IV. Similarity and
A note on similarity and categorization
A word is in order about the relation of similarity to several terms
that have appeared fairly often in the avian literature including "categorization",
"categorical perception" and "concept". Much has been written
about all of these; a particularly lucid treatment is given by Smith and
Medin (1981). Other chapters in this book introduce the reader to
some of this material in the avian context; see especially chapters by Huber
(2001), Urcuioli, (2001) and by Young &
Categorization is necessary because the world offers a limited number
of behavioral options. Perhaps each seed on the ground is unique,
but a bird has to decide what to do with it. When a bird eats it
we can say that the item has been categorized as "food". From
the present point of view, similarity underlies this categorization;
the bird computes the similarity of the item to previously experienced
(or innate?) representations that have the "food" tag; if the computed
value exceeds a threshold or criterion, the bird eats.
Items that are categorized in this way are apt to be similar
to each other by the measures treated in earlier sections, for example
falling into the same cluster. However, the presence of such groups
does not alone justify our saying that an animal categorizes the grouped
items together. After all, computer programs will produce clusters
from purely random data. Only if an animal does something different
in the presence of members of one cluster than it does in the presence
of members of other clusters does it seem appropriate to use the summary
term “categorization”. For example, the calls of Dooling’s
budgerigars fell into “alarm”, “contact” and other similarity clusters,
but in addition the birds responded in the same way to members of a given
cluster, and this response differed between clusters. Even
if these conditions are met, one need not assume an internal process
that represents a category as such, as opposed, for example, to a
set of stored exemplars that participate in the criterion/decision
process suggested above. (See Smith & Medin (1981) for a discussion
of different representations of categories.)
The case is somewhat different for “categorical perception”, where the
human observer has the distinct impression of perceiving an object
in one way or another, as opposed to behaving in one way or
another to objects that may function similarly but are perceptually distinct.
Examples are provided by reversible figures and phonemic differences (e.g.
the switch from “ba” to “pa”), where at some point during a gradual change
in the physical stimulus (or an attentional shift by the observer) there
is a sudden flip from one appearance to another. There should be
high similarity between items on the same side of the "flip" discontinuity,
and much lower similarity between items on different sides of it.
It is difficult to establish categorical perception in humans; I know of
no established instance in non-human subjects.
According to this analysis, categories are defined by a common behavioral
response, and similarity plays a role in determining the occurrence of
this response. Just what additional meaning, if any, "concept"
adds to "category" for non-verbal species has been controversial.
For example, just because a bird categorizes some items as food, there
seems no reason to say that it "has a concept of food", and
it is difficult to say just what one would have to observe to justify
this statement. To add to the confusion, some writers
appear to use "category" and "concept" interchangeably; more
often, "concept" is used when one is referring to abstract categories,
which cannot be defined in terms of perceptual similarity alone (see, for
example, the discussion by Young & Wasserman, this book, of the "same-different
concept"). The resolution of this matter appears to lie elsewhere
than in a discussion of perceptual similarity, and the reader is invited
to consult the references given. For a specific discussion of criteria
for pigeon concepts, see Lea (1984).
Next Section: References