Ultrasonic courtship signals, in the same range as the hearing of bats,
increase mating success in both ctenuchid (Sanderford and Conner 1995;
Simmons and Conner 1996) and wax moths (Jang and Greenfield 1996), and
in some ctenuchids the males and females conduct an ultrasonic dialog. It
seems probable that the use of ultrasonics has nothing to do with the precise
message being sent but was a convenient channel to use because it already
existed, albeit for another purpose.
6.4.2. Neural Networks and Response Biases
The examples above illustrate that if an auditory receiver is adapted for one
function, this can bias how it then becomes adapted for another function.
A similar effect could also occur across time but within the same function.
As we have been discussing, long-distance cues are often important components of a species’ mate-recognition system. When an ancestral species
splits into two daughter species, the two daughter species have different
recognition systems: the mate-recognition signals will differ between the
species, and each species will be biased toward responding to the conspecific signal. Thus, at a minimum, one daughter species evolved a new recognition system (i.e., signal plus receiver properties) and the other maintained
the ancestral recognition system, or both diverged from the ancestral signal
and receiver.
There are probably a large number of computational strategies by which
a receiver can bias its response to the conspecific signal. The strategy it
chooses might be dependent on how ancestors of this receiver achieved the
same task. For example, if within a lineage of animals the mate-recognition
signals of species could always be discriminated by signal duration or by a
more subtle multivariate comparison of a multitude of spectral parameters,
we might expect the receiver to be biased toward using temporal parameters for recognition, much as the moths discussed above utilized ultrasonics for communication. This should be true as long as such a strategy
could achieve the task.
Phelps and Ryan (1998) recently addressed this issue of historical biases
of receivers by combining studies of artificial neural networks with their
empirical studies of mate recognition in túngara frogs. Initially, they trained
recurrent artificial neural networks to recognize a túngara frog call in 20
replicate populations. In each population, they retained the network that
best discriminated between the call and noise. They then determined their
responses to a variety of signals, such as heterospecific calls and purported
ancestral calls, with which túngara frogs had been tested. Neither the networks nor the frogs had any previous experience with these signals. The
responses to these signals therefore are considered response biases because
they are incidental rather than being the target of selection. There was a
strong correlation between the response biases of the frogs and the networks. Thus, whatever computational strategies these two systems were
5. Selection on Signals
263
increase mating success in both ctenuchid (Sanderford and Conner 1995;
Simmons and Conner 1996) and wax moths (Jang and Greenfield 1996), and
in some ctenuchids the males and females conduct an ultrasonic dialog. It
seems probable that the use of ultrasonics has nothing to do with the precise
message being sent but was a convenient channel to use because it already
existed, albeit for another purpose.
6.4.2. Neural Networks and Response Biases
The examples above illustrate that if an auditory receiver is adapted for one
function, this can bias how it then becomes adapted for another function.
A similar effect could also occur across time but within the same function.
As we have been discussing, long-distance cues are often important components of a species’ mate-recognition system. When an ancestral species
splits into two daughter species, the two daughter species have different
recognition systems: the mate-recognition signals will differ between the
species, and each species will be biased toward responding to the conspecific signal. Thus, at a minimum, one daughter species evolved a new recognition system (i.e., signal plus receiver properties) and the other maintained
the ancestral recognition system, or both diverged from the ancestral signal
and receiver.
There are probably a large number of computational strategies by which
a receiver can bias its response to the conspecific signal. The strategy it
chooses might be dependent on how ancestors of this receiver achieved the
same task. For example, if within a lineage of animals the mate-recognition
signals of species could always be discriminated by signal duration or by a
more subtle multivariate comparison of a multitude of spectral parameters,
we might expect the receiver to be biased toward using temporal parameters for recognition, much as the moths discussed above utilized ultrasonics for communication. This should be true as long as such a strategy
could achieve the task.
Phelps and Ryan (1998) recently addressed this issue of historical biases
of receivers by combining studies of artificial neural networks with their
empirical studies of mate recognition in túngara frogs. Initially, they trained
recurrent artificial neural networks to recognize a túngara frog call in 20
replicate populations. In each population, they retained the network that
best discriminated between the call and noise. They then determined their
responses to a variety of signals, such as heterospecific calls and purported
ancestral calls, with which túngara frogs had been tested. Neither the networks nor the frogs had any previous experience with these signals. The
responses to these signals therefore are considered response biases because
they are incidental rather than being the target of selection. There was a
strong correlation between the response biases of the frogs and the networks. Thus, whatever computational strategies these two systems were
5. Selection on Signals
263