Jody Hey Evolutionary GeneticsProfessor - Department of Genetics - Rutgers University |
|
Words and Distinction
- Jody Hey
Results
Fig 1.
I computer program was written to count up distinct words and generate histograms of word frequency (i.e. # of words used just once, twice etc., throughout each complete sample group).. Despite lower total word counts (2669 for group B compared to 3063 for group N), the number of distinct words (used at least once) was considerably higher in the B group (624 for the B group compared to 383 for the N group). These numbers are totals, and not adjusted for the difference in the number of responses in each group (45 for B, 41 for N).. The difference between the two frequency distributions shows a clear pattern, with the low frequency classes having markedly higher counts in the B group than in the N group, while the N group was consistently higher in the highest frequency classes. The difference in the singleton class (words that appeared only once) is especially striking, see Fig 2. The Shannon-Wiener diversity index for group N was 4.57, while it was 5.27 for group B.
Fig 2. The purpose of this little study was to generate some kind of foothold for thinking about the ways that scientists give descriptions to things, and how these descriptions vary with the blurriness of their subject. Language is an inherently digital medium of communication. Whether written or spoken, words are made up of members drawn from a distinct finite set of subunits (letters or characters, and phonemes respectively). Also it is easy to think of noun-like words has having some kind of one-to-one correspondence with their referents. Thus I began this study with an a priori guess - a wrong one I see now. I supposed that the N picture was more ordered and simpler - certainly it was easier to draw - and that respondents would more easily find words to describe it. Thus I expected shorter responses in the N group. Instead the responses for the B group had a lower mean length. The main effect I had not foreseen was the efficient use of metaphor by the B group respondents. Almost every description appealed to another image that the respondent supposed was likely to be familiar to a reader. Here are some examples:
… Looking down on two hills, or two piles of snow, with a valley in between them and plains around them … …someone's nude bottom placed on a Xerox machine… … A car coming towards you. Two headlights on a dark, foggy night… ... the molecular force field of a molecule of sodium chloride in solution ... Sometimes multiple metaphors were employed: …somewhat dumbbell shaped . . . like a gourd. The only other thing I can recall from memory that this looks like is that of an early stage of a budding event for some fungus . . .
I think the use of metaphor largely explains the
frequency distribution difference (Fig 2), as there were many different
metaphors and even among similar ones, the choice of adjectives varied
considerably.
The two groups of respondents also differed in
the extent to which they referred explicitly to their perceptions of the
graphic. In particular, several in the B group began with something like
"I see ...". Below is a table listing counts for a few words that
were used in reference to the respondents perception
The reason for the wide variance among the N group responses is not clear. Reading through them the impression I received was that these respondents approached their descriptions as if they perceived it to be an orderly task that one could pursue with any level of precision. Thus in response to the instructions (.. give a reader who has not seen the picture a fairly good mental image .. ) the variance in the descriptions seems to reflect variance in the detail that respondents saw as necessary to meet the requested criteria. Here are fairly typical short and longer descriptions from this group: This picture shows two white shapes on a black background. In the lower left is a square. In the upper right is an oval, with the longer axis along the horizontal. The upper right corner of the square is nearly touching the left edge of the oval. This is an image of two white figures on a black rectangular background, the long axis of which is oriented horizontally. The length of the rectangle is about 50 percent greater than the width. A white ellipse, with its long axis parallel to that of the rectangular background, lies near the upper right corner; its length and width (along the major and minor axes) are about 60 to 70 percent as long as the length and width of the rectangle. A white square lies near the lower left corner of the rectangle. Its upper right corner abuts the ellipse at a point slightly below the left terminus of the ellipse's long axis. But this explanation of variance among the N group does not by itself explain the difference between the two groups. Why did not the blurry picture respondents also react with similar variance? My best guess extends from the difference in the use of metaphors. Once a B group respondent had committed them self to finding a metaphor, they seemed not to pursue precision with the same vigor - nor variance in vigor - as did those in the N group. This small study probably does not merit too much interpretation. I will add a couple comments about my sense of the results and what it seems the respondents have done to generate them. This sense is that the individuals in the two groups of respondents felt differently about their goal. The N group seemed to be addressing a straightforward task in an orderly manner, though with lots of variation on how in depth that task should be. The B group responses seemed to suggest that the writers had struggled a bit for a description, and then, having found something that would serve, avoided dwelling on more detail. An important issue that this study does not address is how successful have the two groups been in generating "a fairly good mental image" of the graphics. For most of the N group descriptions I don't doubt that a reader could draw a figure that held the basics of the original. But this may be true of the B group as well. As any poet knows, metaphor is efficient and powerful stuff. An interesting followup study would be to ask some people to draw diagrams on the basis of some of the texts of the respondents (and ask a different group to judge the correspondence of the new diagrams with the originals). Another question that a psychologist might ask is 'what did the respondents think they were doing?'. In my original email to people I said very little about the purpose. Academics think they know a test when they see one, and so the responses may have been different if the instructions had been, say, more positively misleading about the real goal. Many respondents did convey, in their description of the graphic or in ancillary text, that they wondered what kind of test they were taking. But I am not too bothered by this, as most scientists also value clever, pithy explanations in the course of regular scientific discourse. Here are a couple quick thoughts on the design of the experiment. The 'treatment' can be seen as no-blurring/blurring. With one caveat, I think this aspect of the study was simple and reasonable. My major concern is that not everybody's' viewers worked the same way or generated diagrams of the same size. Probably this is not a cause of the patterns that I have described. For example, the non-blurry diagram, which was probably more easily drawn by most viewers, is the image that generated the higher variance. Also, there were very few responses that referred to the absolute size of the figure, or to colors that were not black white or gray. The sampling scheme is a bit more questionable. The population from which the respondents came is peculiar (my email list), and the responses were self selected. Only a minority of folks responded, of course. Quite a number of email addresses had expired or were out of date in some way, and many people did not bother to respond. However it is encouraging that nearly equal numbers responded in the two groups, suggesting that there is not a strong interaction between treatment (B or N) and the likelihood of responding.
If anyone knows of useful references on topics related to those addressed here, please send them to me. If anyone has comments on the study, that I or
others would be interested in, please send them to me.
The data is available in the following files. Non-Blurry picture word counts If anyone is interested in completing the experiment
please let me know. By that I mean if they foresee an oppurtunity
to assess how well people draw graphics using the descriptions in the data
files, and how well these graphics match the originals, then I would
be quite interested.
Thanks very much to all those people who responded. Jody Hey
References Williamson T. (1994) Vagueness . London : Routledge Quine, W.V.O. (1960) Word & Object. John Wiley & Sons, NY.
To email comments, click here hey@biology.rutgers.edu ã 1999 Jody Hey |
