Experimental conditions
Another experiment was carried out to try to better define the criteria used by the subjects to associate photos with flavors. An example of a response to this test is given above. To complete or improve the approach used in the experiment with the zakouskis, I decided to modify it in the following way:
- zakouskis (whose forms can induce associations) are replaced by beers (‘bières’ in French);
- the proposed beers were first analysed starting from tasting notes given by a renowned zythologist -Christophe Gillard. They were in fact selected from 20 beers analyzed in detail in a book on regional artisanal beers. This allowed me to select a subset of 4 beers that are organized in terms of distance in two homogeneous sub-groups: the beers « Rulles Estivale » and « Simcoe Lager » are two fairly close « blondes » and the Bellevaux Black and the Big Mama Stout two stouts relatively similar ;
- the photos are pictures of dolls (‘poupées’ in French) taken during an exhibition;
- the dolls were also chosen to match the structure of the 4 beers : some pairs of dolls are close, others are more distant.
The 10 selected pictures can be found in the figure above.
Based on the calculated distances between the beers on the one hand and between the photos on the other hand, an optimal fit can be found using my algorithm. In fact, here two associations are equally optimal : the mapping (2,1,3,8) – which associates photo 2 with the Big Mama, 1 with the Simcoe, 3 with the Bellevaux and finally 8 with the Rulles, and the mapping (3,1,2,8) – which by the way uses the same photos but in a different order. The distances between the beers are given in the following table:
Main results
- The choices made by the subjects seem less heterogeneous than in the experiment with the zakouskis and some coupling are identical: associating the photo 5 with the Big Mama or the photo 10 with the Rulles for example was done by several subjects. But even if some photos have been selected primarily for certain beers, the choices made remain substantially different. The observed homogeneity is only slightly higher than would be observed if the associations were random.
- It is also interesting, considering the seed effect observed in the experiment with the zakouskis, to focus on those who in their sequence associate at least once the same beer with the same doll : 6 subjects retained photo 5 for the Big Mama Stout and 8 subjects associated photo 10 with the Rulles. Are the choices of these subjects more homogeneous? The answer is yes, at least that is what is observed on these two small subsets of subjects.
- Do the subjects seek as the mapping device to associate with similar beers, similar photos ? Clearly not all of them : 8 subjects out of 13 seem to respect that principle. And the choices made are quite different from what the algorithm proposes. There are therefore subjects who do not seek or fail to respect the order of distances – similar photos are not associated with similar beers as is done in the search for the optimum – or whose perceptions of similarities between beers and / or between photos do not match what is measured.
- In fact, as for the tests with zakouskis and beers, different logics to carry out the maping tasks seem to be used. The comments made after the tests are enlightening. I was told for instance :”This beer is strong, you could get drunk ; therefore I have associated it with a doll that represents a man a little drunk “. Or, by the same person: “This beer is acidulous, a little feminine; I have associated it with a doll with a dress “. Another tested person seems to work with more abstract associations: “If the beer is light, I look for white dolls, on the other hand if it is strong, I turn rather towards darker evocations, images of intestines which symbolize pain “.