extent of peripheral and central specialization is species- and behaviordependent and, at least in some cases, can also be influenced by sex (e.g.,
Narins and Capranica 1976). It is noteworthy that peripheral and central
specializations were predicted in the vigorous debates of the early ethologists (Lehrman 1953). Obviously, each species must vocalize in a range that
it can hear, but this does not predict the observed peripheral specializations.
For example, in several species of anurans, different frequency ranges are
represented in different peripheral auditory organs (Capranica 1978).
In some species of bats, the important second-harmonic region of the echolocation signal is highly overrepresented in the cochlea (Bruns and
Schmieszek 1980). Peripheral specializations have also been observed in
reptiles and birds (Manley 1990).
It follows that if the evolution of vocal signals can shape the action and
distribution of peripheral receptors to match the vocal signal (or vice versa),
then there are likely to be central effects as well. Indeed, in numerous
systems, central neurons have been observed that respond selectively to
specific spectral and temporal features of conspecific or autogenous (selfproduced) vocalizations or are selective for conspecific vocalizations
within a repertoire of sounds (Leppelsack and Vogt 1976; Mudry et al. 1977;
Suga et al. 1978, 1979; Scheich et al. 1979b; Margoliash 1983; Rose and
Capranica 1983; Rose et al. 1988; Rauschecker et al. 1995; Bodnar and
Bass 1997, 1999; Crawford 1997; see Ewert et al. 1983b; Feng and Schellart
1999). In many cases, the specializations involve processing for spectral
combinations and temporal sequences of sounds. These are reviewed
separately in the next sections.
The feature-detector concept does not require stimulus and response
invariance. Feature-detector cells may exhibit dynamic modulation of
response properties (Zhang et al. 1997), in some cases in response to
changes in behavioral state (Dave et al. 1998a), and more permanent plastic
changes in response profiles (Knudsen 1985; Doupe 1997). These dynamic
properties may complicate the analysis of the feature representations, but
if the dynamic process is lawful, then it does not invalidate the concept of
feature detection (cf. Manley and Müller-Preuss 1978). In the best-studied
cases, the dynamic changes observed have been closely correlated with
changes in behavior.
2.1.2.1. Parallel Hierarchical Pathways
Feature detectors imply hierarchical organization, but there is also extensive evidence that within a sensory system several parallel pathways may
exist, each hierarchically processing different sets of sensory cues associated with different behavioral tasks (e.g., different aspects of recognition or
localization). Such data have provided further support of the featuredetector hypothesis. The highly specific behavioral deficits in human
patients with certain brain lesions are particularly compelling in this regard
334
T.Q. Gentner and D. Margoliash
Narins and Capranica 1976). It is noteworthy that peripheral and central
specializations were predicted in the vigorous debates of the early ethologists (Lehrman 1953). Obviously, each species must vocalize in a range that
it can hear, but this does not predict the observed peripheral specializations.
For example, in several species of anurans, different frequency ranges are
represented in different peripheral auditory organs (Capranica 1978).
In some species of bats, the important second-harmonic region of the echolocation signal is highly overrepresented in the cochlea (Bruns and
Schmieszek 1980). Peripheral specializations have also been observed in
reptiles and birds (Manley 1990).
It follows that if the evolution of vocal signals can shape the action and
distribution of peripheral receptors to match the vocal signal (or vice versa),
then there are likely to be central effects as well. Indeed, in numerous
systems, central neurons have been observed that respond selectively to
specific spectral and temporal features of conspecific or autogenous (selfproduced) vocalizations or are selective for conspecific vocalizations
within a repertoire of sounds (Leppelsack and Vogt 1976; Mudry et al. 1977;
Suga et al. 1978, 1979; Scheich et al. 1979b; Margoliash 1983; Rose and
Capranica 1983; Rose et al. 1988; Rauschecker et al. 1995; Bodnar and
Bass 1997, 1999; Crawford 1997; see Ewert et al. 1983b; Feng and Schellart
1999). In many cases, the specializations involve processing for spectral
combinations and temporal sequences of sounds. These are reviewed
separately in the next sections.
The feature-detector concept does not require stimulus and response
invariance. Feature-detector cells may exhibit dynamic modulation of
response properties (Zhang et al. 1997), in some cases in response to
changes in behavioral state (Dave et al. 1998a), and more permanent plastic
changes in response profiles (Knudsen 1985; Doupe 1997). These dynamic
properties may complicate the analysis of the feature representations, but
if the dynamic process is lawful, then it does not invalidate the concept of
feature detection (cf. Manley and Müller-Preuss 1978). In the best-studied
cases, the dynamic changes observed have been closely correlated with
changes in behavior.
2.1.2.1. Parallel Hierarchical Pathways
Feature detectors imply hierarchical organization, but there is also extensive evidence that within a sensory system several parallel pathways may
exist, each hierarchically processing different sets of sensory cues associated with different behavioral tasks (e.g., different aspects of recognition or
localization). Such data have provided further support of the featuredetector hypothesis. The highly specific behavioral deficits in human
patients with certain brain lesions are particularly compelling in this regard
334
T.Q. Gentner and D. Margoliash