which communication typically occurs. It has been appreciated for some
time that responses to natural vocalizations often cannot be predicted from
responses to simpler stimuli. Nevertheless, there are many fewer experiments in which neuronal responses are recorded to complex sounds presented in the kinds of time-varying, biologically realistic sequences animals
encounter while communicating.
Much research on sensory mechanisms has been guided by the matched
spectral filter or “call-detector” models. Although these models are conceptually appealing and have received experimental support (Gentner and
Margoliash, Chapter 7), they are insufficient to provide a full view of the
mechanisms underlying acoustic communication. Two problems that have
emerged on a behavioral level are that animals listen to and respond to
sounds other than mating calls and that they tolerate a great deal of individual variation in call parameters. In some instances, animals prefer “supernormal” signals, with features such as duration, rate, and intensity that are
outside the normal range of variation found in natural conspecific signals
(Ryan 1990; Ryan and Keddy-Hector 1992). A neural response system
based on sharply tuned mating-call detectors cannot easily account for such
behavioral biases.
Original formulations of call-detector models emphasized detection of
spectral features of sounds to the exclusion of temporal processing mechanisms. Temporal codes based on phase-locked responding by peripheral
neurons or autocorrelation-type mechanisms at more central levels play a
significant role in extracting behaviorally important features of complex
sounds (Schwartz and Simmons 1990; Simmons et al. 2000). Temporal “template” models based on cross-correlation have been examined most extensively in bat echolocation (Simmons et al. 1995). Unfortunately, these
models have not been used extensively in analysis of other communication
systems, even though temporal cues play crucial roles in guiding behavior.
In addition, spectral and temporal features can interact in complex, nonlinear ways (Gerhardt 1992). Understanding the neural coding of signals
that covary in both spectral and temporal properties is still incomplete, but
it presently seems likely that a joint time-and-frequency mechanism may
be common to a variety of different auditory tasks. Computational techniques for estimating complex stimulus–response functions, such as the
spectro-temporal receptive field, have been available for some time
(Aertsen and Johannesma 1981) but are not widely used in the neuroethological literature (Theunissen et al. 2000). More widespread use of this and
other computational techniques should yield additional insights into the
neural code underlying perception of complex sounds. Another challenge
for neuroethology is to derive realistic computational models of central
auditory processing based on these complex neuronal response properties.
Call production and perception are inextricably linked in theories of ultimate causation of communication (Ryan and Kime, Chapter 5). On the
proximate level, the link between production and perception (sensorimo10
A.M. Simmons
time that responses to natural vocalizations often cannot be predicted from
responses to simpler stimuli. Nevertheless, there are many fewer experiments in which neuronal responses are recorded to complex sounds presented in the kinds of time-varying, biologically realistic sequences animals
encounter while communicating.
Much research on sensory mechanisms has been guided by the matched
spectral filter or “call-detector” models. Although these models are conceptually appealing and have received experimental support (Gentner and
Margoliash, Chapter 7), they are insufficient to provide a full view of the
mechanisms underlying acoustic communication. Two problems that have
emerged on a behavioral level are that animals listen to and respond to
sounds other than mating calls and that they tolerate a great deal of individual variation in call parameters. In some instances, animals prefer “supernormal” signals, with features such as duration, rate, and intensity that are
outside the normal range of variation found in natural conspecific signals
(Ryan 1990; Ryan and Keddy-Hector 1992). A neural response system
based on sharply tuned mating-call detectors cannot easily account for such
behavioral biases.
Original formulations of call-detector models emphasized detection of
spectral features of sounds to the exclusion of temporal processing mechanisms. Temporal codes based on phase-locked responding by peripheral
neurons or autocorrelation-type mechanisms at more central levels play a
significant role in extracting behaviorally important features of complex
sounds (Schwartz and Simmons 1990; Simmons et al. 2000). Temporal “template” models based on cross-correlation have been examined most extensively in bat echolocation (Simmons et al. 1995). Unfortunately, these
models have not been used extensively in analysis of other communication
systems, even though temporal cues play crucial roles in guiding behavior.
In addition, spectral and temporal features can interact in complex, nonlinear ways (Gerhardt 1992). Understanding the neural coding of signals
that covary in both spectral and temporal properties is still incomplete, but
it presently seems likely that a joint time-and-frequency mechanism may
be common to a variety of different auditory tasks. Computational techniques for estimating complex stimulus–response functions, such as the
spectro-temporal receptive field, have been available for some time
(Aertsen and Johannesma 1981) but are not widely used in the neuroethological literature (Theunissen et al. 2000). More widespread use of this and
other computational techniques should yield additional insights into the
neural code underlying perception of complex sounds. Another challenge
for neuroethology is to derive realistic computational models of central
auditory processing based on these complex neuronal response properties.
Call production and perception are inextricably linked in theories of ultimate causation of communication (Ryan and Kime, Chapter 5). On the
proximate level, the link between production and perception (sensorimo10
A.M. Simmons