Logan:  Basic Principles of Learning

                    PRINCIPLE: STIMULUS GENERALIZATION

          A response whose probability of occurrence has been increased 
          by any conditioning procedure will also tend to occur in the 
          presence of stimuli similar to that which prevailed at the 
          time of conditioning.

     The Principle of Stimulus Generalization has been observed in every
conditioning context.  For example, during conditioning, a tone could 
precede food delivery in the classical paradigm, could signal the 
availability of food reinforcement for emitting an operant response, or 
could indicate whether running down a runway would result in food reward.
Subsequently, the extent to which the tone stimulus has gained selective 
control over behavior is best measured by changing that stimulus to 
determine whether there is a stimulus generalization decrement.  If the 
response is weaker as a result of a stimulus change, then clearly the 
stimulus was exerting some degree of control over the response.

     There are many factors that affect the amount of generalization.  
Generally speaking, generalization is greater after irregular conditions 
of reinforcement, be that partial reinforcement or a variation in the 
amount, delay, or quality of the reinforcing event.  Occasional punishment 
of a rewarded response also increases generalization.  The implication of 
the Principle of Stimulus Generalization is a gradient showing the amount 
of generalized response tendency as a function of the similarity of a
test stimulus to the training stimulus.  

     Although the empirical Principle of Stimulus Generalization is 
generally accepted and widely employed in various contexts, there is no
compelling theoretical account of the relevant phenomena.  It would be 
simple enough to say that two tones sound somewhat alike, or two visual 
displays look somewhat alike, but it is not at all apparent why irregular
conditions of reinforcement should result in greater generalization.
Specifically, it would have to be assumed that partial reinforcement of 
one tone makes a second tone somehow sound more like the conditioned tone, 
but this is hardly intuitively reasonable.  Quite probably the account 
will ultimately relate to the fact that the factors that influence
generalization also influence persistence in the same way.

       Generalized response tendencies have relatively little persistence; 
extinction with a generalized stimulus proceeds much more rapidly than to
the original stimulus.  Even more important than reinforcement variability 
in increasing generalization and persistence is variation in the stimulus
situation during learning. If a response has been conditioned to a wide
variety of stimuli, it will naturally generalize broadly to an even wider 
range of stimulus situations.  (Note that reinforcement variability may be 
seen as a special case of stimulus variability, and hence persistence may 
be viewed as a special case of generalization to conditions that occur 
during extinction.)

     There are several special cases of stimulus generalization, one of 
which is called CROSS-MODAL generalization.  This refers to a tendency to
respond to a stimulus presented in one sensory modality after learning 
occurred to a stimulus in another modality.  For example, a person might 
learn the morse code of dots and dashes seen as a series of flashing lights, 
and then be able to translate the code heard as a series of clicks.  An
especially important case of cross-modal generalization in human behavior 
is called DYNAMIC TRANSMISSION.  This refers to the tendency to
generalize between objects or events and the words used to refer to them.
It is this fact that makes verbal instructions possible; if you are given
directions to turn left at the next traffic light, you must generalize
from the spoken word "traffic light" to the actual sight of the light.

     As with most principles of learning, dynamic transmission is not
always adaptive.  Some people cringe at the mere mention of the word
"snake," and many people are upset by the use of four-letter words that
are only sounds in the air or marks on a page.  Which is worse, attaching
a label such as "slow," "homely," or "stutterer" may cause the very
outcome being described.

     Stimulus generalization is most apparent in everyday life when it can 
be described as either overgeneralization or undergeneralization.  A student 
who concludes that all mathematics courses are terrifying as a result of 
some early difficulties with arithmetic has persistent overgeneralized
reactions.  A student who does not see the basic Principles of Learning
operating in day-to-day activities reflects an undergeneralization from 
the classroom to the outside world.  The principle invades even our own 
self-image; the "born-loser" syndrome is an overgeneralization based on 
a few unfortunate experiences that may, as a matter of fact, turn into a 
self-fulfilling prophecy.

TERMS: Stimulus control, stimulus generalization decrement,
generalization (cross modal, stimulus), generalization gradient, stimulus
similarity.