Stimulus Generalization

Psychophysical discrimination experiments can estimate the smallest physical difference between two stimuli that is detectable. Many of the important facts and concepts of hearing have been obtained from discrimination experiments on many vertebrate animals (reviewed by Fay, 1988; 1992a). Discrimination thresholds and their functional relationships to various acoustic variables essentially reveal the limits of resolution of auditory systems and may suggest the underlying processing and decision mechanisms. But psychoacoustical discrimination thresholds do not necessarily tell us about the nature of an organism's the sense of hearing. For human listeners, for example, a frequency discrimination threshold seems related to the sense of pitch which we experience in daily life, since a frequency change can lead to a perceived change in pitch. However, pitch is defined independently by our daily experiences and using scaling methods. Some of these methods involve pitch matching or other paradigms in which listeners are asked to judge the perceptual distance (similarity and dissimilarity) between stimuli that may differ in spectrum, time waveform, envelope, etc. These judgments are used to define and describe the dimensions of pitch perception.

In investigating the neural mechanisms of hearing, we use animal models for physiological and anatomical studies. In order to help determine how physiological responses may be used in synthesizing perceptions, we use psychophysical measures as well. However, discrimination thresholds do not necessarily tell us whether a species studied has perceptual dimensions that correspond to our own (e.g., pitch, loudness, timbre, noisiness, etc.), and thus do not fully address the question of the relationships between physiological responses and the sense of hearing that they may serve. In order to address these and other fundamental questions in hearing, we need to use additional behavioral methods that estimate perceptual distances or similarities among stimuli for the species studied. In this way we may be better able to use the results of physiological and anatomical studies on animal models to better understand the mechanisms of human hearing.

One behavioral phenomenon that has been used in this way is termed "stimulus generalization." First described and interpreted by Pavlov (1927) using classical conditioning, stimulus generalization has been studied extensively (Mostofsky, 1965). In general, stimulus generalization is "the behavioral fact that a conditioned response formed to one stimulus may also be elicited by other stimuli which have not been used in the course of conditioning" (Hilgard and Marquis, 1940).

In a generalization experiment, animals are conditioned to a given simple or complex stimulus, and then tested to novel stimuli that may or may not have elements or dimensions in common with the conditioning stimulus. If the animal does not respond to the same degree to the test stimuli (i.e., if it fails to generalize), it can be concluded that there is an essential difference between the perceptions evoked by the conditioning and test stimuli. In common language, one might say in this case that conditioning and test sounds "sound different." A monotonic gradient of response along a particular acoustic dimension further suggests the existence of a corresponding perceptual dimension (Guttman, 1963), and the form of the generalization gradient may reveal quantitative relations between the acoustic and internal dimensions (Shepard, 1965).

We have applied these methods to study the sense of hearing in goldfish using classical, or Pavlovian conditioning. In the simplest experiments, animals are classically conditioned to suppress respiration in response to a tone of given frequency and then tested for response to equally detectable tones of novel frequencies. Figure 1 shows some typical, trial-by-trial results. In each case, the conditioning frequency evoked the largest response (respiratory suppression). The suppression response to novel tone frequencies declines monotonically with increasing frequency differences between the conditioning and test stimuli. Since the conditioning is not differential, or does not explicitly train the animal to make distinctions among frequencies, the generalization results can be interpreted as revealing the operation of natural criteria or a perceptual dimension that is evoked by the initial conditioning and then used in evaluating novel stimuli presented in the same context.

Figure 2 shows the results for several experiments in which the conditioning stimulus was a pure tone of given frequency (indicated for each group by the vertical line), and then tested for generalization to novel pure tones with frequencies above and below the conditioning frequency (Fay, 1970, 1992b). Symmetrical, monotonic gradients are obtained in every case. In these experiments, goldfish behave as if they had a perceptual dimension similar to pure tone pitch as it has been described for human listeners.

So far, our generalization experiments have suggested the existence of perceptual dimensions not unlike pure-tone pitch (Fay, 1970; 1992b), periodicity pitch and roughness (Fay, 1972; 1994), and timbre (Fay, 1995). Figure 3 shows the results from a periodicity pitch experiment. Groups of animals were conditioned to a band-pass pulse repeated at a given rate (indicated by the vertical lines above each gradient) and then tested for generalization to novel pulse rates (Fay, 1994). In this experiment, the center frequency and bandwidth of the test stimuli did not differ from that of the conditioning stimulus. Again, symmetrical, monotonic gradients are obtained. Here, goldfish behave as if they had a perceptual dimension similar to periodicity pitch as it has been described for human listeners.

Figure 4 shows results from a set of experiments indicating that goldfish simultaneously acquire information about the repetition rate and the spectral envelope of pulsed sounds (Fay, 1995). In panel A, for example, animals were conditioned to a pulse having low-frequency energy repeated at about 20 pulses per sec. They were then tested for generalization to the same pulse repeated at different rates, and to a high-frequency pulse repeated at the same set of rates. Generalization gradients for both pulse types decline as the test pulse rate rises above the conditioning rate, but there is less generalization, overall, to the pulse with its spectral envelope in a higher frequency range. The other panels of Fig. 4 show that the same pattern of results is obtained for conditioning to different pulse rates (compare panels A and B), and for conditioning to the alternative pulse type (compare top panels with bottom panels). These results indicate that goldfish behave as if they have perceptual dimensions similar to those of periodicity pitch and timbre as described for human listeners.

Figure 4.Generalization gradients for 4 group (N=4) in pitch and timbre exp. Example: In panel A, animals conditioned to the low-frequency pulse (center frequency 238 Hz), and then tested for generalization to the same pulse repeated at different rates (abscissa), or to the high-frequency pulse (center frequency 625 Hz) also repeated at various rates.
The low-frequency pulse is best shown as inset in panel A.
This high-frequency pulse is shown as inset in panel C.

More recently, generalization experiments have demonstrated that goldfish and human listeners share the perception that tone bursts having slow rise times and rapid decays ("ramped sinusoids") have more in common with pure tones than the same tone bursts played backward ("damped sinusoids") (Fay et al, 1996). A neural coding explanation derived from a computational model of early processing in the human auditory system (the "Auditory Image Model" of Patterson et al., 1992) received support from physiological measurements of the neural representations of these stimuli in the goldfish auditory nerve.

In general, results from generalization experiments have shown that goldfish order stimuli in accord with particular physical dimensions (e.g., repetition rate, spectral location, envelope dynamics), in essentially the same ways that human listeners do. Regardless of how we wish to label these sorts of sensory behaviors, it now seems likely that they are a shared and probably primitive feature of the way that many, if not all, vertebrates perceive acoustic events.
Our next generalization experiments on goldfish will focus on the pitch perceptions evoked by iterated ripple noise in human listeners (Yost, 1996), and the time-domain neural processing that is hypothesized to underlie them (Yost et al., 1996).

Fay, R.R. (1970) Auditory frequency generalization in the goldfish. J. Exp. Anal. Behav. 14, 353-360.
Fay, R.R. (1972) Perception of amplitude modulated auditory signals by the goldfish. J. Acoust. Soc. Amer. 52, 660-666.
Fay, R.R. (1988) Hearing in Vertebrates: A Psychophysics Databook. Hill-Fay Associates, Winnetka, IL.
Fay, R.R. (1992a). Structure and function in sound discrimination among vertebrates. In: The Evolutionary Biology of Hearing, Webster, D., Fay, R., and Popper, A., eds., Springer-Verlag, New York, pp. 229-263.
Fay, R.R. (1992b) Analytic listening by the goldfish. Hear. Res. 59, 101-107.
Fay, R.R. (1994) Perception of temporal acoustic patterns by the goldfish (Carassius auratus). Hear. Res. 76, 158-172.
Fay, R.R. (1995) Perception of spectrally and temporally complex sounds by the goldfish (Carassius auratus). Hear. Res. 89, 146-154.
Fay, R.R., Chronopolous, M. and Patterson, R. (1996) The sound of a sinusoid: Perception and neural representations in the goldfish (Carassius auratus). Auditory Neuroscience, 2, 377-392.
Guttman, N. (1963) Laws of behavior and facts of perception. In: S. Koch (Ed.) Psychology: A Study of a Science vol. 5. McGraw-Hill, New York.
Hilgard, E., and Marquis, D. (1940) Conditioning and Learning. New York: Appleton-Century.
Mostofsksy, D. (1965) Stimulus Generalization. Stanford University Press, Stanford, CA.
Patterson, R.D., Robinson, K., Holdsworth, J.W., McKeown, D., Zhang, C., and Allerhand, M. (1992). Complex sounds and auditory images. In: Auditory Physiology and Perception, Cazals, Y., Demany, L., and Horner, K., eds., Pergammon Press.
Pavlov, I.P. (1927) Conditioned Reflexes. translated by G.V. Anrep. London: Oxford University Press.
Shepard, R.N. (1965) Approximation to uniform gradients of generalization by monotone transformations of scale. In: D. Mostofsky (Ed.) Stimulus Generalization. Stanford University Press, Stanford CA.
Yost, W.A. (1996) Pitch of iterated rippled noise. J. Acoust. Soc. Am. 100, 511-518.
Yost, W.A., Patterson, R., Sheft, S. (1996) A time-domain description for the pitch strength of iterated rippled noise. J. Acoust. Soc. Am., 99, 1066-1078.

Stimulus Generalization and the Sense of Hearing in Goldfish