The evidence for feature detectors of specific signals is extensive, especially in acoustical-vocal systems (see below), which have been extensively
studied from the perspective of natural stimuli. The evidence in other
systems, especially vision, is not as complete, in spite of pioneering work on
the frog visual system (Lettvin et al. 1959; Ewert et al. 1983a). Recent work
suggests that combinations of visual feature detectors (columns) are a part
of the process of object recognition (Tsunoda et al. 2001). Resolution of the
issue of visual feature detectors may await more consistent application of
neuroethological principles to vision research. There is strong suggestive
evidence and considerable dispute, for example, regarding the existence
of feature detectors in the context of visual face recognition in monkeys
(e.g., Fujita et al. 1992; Gross 1992; Wang et al. 2000). Yet, far more of visual
processing than face and hand recognition—in the case of monkeys, for
instance, for fruits and other food sources, classes of predators, and classes
of habitat—may be dominated by specialized mechanisms than is generally
appreciated. Visually guided behaviors for these classes of stimuli provide
logic for searching for corresponding biases in the visual system. The welldocumented referential call system based on classes of predators in vervet
monkeys (Seyfarth et al. 1980a) suggests an interesting parallel that could
guide vision research.
Finally, if neurons in some pathways act as local feature detectors for
certain classes of stimuli, then it becomes important to specify how subsets
of those neurons interact to produce the percept. There is considerable
theoretical work on population coding and the role of synchronous neuronal activity to dynamically “bind” simpler response profiles into global
percepts (Singer and Gray 1995; see deCharms and Zador 2000) and some
computational work that might suggest how simple subfeatures are combined to detect objects (Lee and Seung 1999). This is an area where a neuroethological approach could be quite advantageous but so far has only
begun to be applied.
2.1.2. The Organization of Feature Detectors
In early single-neuron studies of sensory systems, neuroethologists observed specializations of the auditory system related to vocalizations of
interspecific predators and courtship signals (Capranica 1965; Roeder
1966). These observations were interpreted in terms of a hierarchical organizational scheme of sensory systems, with neurons higher in the hierarchy
having more selective responses related directly to behavioral output.
Extensive data, collected from many systems following these early studies,
now support the existence of both hierarchical organization schemes and
feature-detector cells. The salient observations included tuning of peripheral responses to a behaviorally relevant range of parameters and cells at
higher levels of a sensory hierarchy sensitive to particular combinations of
spectral and temporal components of behaviorally relevant sounds. The
7. Neuroethology of Vocal Communication
333
studied from the perspective of natural stimuli. The evidence in other
systems, especially vision, is not as complete, in spite of pioneering work on
the frog visual system (Lettvin et al. 1959; Ewert et al. 1983a). Recent work
suggests that combinations of visual feature detectors (columns) are a part
of the process of object recognition (Tsunoda et al. 2001). Resolution of the
issue of visual feature detectors may await more consistent application of
neuroethological principles to vision research. There is strong suggestive
evidence and considerable dispute, for example, regarding the existence
of feature detectors in the context of visual face recognition in monkeys
(e.g., Fujita et al. 1992; Gross 1992; Wang et al. 2000). Yet, far more of visual
processing than face and hand recognition—in the case of monkeys, for
instance, for fruits and other food sources, classes of predators, and classes
of habitat—may be dominated by specialized mechanisms than is generally
appreciated. Visually guided behaviors for these classes of stimuli provide
logic for searching for corresponding biases in the visual system. The welldocumented referential call system based on classes of predators in vervet
monkeys (Seyfarth et al. 1980a) suggests an interesting parallel that could
guide vision research.
Finally, if neurons in some pathways act as local feature detectors for
certain classes of stimuli, then it becomes important to specify how subsets
of those neurons interact to produce the percept. There is considerable
theoretical work on population coding and the role of synchronous neuronal activity to dynamically “bind” simpler response profiles into global
percepts (Singer and Gray 1995; see deCharms and Zador 2000) and some
computational work that might suggest how simple subfeatures are combined to detect objects (Lee and Seung 1999). This is an area where a neuroethological approach could be quite advantageous but so far has only
begun to be applied.
2.1.2. The Organization of Feature Detectors
In early single-neuron studies of sensory systems, neuroethologists observed specializations of the auditory system related to vocalizations of
interspecific predators and courtship signals (Capranica 1965; Roeder
1966). These observations were interpreted in terms of a hierarchical organizational scheme of sensory systems, with neurons higher in the hierarchy
having more selective responses related directly to behavioral output.
Extensive data, collected from many systems following these early studies,
now support the existence of both hierarchical organization schemes and
feature-detector cells. The salient observations included tuning of peripheral responses to a behaviorally relevant range of parameters and cells at
higher levels of a sensory hierarchy sensitive to particular combinations of
spectral and temporal components of behaviorally relevant sounds. The
7. Neuroethology of Vocal Communication
333