1. INTRODUCTION
A high proportion of animals must actively move about their environment in order to feed, breed and rest. It follows that knowledge of its current position relative to other important locations and of the directions among locations provides a selective advantage to an animal as such information is critical to survival and reproduction. Movements can occur over distances ranging from a few centimetres to thousands of kilometres and animals employ a wide variety of sensory cues to find their way over these different scales.
Over short distances, orientation behaviours are usually guided by sensory cues directly associated with the animal's goal. When, however, moving over greater distances animals have to utilize more complex orientation mechanisms. Under these circumstances, many animals rely on magnetic and/or celestial compass systems to decide on a direction for movement. True navigation, as seen in homing pigeons and adult migratory birds, is the most sophisticated form of orientation behaviour and requires the ability to determine both position and direction. A map is hypothesised to be used to determine the animal's position in relation to its goal, to update position during movement and compensate for possible displacement due to winds, currents, obstacles or other disturbances in the course taken by the animal. A compass is then thought to translate the animal's position into a direction for movement. Although the compass step is relatively well understood, there is little agreement on the sensory cues used to determine position in true navigation. Visual, acoustic, olfactory, and magnetic cues all have been implicated in position determination, with research currently being focused on the latter two mechanisms.
Orientation and navigation behaviours have traditionally been studied in the laboratory by placing animals in a circular orientation arena or maze, within which they are free to orientate in the presence of one or more sensory cues. These studies depend on the animal spontaneously producing orientation behaviour in the presence of the experimental cue(s) as well as the critical assumption that the behaviour produced accurately reflects what the animals would be doing in the field. When, however, an animal fails to respond in an orientation experiment to either a stimulus or a change in stimulus, it is not clear whether the animal either failed to perceive the stimulus, or perceived the stimulus but chose to ignore it. Even if there is a response, its strength may be influenced by the animal's current motivational state to perform such a behaviour. The difference between the inability to detect a sensory stimulus and failure to respond when the stimulus is detected particularly becomes an issue when detailed studies of sensory function (e.g., measurement of threshold sensitivity) are attempted.
In field studies of orientation and navigation mechanisms, it is usually only possible to manipulate the input from one external stimulus at a time. This leaves open the possibility that other external stimuli may provide navigational information which the animals will use instead of the experimental stimulus. In much the same way as in laboratory orientation experiments, it is thus difficult to know whether failure of an animal to respond in the predicted manner in the field is due to a failure to use the experimental stimulus in the manner predicted or the animal switching to other stimuli that provide equivalent information about position and/or direction. Removal of one or more of these alternative stimuli through sensory impairment may, however, to lead to failure to produce orientation behaviour at all. As a consequence, it is very difficult to determine whether the use of particular stimuli is in fact necessary or not necessary for successful navigation.
Conditioning is a laboratory-based technique that controls both the sensory stimuli available to the animal and the animal's motivation to perform the behavioural task in response to the stimuli. Conditioning techniques have been employed by psychologists for more than half a century to study sensory perception as well as the processes underlying learning. In recent years there has been increasing recognition that the control of motivation to respond presents an opportunity to increase the power of laboratory experiments investigating perception of external stimuli that might be used in navigation. This review seeks to facilitate that development by providing a general introduction to conditioning and considerations involved in designing robust conditioning experiments. We thus hope to bridge the gap resulting from the historically limited interaction between biologists and psychologists, which has lead over time to a divergence in terminology and experimental practices. Furthermore, we will provide some current perspectives for the application of conditioning in the field of animal orientation.
2. CONDITIONING AS A TECHNIQUE
Conditioning, which records behavioural or neural responses in intact and awake organisms in response to sensory stimuli, is a technique commonly used in both animal psychology and animal psychophysics. Animal psychology (sometimes also called comparative psychology) studies the process of learning, with learning defined as a change in behaviour as the result of experience with sensory stimuli in an experimental situation. In contrast, the emphasis in animal psychophysics lies with the sensory input itself, the aim being the analysis of sensory perception unaffected by motivation or payoff. Thus, while sensory systems are frequently utilized in animal psychology as a window into the brain to explore the general processes underlying learning, animal psychophysics uses what is known about learning to study the properties of the window itself, such as measures of detectability and sensory thresholds.
2.1. Classical/Pavlovian Conditioning
The first conditioning technique was pioneered by the Russian physiologist I.P. Pavlov (1849–1936). He noticed that dogs would not only salivate reflexively when receiving appropriate stimulation through food in their mouth, but the salivation reflex was also elicited when they saw food or a familiar food dish, or heard the steps of the caretaker. That is, they would salivate in anticipation of food, provided they had previous experience with the external stimulus (the food type, the food dish, or the caretaker) prior to being fed. Thus, an association formed between the food and the external stimulus as a consequence of the external stimulus reliably predicting the delivery of food.
Pavlov found that he could form a new conditioned response from an existing reflex by pairing a previously neutral stimulus (i.e., a stimulus that had until then no observable behavioural consequence) with a stimulus that already elicited a response. During initial tests, Pavlov's dogs were presented with the sound of a bell or a flashing light as the to-be-conditioned stimulus (CS), which elicited an orientating response (OR) in form of head lifting or pricking of ears. The sound or light then was immediately followed by the presentation of food (or in a related set of experiments, injection of a weak solution of acid into the dog's mouth) as the unconditioned stimulus (US). The US would evoke an unconditioned response (UR) – seeing the food, or tasting the acid, would cause the dog to salivate. As such presentations were repeated, the sound of the bell or the flashing light eventually no longer elicited an OS with the dog having become accustomed to the presence of the CS. The tone or light (CS) now immediately preceded the food or acid (US) with the US still causing salivation (UR). After several such conditioning sessions, the tone or light (CS) on its own caused the dog to salivate during the testing phase, the salivation reflex now being termed a conditioned response (CR). Thus, control of the salivation response had been transferred from the food or acid stimulus to the tone or light stimulus.
Classical conditioning provided the first reliable and objective method for systematically measuring sensory thresholds by varying some property or dimension of the conditioned stimulus until it no longer elicited the conditioned response. While animal psychophysicists eventually became aware of several limitations to this technique, it still plays an important role in the analysis of animal learning to this day. The responses most commonly used are heart rate, respiration rate, skin resistance, eye blink, and glandular secretion, while the most frequently used unconditioned stimuli are food in appetitive conditioning or mild electric shocks as painful/unpleasant stimulation in aversive conditioning.
2.2. Operant/Instrumental Conditioning
In 1911, E.L. Thorndike (1874–1949), an animal psychologist interested in the associative processes underlying animal learning, began to investigate what he termed trial-and-error-learning. A hungry animal, such as a cat, was locked in a puzzle box with food outside the box being visible to the animal inside. To escape and obtain the food, the animal had to learn through trial and error to open the door of the box by operating a catch, pressing a lever, or pulling a string. The time until the subject managed to escape the box decreased rapidly with increasing experience with the problem. Thorndike was thus the first researcher to investigate how an animal's non-reflexive behaviour could be modified as the result of an animal's experience.
In the 1930s, B.F. Skinner (1904–1993) further developed Thorndike's ideas of studying animal learning by training his animals in a small experimental chamber, where he increased the frequency of a behaviour through response-contingent reinforcement. In the chamber, known as the Skinner box, the experimental subjects had to learn to operate a manipulandum or mechanical device (e.g., to press a lever (rats) or to peck a key (pigeons)), to obtain a reward (e.g., food or water) or to avoid aversive stimulation (e.g., mild electric shock, air blast, or a time penalty during which access to food is denied). Skinner's animals learned to perform a response when it was followed by a positive reinforcer and/or when it avoided a negative reinforcer, in either case using as the only cue(s) the sensory stimulus or stimuli presented in the experimental situation.
This form of conditioning is called operant or instrumental conditioning, named so because the subject operates on its environment and its behaviour is instrumental in obtaining/avoiding the reinforcer/punisher. That is, in operant conditioning the reinforcer will occur in the presence of a specific stimulus if, and only if, the conditioned response occurs. Hence, the delivery of the reinforcement in operant conditioning is no longer controlled by the experimenter as it was in classical conditioning, but depends entirely on the experimental subject's performance of the designated response (stimulus→response→reinforcement, termed a three-term contingency by Skinner).
Today, the most commonly used apparatus for operant conditioning is still the Skinner box, a small, sound- and light-insulated chamber. It usually contains a system to dispense food rewards and/or deliver aversive stimuli, devices to provide the desired sensory stimulation, and a manipulandum (e.g., pecking key, lever bar, treadle, or paddle) for recording responses by the animal. Operant conditioning procedures are generally fully automated to provide precisely timed control of complex stimulus and reinforcement sequences while at the same time reducing variability in the recorded data by avoiding cueing or observer bias. Cueing (Clever Hans phenomenon) occurs when an animal relies on extraneous cues provided by the observer or the experimental situation rather than the stimuli being tested. Observer bias concerns the issue of subconscious reporting or preferring of favourable over unfavourable responses by the observer. Furthermore, automation circumvents time-consuming and potentially stressful handling of the animal.
Conditioning experiments are preceded by several training stages during which the animal learns the response and becomes accustomed to the experimental situation. First, the animal learns to feed from the device delivering the food reward. As this is most commonly a magazine dispensing food pellets (rats) or grain (pigeons), this step is often referred to as magazine training. Food delivery occurs at irregular intervals to ensure that the animal is always performing a different behaviour when the food is given to avoid any chance association between an undesired behaviour and the food reward that might increase the frequency of this undesired behaviour. During the next training phase, response shaping or response acquisition, the animal learns to perform the designated instrumental response with any behaviour directed towards, or in the vicinity of, the response-recording device (manipulandum) being rewarded. Eventually, in a technique known as successive approximations, only responses that are increasingly similar to the ultimately desired response are positively reinforced. Regular, immediate, and frequent reinforcement is important at this stage of training. Most commonly used responses are lever pressing (rats) and key pecking (pigeons), but other species have been trained to perform a range of tasks (e.g., lever-pulling, panel-pressing, objects-grasping, pulling of strings attached to objects or doors, swimming through hoops or digging through sand). Animals are usually motivated by being deprived of the substance used as the reward (e.g., 80–85% free-feeding body weight or limited access to water).
2.3. Procedural Aspects of Operant Conditioning
Once the desired instrumental response is performed reliably, the subject's performance is measured. In discrete-trial conditioning, the frequency or latency of the emitted responses is recorded during a specified testing period. This period is often indicated by a ready signal such as a trial light, and limited by the subject emitting a certain number of responses or by a predetermined time interval. During the inter-trial interval (ITI), the experimental chamber or the pecking key is often darkened, or the manipulandum temporarily removed. The length of the ITI is either fixed or based on a geometric mean. Geometric means, in contrast to arithmetic means, ensure uniform probability of when the next trial will start thus maintaining a uniform level of the arousal in the animal. In free-operant conditioning, the manipulandum/manipulanda is/are always available to the animal, so that responses may be emitted freely and continually, with the rate of responding over time being of interest. In choice procedures, a variant on discrete-trial or free-operant procedures characterized by more than one response being available, the experimenter looks for a change in the proportion by which one alternative is chosen over the other available alternative(s) or for the percentage of correct choices made.
In general, operant conditioning is said to have occurred when the subject's behaviour changes as a consequence of the contingencies of reinforcement. That is, the animal performs the rewarded response and/or refrains from performing the punished response as a consequence of experience with the systematic relationship between the presented stimulus/stimuli, the response, and the predetermined reinforcing consequence. Behaviour can be increased by the response either being followed by an appetitive event (positive reinforcement), or it being followed by the absence of an aversive event (negative reinforcement). Correspondingly, behaviour will decrease, if a response is either followed by an aversive event (positive punishment), or if it is followed by the absence of a positive event (negative punishment, also called omission training). Thus, the term reinforcement refers to behaviour increase, and punishment to behaviour decrease, while the terms positive and negative refer respectively to the contingency of producing an event or removing an event.
Positive and negative reinforcers are usually delivered according to a particular schedule of reinforcement, the purpose of which is to control the level of the animal's motivation to respond throughout the session. During the initial phase of response acquisition, a continuous reinforcement schedule (CRF) is normally used, where every correct response is reinforced. This generally produces a rapid increase in response rate, but has the disadvantage of fast satiation and associated loss of motivation. These effects reduce the amount of training or testing that can be done during a conditioning session. In addition, behaviourists, such as Skinner, are interested in the effects that different schedules of reinforcement have on behaviour and learning. Therefore, once responding is established, a partial reinforcement schedule is used, which reinforces only a portion of the correct responses.
There are several types of partial reinforcement schedules. In simple reinforcement schedules only one schedule is in effect in a session. Several simple schedules can be combined within a session to produce complex reinforcement schedules. The most commonly used simple schedules are ratio schedules, in which the next reinforcer is delivered on the basis of the number of responses emitted, and interval schedules, in which a response is reinforced on the basis of elapsed time. On a fixed-ratio (FR) schedule a response is reinforced when a specified number of responses has been emitted since the last reinforcer (e.g., FR 10=every 10th response reinforced), while on a variable-ratio (VR) schedule the number of responses required varies within a given range around a specified average value (e.g., VR 10=on average every 10th response reinforced). Correspondingly, on fixed-interval or variable-interval schedules, a reinforcer is delivered for the first response respectively after a fixed time interval (e.g., FI 20 s=reinforcement of first response after every 20-s interval) or after a variable time interval, which comes from a distribution varying around a mean (e.g., VI 20 s=reinforcement of first response after average 20-s interval).
2.4. Conditioned Discrimination
In conditioned discrimination procedures differential responsiveness to different stimuli is studied by utilizing either classical or operant conditioning approaches. This approach is based on the principle that an animal is only able to respond differently to different sensory stimuli, if it is able to detect these stimuli and discriminate among them, and if the contingencies of reinforcement differ between the stimuli.
Pavlov's research on classical conditioning was used by early animal psychophysicists to determine sensory thresholds in differentiation experiments. Typically two stimuli were presented repeatedly with one of the stimuli being paired, and the other not paired, with the unconditioned stimulus. However, the development of operant conditioning procedures opened up more ways of controlling behaviour through manipulations of its consequences than was possible with the relatively inflexible behaviour utilized in classical conditioning. In operant conditioning, the probability of the occurrence of a response to one stimulus is high if it has been associated with positive reinforcement in the past, but will be low in the presence of another stimulus formerly associated with no reinforcement or punishment. Thus, differential responding can be established under conditions of differential reinforcement in the presence of several different stimuli. Since differential behaviour is only possible if the animal has the sensory capacity to discriminate between the stimuli associated with the differential responses, this approach is frequently applied to assess the sensory capacities of different animal species. As the most important and widely used technique of modern animal psychology, it is also used to investigate the processes underlying discriminative learning.
There are two types of conditioned discrimination techniques based on the principles of operant conditioning. The first is the unitary, go/no-go, single-response, or single-choice procedure, during which the animal learns through positive and/or negative reinforcement to perform a single instrumental response or to operate a single manipulandum in the presence of one stimulus (S+), and to refrain from performing the response in the presence of a second, different stimulus (S−). S+ and S− are called discriminative stimuli because they are associated with differing reinforcement contingencies. The absence of a stimulus (background stimulation), can also serve as either the S+ or S− stimulus. For free-operant schedules, the measure of discrimination performance is either the latency of change between responding and not responding after stimulus change or the difference in time spent responding to the S+ stimulus compared to S−. For discrete-trial go/no-go procedures, the response rates to the S+ and S− stimuli are compared with the response frequency being greater to the stimulus associated with the reward. Occasionally, the percentage of trials on which response is correctly emitted is measured.
There are two methods for establishing unitary discrimination:
• a baseline response rate is developed with reinforcement in the presence of the future S+ stimulus, after which a second stimulus (S−) is introduced to signal the absence of a reward (extinction) or even punishment; or
• equal reinforcers are given in the presence of both S+ and S− stimuli, after which reinforcement is discontinued (and possibly even punishment introduced) in S− trials.
In both cases, if the stimuli are discriminated by the animal, it will come to respond almost exclusively to the S+ presentations. The procedure is able to give a graded, ordinal measure of the degree to which the animal discriminates the two stimuli. While this procedure frequently suffers from a response bias (animals are more likely to produce than to withhold a response), it is certainly sufficient to demonstrate that animals can discriminate stimuli.
The second type of conditioned discrimination procedure is the matching-to-sample (MTS) procedure, which may either be symbolic (SMTS) or non-symbolic (MTS) with the choice stimuli, respectively, being different from or the same as the stimuli being assessed. This procedure is a choice procedure (sometimes also referred to as a two-choice procedure to distinguish it from the single choice made in the unitary approach described above). Its symbolic variant is also used in psychophysical research as the Yes-No signal-detection procedure. Generally, in this procedure one of two stimuli to be discriminated is presented at the start of a trial. An observing response to the presented stimulus produces a choice between two other responses, signalled by two stimuli (e.g., red and green illuminated pecking keys), or maybe just in the form of a left and right manipulanda. One of the available responses will provide a reinforcer (a correct response, typically reinforced with food or water), while the other will not (an incorrect response or error, typically punished by a time-out, during which no food/water can be obtained, or even by an aversive event such as a mild electric shock or airblast). There are many variations on this procedure – for example, the stimulus may be left on during the choice phase or be extinguished after the observing response.
This type of procedure provides a 2×2 matrix of data (correct and error responses following each of the two signalling stimuli). Correct and error responses may be combined to give a single measure of percentage correct, or may be analyzed using signal-detection theory to provide two independent measures, one being stimulus detectability (d') or discriminability (log d), the other being a measure of the degree to which the reinforcers affect responding to the two choices (criterion or β in signal-detection theory, or response bias in behavioural approaches to signal detection). Detection measures, rather than percentage (−) correct measures, are preferred for the detailed analysis of animal and human sensory psychophysics because percentage-correct measures of stimulus effects are often seriously contaminated by response bias. However, as with the unitary procedure (which can also be subjected to signal-detection analysis), the demonstration of control by a stimulus dimension (rather than the psychophysics of a stimulus dimension) can be generally convincingly achieved using just percentage-correct measures.
While the method of stimulus presentation in unitary procedures is, by definition, successive, during choice procedures the discriminative stimuli can be presented either successively or simultaneously. In the successive situation, one of several discriminative stimuli is presented in isolation, whereas in the simultaneous situation two or more discriminative stimuli are displayed at the same time, each usually in close proximity to a manipulandum (e.g., two pecking keys, one illuminated red and the other green). In the simultaneous case, responses to the S+ manipulandum are rewarded while responses to any manipulandum associated with S− are unreinforced or punished (e.g., red light signals rewarded stimulus→press lever which is associated with red light).
For both successive and simultaneous presentations in choice procedures, the stimuli will be discriminated by the animal, as evident by the number of correct choices rising above chance level, if the stimuli presented are discriminable and if the reinforcement conditions provide sufficient motivation for a differential response. The reinforcers have two essential functions:
• to maintain the overall detection performance, and
• to provide differential reinforcement with respect to the signaling stimuli.
Detection procedures usually, but not invariably, maximize the latter by arranging reinforcers only for correct responses, and extinction (and often blackout punishment) for errors. If equal numbers of reinforcers were arranged for both correct and error responses following both signalling stimuli, an animal would show no discrimination between the two stimuli even though it was able to discriminate the stimuli excellently when errors were not reinforced.
In both unitary and choice procedures, the stimuli to be discriminated are presented in a non-predictable order to prevent the animal from using the order of stimulus presentation as an alternative cue for determining the correct response. Frequently quasi-random sequences are applied to mitigate the effects of the long sequences of the same trial type that are possible when there are only two trial types presented in random order. In full psychophysical choice procedures, the relative frequencies of presentation of the stimuli (stimulus-presentation probability, SPP) or the probability of reinforcers for the two correct responses are often varied across experimental conditions. This produces a receiver-operating characteristic curve (ROC) or a pair of generalized-matching lines to characterize fully the animal's sensory detection system.
In choice experiments, the conditioned response is normally performed only once per stimulus presentation (e.g., making the choice of which arm of a T-maze to enter, which door to jump at in a jumping stand, or which of two choice keys to peck). Other designs can allow multiple responses per stimulus presentation (e.g., key pecking and lever pressing) measuring discrimination performance either on free-operant or discrete-trial schedules analogous to the unitary approach. Measuring multiple rather than single responses can provide additional information about a tendency to respond, which is useful, for example, when investigating whether responses are made more frequently or more rapidly under some conditions compared to others.
Both unitary and choice conditioning procedures have an initial pre-training phase, during which the animal learns to obtain the reward (e.g., food) in the novel experimental situation. This is followed by a phase of response acquisition where the conditioned response(s) is/are learned in the presence of the appropriate stimuli. Once the conditioned response(s) is/are reliably emitted, actual discrimination training begins with the discriminative stimuli being associated with the particular reinforcement contingencies. During the pre-solution period, i.e., before any stimulus control appears, animals usually do not behave randomly, but rather tend toward behavioural predispositions usually related to naturally occurring behaviours. In addition, whenever animals are presented with multiple manipulanda simultaneously, subjects often develop position habits, persistently choosing one manipulandum over the other irrespective of stimulus presentation. Thus, discrimination training is at least partially a process of encouraging the subject to abandon such undesirable behavioural patterns, and to attend to the stimuli that signal the reinforcement contingencies. Discrimination learning is said to be occurring when the discriminative stimuli come to control the animal's behaviour differentially and is complete once the stimuli consistently control different reactions that are quantitatively stable, with non-responding also being one possible response.
In learning to discriminate between two or more stimuli, the animal has to attend to the relevant aspects of the stimuli and to respond appropriately to these differences. Three main factors influence the speed of learning and the level of the animal's discrimination performance. The first factor is the discriminability of the stimuli presented (stimulus differential), i.e., how close they are to one another on one or more stimulus dimensions. Generally, the smaller the difference between the discriminative stimuli and the closer the difference is to the animal's species-specific difference threshold for the sensory system involved, the more difficult the discrimination task for the animal. Small stimulus differentials are learned more slowly, and asymptotic measures of performance (e.g., d' or log d) are lower.
Secondly, as mentioned above, the degree of differential reinforcement also plays a vital role in bringing the behaviour under stimulus control. This is because it provides the incentive for the animal to respond differentially to the different stimuli presented. Consequently, increasing the delay between stimulus and choice (delayed matching-to-sample, a memory paradigm) or between choice and reinforcement (delay of reinforcement), separates the discriminative stimuli, the choice responses, and the reinforcers in time (or space due to experimental design), and thus tends to slow discrimination learning, and lowers discrimination performance. However, decreased motivation to respond (through satiation) generally does not affect measures of discrimination in choice procedures, though it may affect unitary discrimination measures.
Thirdly, the method of stimulus presentation can also be of importance. Simultaneous presentations are generally more effective than successive presentations of stimuli; possibly because the former is less dependent on memory (the animal is able to compare the alternative stimuli at the same time). This is particularly true if the conditioned response is made in close spatial association with the stimulus (e.g., lighted pecking key or symbol on jumping door) as, under these conditions, the simultaneous method produces more accurate discrimination performance. On the other hand, if the response is removed from the locus of stimulation, the successive method tends to produce better discrimination performance.
As far as the similarity of the discriminative stimuli is concerned, simultaneous procedures have been shown to produce higher levels of percentage correct, d' (or log d), than successive procedures for difficult discrimination tasks. However, when stimuli are easy to discriminate, there is no statistical difference between these two methods. Naturally, whether simultaneous presentation is suitable, or even possible, depends on the physical characteristics of the stimulus dimension investigated – visual stimuli, for example, lend themselves easily to simultaneous presentation while magnetic stimuli do not. Also, learning a difficult discrimination can be facilitated by first learning an easy, but related, discrimination where the stimuli are more distant from one another along the stimulus dimension.
2.5. Equipotentiality Premise
Pavlov believed that any cue the animal could sense could be used as a conditioning stimulus for any reflex response and Skinner stated that any animal could be conditioned to perform any behaviour of which it was physically capable. This idea that the experimenter's choice of stimuli, responses, reinforcement, and subject species is relatively unimportant for the purpose of conditioning (equipotentiality premise) was a critical assumption in animal psychology for many decades. However, it is now accepted that this was too broad a generalization and that there are serious biological constraints on the learning of conditioned responses. That is, even though most likely some general learning principles underpin all conditioning, the choice of stimulus, response, and reinforcer may significantly affect the ease and speed with which conditioned learning occurs, and are thus critical to success for a given species.
3. CONDITIONING AS AN APPROACH TO ORIENTATION BEHAVIOUR
The homing pigeon is one of the model species in both animal psychology and animal orientation. It is thus not surprising that it was the species of choice for early conditioning experiments investigating the sensory cues involved in animal orientation. For example, pigeons were successfully trained in classical conditioning experiments to discriminate between a linearly polarized light source with a rotating or a stationary axis of polarization (Kreithen and Keeton, Reference Kreithen and Keeton1974). They also learned to detect sounds as low as 0·05 Hz (infrasounds) (Kreithen and Quine, Reference Kreithen and Quine1979). In contrast, numerous attempts made to condition pigeons to a variety of magnetic cues (review: Wiltschko and Wiltschko, Reference Wiltschko and Wiltschko1995) produced only negative results or data too weak to be convincing. These repeated failures were at odds with the discovery of a magnetic compass in pigeons and migratory birds (and consequently were used to argue against the possibility of magnetic map navigation (review: Wiltschko and Wiltschko, Reference Wiltschko and Wiltschko1995).
Magnetic conditioning studies with a variety of other species have since then demonstrated sensitivity to magnetic fields in both invertebrates and vertebrates (honey bee: Walker and Bitterman, Reference Walker and Bitterman1985; yellow-fin tuna: Walker, Reference Walker1984; rainbow trout: Walker et al., Reference Walker, Diebel, Haugh, Pankhurst, Montgomery and Green1997; Haugh and Walker, Reference Haugh and Walker1998; short-tailed stingray: Walker et al., Reference Walker, Diebel, Kirschvink, Collin and Marshall2003). They also revealed some unexpected requirements to be critical for magnetic conditioning to succeed. In particular, conditioning was only achieved when the magnetic stimuli used were spatially distinctive and the behavioural response required movement by the animal.
Based on this discovery, a recent conditioning study with pigeons, which for the first time simultaneously fulfilled both requirements with this species, successfully trained the birds to discriminate the presence and absence of a spatially-variable magnetic intensity anomaly within an experimental chamber (Mora et al., Reference Mora, Davison, Wild and Walker2004). The pigeons were required to spend time walking between two feeder platforms, during which they were exposed to the presence or absence of spatial variations in the magnetic field within the chamber. They then had to choose the feeder associated with the field present in each trial.
Magnetic conditioning is thus an example of how the choice of stimulus, response, and reinforcer is crucial for the animal to be able to learn the conditioning task and consequently an example of why the equipotentiality premise does not hold. This consideration is of particular importance to orientation researchers. It may not always be possible to utilize model species typically used in animal psychology because they may not have been demonstrated to possess the sensory systems of interest for a given orientation behaviour (e.g., magnetic and electric senses). Thus conditioning protocols developed over decades, for example, for pigeons with standard visual or auditory cues will have to be carefully adapted to fit the type of sensory stimulus presented as well as the biology of the particular species investigated. A species' biology has to be especially carefully considered when choosing the behavioural response the animal is required to perform. For example, while a goldfish (Walker and Bitterman, Reference Walker and Bitterman1986) is able to protrude its mouth to press a button, rainbow trout are not able to do so, but are suited to striking a target with their snout (Walker et al., Reference Walker, Diebel, Haugh, Pankhurst, Montgomery and Green1997).
Two additional points should be kept in mind when attempting conditioning studies. Firstly, conditioning data have to be collected and presented in such a manner as to demonstrate learning and to show therefore that stimulus control has been achieved. Conditioning is either based on reflex behaviour (classical conditioning) or on learned behaviour (operant conditioning) and it is important to determine just which of these two categories applies to the response to be used. During classical conditioning, reflex behaviour consists of an autonomic reaction (e.g., salivation or heart rate change) which is then linked to a stimulus whereas learning during operant conditioning is defined as a change in behaviour as the result of experience with the positive and/or negative reinforcement contingencies of the experimental situation. In both cases, successful conditioning is demonstrated as a change in the measure of behaviour (e.g., latency to eye blink or key press) over time. Ideally the effectiveness of a conditioning protocol is tested with a stimulus the animal is known to perceive before attempting to demonstrate that the conditioned response can be brought under the control of the stimulus of interest.
In the light of these requirements, it is difficult to conclude that a response has been conditioned or trained in the absence of data demonstrating a change in behaviour over time (e.g., Muheim et al., Reference Muheim, Edgar, Sloan and Phillips2006a; Freire et al., Reference Freire, Munro, Rogers, Wiltschko and Wiltschko2005; Voss et al., Reference Voss, Keary and Bischof2007; Gegear et al., Reference Gegear, Casselman, Waddell and Reppert2008). A recent study using groups of fruitflies to discriminate magnetic field stimuli (Gregear et al., 2008), also raises the potentially problematic issues of demonstrating conditioned learning when:
• the conditioned behaviour may not be independent of any probabilities of aggregation,
• the effect(s) of social facilitation in large groups is/are unknown, and
• the appropriate choice of sampling unit for replication (i.e., whole group versus individual animals) may not be clear.
Secondly, presentation of the data should illustrate acquisition of the conditioned response over time rather than totalling results of correct and incorrect choices over the entire experimental series (e.g., Freire et al., Reference Freire, Munro, Rogers, Wiltschko and Wiltschko2005; Voss et al., Reference Voss, Keary and Bischof2007). The acquisition phase, besides being the crucial evidence of learning, may reveal important differences between stimulus settings in terms of the speed or strength of the conditioning achieved. Analyzing changes in response behaviour over time is particularly useful in threshold studies. This is because it provides a congruent criterion for recognizing acquisition as well as demonstrating recovery of discrimination performance as the stimulus magnitude is progressively decreased. Simple summing of correct responses is not only unsuitable as such a criterion, but is also highly inefficient requiring many more trials than necessary to demonstrate successful discrimination.
4. CURRENT PERSPECTIVES ON STUDYING ORIENTATION BEHAVIOUR
Multiple challenges remain in the study of sensory perception during orientation. Firstly, there are sensory systems which are still too poorly understood for investigation based solely on approaches traditionally used by orientation researchers. For example, magnetic sensitivity in animals was proposed in the late 1800s. Yet the existence of a magnetic sense and its use in orientation behaviours was not widely accepted until only a few decades ago and many fundamental questions regarding the mechanisms of magneto-reception remain unsolved. Secondly, stimulus properties relevant to orientation and navigation may not be easily specified or controlled, especially in field situations. It is difficult, for example, to separate changes in intensity from changes in direction when manipulating magnetic fields in experimental situations. Finally, these challenges may be further compounded by a species that seems likely to possess a particular sense, but which is too difficult to study experimentally in the field (e.g., marine mammals). We suggest, therefore, that use of conditioning techniques will be valuable in determining which sensory cues can be perceived by an animal, and thus would be available for orientation tasks, as well as an animal's level of sensitivity to a particular type of stimulus.
Sensitivity is of particular importance when theoretical models of map and compass orientation are developed because studies of sensory limits will identify restrictions at the neurophysiological and consequently behavioural levels and thus determine how realistic are such models. For example, theories of magnetic map navigation have been based on the sensitivity of homing pigeons to magnetic intensity, which has been indirectly estimated from field studies to be in the range of 10–50 nT/km (e.g., Gould, Reference Gould1980; Kirschvink and Gould, Reference Kirschvink and Gould1981). However, such levels of sensitivity have yet to be behaviourally demonstrated in pigeons, or any other vertebrate species, in a controlled experimental situation and it is not clear whether sensitivity thresholds would be the same or similar for different species. Nevertheless, such information is required to test and further refine current iron-based receptor models (e.g., Fleissner et al., Reference Fleissner, Stahl, Thalau, Falkenberg and Fleissner2007a&b; Walker, Reference Walker2008), which predict high sensitivity to magnetic intensity.
The sensitivity of both magnetic and celestial compasses is also currently unknown. Recent publications report that domestic chickens (Freire et al., Reference Freire, Munro, Rogers, Wiltschko and Wiltschko2005) and zebrafinches (Voss et al., Reference Voss, Keary and Bischof2007) are the first species conditioned to discriminate magnetic compass directions. However, clear evidence for the acquisition of the response and thus for stimulus control of the response behaviour (see above) were not presented in these studies. Such data will be required before attempts can be made to establish the smallest change in the vector axis detectable by these birds. Similarly, it is well accepted that the sun compass is time-compensated for the sun's 15°/hour movement across the sky. However, it is not clear how accurately the sun's azimuth, which is used to determine compass direction, can be detected by animals. A similar question of sensitivity arises for night-migratory birds using a star compass, which is based on the detection of the centre of star rotation.
Orientation behaviour based on a polarized light compass is another example where conditioning studies could potentially answer questions related to sensory perception. For this compass, diurnal changes in the pattern of skylight polarization are directly related to the location and movement of the sun (review: Muheim et al., Reference Muheim, Moore and Phillips2006b). However, very little is known about what aspects of polarized light migratory birds and other animals are able to perceive. Perception and thus usefulness of polarized light cues to the animal are likely to be determined by two parameters:
• the level of polarization of the light observed and
• the orientation of the e-vector, i.e., the orientation of the plane of polarization, in relation to the horizon.
Neither the minimum level of polarization required for perception nor the smallest difference in the orientation of the e-vector detectable to migratory birds is currently known. At sunrise and sunset, the orientation of the e-vector is horizontal at the sun's azimuth and vertical at ±90° to the sun's position. Not only does this difference in e-vector orientation decline with altitude above the horizon, but under overcast conditions the highest level of polarized skylight transmitted (5–10%) is found in the area near the horizon (Hegedüs et al., Reference Hegedüs, Åkesson and Horváth2007). Consequently, polarized light near the horizon would provide the clearest and most reliable directional cue for compass calibration, as suggested by Muheim et al. (Reference Muheim, Åkesson and Phillips2007). However, for more targeted experiments, it would be essential to know whether sensitivity to e-vector differences changes with the level of polarization (e.g., 100% versus 70%) and whether polarization sensitivity varies between species. Only then can adequate stimulation be assured during orientation arena experiments.
Detailed conditioning studies of polarized light vision in migratory birds would also likely assist in resolving some recent conflicting findings. A study with Savannah sparrows (Muheim et al., Reference Muheim, Åkesson and Phillips2007) suggested that polarized light cues at sunrise and sunset provide the primary calibration reference for the compass system of migratory birds with polarized light cues near the horizon being particularly important. Contrary findings were then reported for Australian silvereyes (Wiltschko et al., Reference Wiltschko, Munro, Ford and Wiltschko2008a), which led to an intense discussion as to the generality of the Savannah sparrow results (Muheim et al., Reference Muheim, Åkesson and Phillips2008; Wiltschko et al., Reference Wiltschko, Munro, Ford and Wiltschko2008b). One of the potential causes is a difference in methodology with Wiltschko et al. (Reference Wiltschko, Munro, Ford and Wiltschko2008a, Reference Wiltschko, Munro, Ford and Wiltschkob) arguing that Muheim et al. (Reference Muheim, Åkesson and Phillips2007) used unnaturally strong polarized light (100% instead of 70%) for their vertically and horizontally polarized light cues. However, as discussed above, too little is currently known about the birds' sensitivity to such polarization cues to come to a resolution based on the above studies or efficiently design further orientation experiments.
As a final point, conditioning as an experimental technique can also reveal through impairment experiments important information about the location and likely mechanism of the receptors for less-well understood sensory systems (e.g., magnetic sense). Because the animal's motivation to demonstrate its sensory abilities can be controlled, successful impairment of a conditioned task is readily apparent by a drop in discrimination performance to chance level. For example, conditioned discrimination of a magnetic anomaly was successfully impaired in pigeons through attachment of a magnet atop the cere, anaesthetization of the nasal cavity and sectioning of the ophthalmic branch of the trigeminal nerve (Mora et al., Reference Mora, Davison, Wild and Walker2004). Thus, pigeons have a magneto-receptor that is located in the nasal region, with the receptor most likely being iron-based, and transmitting sensory information to the brain via the trigeminal nerve.
5. CONCLUSIONS
In summary, conditioning studies can reveal how animals perceive the world by providing information about sensory systems in terms of existence, sensitivity, location and (indirectly) detection mechanisms. They cannot investigate how sensory information is used by the animal to determine its behavioural output, e.g., how an animal's brain processes sensory cues to decide on orientation behaviour displayed in the field. We therefore suggest that they be used in a complementary fashion to traditional field releases and orientation arena experiments and, wherever possible, in collaboration with animal psychologists to ensure experimental designs that are implemented effectively and efficiently.
