While reading Barthes’ essay The rhetoric of the image (1), I came across the following passage which left me intrigued:
“…all images are polysemous; they imply, underlying their signifiers, a “floating chain” of signifieds, the reader able to choose some and ignore others. Polysemy poses a question of meaning and this question always comes through as a dysfunction, even if this dysfunction is recuperated by society as a tragic (silent, God provides no possibility of choosing between signs) or a poetic (the panic “shudder of meaning” of the Ancient Greeks) game; in the cinema itself, traumatic images are bound up with an uncertainty (an anxiety) concerning the meaning of objects or attitudes. Hence in every society various techniques are developed intended to fix the floating chain of signifieds in such a way as to counter the terror of uncertain signs; the linguistic message is one of these techniques.” (2)
To the extent that an image can have multiple interpretations, Barthes seems to be suggesting that in the absence of any clues as to what the intention of the artist is, it is likely that the infinite possibilities of meaning that could be attributed to such photographs would end up causing uncertainty and even anxiety on the viewer. It is possible that such uncertainty, and this is my own hypothesis, would cause the viewer to reject, dismiss or ignore an image without clues on meaning.
I tried to test this hypothesis in a crude way. My intention was not do something very rigorous about this, but just to see if a simple test would yield results in the direction that I was expecting. In the same essay, Barthes suggest that nowadays images are nearly always accompanied by textual references, and that such references can be a way of anchoring meaning in images in such a way as to reduce the anxiety caused by their polyseism (3).
In this experiment, I took seven images in black and white and stripped them of nearly all metadata (only the copyright information was left, buried into the exif data, to identify the images as mine). The images I used for this experiment were a mixture of quasi-abstracts and every day objects taken under poor lighting conditions (eg at night, using either moonlight or diffused / direct street light. It would have been possible to identify most of the objects in the images, but not necessarily why an image of them had been taken. The images are shown below:
For this experiment, I created two new accounts in 500px.com, a social media website for photography. In this website, every new picture uploaded into the public profile goes through various streams that are visible to every visitor of the site. After posting, an image first goes into the “fresh” stream, and if it receives a sufficiently high score (given by the number of favourites / likes or comments left by other users), it will then move to a second stream called “upcoming” and, if it continues to receive likes, to a third stream called “popular”. Within each stream, pictures will move up and down in the order in relative terms to each other (ie the images with the highest number of likes and comments will stay at the top of each stream and consequently, be more visible to users. The advantages of this system (for the purposes of my experiment) is that the images posted were guaranteed to be visible to a large number of users, as opposed to other types of social media where the visibility of content would be a function of the number of followers or friends that a particular user may have.
The first account that I created had no personal details of myself, no name, no visible email address and no picture profile. My first and last names were substituted by just a couple of underscores. Other than these two signs, there were no other letters in my profile, which looked like this:
I then proceeded to upload to this new account each of the aforementioned seven pictures. In addition to being stripped of any information that could give clues as to how they were taken, they did not contain any location information, title, caption or description and they were not placed in any specific photographic category or gallery. All images were posted around the same time in the morning, for seven consecutive days. When clicking on a particular image, the viewer would see no textual clues accompanying the image, as shown in the example below
After uploading the seven pictures, I created a second account in the same site, this time filling part of the profile information, including my name and location, as well as adding a picture profile. I then posted again the 7 images, one at a time every day for a week, but on this occasion I left all the image metadata intact, including keywords, camera, lens and exposure details, as well as the location where each picture was taken. I also gave a title to each image and classified them into a specific category. When clicking the picture, the viewer would find something like this
The idea I wanted to test was if by adding more contextual information about a picture, including the name of its author, title, description, location where taken, etc, the reaction from viewers was more positive, as measured by the number of likes, comments and favourites that the images got. I am also conscious of the possibility that the additional information added to each image may have enhanced its discoverability (ie, ability of a person looking for that type of image to find it), so for each image I have also calculated the ratio of likes to views. All things being equal, whan I am trying to measure is whether more people liked the image with additional information after viewing it, compared with the pure image without contextual information. Here is a summary of the results:
|Total number of likes||Ratio of likes per views|
|Image||Without additional context||With additional context||Without additional context||With additional context|
The results were slightly surprising. It is true that when adding more contextual information to each picture the absolute number of likes per image was noticeably higher than without the extra details, almost double the average number of likes per picture, but when measuring the percentage of likes against the total number of views per images, the results essentially inverted and the number of likes per view was significantly higher in the cases where the image was presented without any additional contextual information, than when additional information was given. A greater percentage – nearly 1 in 3 – of those who viewed the images without contextual data liked the images, than those who viewed with the full context added, where only 1 every 6 people liked the images.
This outcome could be down to chance, but it may also be a reflection of how people look at images online these days. When I do it myself, I primarily look for something that hooks me and if that happens, I keep looking a the image and may like it. It is possible that most people who were attracted to the images reacted in the same way. In this case, it is also likely that context has only helped to increase the exposure of the image (hence the larger amount of views) without necessarily providing any enhancement to its attractiveness.
Many of the images had clear shapes on them to guide the viewer as to what they were looking at. I feel now that I would like to repeat this experiment by using only abstract images, without determined shapes, to make it more difficult for the viewer to relate to them; to measure if the reactions are the same with and without contextual information.
(1) Barthes, R., 1977. Image, Music, Text. HarperCollins UK. Pp 32-51
(2) Idem, Pp 38-39
(3) Idem, p 39