Six adult pigeons (Columba livia) of local homing stock served in this experiment, though not all pigeons were used in all phases of the experiment. They were housed in individual cages (40 × 40 × 45 cm) in a well-ventilated room that was kept on a 12–12 light–dark cycle and were maintained at 85% of their free-feeding weights. Water was freely available during the experimental sessions.

We attached a horizontal conditioning platform (10 cm wide × 5 cm deep × 1 cm thick, solid steel) to the standard home cages of the various pigeons. The subjects had access to it through a wall opening in the cage that normally gave access to the food trough (Fig 1). During the experimental sessions, the platform was indirectly lit through a window with ~300 lux daylight through a photometer controlled blind. A solenoid-operated dispenser [13], affixed 15 cm above the platform, delivered three to six millet seeds through tubing into a 3 cm diameter receptacle glued onto one side of the platform. A 2.5 cm diameter impact-sensitive capsule was glued on the other side of the platform with its center 4 cm away from that of the capsule; a 2.5 cm diameter piezo buzzer was attached to the underside of the platform. Affixed on top of the capsule was a 2 mm thick white plastic disc in the center of which was a 3 mm diameter light-emitting diode. The disc-plus-diode unit was interchangeable so that in successive experimental phases we employed a green (565 nm), a red (660 nm), an infrared (810 nm), a deeper infrared (880 nm), a blue (480 nm), and an ultraviolet (375 nm) light-emitting diode. The diodes were connected through a 100 Ω resistor to a precision adjustable gain amplifier. It was used to set the on-state radiances of the diodes according to their respective voltage-radiance calibrations (see Acknowledgments and later specifications). The amplifier was supplied by one or the other of two voltage-gated precision square-wave, adjustable pulse-width, voltage-controlled frequency oscillators. The voltage outputs of the amplifier were monitored with a calibrated oscilloscope. A personal computer with interface cards served to program the stimulus presentations and to register the pecking responses of the subjects.

The pigeons were auto-shaped to peck the green 565 nm diode. Six second periods during which it was lit alternated with 20 s intervals when it was unlit. Two seconds before the lit periods ended, a reward was automatically delivered; if the bird pecked during the first 4 s of the lit periods, the reward was delivered immediately. Pecks during the unlit intervals had no consequence. A session consisted of 50 such lit–unlit cycles and there were 2 sessions per day. As soon as the pigeons began to peck regularly, the automatic rewards were discontinued and the lit periods were reduced to 4 s. The number of pecks necessary for a reward delivery was gradually increased from one to eight over several sessions. Moreover, while the former lit periods were converted into S+ periods that, at 200 Hz, would certainly be perceived by pigeons as non-flickering [11,12], the previously unlit periods were turned into lit, S- periods that flickered at 30 Hz. Both S+ and S- periods were reduced stepwise to 4 s in duration, interposing 1 s unlit intervals between the successive lit trials. The 1-to-1 on–off flickering of both S+ and S- ensured that they were of equal mean radiance. Next the sequencing of the 25 S+ and the 25 S+ trials was quasi-randomized [14]. Each peck issued during an S- period now generated a 50 ms click sound; furthermore these pecks prevented the S- period from terminating before 4 s had elapsed without a peck. Over successive sessions the reinforcement ratio was modified until it was a variable one of three pecks on average (VR3), with a range of one to five pecks. Pecks during positive and negative stimulus periods were separately counted. When 90% or more of the pecks within a session occurred during the S+ periods the birds were deemed ready for the next stage.

Using four of the pigeons that had undergone this pre-training we determined their flicker fusion frequency employing a green 565 nm light-emitting diode set at a light intensity of 3 μW/sr. It was chosen to yield an about equal luminance of 2 log cd/m² as the monitor-displayed patterns in the experiment mentioned above. The session now consisted of 2 runs of 12 blocks of randomly sequenced 100 S+ and 100 S- trials, all of these separated by 1 s intervals. There were two such sessions per day except weekends. Using a similar platform, Xia et al. [13] had found that the visual discrimination learning of pigeons was not adversely affected by a similarly massed trial conditioning procedure. During S+ trials the diode was always driven with 200 Hz pulse trains. Within each staircase run, starting from S- 40 Hz, the frequency of the S- was increased by a 5 Hz step after every 4 blocks of trials until the last 4 blocks of the trials had a 100 Hz S-. One hundred quasi-randomly selected trials were training trials as described earlier, with pecks during S+ being VR2 rewarded and pecks during S- periods trials being penalized with a click and an extension of the stimulus period. The remaining 100 trials of the block were probe trials in which pecks during the first 2 s were separately counted as correct (S+) or incorrect (S-) but had no other consequences while pecks were reinforced as above within the remaining 2 s. A criterion of at least 150 probe pecks per run was enforced: when occasionally a pigeon did not fulfill it, the run was repeated. After 2.5 to maximally 4 sessions per pigeon we had collected 5 runs that conformed to this criterion for each of the 4 pigeons.

Using the same procedure we next proceeded to determine the flicker fusion frequency corresponding to the red 660 nm light-emitting diode set to yield 100 μW/sr. The on-state radiances of this and the other light-emitting diodes to be mentioned was adjusted to yield an about equal pigeon-subjective brightness by referring to the pigeon’s known photopic spectral sensitivity [15]. However, only three pigeons were used in this instance, and only four runs that met the 150-peck criterion were exacted, these demanding 2 to 2.5 sessions per pigeon. Next we did the same using the blue 470 nm diode set to yield 100 μW/sr, and the same three birds. Three sessions were required for each pigeon to meet the minimum-response criterion. In this instance the S- ascended in 7 steps of 5 nm from 30 Hz to 60 Hz.

After completing the earlier fusion frequency determination with the green 565 nm diode, we implemented three supplementary sessions with two of the pigeons. These sessions each consisted of two runs, abbreviated to five blocks of trials going from S- 75 Hz in steps of 5 Hz to 95 Hz, while the S+ continued to be 200 Hz. This examined whether overtraining would improve the discriminative performance threshold levels. Two other pigeons were first run for three additional sessions with the stimuli S- 65 Hz (fixed) and S+ 200 Hz, both set to an 1-to-4 on–off pulse ratio with an on-state radiance intensity increased to 12 μW/sr. The latter adjustment ensured that the mean luminous intensity of the stimuli was about equal to that of the 1-to-1 on–off ratio stimuli used before. Finally, after having determined the frequency fusion of the red 660 nm diode (see above), two pigeons underwent two supplementary sessions with the diode’s radiance increased to 820 μW/sr, and with frequencies set at S+ 200 Hz and S- 60 Hz, both with a 1-to-1 on–off pulse ratio.

Though not particularly relevant to the monitor display of stimuli issue occupying us here, these wavelength ranges are otherwise quite relevant to pigeons [15]. The peck transducer was accordingly furnished first with a near-infrared 810 nm light-emitting diode with the on-state radiance set to 6 mW/sr. Two pigeons first underwent pre-training sessions with S- 20 Hz and S+ 200 Hz stimuli. When they achieved the criterion of 90% pecks correct in that stimulus condition they proceeded to three testing runs fashioned as before but involving seven ascending S- 5 Hz steps from 35 Hz to 70 Hz. The same two pigeons were then trained to the same criterion using a 880 nm diode set at 10 mW/sr and tested in three runs between 15 Hz and 50 Hz, ascending in 5 Hz steps. Finally the same two birds were tested with the ultraviolet 375 nm light-emitting diode with its on-state radiance set to 1 mW/sr. There were five runs with seven S+ steps from 30 Hz and 65 Hz.

Three adult human subjects of normal or corrected vision served as observers. An abbreviated procedure was used to determine their fusion frequencies when viewing the light emitted by the various diodes set at the same on-state radiances as used with the pigeons and at reading distance (~25 cm). In four successively ascending and descending staircase runs, between 30 Hz and 80 Hz at 5 Hz per step they were required to intermittently report whether the stimulus flickered (‘yes’), perhaps flickered (‘maybe’), or did not flicker (‘no’). Having thus determined their approximate fusion frequency they were asked to set their ‘just yes’ (= fusion) frequency by adjusting the frequency setting knob of the oscillator themselves. All this could not usefully be done with the 375 nm ultraviolet and the 810 nm infrared diodes because their light was at most only visible as a very dim violet or red glow to humans. With the well visible diodes, human flicker sensitivity could by the way be much incremented by reducing the ambient lighting and recurring to repeated off-diode fixation saccades (formation of afterimage ‘ribbons’: [16]). Whether pigeons could perhaps have analogously benefited from comparable afterimages and head and eye saccades [17,18] we do not know.

Four pigeons of homing stock kept and treated as in Experiment I were used but no water was available to them during the briefer experimental sessions.

A plywood chamber measuring 32 cm × 26 cm × 26 cm, was equipped with a 2.5 cm diameter transparent key centered 15 cm above the grid floor and recessed 1.5 cm into the back wall. The front wall and roof were made of wire grid. A solenoid feeder fixed outside delivered 6 to 9 millet seeds per activation through a plastic tube to an inside receptacle of 5 cm × 4 cm × 1.5 cm affixed 8 cm below the key. A frosted miniature 5 W house-light placed 6 cm above the key served to illuminate the chamber but no direct light from it fell upon the recessed pecking key. The same monitor (black–white, P4 phosphor) as used by Delius and Lohmann (1985, unpublished observations) was placed so that the center of its screen was located 0.5 cm behind the key. The same home computer used by these experimenters was also employed to generate a light square of 8 × 8 pixels, 5 mm × 5 mm, on a dark background precisely centered behind the pecking key using the monitor’s x-axis and y-axis adjustment potentiometers. This square pattern could be generated either with a left-to-right (= conventional) writing cathode tube beam or alternatively with a right-to-left (= unconventional) writing beam. The writing direction reversal was achieved with a switch-over, two-pole, sound-proofed relay that inverted the supply polarities to the horizontally deflecting cathode tube solenoids. Examined with a 8× magnifying lens–see above–, the normally displayed square revealed a left border light line shading and a right border dark line shading whereas the reverse displayed square exhibited a left border dark line shading and right border light line shading: see Fig 2. Though both were supposed to be identically symmetrical, the square shapes were in reality dissimilarly asymmetric. This was due to the high-frequency response limitations–resulting from unavoidable inductances and capacitances–of the brightness (= intensity, ‘z-axis’) controlling circuit of the cathode ray causing a so-called signal ‘ringing’ at the stimulus borders. Naturally the artifacts affected all vertical symmetric, as well as all asymmetric patterns displayed, but whereas they broke up the symmetry of the former they only added to the asymmetry of the latter. A noisy (‘chattering’) alternating current solenoid driven vane was interposed between the monitor face and the pecking key for 4 s during all inter-trial intervals. This black shutter prevented the bird from seeing the momentarily erratic monitor display caused by the switchover and to drown the faint switch-relay sound. A personal computer equipped with a digital interface card was programmed to control and record all the experimental states and events.

The pigeons were shaped to peck the key displaying the S+ square patterns detailed below. For this they were exposed to a succession of schedules adapted from those used for a similar purpose in Experiment I. Each of the birds underwent 10 to 13 such sessions. During the last three sessions at least, the schedule incorporated a discrimination training between S+ and S-. For two birds the S+ was the left–right generated square and the S- the right–left generated square; for the other two birds the allocation was the reverse.

Sessions now consisted of 30 S+ trials and 30 S- quasi-randomly sequenced trials [14]. The individual trials were separated by 4 s shutter closure and any pecks to the key during these inter-trial intervals led to a reset of the 4 s timer. Whenever required, according to the quasi-random sequencing, the switchover relay was activated or deactivated 1 s after the shutter had been operated. S+ trials lasted 25 s, of which the first 6 s (probe periods, on 20 random trials out of the 30) merely served to count pecks. Once the probe period was over, a grain reward was given with 0.5 probability if at least two pecks had been issued during the probe period. After the probe period, or right from the beginning of the other 10 probe-less trials, any pecks were reinforced on a VR10 schedule, i.e., 10 pecks on average, with a range of 5 to 15, led to reward. S- trials lasted a minimum of 10 s, of which again the first 6 s (probe periods, 20 random trials out of the 30) merely served to count pecks. Once the probe period was over, if two or more than pecks had accumulated during the probe period, there was 1 s time out (house-light off and a beep noise on). After the probe period, or right from the beginning of the other 10 probe-less trials, any pecks were punished on a VR 10 schedule by yielding a 1 s time out and initiating an at least 10 s continuation of the S- trial. This latter condition implied that from the 15th s onwards, peck responses led to a lengthening of the S- trial, thus instituting an extinction-ensuring condition. The discrimination performance was assessed solely on the basis of pecks issued during the probe periods.

Three pigeons of homing stock were kept and treated as described for Experiment I except that they had no access to water during the experimental sessions.

A 33 cm × 34 cm × 33 cm metallic chamber was employed. It had a horizontal 12 cm × 9 cm platform, 5 cm above the floor, set back as an alcove 12 cm high into the chamber’s back wall. The platform incorporated two side-by-side pecking keys, 2.2 cm in diameter, and with their centers 9 cm apart (Fig 3). Visual stimuli were back-projected onto the underside of the translucent keys with an automated slide projector via a front-silvered mirror angled at 45 degrees. The display on the keys was controlled by a pair of rotatory solenoid-driven vane-shutters. On the platform behind the keys were two circular 2 cm dishes into which two solenoid-operated food dispensers could deliver 6 to 10 millet seeds. The chamber was provided with a house-light and the alcove with a reward light [42] The projector used special slides that bore two side-by-side circular 10 mm diameter openings and three 5 mm diameter coding holes along the lower ledge. These openings corresponded to the keys and were furnished with film images of the stimuli that were fixed with transparent sticking tape. With the help of a zoom projection lens the images could be precisely centered on the platform keys. The coding holes could be occluded individually with black masking tape. Three photocells built into the projector registered the slide coding. The experiment was controlled by a personal computer furnished with an interface card and programmed in PSYCHOBASIC [3]. Note that the conditioning procedures employed here were not intended to emulate those of Delius and Nowak [1], who used a successive rather than simultaneous discrimination procedure, but were instead based on programs that had more recently proven effective in our laboratory.

Three adult pigeons of local homing stock were deprived to 85% bodyweight and were shaped to peck the keys for food reward. Incidentally, a second group of pigeons was to have been trained and tested analogously with asymmetric patterns positive and symmetric patterns negative, but this undertaking was cut short by the intervening laboratory closure (see Introduction). Twice-daily sessions were run throughout the experiment. A symmetric pattern–two different alternative symmetric stimuli being randomly used–was projected alternatingly on the right or left key for 8 s, the other key remaining dark. If the key bearing the pattern was pecked during the stimulus period, six to eight millet grains were immediately dropped into the receptacle next to it; but in any case, a millet reward was automatically issued into that same receptacle at the end of the stimulus period. The issue of rewards was invariably accompanied by a 3 s reward light onset. A pause of 20 s followed during which both keys remained dark and any key pecks yielded a 50 ms beep sound. Sessions consisted of 60 such cycles. After six such sessions the automatic rewards were discontinued, only pecks yielded reward, and the stimulus presentation gradually reduced to 5 s over the next 6 sessions.

During discrimination training, pairs of stimuli were presented on each trial. An S+ symmetric pattern and an S- asymmetric pattern were allocated to right and left keys on a quasi-random basis [14]. The S+ and S- patterns constituting a given pair were quasi-randomly selected from those used during pre-training, which are those shown in the upper row of Fig 3. These stimuli stayed on until the pigeon pecked one of the two keys. If the key bearing the S+ pattern was pecked, a reward was delivered into the receptacle next to it, followed by a 5 s pause with no stimuli. If the key bearing S- was pecked, then this led to the beep sound and a timeout period with the house-light off for 5 s; if the dark key was pecked, this extended the timeout by another 3 s; the next trial was then a repeat correction trial. The inter-trial intervals were 5 s pauses with no stimuli displayed on the keys and there were now 80 trials per session. Any pecks at a dark key also yielded the beep sound. The reward schedule for pecks to the S+ was gradually increased to VR3 (range 1–5 pecks) while the set of patterns was increased by adding two symmetric–asymmetric pairs of patterns per session. The latter, although different from them, were in the same style as the ones shown in Fig 3 and were borrowed from Delius and Nowak (cf. Fig 1 in [1]). The increase of the pattern set continued until the pigeons were routinely dealing with a total of 24 different patterns, 12 of each kind, which were randomly rearranged into new pairs before every second session. When a pigeon completed 2 consecutive sessions with 80% or more of its pecks at the S+ symmetric patterns at this stage, it proceeded to the next stage.

Within each of the subsequent sessions there were 16 out of the 80 trials in which pecks had no consequence for the pigeons. Eight of these trials included the unreinforced presentation of an old symmetric–asymmetric pair of patterns (dummy tests); the other eight trials involved the unreinforced presentation of a novel symmetric–asymmetric pair of patterns (true tests). The pecks delivered during the individual trials of these two types of tests were separately counted as correct (directed at S+) or incorrect (directed at S-). Any trial that yielded more S+ than S- pecks counted as a correct trial; any trial that yielded equal numbers (only a few such trials occurred) or more S- than S+ pecks counted as an incorrect trial. Any test that yielded more correct than incorrect trials counted as correct. Any test that yielded as many (i.e., four of each) or more incorrect than correct trials, counted as incorrect. The purpose of the dummy test pairs was simply to reduce the chance that pigeons would come to associate non-reinforcement with stimulus pattern novelty, which might have depressed test pecking rates. A session always began with 10 training trials and ended with 10 training trials. The 16 test trials were inserted quasi-randomly among the remaining 44 training trials.

In a total of 34 such sessions we successively introduced one by one: (a) 8 novel test pairs, (b) another 4 such pairs, and (c) a further 4 novel test pairs. Series (c) was special insofar that it incorporated a coding reversal of all the symmetric–asymmetric stimulus slides used in the relevant sessions accompanied by corresponding computer reprogramming: the exercise was meant to control the possibility that the optical slide coding was somehow a cue that was accessible to the pigeons (see General Discussion below). It was followed by sessions involving the introduction of (d) six novel test pairs incorporating oblique and vertical axis symmetry patterns, (e) six novel test pairs of piled bar design patterns, and finally (f) six test pairs involving novel colored symmetric and asymmetric patterns. After a novel pattern had served as a test pattern during one session, it was routinely added to the set of training patterns used in subsequent sessions so that towards the end of the experiment, i.e., during the color test (f), the pigeons were dealing with a total of 52 different white-on-black symmetric and asymmetric stimuli.