Fig. 2, Experiment 2). Consistent with the results of previous
studies, sweetness was consistently associated with red and pink hues.
However, the associated colours tended to be darker than those used in
Fig. 2. 2D density plots of colour associations for each basic taste in Experiments 1 (left-hand), 2 (middle), and 3 (right-hand). Lighter and yellow contour lines
represent a higher density of responses. (For interpretation of the references to color in this gure legend, the reader is referred to the web version of this article.)
Table 1
Relative treatment effects in Experiment 2.
Taste x-coordinates (hue) y-coordinates (brightness)
Bitter 0.448 0.649
Sour 0.427 0.483
Sweet 0.697 0.387
Salty 0.464 0.366
Umami 0.464 0.615
C. Velasco et al.
Food Quality and Preference 112 (2023) 105009
5
previous studies as well, as shown in Experiment 1. In addition, and also
consistent with previous research, sourness was associated with green
colours and, to a lesser extent, with yellows. Saltiness was associated
with dark yellow and dark blue hues. The participants associated
bitterness with dark yellow-greenish and light orange colours. Umami
did not exhibit highly precise associations, though the participants
seemed to broadly associate it with brown and green–blue hues.
Whilst these results suggest, in one way, that the colours matched to
tastes in previous studies (e.g., red and sweetness) also occur when a
broader colour space is used, other colours also appear (e.g., bitterness
and yellow-greenish and light orange). This might be indicative of
changes associated with the reference point (e.g., colour space) used by
the participants to respond to the matching task. As such, in Experiment
3 we aimed to replicate the ndings of Experiment 2, and we also
evaluated whether participants used a specic reference object in order
to guide their associations.
4. Experiment 3: Replication of matching task replication and
semantic analysis
4.1. Participants
A total of 250 native English speakers from the US (122 males, 124
females, 4 other), aged 18–57 years (M =34.69 years, SD =8.04), took
part in the study. The participants were recruited from Prolic (https:
//www.prolic.com/) and were compensated with GBP 0.45.
4.2. Design, materials, and procedure
Experiment 3 followed the same design and procedure as Experiment
2, with the addition of a question to assess the participants’ semantic
mappings underlying their colour-taste associations. More specically,
after participants selected the colour they most strongly associated with
each basic taste, they were asked, in an open-ended question, what they
had thought of when making that association with the specic taste (e.
g., What did you think of when making this association with
bitterness?).
4.3. Analyses
We conducted the same analyses and visualisations as in the previous
study. Furthermore, participants’ semantic mappings were analysed via
a semantic network analysis. We created co-occurrence network graphs
of keywords for each basic taste. First, the data was pre-processed by
converting all words in participants’ responses to lowercase, removing
symbols, and correcting for word elongations and additional spaces.
Then, participants’ responses were tokenized and stop words were
removed. Next, the pairwise count of all keywords was computed. To
create the co-occurrence network graphs, a threshold of at least three
occurrences of keyword pairs was selected. To perform the semantic
network analysis, the {tidyr}, {widyr}, {textclean}, {igraph}, and
{ggraph} R packages were used.
4.4. Results and discussion
The visualisation of participants’ colour-taste associations revealed
consistent results with the previous study (see Fig. 2, Experiment 3). As
in Experiment 2, a signicant main effect of taste was observed for both
x-coordinates (hue), W(4) =111.41, p <.001, and for y-coordinates
(brightness), W(4) =189.83, p <.001, suggesting, once again, the ex-
istence of differences in both hue and brightness as a function of taste
quality. The x-coordinates were signicantly higher for sweetness than
for the other tastes (p <.001), and higher for umami than for sourness (p
=.048). No other signicant differences were observed (ps >0.071). In
terms of the y coordinates, just as in Experiment 2, bitterness and umami
were signicantly higher than saltiness, sourness, and sweetness (ps <
0.001). No difference was observed between bitterness and umami (p
>.999). Sourness was signicantly higher than saltiness (ps <0.001) but
not sweetness (p =.974). The coordinates were higher for sweetness than
for saltiness (p <.001).
The semantic network analysis revealed that participants tended to
arrive at colour–taste associations via intermediate mappings to specic
entities (see Fig. 3). In the case of sweetness, two prominent patterns
emerged. The participants seemed to associate sweetness with red col-
ours because of fruits such as cherries, strawberries, and apples (though
they are also green), as well as candies. In addition, they appeared to
associate sweetness with pink colours because of candies. In regard to
saltiness, the association with blue seemed to be driven by mappings to
the ocean, whereas the association with yellow seemed to be related to
potato chips/fries. For sourness, participants appeared to associate with
the colour green due to apples and apple candies and with yellow due to
lemons. As per bitterness, associations with brown seemed to be related
to dark chocolate, whereas those with green hues were driven by veg-
etables, and those with yellow colours were picked up on mappings to
lemons. Finally, colour associations with umami revolved around soy
sauce, mushrooms, and meat.
The ndings of this experiment largely replicate the results of
Experiment 2, in terms of the taste-colour associations that were docu-
mented. In addition, this experiment reveals that, when asked, partici-
pants appear to base their answers on a reference object and that there
are certain commonalities between said objects. To study how consistent
the results of Experiment 3 are with the images found online (Experi-
ment 1), the similarity of the content of images associated with tastes
and that of images associated with specic categories appearing in the
words found in Experiment 3 was determined.
5. Experiment 4: Semantic information in online pictures
5.1. Method, procedure, and analyses
Estimating content associations. First, the words from Experiment 3
were clustered to determine the most relevant categories grouping the
elements that were mentioned by the participants. Before clustering, all
the words that were in a plural form were replaced by their singular
forms. Only food-related nouns were included, as participants also listed
taste and colour-related words; this was performed manually. A total of
223 words were then grouped into 11 categories. An initial clustering
was performed by applying the DBSCAN algorithm (Ester et al., 1996)
using the minimum path similarity in WordNet (Miller, 1995) between
two words as a distance metric, with which the number of clusters was
determined. This approach was used as DBSCAN does not require to pre-
dene a number of clusters, but it determines such a number from data.
The resulting clustering included 11 clusters with the parameter epsilon
set to 0.98, and a minimum of 2 data points per cluster. Afterwards, the
clusters were labelled and rened manually by assigning words not in
WordNet or for which the denition did not match the context of the
best-matching cluster given alternative meanings of the word.
The nal clusters were: Asian food, citrus, fruit, hot beverage, meat,
salt, sauce, snack, soup/broth, sweets, and vegetable (See Appendix B
for the words in each cluster).
Following the same procedure as for the tastes in Experiment 1,
images were downloaded using the cluster names as search terms. All of
the images were then compared to determine the most probable group
associated with a taste image. Labelling the i
th
image associated with a
given taste as t
i
and the j
th
image in one of the dened food-related
groups as g
j
, we can dene a distance D(t
i
, g
j
), determining how visu-
ally different the two images are. Then, for every taste image t
i
, the
argmin
j
(D(t
i
, g
j
)) is determined by taking into account the images of all
the food-related groups. To compare the images, the image encoder in
the CLIP model introduced by Radford et al. (2021) as a feature extractor
was used. Specically, we use the last layer of a pre-trained ViT-B/32. As
a distance metric D, the normalised squared Euclidean distance was
C. Velasco et al.
Food Quality and Preference 112 (2023) 105009
6
used. With this information, we create a distribution over food-related
groups per taste by calculating a normalized histogram of the number
of images of a given taste with a minimum distance to an image in a
specic food-related group.
5.2. Results and discussion
Content associations: Images associated with bitterness were visually
closer, in terms of the normalized Euclidean distance, to pictures of
vegetables, hot beverages, citrus and fruit. Images associated with
saltiness were visually closer to images associated with snacks, and salt.
Images associated with sourness were not visually similar only to images
of a specic group, but rather to several different categories, mainly
images of citrus, fruits, snacks, and vegetables. Images associated with
sweetness were primarily similar to those of sweets, snacks and fruits.
Finally, images associated with umami were visually closer to those of
Asian food, soup/broth, and meat.
The results of Experiment 4 are similar to those reported in Experi-
ment 3. The distributions over clusters per taste in Experiments 3 and 4
are presented in Fig. 4. To compare the results of these two experiments,
the Spearman’s correlation was calculated to determine whether the
order of frequency of the elements associated with a given taste was
similar between the two sets of distributions. A correlation was esti-
mated for each taste: Bitter: r(9) =0.75, p <0.025, salty: r(9) =0.34, p
>0.1, sour: r(9) =0.64, p <0.05, umami: r(9) =0.54, p <0.1, and
sweet: r(9) =0.68, p <0.05. These results indicate similarities between
the distributions, which can also be observed in Fig. 4. They are sig-
nicant except for salty and umami. The main differences are attribut-
able to more variety in the images per taste than in words. These reect
in the occurrence of sauce in pictures but not in the words from
Fig. 3. Keyword co-occurrence network graphs of colour associations for each basic taste in Experiment 3. The thickness of the link between pairs of words denotes
the occurrence frequency of the word pair.
C. Velasco et al.
Food Quality and Preference 112 (2023) 105009
7
Experiment 3 for the bitter taste. The same happens for Asian food and
soup/broth for salty and sour. In the case of salty, the images also focus
more heavily on ‘snack’, while words from participants are more focused
on salt. Similarly, there is a higher occurrence of sauce in images than in
words for umami. In sweet, the main differences are due to a higher
proportion of fruit in the words and the occurrence of hot beverages in
images but not in words mentioned by participants.
The results of Experiment 1–4 reveal a number of insights. First,
colour-taste correspondences and source objects can be evaluated
through both image processing analysis and matching tasks. Second, the
colours that correspond to specic taste categories are more varied than
documented before. Third, people appear, to a certain degree, to match
the colours that correspond to a specic source object, which may help
to explain the level of variability in the data. Assuming that the asso-
ciative learning account of taste-shape correspondences is correct to a
certain degree, it is reasonable to expect variations in the data as the
saliency of objects involving colours and tastes that people are exposed
to can vary. As such, patterns of association may build on a number of
dynamic changes in people’s environments. For example, fruits, which
were a common object mentioned in the matching tasks, and which
appeared in the image processing analysis, also change over time, and
people may interiorize these regular transformations. To further eval-
uate the associative learning account of taste-colour correspondences, in
Experiment 5, we evaluate the extent to which changes in the ripening
process of fruits relate to variations in the colours and tastes attributed
to the fruit. The reason for exploring this is that variations in fruit
ripening can involve changes in both colour (e.g., green, red) and also
taste (sour/bitter, sweet; Foroni et al., 2016; Maga, 1974).
6. Experiment 5: Internalization of statistical regularities
6.1. Participants
A total of 201 native English speakers took part in the study.
Nevertheless, data from four participants with incomplete or erroneous
answers (e.g., answer age with a string of characters) were removed
from the analyses. The nal sample consisted of 197 participants (97 he/
him, 97 she/her, 2 they/them, and 2 other) aged 18–80 years (M =
39.70 years, SD =14.44). Participants were recruited from Prolic
(https://www.prolic.com/) and were compensated with GBP 0.67.
6.2. Design, materials, and procedure
The experiment was programmed and conducted on Qualtrics (htt
ps://qualtrics.com). The study followed a one-way repeated measures
designed with factor fruit stage (unripe, ripe, and overripe). The
dependent variables were colour coordinates (see Fig. 5), and tastes
(sweet, sour, salty, bitter, and umami) and liking. Note, though, that the
alternatives given to the participant were only a subset of those in the
previous experiments, considering the areas that the associations
covered.
At the beginning of the experiment, the participants were asked to
report their age, gender, and frequency of fruit consumption (daily, 4–6
times a week, 2–3 times a week, once a week, less often, I don’t eat fruit).
Followed by that, the participants were presented with three fruit stage
conditions in a random order. In this part of the study, the participants
were told “Imagine an [unripe OR ripe OR overripe] fruit and answer the
questions following.” The rst question consisted of indicating the
colour that they associated with the fruit by specifying the coordinates
in the colour space (see Fig. 5). After that, the participants were asked to
evaluate how they associated the tastes with the fruit, and the extent to
Fig. 4. Distributions over clusters per taste. A) From words in Experiment 3. B) From image similarities in Experiment 4.
Fig. 5. Colour space used in the tasks of Experiment 5.
C. Velasco et al.
Food Quality and Preference 112 (2023) 105009
8
which they thought they would like the fruit. These variables were
measured in 10-point visual analogue scales. Finally, the participants
were asked whether they have thought of a specic fruit during and to
report it.
6.3. Analyses
To analyse the results, several robust one-way analyses of var-
iance–type statistic (ANOVA-type statistic) were conducted with
ripening stage as factor, and ×and y colour coordinates, as well as
tastes, as dependent variables. The analyses were conducted in the R
statistics environment, via the {nparLD} package.
6.4. Results and discussion
The stage of fruit ripening exerted a signicant inuence on all
variables (see Table 2). The values for hue were lower for the overripe
stage relative to the ripe and unripe ones (ps <0.002), though no dif-
ference was observed between the ripe and unripe stages (p =.180). In
contrast, the values for brightness were lower for the overripe stage
relative to the ripe and unripe stages (ps <0.001). No difference was
observed between the ripe and unripe stages (p =.150, see Fig. 6, for a
visualization of the colour associations).
The sweetness and liking ratings were higher for the ripe, followed
by the overripe, and nally the unripe stage (ps <0.027). The sourness
and bitterness ratings were higher for the unripe, followed by the
overripe, and the ripe stage (ps <0.001). In terms of the salty ratings, the
ratings were higher in the overripe and unripe stages than in the ripe
stage (ps <0.035), while no difference was observed between the
overripe and unripe stages (p >.999). We also performed correlations
between the variables in order to assess the relationship between them
(see Appendix A).
These results suggest that the ripening stage of fruits inuences both
colour and taste perception. This study provides initial support for the
idea that the way in which a fruit changes might involve both colour and
taste information in a way that matches, but only to a certain degree,
colour-taste correspondences. Indeed, overripe differed from ripe and
unripe in terms of colour but no difference was found between ripe and
unripe. However, taste perception did change. Ripe fruits were
perceived as signicantly sweeter than the others and unripe as signif-
icantly more bitter and sour than the others. These results suggest that
colour-taste correspondences might rely on specic identities or mean-
ings for taste-shape correspondences more strongly than specic within-
object variations.
One possible limitation of this study though is the fact that people
might have relied on specic fruits while responding to the task (e.g.,
Foroni, Pergola, & Rumiati, 2016). Whilst we wanted to capture the
overall sense that people might have of fruits at different ripping stages,
we cannot rule out this. For example, some fruits do not have as much of
a visually transparent ripening process such as like watermelon (where
other factors may play a role, such as size) compared to bananas (at least
externally, before opening the fruit).
7. General discussion
The present research focused on replicating and extending research
on colour–taste correspondences by using new analytical methods that
would reveal a more ne-grained understanding of taste–colour corre-
spondences. In addition, it focused on investigating the potential role of
the associative learning account in these correspondences and whether
people use source objects and variations captured in the statistical reg-
ularities in of food contexts to guide their associations. While the
research has documented crossmodal correspondences between tastes
and colour hues, there is still a need to clarify why people associate these
features. Spence and Levitan (2021) outlined several possible mecha-
nisms, including crossmodal similarity, associative learning, and
emotional mediation. Overall, the ndings suggest that the statistical
account may play a key role in the formation of certain taste–colour
correspondences at the object category level (e.g., fruits), but not
necessarily at the case-specic level (e.g., banana), and that such regu-
larities would seem to be present in those online images that are asso-
ciated with tastes. However, other factors are likely also at play.
Five experiments were conducted to revise, replicate, and extend
previous ndings on colour–taste correspondences, evaluating the
relationship between specic food categories and colours, and assessing
people’s colour and taste associations with fruits at different stages of
the ripening process. The experiments used image processing analysis,
colour–taste matching tasks, and questions to assess whether specic
source objects guided associations. The results of Experiment 1 revealed
that those online images associated with sweetness tend to have pre-
dominantly red hues, while salty tastes are associated with yellow and
orange hues. Sourness was associated with dark oranges, yellows, and
reds, and bitterness was associated with a broader range of hues,
including green, yellow, orange, and red. Umami was associated with
brighter reds and dark browns. Experiment 2 found that participants
tended to associate different basic tastes with characteristic colours
ranges, that is, sweetness with red and pink hues (darker than previous
studies, Spence and Levitan, 2021), sourness with green and yellow
hues, saltiness with dark yellow and dark blue hues, bitterness with dark
yellow-greenish and light orange hues, and umami with brown and
green–blue hues.
The results of Experiment 3 revealed that participants arrived at
colour–taste associations via intermediate mappings to specic object
entities. Sweetness was associated with red and pink colours due to fruits
and candies, while saltiness was associated with blue and yellow colours
due to the ocean and potato chips/fries, respectively. Sourness was
linked to green and yellow due to apples and lemons, while bitterness
was associated with brown and green due to dark chocolate and vege-
tables. Umami was broadly associated with foods such as soy sauce,
mushrooms, and meat.
The results of Experiment 4 revealed that different tastes were
associated with different types of images. Bitterness was found to be
associated with hot beverages, vegetables and citrus, while saltiness was
associated with snacks and salt. Sourness had broader associations with
citrus, fruits, snacks, and sweets, while sweetness was mainly associated
with sweets, snacks and fruits. Umami was associated with Asian soup,
broths and meat. Furthermore, the results were compared to those of
Experiment 3, and the outcome indicated that the order of frequency of
the elements associated with tastes are congruent between the images
found online, and the answers from participants, except for salty and
umami.
Finally, the results of Experiment 5 revealed that the stage of fruit
ripening (i.e., unripe, ripe, overripe) had a signicant inuence on the
colours and tastes associated with the fruit. Hue values were lower for
overripe fruits compared to ripe and unripe ones, while brightness
values were higher for overripe fruits. Sweetness and liking ratings were
higher for ripe fruits, while sourness and bitterness ratings were higher
Table 2
One-way ANOVA-type statistics associated with the x- (hue) and y-coordinates
(brightness), as well as each of the tastes, as a function of fruit ripening stage, in
Experiment 5.
Variable ANOVA RTEs
df W p Unripe Ripe Overripe
x coordinates – Hue 2 73.08 <0.0001 0.614 0.483 0.403
y coordinates -
Brightness
145.78 <0.0001 0.445 0.390 0.666
Sweet 459.41 <0.0001 0.252 0.683 0.565
Sour 96.60 <0.0001 0.630 0.385 0.484
Salty 18.35 0.0001 0.513 0.460 0.527
Bitter 207.42 <0.0001 0.660 0.343 0.497
Liking 636.61 <0.0001 0.334 0.785 0.381
C. Velasco et al.
Food Quality and Preference 112 (2023) 105009
9
for unripe fruits. Salty ratings were higher in the overripe and unripe
stages compared to the ripe stage.
In general, these results replicate previous research on taste-colour
correspondences (e.g., Saluja & Stevenson, 2018; Spence & Levitan,
2021; Velasco et al., 2016) and extend them by providing a more ne-
grained hue and brightness scale. For example, Motoki, Takahashi,
and Spence (2021) found that coffee shop images with reddish and
lighter colours were linked to sweeter coffee expectations, while
greenish and darker images were associated with more bitter/tasty
coffee expectations (see also Huang et al., 2019, 2020).
Our results suggest that the assessment of colour–taste correspon-
dences and source objects can be conducted using both image processing
analysis and matching tasks. Notably, the ndings revealed a greater
variety of colours associated with specic taste categories than previ-
ously reported. What is more, our study provides, to the best of our
knowledge, for the rst time, a clear link into the sort of object cate-
gories that may guide said associations (Experiments 2–4). Individuals
exhibited a tendency, to some extent, to match colours corresponding to
specic source objects, which potentially accounts for the observed data
variability. Assuming the potential validity of the associative learning
theory regarding taste-shape correspondences, it is reasonable to
anticipate uctuations in the data due to variations in the saliency of
colours and tastes to which individuals are exposed. Consequently,
patterns of association could arise from the dynamic changes present in
people’s environments. Whilst Experiment 5 offered preliminary evi-
dence supporting the notion that the transformation of a fruit involves
the perceptual integration of colour and taste information, it only
aligned to a certain degree with the colour-taste correspondences
documented in previous research. These ndings suggest that colour-
taste correspondences may rely more heavily on specic identities or
meanings associated with taste-shape correspondences rather than
variations within them.
Whilst the various results presented here are indicative of an asso-
ciative learning mechanism of this type of correspondence (Higgins &
Hayes, 2019), we cannot rule out other potential explanations. Indeed,
recent research on colour–odour correspondences suggests that colour-
odour matching is not affected by verbal interference and that odour-
colour matches are more accurate for familiar scents and correctly
identied odours (Speed et al., 2023). This led Speed and her colleagues
to propose that semantic associations play a role in odour–colour asso-
ciations, but the act of labelling an odour in the moment does not seem
to have a signicant impact. As such, other mechanisms may be involved
in the matching process. This is, perhaps, something that may inform the
results of the present study. Although we can observe that associative
learning appears, to a great extent, to guide their associations, this might
not necessarily be the only mechanism at play (see also Barbosa Escobar
et al., 2023, for another example in the context of temperature-based
correspondences).
The implications of these ndings suggest that colour plays an
important role in shaping our perceptions of taste. Marketers and food
manufacturers may benet from understanding the colour-taste associ-
ations and using them in their product design and packaging (Velasco
and Spence, 2018). Moreover, chefs and food designers can use colour to
create visually appealing dishes that enhance the perception of taste
(Spence et al., 2022). Understanding the impact of fruit stage on colour
and taste could also inform decisions about when to harvest fruit for
optimal avour. Furthermore, our results can inform potential solutions
leveraging colour-taste crossmodal effects and their relationship with
ripening stage to reduce food waste of aesthetically imperfect fruits (van
Giesen & de Hooge, 2019). Overall, the results of the present study
highlights the complexity of the relationship between taste and colour
and the importance of considering both factors in the design and con-
sumption of food.
7.1. Limitations and future directions
Several limitations should be acknowledged in this study. First, the
participants were from a specic country, and online images were linked
to the same location. Hence, if colour-taste associations were to exhibit
some degree of variation across different countries/cultures, especially
when these associations are mediated by source objects, a cross-country
study would be expected to highlight this. This, too, suggests that future
research might aim to evaluate the effect of source objects that come up
in the searches, with broader and more extensive images, in order to
assess object- or context-specic colour-taste correspondences. Second,
the colour palette that participants in the different experiments were
exposed to may not have been strictly the same, as screens differ in the
extent and accuracy of the gamut of colour that they can display.
Nevertheless, to precisely control for colour accuracy requires screens
with colour accuracy certications (e.g., Adobe RGB), which limits the
sample size of any potential experiment. Third, saturation, which is
another key colour feature, was not evaluated as a factor and as such,
future research may look at taste-colour saturation associations capi-
talizing on the methods presented here.
Such limitations can also be connected to the differences encoun-
tered among Experiments 1–3. The colours found in Experiment 1 are
consistent with those in Experiments 2 and 3 in most cases. Yet, the
results might differ signicantly as the exact mapping of hue values from
Fig. 6. Colour – fruit ripening state associations in Experiment 5. The letter represents the average of the stage of the fruit as a function of X (hue) and Y (brightness)
and the ellipses represent the corresponding standard deviation.
C. Velasco et al.
Food Quality and Preference 112 (2023) 105009
10
pixels in images does not consider the possible variability in the
perceptual mapping between colour associations and colour choices that
participants make. So, for example, certain shades of red or pink asso-
ciated with sweetness can be linked to many hue values by participants
and are particularly centred on the left area of the colour palette they
were presented with, yet the actual values in images map to other areas
of the palette that can also be semantically associated with red colours.
This makes the comparison more difcult given that the mappings are
not equivalent and might be measuring different attributes.
Moreover, those differences can imply that the representation in
terms of hue is not aligned with the actual colour perception of partic-
ipants, which would make using colour categories more meaningful than
hue values for these comparisons. On the other hand, the results of
Experiment 4 align with those of Experiment 3. So, the objects that
people associate colours with are consistent with those in images related
to specic tastes. That suggests that the observed intermediate mappings
are more consistent than colours, possibly because the mapping to a
single colour is not optimal given the different hues an object might
have, while food categories are more consistent and stable. This might
imply that asking participants about the food items they associate with
specic tastes, and then analyse the colours in images associated with
those objects might indicate the taste-colour associations better, or at
least in a more stable way, than directly asking participants about col-
ours they associate with taste.
CRediT authorship contribution statement
Carlos Velasco: Conceptualization, Data curation, Formal analysis,
Funding acquisition, Investigation, Methodology, Project administra-
tion, Validation, Visualization, Writing – original draft, Writing – review
& editing. Francisco Barbosa Escobar: Conceptualization, Data cura-
tion, Formal analysis, Methodology, Visualization, Writing – original
draft, Writing – review & editing. Charles Spence: Conceptualization,
Methodology, Writing – original draft, Writing – review & editing. Juan
Sebastian Olier: Conceptualization, Data curation, Formal analysis,
Investigation, Methodology, Validation, Visualization, Writing – orig-
inal draft, Writing – review & editing.
Declaration of Competing Interest
The authors declare that they have no known competing nancial
interests or personal relationships that could have appeared to inuence
the work reported in this paper.
Data availability
Data will be made available on request.
Acknowledgements
We would like to thank the Research Fund of the Department of
Marketing, BI Norwegian Business School, for funding the data collec-
tion of the experiments. In addition, we would like to thank Sophie
Donohugh, for her support with the execution of Experiment 5.
Appendix A
Sample images per taste gathered for Experiment 1, and the corresponding estimated saliency maps.
Appendix B
Words included in clusters in Experiment 4.
C. Velasco et al.
Food Quality and Preference 112 (2023) 105009
11
Asian food sushi, Asian, Japanese, seaweed, Chinese, wasabi
Broth miso, broth, soup, ramen
Citrus grapefruit, lemon, orange, lime
Fruit cherry, berry, raspberry, strawberry, fruit, melon, apple, grape
Hot Beverage coffee, tea
Meat meat, roast, sh, steak, chicken, beef, meaty
Salt salt
Sauce soy, sauce
Sweet sugar, cake, sweet, tart, candy, dessert, pie, cotton, cupcake,donut, brownie, lollipop
Vegetable green, arugula, vegetable, kale, broccoli, brussel sprout, kale, leaves, leafy, pickle, mushroom, bitter melon, broccoli
Appendix C
C. Velasco et al.
Food Quality and Preference 112 (2023) 105009
12
References
Auvray, M., & Spence, C. (2008). The multisensory perception of avor. Consciousness
and Cognition, 17(3), 1016–1031.
Avery, J. A., Liu, A. G., Ingeholm, J. E., Riddell, C. D., Gotts, S. J., & Martin, A. (2020).
Taste quality representation in the human brain. Journal of Neuroscience, 40(5),
1042–1052.
Barbosa Escobar, F., Velasco, C., Byrne, D. V., & Wang, Q. J. (2023). Assessing
mechanisms behind crossmodal associations between visual textures and
temperature concepts. Journal of Experimental Psychology: Human Perception and
Performance, 49(6), 923–947.
Barbosa Escobar, F., Wang, Q. J., Corredor, A., & Velasco, C. (2022). The taste of visual
textures. Food Quality and Preference, 100, Article 104602.
Delwiche, J. F. (2012). You eat with your eyes rst. Physiology & Behavior, 107(4),
502–504.
Di Stefano, N., & Spence, C. (2023). Perceptual similarity: Insights from the crossmodal
correspondences. Reviews in Philosophy & Psychology. https://doi.org/10.1007
/s13164-023-00692-y.
Ester, M., Kriegel, H. P., Sander, J., & Xu, X. (1996). August). A density-based algorithm
for discovering clusters in large spatial databases with noise. KDD, 96(34), 226–231.
Foroni, F., Pergola, G., & Rumiati, R. I. (2016). Food color is in the eye of the beholder:
The role of human trichromatic vision in food evaluation. Scientic Reports, 6, 37034.
Higgins, M. J., & Hayes, J. E. (2019). Learned color taste associations in a repeated brief
exposure paradigm. Food Quality and Preference, 71, 354–365.
Huang, F., Huang, J., & Wan, X. (2019). Inuence of virtual color on taste: Multisensory
integration between virtual and real worlds. Computers in Human Behavior, 95,
168–174.
Huang, F., Qi, Y., Wang, C., & Wan, X. (2020). Show me the color in your mind: A study
of color-avor associations in virtual reality. Food Quality and Preference, 85, Article
103969.
Hurling, R., & Shepherd, R. (2003). Eating with your eyes: Effect of appearance on
expectations of liking. Appetite, 41(2), 167–174.
Hutchings, J. B. (1977). The importance of visual appearance of foods to the food
processor and the consumer. Journal of Food Quality, 1, 267–278.
Ikeda, K. (2002). New seasonings. Chemical Senses, 27(9), 847–849. (Reprinted from
1909. Doi: 10.1093/chemse/27.9.847).
Jiang, M., Huang, S., Duan, J., & Zhao, Q. (2015). Salicon: Saliency in context. In
Proceedings of the IEEE conference on computer vision and pattern recognition (pp.
1072-1080).
Koch, C., & Koch, E. C. (2003). Preconceptions of taste based on color. The Journal of
Psychology, 137(3), 233–242.
Kroner, A., Senden, M., Driessens, K., & Goebel, R. (2020). Contextual encoder–decoder
network for visual saliency prediction. Neural Networks, 129, 261-270.
Maga, J. A. (1974). Inuence of color on taste thresholds. Chemical Senses, 1(1),
115–119.
Marks, L. E. (1978). The unity of the senses: Interrelations among the modalities. New York,
NY: Academic Press.
Marmolejo-Ramos, F., Elosúa, M. R., Yamada, Y., Hamm, N. F., & Noguchi, K. (2013).
Appraisal of space words and allocation of emotion words in bodily space. PLoS
One1, 8(12), 81688.
Miller, G. A. (1995). WordNet: A lexical database for English. Communications of the
ACM, 38(11), 39–41.
Motoki, K., Takahashi, A., & Spence, C. (2021). Tasting atmospherics: Taste associations
with colour parameters of coffee shop interiors. Food Quality and Preference, 94,
Article 104315.
O’Mahony, M. (1983). Gustatory responses to nongustatory stimuli. Perception, 12,
627–633.
Parise, C. V., Knorre, K., & Ernst, M. O. (2014). Natural auditory scene statistics shapes
human spatial hearing. Proceedings of the National Academy of Sciences of the USA,
111, 6104–6108.
Piqueras-Fiszman, B., & Spence, C. (2015). Sensory expectations based on product-
extrinsic food cues: An interdisciplinary review of the empirical evidence and
theoretical accounts. Food Quality & Preference, 40(A), 165–179. https://doi.org/
10.1016/j.foodqual.2014.09.013
Radford, A., Kim, J. W., Hallacy, C., Ramesh, A., Goh, G., Agarwal, S., … Sutskever, I.
(2021). Learning transferable visual models from natural language supervision. PMLR.
pp. 8748-8763.
Saluja, S., & Stevenson, R. J. (2018). Cross-modal associations between real tastes and
colors. Chemical Senses, 43(7), 475–480.
Schifferstein, H. N. J., & Tanudjaja, I. (2004). Visualizing fragrances through colors: The
mediating role of emotions. Perception, 33(10), 1249–1266. https://doi.org/
10.1068/p5132
Speed, L. J., de Valk, J., Croijmans, I., Huisman, J. L., & Majid, A. (2023). Odor-color
associations are not mediated by concurrent verbalization. Cognitive Science, 47(4),
e13266.
Spence, C., Wan, X., Woods, A., Velasco, C., Deng, J., Youssef, J., & Deroy, O. (2015). On
tasty colours and colourful tastes? Assessing, explaining, and utilizing crossmodal
correspondences between colours and basic tastes. Flavour, 4, 23.
Spence, C. (2011). Crossmodal correspondences: A tutorial review. Attention, Perception,
& Psychophysics, 73, 971–995.
Spence, C. (2017). Gastrophysics: The new science of eating. London, UK: Viking Penguin.
Spence, C. (2020). Olfactory-colour crossmodal correspondences in art, science, &
design. Cognitive Research: Principles & Implications (CRPI), 5, 52. https://doi.org/
10.1186/s41235-020-00246-1
Spence, C. (2022). Exploring group differences in the crossmodal correspondences.
Multisensory Research, 35(6), 495–536.
Spence, C. (2023). Explaining visual shape-taste crossmodal correspondences.
Multisensory Research, 36, 313–345. https://doi.org/10.1163/22134808-bja10096
Spence, C., & Di Stefano, N. (2022). Coloured hearing, colour music, colour organs, and
the search for perceptually meaningful correspondences between colour and pitch. i-
Perception, 13(3), 1–42. https://doi.org/10.1177/20416695221092802
Spence, C., & Levitan, C. A. (2021). Explaining crossmodal correspondences between
colours and tastes. i-Perception, 12(3), 20416695211018223.
Spence, C., Motoki, K., & Petit, O. (2022). Factors inuencing the visual appeal of food:
From energy-density to the aesthetics of gastroporn. Food Quality & Preference, 102,
Article 104672. https://doi.org/10.1016/j.foodqual.2022.104672
Spence, C., Smith, B., & Auvray, M. (2015). Confusing tastes and avours. In D. Stokes,
M. Matthen, & S. Biggs (Eds.), Perception and its modalities (pp. 247–274). Oxford,
UK: Oxford University Press.
Stevenson, R. J. (2009). The psychology of avour. Oxford, UK: Oxford University Press.
Stevenson, R. J. (2014). Flavor binding: Its nature and cause. Psychological Bulletin, 140
(2), 487–510.
Sun, J., & Gauthier, I. (2023). Does food recognition depend on color? Psychonomic
Bulletin & Review, 1–11.
Taveras-Cruz, Y., He, J., & Eskew, R. T. (2022). Visual psychophysics: Luminance and
color. In N. Santhi & M. Spitschan (Eds.), Progress in Brain Research (1
st
ed., Vol. 273,
Issue 1, pp. 231-256). Elsevier.
Tomasik-Krotki, J., & Strojny, J. (2008). Scaling of sensory impressions. Journal of
Sensory Studies, 23, 251–266.
Turoman, N., Velasco, C., Chen, Y. C., Huang, P. C., & Spence, C. (2018). Symmetry and
its role in the crossmodal correspondence between shape and taste. Attention,
Perception, & Psychophysics, 80, 738–751.
van Giesen, R. I., & de Hooge, I. E. (2019). Too ugly, but I love its shape: Reducing food
waste of suboptimal products with authenticity (and sustainability) positioning. Food
Quality and Preference, 75, 249–259.
Velasco, C., Michel, C., Youssef, J., Gamez, X., Cheok, A. D., & Spence, C. (2016).
Colour–taste correspondences: Designing food experiences to meet expectations or to
surprise. International Journal of Food Design, 1(2), 83–102.
Velasco, C., & Spence, C. (Eds.). (2018). Multisensory packaging: Designing new product
experiences. Springer.
Velasco, C., Woods, A. T., Deroy, O., & Spence, C. (2015). Hedonic mediation of the
crossmodal correspondence between taste and shape. Food Quality and Preference, 41,
151–158.
Walker, P. (2016). Cross-sensory correspondences: A theoretical framework and their
relevance to music. Psychomusicology: Music, Mind, and Brain, 26(2), 103–116.
Wan, X., Woods, A. T., van den Bosch, J., McKenzie, K. J., Velasco, C., & Spence, C.
(2014). Crosscultural differences in crossmodal correspondences between tastes and
visual features. Frontiers in Psychology: Cognition, 5, 1365.
C. Velasco et al.
