Animal Cognition

, Volume 16, Issue 5, pp 737–753

Pigeons learn virtual patterned-string problems in a computerized touch screen environment

Authors

    • Department of PsychologyThe University of Iowa
  • Yasuo Nagasaka
    • Laboratory for Adaptive IntelligenceBrain Science Institute, RIKEN
  • Leyre Castro
    • Department of PsychologyThe University of Iowa
  • Stephen J. Brzykcy
    • Department of PsychologyThe University of Iowa
Original Paper

DOI: 10.1007/s10071-013-0608-0

Cite this article as:
Wasserman, E.A., Nagasaka, Y., Castro, L. et al. Anim Cogn (2013) 16: 737. doi:10.1007/s10071-013-0608-0

Abstract

For many decades, developmental and comparative psychologists have used a variety of string tasks to assess the perceptual and cognitive capabilities of human children of different ages and different species of nonhuman animals. The most important and widely used of these problems are patterned-string tasks, in which the organism is shown two or more strings, only one of which is connected to a reward. The organism must determine which string is attached to the reward and pull it. We report a new way to implement patterned-string tasks via a computerized touch screen apparatus. Pigeons successfully learned such virtual patterned-string tasks and exhibited the same general performance profile as animals given conventional patterned-string tasks. In addition, variations in the length, separation, and alignment of the strings reliably affected the pigeons’ virtual string-pulling behavior. These results not only testify to the power and versatility of our computerized string task, but they also demonstrate that pigeons can concurrently contend with a broad range of demanding patterned-string problems, thereby eliminating many alternative interpretations of their behavior. The virtual patterned-string task may thus permit expanded exploration of other species and variables which would be unlikely to be undertaken either because of inadequacies of conventional methodology or sensorimotor limitations of the studied organisms.

Keywords

Patterned-string taskPigeonsTouch screenPerceptionCognition

Introduction

Developmental and comparative psychologists have, for more than 80 years, deployed an assortment of so-called string tasks (and other closely related methods) to assess the perceptual and cognitive standing of human children of different ages (Brainard 1930; Gesell 1928; Mathieson 1931; Piaget 1937; Richardson 1932; Willatts 1999) as well as different species of nonhuman animals. Those animals include a variety of birds and monkeys, plus rats, raccoons, cats, dogs, wolves, elephants, marmosets, orangutans, gorillas, and chimpanzees (see Table 1 for a listing of research reports; for selective reviews of the literature, see Deaner et al. 2006; Dewsbury 2000; Huber and Gajdon 2006; Emery 2006; Thorpe 1964).
Table 1

Summary of string tasks and other related problems studied in nonhuman animals

Species

Author

Support structure

String arrangement

Birds

 Pigeon (Columba livia)

Schmidt and Cook (2006)

Ribbons

Perp.

 New Caledonian Crow (Corvus moneduloides)

Taylor et al. (2010)

Strings

Perp., Nonperp., Crossed

 Common Raven (Corvus corax)

Albiach-Serrano et al. (2012)

Paper, paint

Crossed, Other

 Common Raven (Corvus corax)

Bagotskaya et al. (2011)

Strings

Perp., Nonperp., Crossed, Other

 Common Raven (Corvus corax)

Pfuhl (2012)

Strings

Perp.

 Common Raven (Corvus corax)

Heinrich and Bugnyar (2005)

Strings

Single

 Carrion Crow (Corvus corone)

Albiach-Serrano et al. (2012)

Paper, paint

Crossed, Other

 Hooded Crow (Corvus cornix)

Bagotskaya et al. (2011)

Strings

Perp., Nonperp., Crossed, Other

 Jackdaw (Coloeus monedula)

Dücker and Rensch (1977)

Strings

Perp., Nonperp., Crossed, Other

 Indian Starling (Acridotheres tristis)

Dücker and Rensch (1977)

Strings

Perp., Nonperp., Crossed, Other

 Greenfinch (Carduelis chloris)

Vince (1958, 1961)

Strings

Single

 Goldfinch (Carduelis carduelis)

Seibt and Wickler (2006)

Strings

Single

 Siskin (Carduelis spinus)

Seibt and Wickler (2006)

Strings

Single

 Chaffinch (Fringilla coelebs)

Vince (1958)

Strings

Single

 Canary (Serinus canaria)

Vince (1958, 1961)

Strings

Single

 Great Tit (Parus major)

Cole et al. (2011)

Strings

Single

 Great Tit (Parus major)

Vince (1956)

Strings

Single

 Gray Parrot (Psittacus erithacus)

Pepperberg (2004)

Plastic-link chain

Single

 Blue-fronted Parrot (Amazona aestiva)

Schuck-Paim et al. (2009)

Strings

Single, Perp., Crossed

 Blue-fronted Parrot (Amazona aestiva)

de Mendonca-Furtado and Ottoni (2008)

Cloths

Perp.

 Hyacinth Macaw (Anodorhynchus hyacinthinus)

Schuck-Paim et al. (2009)

Strings

Single, Perp., Crossed

 Lear’s Macaw (Anodorhynchus leari)

Schuck-Paim et al. (2009)

Strings

Single, Perp., Crossed

 Kea (Nestor notabilis)

Auersperg et al. (2009)

Wooden slats

Perp.

 Kea (Nestor notabilis)

Werdenich and Huber (2006)

Strings

Perp., Nonperp., Crossed

 Yellow-crowned Parakeet (Cyanoramphus auriceps)

Funk (2002)

Various supports

Perp.

 Budgerigar (Melopsittacus undulatus)

Dücker and Rensch (1977)

Strings

Perp., Nonperp., Crossed, Other

Mammals

 Rat (Rattus norvegicus)

Tolman (1937)

Strings

Perp.

 Domestic Cat (Felis catus)

Whitt et al. (2009)

Strings

Perp., Crossed

 Domestic Cat (Felis catus)

Trueblood and Smith (1934)

Strings

Perp., Other

 Domestic Cat (Felis catus)

Adams (1929)

Strings

Single, Perp., Nonperp., Other

 Dog (Canis familiaris)

Osthaus et al. (2005)

Strings

Perp., Nonperp., Crossed

 Dog (Canis familiaris)

Range et al. (2011)

Wooden boards

Perp.

 Wolf (Canis lupus)

Range et al. (2012)

Ropes

Perp., Nonperp., Crossed

 Asian Elephant (Elephas maximus)

Irie-Sugimoto et al. (2008)

Trays

Perp.

 Raccoon (Procyon lotor)

Michels et al. (1961)

Small chains

Perp., Crossed, Other

 Ring-tailed Lemur (Lemur catta)

Klüver (1933)

Strings

Perp., Other

 Marmoset (Callithrix jacchus)

Halsey et al. (2006)

Strings

Perp.

 Marmoset (Callithrix jacchus)

Gagne et al. (2012)

Strings

Perp., Nonperp., Crossed, Other

 Cotton-top Tamarin (Saguinus oedipus)

Hauser et al. (1999)

Cloths

Perp.

 Common Squirrel Monkey (Saimiri sciureus)

Cha and King (1969)

Strings

Perp., Nonperp., Crossed, Other

 Common Squirrel Monkey (Saimiri sciureus)

Klüver (1933)

Strings

Perp., Other

 Geoffroy’s Spider Monkey (Ateles geoffroyi)

Warden et al. (1940)

Strings

Perp., Nonperp., Other

 Geoffroy’s Spider Monkey (Ateles geoffroyi)

Harlow and Settlage (1934)

Strings

Perp., Nonperp., Crossed, Other

 Geoffroy’s Spider Monkey (Ateles geoffroyi)

Klüver (1933)

Strings

Perp., Other

 Tufted Capuchin (Cebus apella)

Yocom and Boysen (2010)

Cloths

Perp.

 Tufted Capuchin (Cebus apella)

Warden et al. (1940)

Strings

Perp., Nonperp., Other

 White-headed Capuchin (Cebus capucinus)

Warden et al. (1940)

Strings

Perp., Nonperp., Other

 White-headed Capuchin (Cebus capucinus)

Harlow and Settlage (1934)

Strings

Perp., Nonperp., Crossed, Other

 White-headed Capuchin (Cebus capucinus)

Klüver (1933)

Strings

Perp., Other

 Sooty Mangabey (Cercocebus atys)

Harlow and Settlage (1934)

Strings

Perp., Nonperp., Crossed, Other

 Mona (Cercopithecus mona)

Harlow and Settlage (1934)

Strings

Perp., Nonperp., Crossed, Other

 Spot-nosed Monkey (Cercopithecus nictitans)

Balasch et al. (1974)

Strings

Perp., Nonperp., Crossed, Other

 Rhesus Macaque (Macaca mulatta)

Warden et al. (1940)

Strings

Perp., Nonperp., Other

 Rhesus Macaque (Macaca mulatta)

Harlow and Settlage (1934)

Strings

Perp., Nonperp., Crossed, Other

 Pig-tailed Macaque (Macaca nemestrina)

Harlow and Settlage (1934)

Strings

Perp., Nonperp., Crossed, Other

 Java Macaque (Macaca fascicularis)

Harlow and Settlage (1934)

Strings

Perp., Nonperp., Crossed, Other

 Java Macaque (Macaca fascicularis)

Klüver (1933)

Strings

Perp., Other

 Mandrill (Mandrillus sphinx)

Balasch et al. (1974)

Strings

Perp., Nonperp., Crossed, Other

 Mandrill (Mandrillus sphinx)

Harlow and Settlage (1934)

Strings

Perp., Nonperp., Crossed, Other

 Drill (Mandrillus leucophaeus)

Balasch et al. (1974)

Strings

Perp., Nonperp., Crossed, Other

 Yellow Baboon (Papio cynocephalus)

Harlow and Settlage (1934)

Strings

Perp., Nonperp., Crossed, Other

 Sumatran Orangutan (Pongo abelii)

Albiach-Serrano et al. (2012)

Paper, paint

Crossed, Other

 Orangutan (Pongo pygmaeus)

Mulcahy et al. (2012)

Sticks

Perp.

 Orangutan (Pongo pygmaeus)

Herrmann et al. (2008)

Cloths, ropes

Perp.

 Orangutan (Pongo pygmaeus)

Fischer and Kitchener (1965)

Strings

Perp., Nonperp., Crossed, Other

 Gorilla (Gorilla gorilla)

Albiach-Serrano et al. (2012)

Paper, paint

Crossed, Other

 Gorilla (Gorilla gorilla)

Herrmann et al. (2008)

Cloths, ropes

Perp.

 Gorilla (Gorilla gorilla)

Fischer and Kitchener (1965)

Strings

Perp., Nonperp., Crossed, Other

 Gorilla (Gorilla gorilla)

Riesen et al. (1953)

Strings

Perp., Nonperp., Crossed, Other

 Bonobo (Pan paniscus)

Albiach-Serrano et al. (2012)

Paper, paint

Crossed, Other

 Bonobo (Pan paniscus)

Herrmann et al. (2008)

Cloths, ropes

Perp.

 Chimpanzee (Pan troglodytes)

Albiach-Serrano et al. (2012)

Paper, paint

Crossed, Other

 Chimpanzee (Pan troglodytes)

Herrmann et al. (2008)

Cloths, ropes

Perp.

 Chimpanzee (Pan troglodytes)

Finch (1941)

Strings

Perp., Nonperp., Crossed, Other

Perp. Perpendicular strings. Nonperp. Nonperpendicular (Slanted) strings. Single designates that only one string was used

Common to all of these tasks is that the organism must pull in a reward which is attached to a string (or which rests on a cloth, paddle, or other surface). Pulling on the string is the only way that the reward can be brought into reach and consumed. The results of these studies have been said to have important implications for such vital and diverse psychological processes as: spatial and relational perception; trial-and-error or associative learning; expectation or hypothesis formation; cause-effect or means-end comprehension; tool acquisition and use; imaginational, inferential, or insightful learning; and spontaneous, inventive, or creative problem solving.

The most important and widely used of these many different string problems are the patterned-string tasks. Here, the organism is shown two or more strings, only one of which is connected to a reward. The organism must determine which string is attached to the reward and pull it. Although the lack of methodological standardization and statistical rigor in the literature makes it difficult to arrive at firm generalizations, it seems safe to conclude that patterned-string tasks are easiest when the strings are aligned Perpendicular to the organism; patterned-string tasks are somewhat more difficult when the strings angle away from perpendicularity, but do not cross; and patterned-string tasks are most difficult when the strings cross one or more times.

Despite their relevance to research and theory in psychological science, patterned-string tasks have historically enjoyed only sporadic use. Because these tasks have not been automated: (a) they are extremely laborious to administer; (b) they offer very limited opportunities to systematically study the learning of patterned-string choice behavior as well as the effects on choice behavior of potentially important independent variables such as the length, separation, and alignment of the strings; and (c) they are highly constrained by the sensorimotor capacities of the subjects, the physical setting, and the stimulus materials.

In addition, even the most ambitious and assiduous of research projects have not fully elucidated the nature and complexity of patterned-string task behavior. Consider, for example, the classic study by Harlow and Settlage (1934). These investigators administered 1,000 trials comprising either 10 or 20 different patterned-string arrangements to each of 31 monkeys from 9 different species. Harlow and Settlage were successful in documenting a progressive rise in errors as the tests became more difficult, with more challenging problems generally involving angled and/or Crossed strings: “Simple tests are solved almost immediately by all monkeys (insight), those of intermediate difficulty were solved with some trouble, and more complicated problems were insolvable to all the animals (p. 433).”

However, these different patterned-string problems appear to have been given in order of their presumed difficulty, thereby making it “impossible to determine any absolute order of difficulty of the test items since experience on the earlier tests influences the results on the later ones (p. 428).” This complication plus the very small number (50 or 100) of training trials may have worked against the researchers being able to detect systematic changes in accuracy within each of the 10 or 20 different patterned-string arrangements, leading Harlow and Settlage to conclude that “although the monkeys showed some adaptation to the general situation as the experiment progressed, our data does not show any consistent learning in any particular pattern (p. 428).” Although subsequent researchers have shown little interest in studying the acquisition of patterned-string task behavior—perhaps because they were mainly interested in examining whether inborn causal understanding or prior real-world experience would allow animals to “spontaneously” perform these tasks—this question is an important one deserving explicit experimental inquiry (see Schmidt and Cook 2006, for recent evidence of patterned-string task learning in pigeons).

Here, we report a new way to implement patterned-string tasks via a computerized touch screen apparatus. Pigeons given automated training on such virtual patterned-string tasks successfully learned these problems and exhibited the same basic performance profile as animals given conventional patterned-string tasks. Our pigeons’ virtual string task learning did unfold over many hundreds of trials, which would surely have taxed the patience of most human experimenters had conventional manual methods been implemented. In addition, parametric manipulation of the length, separation, and alignment of the strings reliably affected the pigeons’ virtual string-pulling behavior. Finally, because we could give the pigeons extended training with these problems—especially those extremely difficult ones that might never have been mastered had fewer training trials been provided—we were able to gain fresh insights into the factors that influenced the birds’ transfer performance from one problem to another.

Thus, further development of the virtual patterned-string task may open the door to investigations of many other species and variables which would be unlikely to be undertaken either because of the shortcomings of conventional methodology or because of sensorimotor limitations of the studied organisms.

What must a virtual patterned-string task accomplish?

An effective simulation of a patterned-string task must faithfully capture the essence of the real patterned-string task (see Miyata and Fujita 2011, who devised a clever virtual detour problem for pigeons). Such fidelity requires that several methodological requirements be met.

First, organisms must be afforded the choice of acting on multiple virtual strings which are distinctively different from one another; otherwise, it would be exceedingly difficult for organisms to appreciate the differential “connectedness” of the strings and the objects. Second, each successive action that the organism makes on each virtual string should bring its attached object closer; this relation is tantamount to “reeling in” the object. Third, the objects at the distal end of the virtual strings can only be obtained by acting on the proximal end of the virtual strings; this functional connectedness is essential to the frequent claims that string tests implicate cause-effect or means-end reasoning in organisms’ string-pulling behavior. Finally, the objects themselves must have differential reward value for the organism; otherwise, there would be no reason for the organism to choose one over the other.

Our virtual patterned-string task

Figure 1 depicts our virtual patterned-string task with perpendicularly arrayed strings. The larger (blue) area reaching the top of the screen represents the region of the apparatus that is usually inaccessible to the organism; in it, two strings and two end objects are located. The strings are colored red and green to guarantee their distinctiveness. Each string color and spatial position is randomly associated with the bright (yellow) square (representing the “full” food dish) and the dark (black) square (representing the “empty” food dish) whose spatial locations are also randomized. The smaller (gray) area reaching the bottom of the screen represents the region of the apparatus that is usually accessible to the organism; in it, two identical response “buttons” are located. Each response to each button moves its attached dish one step closer to the smaller (gray) area in much the same way as turning the crank of a fishing reel moves the end of the line closer to the angler. When one of the dishes is finally moved from the larger (blue) area into the smaller (gray) area, responding to it produces food (if the bright, yellow “full” dish is contacted) or no food (if the dark, black “empty dish” is contacted).
https://static-content.springer.com/image/art%3A10.1007%2Fs10071-013-0608-0/MediaObjects/10071_2013_608_Fig1_HTML.gif
Fig. 1

Examples of some of the string configurations that were used in Experiment 1. The strings were always red and green, and their locations varied randomly from left to right. The bright (yellow) square at the top/distal end of the strings represents the “full” food dish, and the dark (black) square at the top/distal end of the strings represents the “empty” food dish. The strings were perpendicular to the corresponding white-square response buttons. Top examples of some of the Short, Medium, and Long strings that were used in Phase 1. Bottom examples of some of the 3 possible distances between the red and green strings: Near, Middle, or Far, introduced in Phase 2. In Phase 2, the length of the string could also be Short, Medium, or Long. In Phase 3, only the Long strings were shown at the 3 possible distances between the red and green strings: Near, Middle, or Far (color figure online)

We believe that the essence of the real patterned-string task is faithfully captured by this virtual patterned-string task. Of course, our belief requires empirical validation. We thus report three experiments with pigeons that explored the fidelity of this virtual patterned-string task.

Experiment 1

In Experiment 1, we trained 4 pigeons on our virtual patterned-string task. All of the virtual strings were aligned Perpendicular to the horizontal line separating the larger (blue) area of the touch screen display (where the full and empty food dishes were first presented) from the smaller (gray) area of the display (where the two choice responses were located)—(see Fig. 1). Under these conditions, we systematically explored the effect of virtual string length and spatial separation on the pigeons’ choice accuracy.

Method

Subjects

The subjects were 4 feral pigeons maintained at 85 % of their free-feeding weights by controlled daily feeding. The pigeons had served in unrelated studies before beginning this project. Immediately prior to the experimental training that is detailed below, we taught the pigeons to peck at the same small circular and square stimuli that were to be presented later on the same background.

Apparatus

The experiment used four 36- × 36- × 41-cm operant conditioning chambers detailed by Gibson et al. (2004). The chambers were located in a dark room with continuous white noise. Each chamber was equipped with a 15-in LCD monitor located behind an AccuTouch® resistive touch screen (Elo TouchSystems, Fremont, CA). The portion of the screen that was viewable by the pigeons was 28.5 × 17 cm. Pecks to the touch screen were processed by a serial controller board outside the box. A rotary dispenser delivered 45-mg pigeon pellets through a vinyl tube into a food dish located in the center of the rear wall opposite the touch screen. Illumination during the experimental sessions was provided by a houselight mounted on the upper rear wall of the chamber. The pellet dispenser and houselight were controlled by a digital I/O interface board. Each chamber was controlled by an Apple® iMac® computer. Programs to run this and later experiments were developed in MatLab® with Psychtoolbox-3 extensions (Brainard 1997; Pelli 1997; http://psychtoolbox.org/).

Stimuli and procedure

On each trial, one red and one green string appeared on the computer screen superimposed on a larger (blue)/smaller (gray) background (see examples in Fig. 1). Two square “dishes” were placed at the distal (top) end of the strings; one dish was “full” (it displayed a 1.5- × 1.5-cm color photograph of mixed grain), whereas the other dish was “empty” (it displayed a 1.5- × 1.5-cm black square). The left–right locations of the red and green strings and the left–right locations of the full and empty dishes were counterbalanced across trials. The pigeons’ task was to move the full dish from the larger (blue) to the smaller (gray) area, in order to obtain actual food pellet reinforcement; doing so required the pigeons to repeatedly peck the correct response button at the proximal (bottom) end of the string, located in the smaller (gray) background area.

Figure 2 shows the sequence of events in the course of a trial. At the start of a trial, the pigeons were presented with an orienting stimulus: a black plus sign on a white background (6.7 × 6.7 cm) in the middle of the screen. After one peck anywhere on this stimulus, one red string and one green string (0.3-cm width) were presented. Centered between the top ends of the strings, an orange circle (1.2-cm diameter) was presented. The pigeons had to peck the orange circle once, so that their attention would be drawn to the top end of the strings, where the full and empty dishes were next presented; 1 s after the presentation of the two dishes, a yellow circle (1.2 cm diameter) was presented midway between the bottom ends of the strings; the pigeons had to peck once at this yellow circle in order to draw their attention to where the virtual string-pulling responses were to be made and to position them directly between those response options. Pecking once at the yellow circle presented two response buttons (white squares, 1.5- × 1.5-cm) at the bottom ends of each string. Pecking at these buttons progressively lowered the corresponding dish attached to the string. Each peck moved the chosen string 1.7 cm downward. The number of pecks required to move the dish into the smaller (gray) area depended on the length of the string (2 pecks for the Short strings, 4 pecks for the Medium strings, and 6 pecks for the Long strings). Once the dish had reached the proximal end (and overlapped the response button), the pigeons had to peck twice at the dish to complete the choice requirement.
https://static-content.springer.com/image/art%3A10.1007%2Fs10071-013-0608-0/MediaObjects/10071_2013_608_Fig2_HTML.gif
Fig. 2

The sequence of events in the course of a training trial. After one peck at the initial white display (Screen 1), one red and one green string were presented on a blue/gray background with an orange circle presented between the top/distal ends of the strings (Screen 2). After one peck at the orange circle, the “full” and “empty” dishes were presented (Screen 3). One second after the presentation of the “full” and “empty” dishes, a yellow circle was presented between the bottom/proximal ends of the strings (Screen 4). One peck at the yellow circle prompted the presentation of two white response buttons at the bottom end of each string (Screen 5). Pecking at these buttons would move the corresponding dish, attached to the string, downward from the upper/larger (blue) area toward the lower/smaller (gray) area (Screen 6; this representation shows the pigeon’s choice of the “full” dish). Each peck moved the dish 1.7 cm. The number of pecks required to move the string into the lower/smaller (gray) area depended on the length of the string (2 pecks for the Short strings, 4 pecks for the Medium strings, and 6 pecks for the Long strings). Once the dish entered the lower/smaller (gray) area, the birds had to peck it twice to receive the scheduled consequence (food or no food; Screen 7), after which the screen went black for 6 s (Screen 8). After a correct choice, a new trial began (Screen 9); after an incorrect choice, a correction trial was given so that the bird was returned to the beginning of the trial sequence (Screen 2) (color figure online)

Differential food reinforcement was used to encourage correct responses. If the final choice response was correct, then the screen blackened, food was delivered, and a 6-s intertrial interval (ITI) ensued, during which the pigeon awaited the next trial. If the final choice response was incorrect, then the screen blackened, no food was delivered, a 6-s ITI ensued, and one or more correction trials were given until the correct response was made. Scores from correction trials and from infrequent incomplete sessions were not used in data analysis.

Phase 1

Pigeons were trained with strings of three different lengths. On any given trial, both strings were of equal length: Short (2 cm), Medium (5.4 cm), or Long (8.8 cm). The two strings were always 4 cm from one another (the smallest spatial separation in our arrangement). Examples of Phase 1 trials can be seen in Fig. 1 (first row). The left–right locations of the red and green strings (2), the left–right locations of the full and empty food dishes (2), the length of the strings (3), and the lateral locations of the two strings (3; on the far left, in the middle, and on the far right) were counterbalanced, so the total number of unique trials was 36. Each block of 36 trials was presented 4 times, so that each daily training session comprised 144 trials. Trial presentation within each block was randomized. This phase lasted 60 days.

Phase 2

Two additional spatial separations between the strings were next introduced to explore the effect of this variable. Thus, the string separations could be: Near (4 cm, as in Phase 1), Middle (8 cm), or Far (12 cm). The three different string lengths and the two different string separations were presented equally often. Examples of Phase 2 trials can be seen in Fig. 1 (second row). The left–right locations of the red and green strings (2), the left–right locations of the full and empty food dishes (2), the length of the strings (3), and the six possible locations of the two strings were counterbalanced, so the total number of unique trials was 72. The six locations of the strings were Near: (a) the two strings on the far left, (b) the two strings in the middle, (c) the two strings on the far right; Middle: (d) one string on the far left and one string in the middle right, (e) one string on the far right and one string in the middle left; and Far: (f) one string on the far left and one string on the far right. Each block of 72 trials was given twice, so that each daily training session comprised 144 trials. Trial presentation within each block was randomized. This phase lasted 16 days.

Phase 3

In an effort to boost discriminative performance to the Long strings (see later results), we further trained the pigeons with the Long strings alone at the three separation levels. The left–right locations of the red and green strings (2), the left–right locations of the full and empty food dishes (2), and the six possible locations of the two strings (described above) were counterbalanced, so the total number of unique trials was 24. Each block of 24 trials was presented six times, so that each daily training session comprised 144 trials. Trial presentation within each block was randomized. This phase lasted 20 days.

Results and discussion

Phase 1

Our initial observation was that the pigeons very quickly learned to switch from pecking the incorrect button to pecking the correct button before they moved the empty dish into its final proximal location in the smaller (gray) area. Indeed, after the first 2 days of training, the pigeons almost never completed a trial by pecking the wrong button (only 3 trials out of more than 30,000 trials over the remaining training sessions). So, instead of using each trial’s final choice response, we used each trial’s first choice response to compute string-pulling accuracy; this measure represents the most conservative index of the pigeons’ discriminative choice behavior.

We grouped all 60 training sessions into 15 blocks of 4 sessions each. Figure 3 (top) depicts choice accuracy across the 15 blocks of training in Phase 1 of Experiment 1. The pigeons’ first choice accuracy increased rapidly for the Short and Medium strings, and it surpassed 80 % by the second and third block of training, respectively; the pigeons’ first choice accuracy for the Long strings increased more slowly, and it did not surpass the 80 % mark until the seventh block. So, the longer the string, the more difficult the task seemed to be for the pigeons. By the last block of training, accuracy reached 96, 96, and 84 % correct for the Short, Medium, and Long strings, respectively.
https://static-content.springer.com/image/art%3A10.1007%2Fs10071-013-0608-0/MediaObjects/10071_2013_608_Fig3_HTML.gif
Fig. 3

Top percentage of correct button responses to the Short, Medium, and Long strings throughout the 15 blocks of training in Phase 1 of Experiment 1. Bottom percentage of correct button responses to the Short, Medium, and Long strings in Phase 2 of Experiment 1, where the distance between the strings could be Near, Middle, or Far. Error bars indicate the standard error of the means

A 3 (string length: Short, Medium, Long) × 15 (training block) repeated measures analysis of variance (ANOVA) on the percentage of correct first choice responses revealed a statistically significant main effect of training block, confirming the changes in accuracy during the 15 blocks of Phase 1 training, F(14,42) = 22.65, MSE = 0.01, p < .01. Overall accuracy was 66 % correct on the first block, it reached 82 % correct on the fourth block, and it remained over 90 % correct after the ninth block. There was also a statistically significant main effect of string length, F(2,6) = 25.87, MSE = 0.07, p < .01. Tukey HSD post hoc comparisons (α = .05) revealed that overall accuracy was significantly lower for the Long strings (76 % correct) than for the Medium (89 % correct) or the Short (92 % correct) strings; accuracy for the Medium strings was numerically lower than for the Short strings, but this difference was not statistically significant. The String Length × Training Block interaction was not statistically significant.

Phase 2

Figure 3 (bottom) illustrates overall discriminative responding in Phase 2 to the Short, Medium, and Long strings as a function of the horizontal separation between them. Increasing the separation between the strings had rather little systematic effect on choice accuracy. The pigeons’ choice accuracy remained high (above 90 % correct) for the Short and Medium strings, and lower (albeit above 80 % correct) for the Long strings, regardless of the two strings being close together or farther apart.

We grouped the 16 training sessions of Phase 2 into 4 blocks of 4 sessions each. A 3 (string length: Short, Medium, Long) × 3 (separation: Near, Middle, Far) × 4 (training block) repeated measures ANOVA on the percentage of correct first choice responses revealed a statistically significant main effect of string length, F(2,6) = 7.67, MSE = 2.03, p < .05, and a statistically significant main effect of string separation, F(2,6) = 7.81, MSE = 0.13, p < .05. There were no other statistically significant main effects or interactions.

Tukey HSD post hoc comparisons (α = .05) revealed that accuracy for the Long strings (83 % correct) was significantly lower than for the Medium (94 % correct) and Short (97 % correct) strings. In addition, accuracy for the Middle separation (94 % correct) was slightly, but significantly higher than for the Near (91 % correct) and Far (90 % correct) string separations; thus, over the range that we varied, increasing the horizontal separation between the strings had no progressive detrimental effect on pigeons’ choice accuracy.

Phase 3

Overall accuracy had been high in Phase 2; however, accuracy for the Long strings was consistently lower than for the Medium and Short strings (Fig. 3, top and bottom). Because later experiments required that we use Long strings, we further trained the pigeons with only the Long strings at the three spatial separations for 20 more days in Phase 3. We grouped the 20 training sessions of Phase 3 into 5 blocks of 4 sessions each. Accuracy to the Long strings did improve from 83 % correct in the last 4-session block of Phase 2 to 92 % correct in the last 4-session block of Phase 3. A 3 (separation: Near, Middle, Far) × 5 (training block) repeated measures ANOVA on the percentage of correct first choice responses revealed a statistically significant main effect of string separation, F(2,6) = 6.19, MSE = 0.53, p < .05; overall accuracy was slightly lower for the Near (87 %) than for the Middle (92 %) and Far (92 %) separations. The main effect of training block was not significant nor was the Separation × Training Block interaction. Most importantly, in the last training block, the pigeons exhibited high accuracy levels regardless of the distance between the strings: 90, 93, and 94 % correct for the Near, Middle, and Far separations, respectively. Apparently, accuracy to the Long strings in the prior phases was hindered by the presence of the easier-to-discriminate Short and Medium strings, but it reached similarly high levels of accuracy by the end of this training phase.

Experiment 2

In Experiment 2, we continued training the same pigeons with the same Perpendicular string configurations that we had given them in Phase 3 of Experiment 1, while we simultaneously explored the effects of Nonperpendicular string configurations on choice accuracy. Some of the Nonperpendicular patterns involved parallel angled strings; other Nonperpendicular patterns either converged from the proximal to the distal ends of the strings or they diverged from the proximal to the distal ends of the strings. All of these Nonperpendicular string configurations prohibited the pigeons from simply dropping down from the full food dish above to the corresponding choice button directly below in order to execute a correct choice response.

Method

Subjects and apparatus

The same 4 pigeons from Experiment 1 served as subjects in Experiment 2; the birds were maintained as before. The apparatus was the same as in Experiment 1.

Stimuli

We used the same basic setting as in Experiment 1; all of the strings were now Long. We introduced trials on which the orientation of the strings varied so that the response buttons and the full and empty dishes could no longer be connected by a vertical line. We created three new Nonperpendicular configurations: Parallel Slanted, Convergent, and Divergent (see all possible configurations in Fig. 4). On Parallel Slanted trials, the separation between the distal and proximal ends was always the same. On Convergent trials, the proximal ends were close together, but the distal ends were far apart, whereas on Divergent trials, the proximal ends were far apart, but the distal ends were close together.
https://static-content.springer.com/image/art%3A10.1007%2Fs10071-013-0608-0/MediaObjects/10071_2013_608_Fig4_HTML.gif
Fig. 4

Top examples of some of the Parallel Slanted, Convergent, and Divergent string configurations studied in Experiment 2. Bottom percentage of correct responses to these string configurations throughout the 8 blocks of training. Error bars indicate the standard error of the means

The Parallel Slanted, Convergent, and Divergent configurations permitted us to assess the robustness of the pigeons’ transfer performance to strings that were arrayed in different orientations from Perpendicular. Moreover, these configurations also allowed us to learn more about the specific factors that may have controlled the pigeons’ performance. Finally, these configurations helped to minimize the novelty of pigeons’ seeing nonparallel, Slanted lines in upcoming Experiment 3 involving Crossed strings.

As can be seen in Fig. 4 (top left), of the six types of Parallel Slanted trials, there were four in which the strings could be close together (first row) and two in which the strings could be far apart (second row). A consequence of the Parallel Slanted strings being close together is that the full dish could be located directly above the incorrect response button or the empty dish could be located directly above the correct response button. Similar misalignments could also occur on some of the Convergent and Divergent trials (see Fig. 4, top right). If the pigeons had come to use collinearity as a cue for responding in Experiment 1—where the correct proximal and distal ends on Perpendicular trials were always vertically aligned—then performance on Nonperpendicular Collinear trials in Experiment 2 might be lower than on Nonperpendicular Noncollinear trials, because only on the former trials did vertical alignment produce erroneous stimulus–response information.

Procedure

The present procedure was similar to that of Experiment 1, except for the addition of the Nonperpendicular string configurations. Each daily session consisted of 48 baseline trials with the Long Perpendicular strings shown at the three different distances: Near, Middle, and Far, each shown 12 times just as in Phase 3 of Experiment 1. There were also 24 Parallel Slanted trials (the six types in Fig. 4, counterbalanced by the two colors of the strings and the two left–right locations of the full and empty food dishes), 12 Convergent trials (the three types in Fig. 4, counterbalanced by the two colors of the strings and the two left–right locations of the full and empty food dishes), and 12 Divergent trials (the three types in Fig. 4, counterbalanced by the two colors of the strings and the two left–right locations of the full and empty food dishes). In total, each daily session consisted of 96 trials. Each pigeon was trained for 32 sessions.

Results and discussion

We grouped the 32 training sessions into 8 blocks containing 4 sessions each. Figure 4 (bottom) shows the pigeons’ choice accuracy across the 8 training blocks for the Nonperpendicular string trials in Experiment 2. By the end of training, accuracy for all types of Nonperpendicular trials was over 75 % correct; however, the first training block disclosed some large behavioral disparities. Accuracy for the Parallel Slanted strings was only 62 % when the strings were close together (Collinear), but it was 83 % when the strings were farther apart (Noncollinear); moreover, when the strings were Convergent or Divergent, accuracy for Collinear trials was only 45 and 53 %, respectively, whereas accuracy for Noncollinear trials was 80 and 90 %, respectively. So, initial performance to the new Nonperpendicular string orientations varied greatly depending on the specific string configurations. It appears that performance was low when the empty dish was located directly above the correct response button or the full dish was located directly above the incorrect response button; when these conflicting collinearities did not occur, the pigeons exhibited high choice accuracy. The increased error rate for the Collinear trials corresponds to the “proximity error” reported in dogs by Osthaus et al. (2005).

An 8 (training block) × 3 (stimulus orientation: Parallel Slanted vs. Convergent vs. Divergent) × 2 (Collinearity vs. Noncollinearity) repeated measures ANOVA on the percentage of correct first choice responses revealed a statistically significant main effect of block, F(7,21) = 8.41, MSE = 0.22, p < .01, confirming that overall accuracy increased over training (69 % correct for the first block and 84 % correct for the last block). There was also a statistically significant main effect of stimulus orientation, F(2,6) = 6.07, MSE = 0.32, p < .05, with accuracy for Divergent trials (75 % correct) being somewhat lower than for Convergent trials (81 % correct) and Parallel Slanted (81 % correct) trials, and a statistically significant main effect of collinearity, F(1,3) = 30.91, MSE = 0.98, p < .01, with accuracy for Collinear trials (71 % correct) being lower than for Noncollinear trials (87 % correct).

The Training Block × Stimulus Orientation interaction was statistically significant, F(14,42) = 2.38, MSE = 0.10, p < .05, indicating that, over training, accuracy differentially changed depending on string orientation. Tukey HSD post hoc comparisons revealed that accuracy for Parallel Slanted and Convergent strings increased significantly from Block 1 to Block 8 (from 72 to 87 % correct for Parallel Slanted trials and from 63 to 85 % correct for Convergent trials), whereas accuracy for Divergent trials did not increase significantly from Block 1 to Block 8 (71 and 78 % correct, respectively).

The Training Block × Collinearity interaction was also statistically significant, F(7,21) = 5.52, MSE = 0.15, p < .01, indicating that accuracy also differentially changed depending on string collinearity. Tukey HSD post hoc comparisons revealed that accuracy for Collinear trials increased significantly from Block 1 to Block 8 (from 53 to 79 % correct), whereas accuracy for Noncollinear trials, already high in Block 1, did not increase significantly from Block 1 to Block 8 (from 84 to 88 % correct).

On Collinear trials, the full dish could be located directly above the incorrect response button or the empty dish could be located directly above the correct response button. Although both cases involve misleading collinearity, it could be that one is more misleading than the other. For example, if the pigeons look at the full dish on top and then move straight down to choose the response button, the case in which the full dish is located directly above the incorrect response button should be the more disruptive.

In order to test this possibility, we conducted an 8 (training block) × 3 (stimulus orientation: Parallel Slanted vs. Convergent vs. Divergent) × 2 (configuration: Full-above-Incorrect vs. Empty-above-Correct) repeated measures ANOVA on the percentage of correct first choice responses on Collinear trials. This analysis revealed a statistically significant main effect of block, F(7,21) = 5.92, MSE = 0.23, p < .01. There was also a statistically significant main effect of stimulus configuration, F(1,3) = 5.01, MSE = 0.03, p < .05; overall accuracy on Full-above-Incorrect trials (68 % correct) was lower than overall accuracy on Empty-above-Correct trials (74 % correct). The Training Block × Configuration interaction was also statistically significant, F(7,21) = 2.42, MSE = 0.23, p < .05, indicating that differences in accuracy between Full-above-Incorrect and Empty-above-Correct trials varied over the course of training. Indeed, Tukey HSD post hoc comparisons revealed that this difference was significant in the first training block, when pigeons were 38 % correct on Full-above-Incorrect trials and 69 % correct on Empty-above-Correct trials; although the trend was present in the second block (59 vs. 67 %), it was not statistically significant, and it further decreased over the remaining blocks.

The pattern of results in Experiment 2 suggests that changes in string orientation per se need not disrupt pigeons’ discrimination performance. Our birds’ accuracy on Noncollinear trials—where novel string configurations were presented, but vertical alignments of the full dish and the incorrect button or the empty dish and the correct button were precluded—remained very high in the first training block. This first block contained only 16 Noncollinear Convergent trials, 16 Noncollinear Divergent trials, and 32 Noncollinear Parallel trials, yet accuracy was 81, 89, and 83 % correct, respectively (see Fig. 4). So, we can conclude that our pigeons exhibited immediate transfer to novel configurations, provided that misleading collinearity information was not available. Only when the new configurations involved misleading collinearity information did the pigeons’ accuracy drop (both on Full-above-Incorrect and Empty-above-Correct trials, although the initial drop was larger on Full-above-Incorrect trials).

Note that in Experiment 1, the correct proximal and distal ends of the strings on Perpendicular trials were always vertically aligned, thereby making collinearity a perfectly predictive cue. But in Experiment 2, collinearity became a misleading cue for some of the new configurations, where vertical alignment produced erroneous stimulus–response information. The pigeons proved to be sensitive to this misleading information and, consequently, their accuracy dropped on Collinear trials. However, this drop in accuracy does not mean that the pigeons were using only collinearity to solve the task. When the pigeons were shown novel configurations that did not involve misleading collinearity (that is, Noncollinear trials), their accuracy was very high from the very beginning of Experiment 2. So, the pigeons were evidently using other cues, possibly the connectedness between the proximal and distal ends of the strings, to be able to respond correctly to the novel Noncollinear configurations.

In the course of training in Experiment 2, the pigeons largely overcame the initially deleterious effects of conflicting collinearity information; as can be seen in Table 2, by the end training, the pigeons predominately chose the correct button regardless of collinearity serving as a cue for the correct response (on Perpendicular trials) or as a cue for the incorrect response (on Nonperpendicular Collinear trials), albeit at lower levels in the latter case.
Table 2

Performance in the last training block of each of the three experiments

 

Experiment 1

Experiment 2

Experiment 3

Mean

SE

Mean

SE

Mean

SE

Perpendicular

 Near

89.67

0.90

92.18

1.37

86.46

2.48

 Middle

93.49

0.89

93.36

1.56

93.75

2.15

 Far

94.27

1.18

97.66

1.34

93.75

3.05

Nonperpendicular

 Noncollinear

  Convergent

  

93.75

3.05

96.88

3.13

  Divergent

  

79.69

5.07

84.38

6.52

  Parallel

  

91.41

2.49

92.19

3.38

 Collinear

  Convergent

  

78.13

3.69

84.38

4.57

  Divergent

  

77.34

3.71

78.13

5.21

  Parallel

  

83.20

2.34

84.38

3.22

Crossed

 Near

    

71.61

2.30

 Middle

    

80.08

2.50

 Far

    

71.09

4.02

Only long string scores are entered for the Perpendicular arrangements

Accuracy on the Perpendicular trials remained high in Experiment 2 (averaging between 92 and 98 % correct; see Table 2). An 8 (training block) × 3 (separation: Near, Middle, Far) repeated measures ANOVA on the percentage of correct first choice responses on Perpendicular trials yielded no significant main effects or interactions.

Experiment 3

In Experiment 3, we gave pigeons one of the most difficult patterned-string tasks: We crossed the strings in an X pattern. We continued training the birds on the Perpendicular and Nonperpendicular string configurations of Experiment 2 for comparative purposes as well as to maintain their previously learned patterned-string performance. At issue was the versatility of the pigeon’s string task performance when they were concurrently required to choose the correct string under all three basic spatial configurations: Perpendicular, Nonperpendicular, and Crossed.

Method

Subjects and apparatus

The same 4 pigeons from Experiments 1 and 2 served in Experiment 3. The birds were maintained as before, and the apparatus was the same as in the prior experiments.

Stimuli

Crossed string trials were now given to the pigeons in addition to the Parallel, Convergent, Divergent, and Perpendicular string trials that they had been given in Experiment 2. The distal and proximal ends of the Crossed strings could be shown at three different distances: Near, Middle, and Far (see examples in Fig. 5, top).
https://static-content.springer.com/image/art%3A10.1007%2Fs10071-013-0608-0/MediaObjects/10071_2013_608_Fig5_HTML.gif
Fig. 5

Top examples of some of the Crossed string configurations studied in Experiment 3. Bottom percentage of correct responses to these string configurations throughout the 8 blocks of probe trials with nondifferential reinforcement (where all final choices were reinforced, regardless of accuracy) and the subsequent 8 blocks of differential reinforcement training (where only correct final choices were reinforced). Note that, in the training-trial phase, the proportion of Crossed trials in each daily session was four times larger than in the probe-trial phase. Error bars indicate the standard error of the means

Procedure

Probe trials: nondifferential reinforcement

The procedure was similar to Experiment 2, except for the addition of trials with Crossed strings. The Crossed trials were initially given as infrequent probe trials on which any final choice response was reinforced (nondifferential reinforcement), and no correction trials were scheduled; we wanted to see if the pigeons would effectively generalize their virtual string-pulling behavior to the novel Crossed string problem.

The left–right locations of the red and green strings (2), which color string overlapped the other (2), and the left–right locations of the full and empty food dishes (2) were all counterbalanced as were all of the possible string locations for the Near (3), Middle (2), and Far (1) trials; so, the total number of unique Crossed string trials was 48. Presentation of all 48 nondifferentially reinforced trials in a single session would have yielded a very high proportion of trials for which no correct response was required; so, we presented the 48 Crossed string trials in 4 sessions containing only 12 Crossed string trials each. Each group of 4 sessions constituted a training block; in total, 8 blocks (or 32 daily sessions) were given.

Each daily session comprised 100 trials. Sessions began with 16 warm-up trials (Perpendicular, Parallel Slanted, Convergent, and Divergent) followed by 84 additional trials, 12 of which were randomly interspersed Crossed string trials: 6 Near, 4 Middle, and 2 Far (resulting from the various possible combinations that each of the distances permitted).

Training trials: differential reinforcement

After completing 8 blocks with nondifferential reinforced Crossed string probe trials, we gave the pigeons 32 days of differential reinforcement training with the Crossed strings, in order to explicitly encourage them to make correct choice responses. In addition, we increased the proportion of Crossed string trials, so that they were presented four times more often than in the nondifferential reinforcement phase. In total, daily sessions comprised 96 trials: 48 Crossed (24 Near, 16 Middle, and 8 Far), 24 Perpendicular, 12 Parallel Slanted, 6 Convergent, and 6 Divergent.

Results and discussion

Figure 5 (bottom) shows the pigeons’ choice accuracy for the Crossed string trials across the 8 probe-trial blocks and across the 8 training blocks in Experiment 3.

Probe trials: nondifferential reinforcement

As can be seen in the left portion of Fig. 5, the pigeons did not immediately choose the correct string on Crossed trials; on the contrary, the pigeons chose the incorrect string most of the time (between 70 and 80 % of Block 1 trials). Clearly, the pigeons did not effectively transfer their virtual string-pulling behavior from the Perpendicular and Nonperpendicular string patterns to the Crossed pattern of string alignment.

The pigeons had, in Experiment 2, come largely to contend with the problem of conflicting collinearity information on Nonperpendicular trials. Note, however, that one dish and one choice button were misaligned on Nonperpendicular Collinear trials in Experiment 2; both dishes and both choice buttons were misaligned on Crossed trials in Experiment 3. This “double dose” of misalignment is the likely reason for the pigeons’ initially choosing the incorrect button in Experiment 3. Still, even without differential reinforcement for doing so, the pigeons began to exhibit increasingly accurate responding in Experiment 3, but they only reached 60 % correct by Block 8.

An 8 (block) × 3 (crossed type: Near, Middle, Far) repeated measures ANOVA on the percentage of correct first choice responses revealed a statistically significant main effect of crossed type, F(2,6) = 5.92, MSE = 0.14, p < .05, because overall accuracy was slightly lower for Far Crossed trials (41 % correct) than for Middle Crossed (51 % correct) or for Near Crossed trials (48 % correct). There was also a statistically significant main effect of testing block, F(7,21) = 6.80, MSE = 0.25, p < .01, confirming the increase in choice accuracy over the 8 blocks: Overall accuracy was 26 % correct in the first block and 60 % correct in the last block. Even when Crossed string trials were not differentially reinforced (the pigeons were always given food at the end of each Crossed string trial regardless of their final button choice), the birds learned that erroneous responses were moving the empty dish nearer to its final location. Indeed, the pigeons usually corrected themselves during the trial, most of the trials ending with the birds making a high percentage of their final responses to the correct button: Far Crossed trials (86 % correct), Middle Crossed trials (88 % correct), and Near Crossed trials (89 % correct).

Despite the improvement from the first to the last block, we also observed that no pronounced changes in accuracy were occurring across Blocks 4, 5, 6, 7, and 8 (54, 49, 54, 54, and 60 %, respectively); so, we decided to increase the proportion of Crossed trials in each daily session and to arrange differential reinforcement on those Crossed trials.

Training trials: differential reinforcement

We expected that even higher levels of first choice accuracy would be attained if the proportion of Crossed trials in each daily session was increased, and pigeons were given differential reinforcement for final button choices; so, 32 differential reinforcement training sessions were next administered. We grouped those sessions into 8 blocks containing 4 sessions each, and we looked at changes in accuracy on the Crossed trials.

The right portion of Fig. 5 suggests that such training was successful in raising the overall accuracy of the pigeons’ first button choices from 58 % in the initial differential reinforcement training block to 74 % in the final training block. An 8 (training block) × 3 (crossed type: Near, Middle, Far) repeated measures ANOVA on the percentage of correct first choice responses revealed a statistically significant main effect of training block, F(7,21) = 2.73, MSE = 0.80, p < .05, confirming that overall accuracy increased over training. There was also a statistically significant main effect of crossed type, F(2,6) = 5.80, MSE = 0.50, p < .05; Tukey HSD post hoc comparisons revealed that overall accuracy for Crossed Middle trials (73 %) was slightly higher than for Crossed Near (66 %) or Crossed Far (66 %) trials. The Crossed Type × Training Block interaction was not statistically significant.

Table 2 discloses that, at the end of Experiment 3, the Crossed string condition supported robust, but generally less accurate choice behavior than the Nonperpendicular string condition (both Noncollinear and Collinear) or the Perpendicular string condition, with responding under these other conditions remaining largely unchanged from the end of Experiment 2.

An 8 (training block) × 3 (stimulus orientation: Parallel Slanted vs. Convergent vs. Divergent) × 2 (Collinearity vs. Noncollinearity) repeated measures ANOVA on the percentage of correct first choice responses on Nonperpendicular trials showed a statistically significant main effect of collinearity, F(1,3) = 16.02, MSE = 0.12, p < .05, with overall accuracy for Collinear trials (84 % correct) being lower than for Noncollinear trials (89 % correct). The Stimulus Orientation × Collinearity interaction was significant as well, F(1,3) = 16.02, MSE = 0.12, p < .05, due to overall accuracy for Noncollinear Divergent trials (81 %) being significantly lower (as disclosed by Tukey HSD post hoc comparisons) than overall accuracy for the other Noncollinear trials, Convergent (95 %), and Parallel (92 %), but not significantly different from overall accuracy for any of the Collinear trials, Divergent (84 %), Convergent (84 %), or Parallel (83 %). No other effects or interactions were significant.

An 8 (training block) × 3 (separation: Near, Middle, Far) repeated measures ANOVA on the percentage of correct first choice responses on Perpendicular trials yielded a significant main effect of separation, F(1,3) = 16.52, MSE = 0.06, p < .01, due to overall accuracy for the Near separation (87 %) being significantly lower (as disclosed by Tukey HSD post hoc comparisons) than overall accuracy for the Middle (92 %) and Far (91 %) separations.

Overall, the pigeons’ virtual string task performance at the end of Experiment 3 exhibited considerable versatility when the birds were concurrently required to choose the correct button under all three basic spatial configurations: Perpendicular, Nonperpendicular, and Crossed.

General discussion

So far, over 1 billion people are estimated to have played “Angry Birds.” In this wildly popular computer game, players operate a virtual slingshot to fire squawking wingless birds at a variety of structures which house snorting green pigs alleged to have stolen the birds’ eggs.

Beyond its amazing entertainment value, “Angry Birds” is actually a game of physical mechanics—one capitalizing on the intricate interplay between projectile motion and structural rigidity. Of course, a virtual slingshot does not much resemble a real one: It is not made of metal or wood, it does not contain an elastic band, and the trajectory of the launched bird is not Perpendicular to the shooter. Nevertheless, the essence of an actual slingshot is effectively captured by this virtual version, thereby contributing to the player’s enjoyment of the game.

Of course, we did not design our virtual string task to entertain our pigeons; rather, we attempted to capture the essence of real patterned-string problems in a virtual environment to see whether our birds would respond in accord with the behavior of animals that are given the actual string task. Across the many different counterbalancings that we concurrently arranged in our final experiment, our pigeons’ choice accuracy averaged 90 % in the Perpendicular condition, 86 % in the Nonperpendicular condition, and 74 % in the Crossed condition. This ordering nicely accords with the general trend that has been reported across decades of research with dozens of animal species and suggests that our virtual patterned-string task did in fact capture essential properties of the actual patterned-string tasks that have so far been studied.

In addition, our pigeons’ high accuracy levels on all three different types of concurrently scheduled patterned-string problems attest to the birds’ flexibility in responding to many different virtual patterned-string configurations—an unprecedented finding in the literature. Those high accuracy levels also render implausible those accounts of our pigeons’ virtual string-pulling behavior that rely solely on any particular alignment of the dishes and choice buttons.

We also found that specific spatial variables strongly affected the birds’ Perpendicular string performance; shorter strings were more conducive to learning than were longer strings. Longer strings not only entail a greater spatial distance between the choice button and the dish, but they also require more pecks and more time to earn food. So, distance, effort, and time may have contributed to our pigeons’ choice behavior. In addition, not all Nonperpendicular string arrangements were created equal; those patterns in which both dishes and both response options were misaligned (Crossed) were more challenging for the pigeons to learn than those patterns in which one dish and one response option were misaligned (Convergent, Divergent, or Parallel).

All of the above patterned-string effects were measured over many trials and considerable periods of time thereby permitting us to properly assess their statistical reliability. It would have been difficult at best to conduct similar investigations with conventional patterned-string training methods. We therefore believe that there may be special merits to investigating patterned-string performance in the virtual world of a computerized touch screen environment.

Of course, it would be reassuring if our many detailed findings with virtual stimuli can be reproduced with actual stimuli. On this score, Schmidt and Cook (2006) have found that pigeons can—over some 300 differentially reinforced discrimination training trials—learn to retrieve a full dish of food by pulling the ribbon to which it was attached in preference to another ribbon that was detached from a second full dish of food by a visible gap. These continuous and discontinuous ribbons were placed in Parallel Slanted Noncollinear alignment with two full food dishes, whereas only one of our two virtual dishes was full and there were no gaps in our virtual strings. Still, the work of Schmidt and Cook (2006) proves that pigeons too can solve actual patterned-string tasks and suggests that these animals are suitable subjects for further study.

Another benefit of our virtual patterned-string task methodology is its amenability to evaluating the role of prior experience in choice performance. In this first effort to assess the fidelity of our virtual touch screen technique to actual string methods, we sequentially exposed pigeons to virtual patterned-string problems of purportedly different levels of difficulty, just as had Harlow and Settlage (1934) with monkeys. One might expect such easy-to-hard training to help pigeons master tasks of increasing levels of difficulty.

The patterns of behavior that we saw when we introduced novel string configurations in Experiments 2 and 3 provided clear evidence of immediate transfer, but only if no conflicting information was conveyed by the proximal and distal ends of the strings, as was the case with the Noncollinear configurations. Such conflicts might have arisen because initial training with the Perpendicular configurations in Experiment 1 permitted the pigeons to simply attend to the vertical alignment of the proximal and distal ends of the strings, thereby preventing robust transfer to the majority of the new string arrangements given in Experiments 2 (Collinear configurations) and 3 (Crossed configurations).

Such attentional biasing might be precluded by training the pigeons with all three types of string configurations from the outset. This possibility can readily be implemented with our virtual patterned-string task, but it would be difficult to do so with conventional methodology because of the large number of spatial configurations that would have to be programmed.

Before concluding, we wish to discuss the matter of the connectedness of the full/empty dishes and the correct/incorrect response buttons in our virtual string task. Of course, in our task, there are “no (real) strings attached.” Some readers might thus question whether we have truly captured the gist of actual string problems, where the connectedness of strings and objects is deemed to be critical (e.g., Mulcahy et al. 2012). We have two points to make in this regard.

First, the virtual strings in our task were essential to our pigeons’ accurate responding in Experiment 3. Figures 1 and 3 illustrate that Perpendicular and Crossed string trials are identical in the vertical alignment of the dishes and buttons; the only thing that distinguishes these trials is the spatial arrangement of the strings (vertically connected on Perpendicular trials, crisscrossed on Crossed trials). Thus, our pigeons’ accurate choice responding in Experiment 3—with these two types of trials randomly intermixed—can only be explained by the birds’ distinguishing the differential connectedness of the dishes and the buttons.

Second, video recordings of our pigeons’ behavior (see Supplementary Online Material) reveal many instances in which the birds scan and bob their heads along the string, often looking toward and pecking at the dish as they move it down the screen. These video records combine with the touch screen pecking data to suggest that our pigeons did appreciate the connectedness of the dishes and the buttons via the strings—all virtually represented in our touch screen task.

We therefore believe that our virtual patterned-string task represents a promising innovation in comparative and developmental psychology. It may permit expanded exploration of other species and variables which would otherwise be unlikely because of inadequacies of conventional string task methodology or sensorimotor limitations of the organisms. We hope that others join us in further assessing the fidelity of our virtual task to the real-world task as well as in elucidating the perceptual and conceptual bases of patterned-string task performance.

Acknowledgments

We would like to thank Sacha Perez for help in preparing the videos in the online supplementary material.

Supplementary material

10071_2013_608_MOESM1_ESM.doc (52 kb)
Supplementary material 1 (DOC 52 kb)

Copyright information

© Springer-Verlag Berlin Heidelberg 2013