The literature provides only moderate evidence of a role for anticipation in mediation of short-term retention in pigeons.  The strongest evidence comes from a spatial task in which pigeons appear able to remember locations not yet chosen.  The differential outcome literature is consistent with mediation by anticipated affective outcomes, but may reflect instead persisting emotional or behavioral reactions elicited by samples correlated with outcomes that differ affectively.  Research using many-to-one sample-to-choice mapping procedures is similarly not definitive with regard to whether the codes mediating retention have anticipatory content.  Moreover, research has failed to obtain evidence that short-term retention can be mediated by either (a) anticipated outcomes which are affectively equivalent or (b) simple response intentions.  Finally, although analysis of confusion errors and interference effects suggests prospective coding in choice matching with lines or colors (at least in procedures involving the simultaneous presentation of three choice stimuli), the small sample size and mixed statistical evidence suggests caution in evaluating the theoretical significance of those results.  Continued research is necessary to clarify the role of anticipation as a mediator of short-term retention in pigeons.

Chapter Outline & Navigation I.   Introduction II.  The Delayed Matching-To-Sample Procedure III. Retrospective and Prospective Coding           The distinction           Evidence for prospective code content               Summary  IV. Outcome Expectancies and Short-Term Retention            The differential outcome procedure           The differential outcome effect             Differential color outcome           Summary V. "Simple" Response Intentions and Short-Term Retention            Simple vs. conditional discrimination procedures           Retention in simple and conditional procedure           Summary VI. Conclusions VII. References

     I. Introduction 

The reader may view as a truism the notion that at least some species of animals are capable of anticipating future events.  Such anticipation, one might suppose, could be triggered by the occurrence of particular stimuli that in the past have been followed reliably by some event of significance (e.g., the occurrence of food).  Similarly, one might suppose further, such anticipation could be evoked by the actions in which an animal is presently engaged to the extent that those actions have in the past been followed reliably by some event of significance.  Pavlovian or classical conditioning experiments represent a laboratory instantiation of situations in which stimuli are established as reliable predictors of other stimuli.  Instrumental or operant conditioning experiments represent a laboratory instantiation of situations in which actions are reliably followed by stimulus consequences.  

Although it may appear obvious that Pavlovian and instrumental learning tasks would result in organisms learning to anticipate future events on the basis of current stimuli or present actions, respectively, the notion that animals can anticipate future events was strongly resisted by the majority of learning theorists from the early 1900s to the late 1960s.  These theorists, often referred to as S-R (stimulus-response) behaviorists, prominent among whom were Thorndike, Watson, Guthrie, Hull, and Spence, expended considerable effort in developing explanations of behavior that eschewed reference to anticipation or expectancy.  The S-R behaviorists argued that an adequate explanation of the behavior of animals, including humans, could be provided without introducing cognitive processes such as anticipations and expectancies.  Instead, they argued, behavior could be explained in terms of simpler psychological processes, prominent among which was the process of new stimulus-response habit or reflex formation.  On this view, learning involves the strengthening and weakening of stimulus-response habits or reflexes as a result of the organism's experience.  

For example, Thorndike (1898) proposed the well known Law of Effect which maintained that an action producing a favorable outcome became more strongly connected  to the stimuli present when that action occurred.  Similarly, an action producing an unfavorable outcome became less strongly connected to the stimuli present when that action occurred.  Hence, the organism's behavior adjusted to the situation through this mechanical (unthinking, automatic) process of strengthening and weakening of stimulus-response habits as a function of experience.  That is, as an organism's experience in a situation increased, the stimuli present would, with increasing probability, trigger reactions that in the past had produced a favorable outcome.  Given that an adequate explanation of behavioral adjustment as a function of experience could be provided in terms of the low-level psychological process of the strengthening and weakening of stimulus-response habits, the S-R behaviorists contended that the introduction of psychologically more complex mechanisms, such as expectations, was superfluous.  

However, some theorists in the pre 1970 era, most notably Tolman and his associates, argued that an adequate explanation of behavioral adjustment could not be based on a law-of-effect process.  In particular, Tolman (1932, 1948) argued that organisms acquire knowledge about an environment as a function of exposure to that environment.  Hence, Tolman maintained that organisms do learn "what-leads-to-what".  In a maze situation, for example, Tolman argued that organisms do not learn a series of stimulus-response habits which get them from the starting point to the goal location, as the S-R behaviorsts maintained.  Instead, Tolman contended, organisms learn about the geographical properties of the maze.  That is, organisms form an internal representation of the maze, a "cognitive map", and learn to expect food to be encountered in the location where it had been encountered previously.  

Much of Tolman's research was devoted to demonstrating the inadequacies of a law-of-effect based stimulus-response explanation of behavior.  In so doing, Tolman hoped to convince other theorists that the introduction of higher-level psychological processes, such as expectations and cognitive maps, was justified - - indeed, necessary - - to explain behavioral adjustment.  One aspect of Tolman's research involved demonstrating that learning could occur even when responses were not followed by favorable or unfavorable outcomes.  For example, Tolman and Honzik (1930b) provided evidence that rats learn the layout of a complex maze even when no favorable outcome was present in the goal box (i.e., the goalbox was empty), learning that clearly could not be explained by a law-of-effect process.  Another aspect of Tolman's research involved demonstrating that rats represent the geographical properties of mazes and learn to expect food in a particular location in that environment.  For example, Tolman and Honzik (1930a) and Tolman, Ritchie, and Kalish (1946a, b) provided evidence that rats learn where the goal is located, rather than the specific responses required to get to the goal. 

Although the research of Tolman and his associates clearly revealed the inadequacy of explanations of behavior based solely on the Law of Effect, that research did not convince the majority of theorists that the introduction of expectancies and cognitive maps as explanatory constructs was required.  Indeed, Hull (1943, 1952), Spence (1956, 1960), and others (often referred to as neo S-R behaviorists) sought adequate explanations of the phenomena discovered by Tolman and his associates without abandoning a strict stimulus-response analysis of behavior.  Although the neo S-R behaviorists introduced explanatory mechanisms in addition to the Law of Effect, they maintained the fundamental proposition that learning involved nothing more than the strengthening and weakening of stimulus-response habits.  However, research in Pavlovian conditioning conducted in the late 1960s challenged this fundamental proposition. 

In particular, Kamin's (1968, 1969) discovery of blocking, Rescorla's (1966, 1968) work on contingency effects, and Wagner's (1969; Wagner, Logan, Haberlandt & Price, 1968) work on relative cue validity effects made clear the inadequacy of a stimulus-response conception of Pavlovian conditioning.  This research revealed that learning occurred in Pavlovian conditioning only when a stimulus was informative with regard to the probability of occurrence of a second stimulus.  A consideration of the details of this research is beyond the scope of the present article, and it sufficient to note that the work of these researchers revealed that organisms in Pavlovian conditioning situations learn something in addition to, or, more likely, other than stimulus-response habits.  Although these data suggest that the "something else" which is learned is an expectancy for the occurrence or nonoccurrence of the second stimulus upon presentation of the first, they do not compel such a conclusion.

More contemporary research in both Pavlovian and instrumental conditioning has, however, produced definitive evidence that organisms do indeed learn "what-leads-to-what"; that is, that organisms do learn to anticipate the consequence of stimuli and/or their own actions.  Particularly definitive evidence for anticipation has been provided by studies in which the value of the consequent has been modified after conditioning.  In these studies, initially a stimulus (e.g., Holland & Rescorla, 1975; Holland & Straub, 1979) or an action (e.g., Colwill & Rescorla, 1985; Dickinson, Campos, Varga, & Balleine, 1996) has been established as a reliable predictor of an outcome (e.g., sucrose pellets or standard food pellets).  Subsequently, the value of the consequent has been reduced (e.g., by satiating the animal on that consequent, or by making the consequent aversive by pairing it with induced illness).  During testing in which the consequent is no longer presented, animals react to the predictor, whether stimulus or action, in a way clearly revealing that they had learned what is predicted by that stimulus or action.  For example, rats that had initially reacted with agitated excitement to a tone paired with sucrose, show little excitement to the tone following devaluation of the sucrose.  Similarly, rats that had earlier eagerly pressed a lever to obtain sucrose pellets, will, after devaluation of sucrose, be reluctant to press the lever.

The evidence is now sufficient to convince the majority of learning theorists that some species of animals are capable of anticipating the consequences of environmental stimuli and/or their own actions.  The authors of this article are memory researchers, and hence are concerned primarily with how learned information is utilized subsequent to its acquisition (Spear & Riccio, 1994).  In particular, our interest is focussed on the mechanisms and processes which permit organisms to retain information over relatively short intervals (i.e., seconds to minutes) and to utilize that information at the end of the interval.  One abiding interest in our laboratory has been whether animals, specifically pigeons, learn to anticipate future events and whether such anticipation can also mediate retention or memory-based performance.  The remainder of this article is concerned with this question.

In the next section (II), we describe a procedure, known as delayed matching-to-sample, which has been particularly useful in analytical investigations of short-term retention in pigeons.  In the following section (III), we distinguish between retrospective and prospective coding (the latter, but not the former, involving anticipation of future events).  In that section, we ask whether there is convincing evidence that pigeons, at least under some circumstances, code prospectively (and, hence, reveal evidence for mediation of retention by anticipated future events).  We then turn (section IV) to a consideration of whether an anticipated outcome (e.g., food or water for correct responding) can mediate retention.  In section V, we consider the issue of whether pigeons can use a "simple" response anticipation (i.e., to respond or not to respond) to guide or mediate short-term retention.  And, finally, in the last section (VI) we summarize the preceding discussion and address some questions and issues raised by this work, and consider the implications of those questions and issues for our understanding of memory processes and mechanisms.

     II. The Delayed Matching-To-Sample Procedure 

The matching-to-sample procedure has been used in numerous investigations of short-term retention in pigeons.  Matching is a conditional discrimination task because the reinforcement contingencies vary from trial to trial.  For example, trials might always involve a choice between pecking a key illuminated with red light or pecking a key illuminated with green light.  These choice stimuli are typically presented on the outer-most keys in a row of three, and their positions are counterbalanced such that, over trials, each stimulus appears equally often on the right and left keys.  On some trials, pecking the red key produces access to food (and hence pecking the green key does not), and on other trials the reverse is the case (i.e., pecking the green key produces access to food and pecking the red key does not).  A conditional cue, typically referred to as the sample stimulus, is presented at the beginning of each trial and indicates the reinforcement contingencies in effect on that trial.  The sample stimulus is typically projected on the center key and, in the identity version of matching, is physically identical to the choice stimulus that must be pecked to produce reinforcement on that trial.  Hence, in the present example, a red sample stimulus indicates that pecking the red, but not the green, choice stimulus will produce reinforcement, whereas a green sample stimulus indicates that pecking the green, but not the red, choice stimulus will produce reinforcement.  Which sample stimulus is presented (and hence which choice stimulus is correct) varies randomly from trial to trial.  In the simultaneous version of matching, the sample stimulus remains available on the center key until the animal responds to one of the choice stimuli (click here for animated illustration).  

After 10 to 15 sessions of training, with each session comprised of 48 trials, most pigeons peck the matching (and hence correct) choice stimulus on 95% or more of the trials.  Because in simultaneous matching the sample stimulus remains available until the choice response has been made, accurate performance requires only that the organism learn the rules relating sample stimulus and correct choice stimulus (e.g., choose green if green sample, and choose red if red sample).  An additional short-term retention requirement can be imposed by temporally separating presentation of the sample stimulus and presentation of the choice stimuli.  This is accomplished by interpolating a short delay or retention interval, typically a few seconds in duration, between termination of the sample stimulus and presentation of the choice stimuli.  Because the sample is not present at the time the choice response is emitted in the delayed version of matching-to-sample, accurate choice requires utilization of information derived during sample presentation and retained during the delay interval (click here for animated illustration).  

III.  Retrospective and Prospective Coding 

The Distinction  

As noted in the preceding section, accurate performance in delayed matching-to-sample requires the retention and utilization of information derived during sample presentation.  In the early 1980s, several investigators recognized that the information or code which mediates short-term retention in delayed matching-to-sample could have either a retrospective or prospective content (e.g., Grant, 1981; Honig & Thompson, 1982; Riley, Cook, & Lamb, 1981; Roitblat, 1980).  In particular, the animal could retain features of the sample stimulus during the delay, a retrospective or "backward-looking" code.  During testing, the retrospective code, in combination with the rules learned during simultaneous matching, would be sufficient to generate accurate choice performance.  Alternatively, the rules learned during simultaneous matching could be activated during sample presentation to generate an anticipatory code.  That is, the code could represent features of the correct choice stimulus, a prospective or "forward-looking" code.  

The distinction between retrospective and prospective (or anticipatory) code content is most readily illustrated in the symbolic (or arbitrary) version of delayed matching.  As opposed to identity matching, in which the correct choice stimulus is physically identical to the sample presented on that trial, in arbitrary matching the relation between sample and correct choice stimulus is established arbitrarily by the experimenter.  For example, as illustrated in the accompanying animation, the samples might be red and green colors and the choice stimuli might be white bars oriented vertically or horizontally:  And, for example, vertical might be correct (and hence pecking it is reinforced) on red-sample trials, and horizontal might be correct (and hence pecking it is reinforced) on green-sample trials (click here for animated illustration).  

In this example, the code could have a retrospective content in that the code preserves the visual features of the sample stimulus (i.e., "redness" or "greenness" on any particular trial).  If the code is active at the time of testing, correct responding could be generated by the learned rules of the matching procedure (i.e., "if red, peck vertical" and "if green, peck horizontal"), (click here for animated illustration).  

Alternatively, the code could have an anticipatory or prospective content in that the code preserves the visual features of the correct choice stimulus (i.e., "vertical" or "horizontal" on any particular trial).  In this case, the learned rules of the matching procedure are invoked during the sample presentation period to generate the anticipatory code.  At the time of testing, the active code is sufficient to generate correct responding (click here for animated illustration).  

Evidence for Prospective Code Content  

In the present article, the primary question of interest is whether there is evidence that pigeons employ anticipatory or prospective codes in delayed matching (and in related procedures that require information to be maintained over relatively short intervals).  In this section we consider evidence for prospective coding obtained in standard and many-to-one matching procedures, and from a procedure requiring retention of spatial information over short intervals.  

In standard matching. We use the term standard matching to refer to identity or arbitrary matching procedures in which (a) the samples are colors, lines and/or forms and (b) each sample stimulus is associated with a unique choice stimulus.  Experiments employing the standard procedure have provided some evidence for prospective code content.  Perhaps the most powerful of this evidence was reported by Roitblat (1980, Exp. 3).  Roitblat's procedure was based on an experiment by Conrad (1964) in which human subjects were employed.  

Conrad (1964) was interested in whether humans would code visually presented letters in terms of their visual properties (e.g., letter B as the visual image of B) or in terms of their acoustic properties (e.g., letter B as the sound of B).  He argued that if subjects code letters visually, then memory errors should reflect the visual similarity between letters (e.g., misremembering B as visually similar R).  If subjects code letters acoustically, then memory errors should reflect the acoustic similarity between letters (e.g., misremembering B as acoustically similar V).  Conrad found that errors were related to acoustic, rather than visual, similarity, suggesting that the letters were coded in terms of their acoustic properties. 

In the case of pigeons performing delayed matching with colors or line orientations, it is reasonable to assume that the code represents visual features of the stimuli.  Hence, the question of interest is not the modality of the code (e.g., visual or acoustic), but rather whether the code preserves visual features of the sample (i.e., retrospective content) or visual features of the correct choice stimulus (i.e., prospective content).  As Roitblat (1980) noted, Conrad's (1964) research suggests a way to determine whether pigeons in delayed matching code features of the sample or code features of the correct choice stimulus.  Specifically, if pigeons code retrospectively, then memory errors should reflect similarity among the samples (e.g., confusing similar orange and red samples rather than dissimilar orange and blue samples).  If, on the other hand, pigeons code prospectively, then memory errors should reflect similarity among the choice stimuli (e.g., confusing similar vertical and nearly vertical choice stimuli rather than dissimilar vertical and horizontal choice stimuli). 

To implement this test, Roitblat (1980) trained birds in an arbitrary matching task involving three samples and three choice stimuli.  Two birds were trained with orange, red, and blue color samples and vertical, nearly vertical, and horizontal line choice stimuli, and the remaining bird received lines as samples and colors as choice stimuli.  As shown in Figure 1, similar pairs of samples were associated with dissimilar pairs of choice stimuli, and dissimilar pairs of samples were associated with similar pairs of choice stimuli.  A delay ranging between .25 and 5.60 s intervened between sample termination and presentation of all three choice stimuli.  
   
To further illustrate the rationale underlying Roitblat's (1980) experiment, consider a trial involving the orange sample and the nearly vertical correct choice.  On this trial type, the birds will err by choosing either the vertical line associated with the blue sample or the horizontal line associated with the red sample.  If the birds are remembering "orange" (retrospective code), they should tend to confuse orange with red and should therefore select the horizontal line.  If the birds are remembering "nearly vertical line" (prospective code), they should tend to confuse nearly vertical with vertical and should therefore select the vertical line.  

The data revealed that confusion between similar choice stimuli increased faster across the retention interval than did confusion between similar samples, although the effect was statistically significant only for the birds receiving color samples and line choice stimuli.  This finding provides evidence for prospective code content, and hence suggests that the anticipation of a color or a line can mediate retention in delayed matching.  Roitblat and Scopatz (1983) maintained that this conclusion was further strengthened by an analysis of proactive interference.  Roitblat and Scopatz employed three pigeons and an identity-matching task with red, green and blue stimuli.  As in Roitblat's (1980) experiment, following sample presentation all three stimuli were presented simultaneously for a choice.  They examined choice on each trial (after the first) as a function of the sample presented and choice made on the immediately preceding trial.  They found that choices on a trial were influenced by the choice made, rather than by the sample presented, on the preceding trial.  Roitblat and Scopatz argued that this result also suggests that pigeons remember "what to do" rather than "what happened" in standard delayed matching (i.e., a prospective code).  

It may be, however, that the Roitblat and Scopatz (1983) finding can be explained without assuming that samples are coded prospectively.  It could be argued, for example, that pecking a choice stimulus activates a representation of that stimulus.  Moreover, the representation of the chosen stimulus might interfere with the retrospective code of the sample presented earlier on that trial (retroactive interference effect).  It then follows that, at the time of choice on the succeeding trial, the animal is more likely to remember the stimulus chosen than the sample presented on the preceding trial.  This persisting memory for the stimulus chosen on the preceding trial would compete with memory for the sample on the current trial for control over choice responding.  Hence, choice on a trial should be related more strongly to the stimulus chosen than to the sample presented on the preceding trial.  It is also possible that sample coding errors contributed to Roitblat and Scopatz's finding.  Again assume that pigeons code samples retrospectively.  Assume further that errors in delayed matching often reflect the misidentification of the sample.  For example, a red sample activates the code "green".  On such a trial, the code "green" would produce choice of the green test stimulus.  If memory of either (or both) the code activated by the sample (i.e., "green") or the stimulus chosen (i.e., green) persisted until choice on the succeeding trial, then choice on that trial should be related to the stimulus chosen (i.e., green) rather than the sample presented (i.e., red) on the preceding trial.  In our view, therefore, one could maintain the view that samples are coded retrospectively and provide an adequate explanation of Roitblat and Scopatz's data.  If so, then their data do not provide definitive evidence for prospective coding.  

It should be noted that Urcuioli and Zentall (1986) obtained data which led them to conclude that pigeons code retrospectively in standard delayed matching.  They employed highly discriminable colors (red and green) and less discriminable lines (horizontal and vertical) as stimuli.  Some birds received color samples mapped to color or line choice stimuli.  Other birds received line samples mapped to color or line choice stimuli.  

Urcuioli and Zentall (1986) reasoned that retention should be a function of the discriminability of the event being retained; that is, highly discriminable events (colors in the present case) should be remembered more effectively than less discriminable events (lines in the present case).  Hence, if pigeons code retrospectively, then birds receiving highly discriminable colors as samples should demonstrate better retention than those receiving less discriminable lines, regardless of the dimension of the choice stimuli.  If, on the other hand, pigeons code prospectively, then birds receiving highly discriminable colors as choice stimuli should demonstrate better retention than those receiving less discriminable lines, regardless of the dimension of the sample stimuli. 

After training, pigeons received two retention tests, one with delays ranging from 0 to 4 s, and a second with delays ranging from 0 to 8 s.  Urcuioli and Zentall (1986) found that sample dimension influenced rate of forgetting, in that retention functions were less steep for pigeons receiving color samples than for those receiving line samples.  In contrast, choice dimension had little effect on rate of forgetting, although accuracy collapsed across delays was higher in pigeons trained with color choice stimuli than in those trained with line choice stimuli.  Urcuioli and Zentall argued that these results suggest that the pigeons coded the samples retrospectively and hence demonstrated better retention when the samples were the more discriminable colors than when they were the less discriminable lines.  

However, Urcuioli and Zentall's (1986) conclusion is open to criticism on both conceptual and interpretational grounds.  At a conceptual level, Roitblat (1993, p. 180) has argued that both prospective and retrospective coding accounts predict that, as found by Urcuioli and Zentall, more difficult to discriminate samples would result in lower accuracy and more rapid forgetting.  Roitblat notes that samples must be identified before a prospective code can be activated, and that more difficult to discriminate stimuli are likely to be identified less accurately and with less strength.  It then follows that more difficult to discriminate samples should result in reduced accuracy and faster forgetting than samples which are more easily discriminated, even if birds code prospectively.  

At a methodological level, it is unclear whether the significant Group x Delay interaction obtained by Urcuioli and Zentall (1986) reflects between-group differences in rate of forgetting or a ceiling artifact.  With regard to the latter, it should be noted that performance on 0-s delay trials was close to 100% in groups trained with color samples (estimating from Figures 1 through 4 in their article, accuracy appears to range from 96 to 98% correct at the 0-s delay).  Hence, it could be that accuracy differences as a function of sample dimension were obscured by a ceiling artifact at the 0-s delay.  At increasingly longer delays, performance dropped below ceiling, allowing between-group differences to emerge.  It may be, therefore, that sample dimension did not influence rate of forgetting, in spite of the significant Group x Delay interaction.  

Some support for this view is provided by examining performance only at longer delays, where the operation of a ceiling artifact is less likely (see Figure 2; the data points in the left panel were estimated from Figure 2, and those in the right panel from Figure 4, in Urcuioli & Zentall, 1986). With either line or color choice stimuli, whether one examines performance at 2- and 4-s delays in the short-delay test (not shown in Figure 2), or performance at 2-, 4-, and 8-s delays in the long-delay test, there is no evidence that forgetting is more rapid on line-sample trials than on color-sample trials.  Indeed, with color choice stimuli (Figure 2 right panel), rate of forgetting appears to be somewhat faster on color- than on line-sample trials.  Although accuracy collapsed across delay length was higher with color than with line samples, this result is easily explained in terms of the differential effectiveness with which easy versus difficult to discriminate samples activate the correct prospective code.  

In many-to-one matching.  As noted previously in this article, matching procedures can be differentiated on the basis of whether the matching relation is based on physical identity or is established arbitrarily by the experimenter.  A second distinction between matching procedures is based on the sample-to-choice mapping ratio.  In the procedures considered so far, that ratio has been one-to-one: each sample stimulus is associated with one and only one correct choice stimulus.  Hence, the number of stimuli in the sample set is equivalent to the number of stimuli in the choice set.  Therefore, the number of codes required to mediate accurate retention in a one-to-one procedure is independent of whether the codes represent features of the sample stimuli (retrospective coding) or the correct choice stimuli (prospective coding). 

Alternatively, in a many-to-one sample-to-choice mapping, two or more sample stimuli are associated with each choice stimulus (e.g., Grant, 1982a, 1982b, 1991; Grant & Spetch, 1993; Santi & Roberts, 1985; Urcuioli, DeMarse, & Zentall, 1994; Zentall, Jagielo, Jackson-Smith, & Urcuioli, 1987; Zentall, Urcuioli, Jagielo, & Jackson-Smith, 1989).  A many-to-one procedure appears particularly conducive to prospective coding because such coding could reduce the number of codes required for accurate performance.  For example, consider a procedure in which three samples are associated with one choice stimulus, and three different samples are associated with a second choice stimulus (a 3:1 many-to-one mapping).  Remembering the sample presented on each trial would require six different codes (one for each of the six different sample stimuli).  Alternatively, remembering the correct choice stimulus on each trial would require only two different codes (one for each of the two different choice stimuli). 

Urcuioli, Zentall, Jackson-Smith, and Steirn (1989, Exp. 2), trained birds in a many-to-one matching procedure with color (red and green) and line (vertical and horizontal) sample stimuli.  All trials involved a choice between vertical and horizontal lines presented on the outer keys.  Vertical was the correct choice on trials involving a red or vertical sample, and horizontal was the correct choice on trials involving a green or horizontal sample (see top panel in Figure 3).  Hence, the red and vertical samples were functionally equivalent (i.e., both were signals to choose vertical) and the green and horizontal samples were functionally equivalent (i.e., both were signals to choose horizontal).  

If the animal coded prospectively, only two different codes would be required.  Specifically, the two samples associated with the vertical choice stimulus would activate the code "vertical" (or "peck vertical") and the two samples associated with the horizontal choice stimulus would activate the code "horizontal" (or "peck horizontal").  If, on the other hand, the animal coded retrospectively, four different codes (one to represent each of the four samples) would be required.  Hence, a many-to-one mapping appears particularly suited to use of prospective coding in which each of the two functionally equivalent samples activates the same code.  

Urcuioli et al. (1989, Exp. 2) did obtain evidence that functionally equivalent samples activate the same code.  They reasoned that if functionally equivalent samples activate the same code, then one might observe transfer of performance between functionally equivalent samples.  To assess this possibility, the many-to-one training phase described previously was followed by a second training phase in which the color samples employed in the many-to-one training phase were associated with new choice stimuli, circle and dots (see middle panel in Figure 3). Following acquisition, the animals received a transfer test in which the line samples from many-to-one training were followed by the circle and dots choice stimuli from one-to-one training (see bottom panel in Figure 3). 
   
If many-to-one training resulted in a color and line activating the same code, and if that same code was used during the subsequent one-to-one training, then presentation of a line sample during transfer testing might tend to produce the same response as that produced by the color sample to which it is functionally equivalent.  Given the design specified in Figure 3, in transfer testing, presentation of vertical would tend to produce the choice associated with a red sample (i.e., circle), and presentation of horizontal would tend to produce the choice associated with a green sample (i.e., dots).  Such a tendency would produce positive transfer (accuracy above 50%) in group consistent and negative transfer (accuracy below 50%) in group inconsistent.  This is precisely what Urcuioli et al. (1989) observed:  During the first 16 trials of transfer testing, accuracy was 71.9% in group consistent and 35.4% in group inconsistent.  Grant and  Spetch (1994) have observed similar transfer effects in subjects initially trained in a many-to-one procedure involving line and temporal (2- and 10-s light presentations) samples.  

Although such transfer results provide strong support for the notion that functionally equivalent samples in a many-to-one mapping activate the same code, these results are not informative in regard to the content of the code.  Roitblat's (1980; Roitblat & Scopatz, 1983) research employing standard matching procedures encourages the view that the code content in many-to-one procedures, at least those employing colors and lines as stimuli, is prospective.  Nonetheless, as pointed out some time ago (Grant, 1982a), the data from experiments employing colors and lines in many-to-one procedures can be explained adequately by the notion that functionally equivalent samples activate the same code, regardless of the content of that code.  It is possible, for example, that samples associated with a vertical choice stimulus activate a retrospective "sample A" code and samples associated with a horizontal choice stimulus activate a retrospective "sample B" code.  It is also possible that samples associated with a vertical choice stimulus active a prospective "vertical" (or "peck vertical") code and samples associated with a horizontal choice stimulus activate a prospective "horizontal" (or "peck horizontal") code. 

Indeed, there is evidence that, at least under some conditions involving "nonstandard" samples (such as the occurrence and nonoccurrence of food), many-to-one training results in codes that do not have prospective content (Kelly & Grant, 1998a; Urcuioli et al., 1994).  In our view, the results with nonstandard samples do not discourage the likely possibility that many-to-one training with standard samples (i.e., colors, lines, and forms) results in codes having prospective or anticipatory content.  Nonetheless, it is our view that studies employing many-to-one procedures, although often providing evidence consistent with use of prospective codes, do not provide conclusive evidence for prospective code content.  

In a spatial task. Particularly definitive evidence for prospective coding was obtained by Zentall, Steirn, and Jackson-Smith (1990) who developed a task to assess memory for spatial location in pigeons.  An array of five pecking keys was illuminated at the onset of a trial.  Choosing (i.e., pecking) a particular key was correct if that key had not been chosen previously on that trial.  An error consisted of choosing a key which had been previously chosen on that trial, and resulted in a 2.5-s blackout after which the keys were again illuminated for a choice.  A trial ended when all five keys had been chosen.  

The data of relevance to the issue of prospective code content came from trials in which a delay was inserted at some point in the choice sequence.  Across trials, the delay occurred after 1, 2, 3, or 4 correct choices had been made.  Zentall et al. (1990) found that the delay caused more errors when it was interpolated after choice 2 than when it was interpolated after choice 1.  This result suggests that the amount of information that the bird had to retain was greater after having made two choices than after having made only one choice.  Hence, this result suggests that the animal begins the trial using a code having retrospective content; that is, early in the trial the animal remembers which locations it has previously chosen.  Interestingly, however, a delay interpolated after choice 3 produced approximately the same amount of forgetting as a delay interpolated after choice 2, suggesting that the memory load was equivalent after either two or three choices had been made.  Equivalent memory load after two or three choices would obtain if pigeons remembered retrospectively after having made two choices (and hence remembered the two locations previously chosen) and remembered prospectively after having made three choices (and hence remembered the two locations not yet chosen).  

Further support for the notion that pigeons switched to prospective coding later in the trial was provided by the finding that a delay interpolated after choice 4 produced less forgetting than a delay interpolated after choice 2 or 3 and, moreover, produced about the same amount of forgetting as a delay interpolated after choice 1. This result is expected if the animal codes prospectively later in the trial because the memory load after choice 4 (i.e., 1 item) is equivalent to that after choice 1 and is less than that after choice 2 or 3 (i.e., 2 items).  Hence, the results of Zentall et al.'s (1990) experiment provides strong evidence that codes having prospective or anticipatory content can mediate retention in pigeons (for a conceptually similar experiment employing rats, see Cook, Brown, & Riley, 1985).  

Summary 

The data reviewed in this section provide some support for the notion that codes having prospective or anticipatory content can mediate short-term retention in pigeons.  The strongest of such evidence is provided by the research of Zentall et al. (1990) using a spatial task.  Evidence from other procedures is, at best, suggestive of prospective mediation.  In particular, although Roitblat's (1980) analysis of confusion errors in standard delayed matching provides evidence for mediation of short-term retention by codes having prospective or anticipatory content, that research involved a small sample size (i.e., three) and the effects suggesting prospective coding were significant only in two of the three subjects.  A replication of Roitblat's experiment would be welcome.  Moreover, the results of Roitblat and Scopatz's (1983) analysis of interference can, in our view, be interpreted without maintaining that prospective coding was mediating retention.  Finally, data from many-to-one matching procedures involving colors, lines, and/or forms support the view that functionally equivalent samples activate the same code.  Although these procedures may result in birds using codes having prospective content, the data do not demand such an interpretation.  In our view, therefore, only the data from Zentall et al.'s experiment provide definitive evidence for mediation of short-term retention in pigeons by codes having anticipatory content.

In the next two sections we consider whether other types of anticipations can mediate short-term retention.  Specifically, we consider whether anticipated trial outcomes (Section IV) and anticipated response patterns (Section V) can mediate short-term retention in pigeons.  

  IV. Outcome Expectancies and Short-Term Retention  

The Differential Outcome Procedure

In the matching procedures considered so far in this chapter, correct responses are always reinforced by the same outcome, typically 3-s access to grain.  In the differential outcome procedure, a procedure developed by Trapold (1970), this is not the case.  Peterson and his associates adapted a differential outcome procedure to the delayed matching task with pigeons (Brodigan & Peterson, 1976; Peterson, Wheeler, & Armstrong, 1978).  In their procedure, color sample stimuli were followed by line choice stimuli and, importantly, correct responses produced one of two outcomes, food or water, with equal probability.  As illustrated in Figure 4,  in the differential outcome group, the type of outcome was perfectly correlated with type of sample and correct choice stimulus.  Hence, for example, choice of vertical after a red sample always produced grain, whereas choice of horizontal after a green sample always produced water.  In the nondifferential outcome group, there was no correlation between sample/choice stimuli and outcome.  That is, choice of vertical after a red sample produced food and water equally often, and, similarly, choice of horizontal after a green sample produced food and water equally often. 
   
The research discussed in Section I of this article suggests that pigeons learn to anticipate trial outcomes after identifying the sample.  Hence, birds in the differential outcome group would learn to expect food if the sample is identified as red (because correct choice on red-sample trials produces food) and to expect water if the sample is identified as green (because correct choice on green-sample trials produces water).  Notice that these pigeons have two potential sources of control over choice responding: a code (either retrospective or prospective) and an outcome expectancy.  With regard to the latter, an expectancy for food should lead to choice of vertical because choice of vertical is always (and choice of horizontal is never) reinforced in the presence of a food expectancy.  An expectancy for water, on the other hand, should lead to choice of horizontal because choice of horizontal is always (and choice of vertical is never) reinforced in the presence of a water expectancy.  

Notice that in the nondifferential outcome group, this second source of control over choice responding is not available because the same expectancy should be present on all trials.  Pigeons in this group should expect both food and water regardless of whether the sample is identified as red or green because both types of trials terminate, after a correct choice, in food and water equally often.  Hence, the same expectancy is present on trials in which choice of vertical is reinforced and on trials in which choice of horizontal is reinforced.  Therefore, the expectancy provides no basis for responding differentially to the choice stimuli.  

Numerous pairs of outcomes in addition to food and water have been employed in differential outcome experiments.  These pairs have included different types of food (e.g., Edwards, Jagielo, Zentall, & Hogan, 1982), food and tone (e.g., Peterson, Wheeler, & Trapold, 1980), different nonzero probabilities of food (e.g., Santi & Roberts, 1985), and illuminated presentations of food and no food (e.g., Peterson et al., 1980; Urcuioli, 1990; Zentall & Sherburne, 1994).  Regardless of which type of outcome pair is employed, matching performance in a differential outcome group is typically superior to that in a nondifferential outcome group.  This finding, known as the differential outcome effect, is illustrated, and its theoretical significance considered, in the next section.

The Differential Outcome Effect  
  
As noted in the previous section, use of differential outcomes typically enhances matching performance.  This enhancement is manifest in two ways.  Specifically, use of differential outcomes facilitates acquisition of the matching task and also enhances retention when delays are interpolated between sample termination and presentation of choice stimuli.  In general, the term "differential outcome effect" is used to refer to either or both of these aspects of facilitation of matching performance.

A recent experiment conducted in our laboratory provides a demonstration of the differential outcome effect.  Kelly and Grant (1998b) trained pigeons on an identity-matching task using horizontal and vertical stimuli.  A 0-s delay intervened between sample termination and onset of the choice stimuli (i.e., choice stimuli were presented immediately after termination of the sample). Correct responses were followed by 3-s of illumination of the feeder, accompanied either by access to food (hereafter referred to simply as "food") or no access to food (hereafter referred to simply as "no food").  Incorrect responses were followed by nothing (i.e., the feeder was not illuminated and food was not available).  As illustrated in Figure 5,  sample type (and hence correct choice) was correlated with outcome in the differential outcome group.  In the nondifferential outcome group, sample type (and hence correct choice) and outcome were not correlated. 
   
Figure 6 shows matching accuracy in group differential and nondifferential during the first 16 sessions of training.  A differential outcome effect is apparent in that accuracy was higher in the differential than in the nondifferential outcome group.  Sessions to meet an acquisition criterion of two consecutive sessions in which accuracy was at least 80% correct with both types of samples also suggested an enhancement in acquisition in the differential outcome group.  Specifically, the nondifferential outcome group required a mean of 24.6 sessions to met the acquisition criterion, whereas the differential outcome group required a mean of only 15.4 sessions. 

After all animals were responding accurately at a 0-s delay, a retention test was conducted in which delays of 0, 2, or 6 s intervened between sample termination and onset of the choice stimuli.  The data from this test, shown in Figure 7, revealed the second aspect of the differential outcome effect: enhanced retention in animals receiving differential outcomes.

The occurrence of the differential outcome effect and, especially, the facilitation of retention by differential outcomes, suggests that short-term retention performance can be mediated by an outcome expectancy.  Moreover, as the data in Figure 7 reveal, these outcome expectancies appear to be particularly memorable, and hence are available to guide choice responding after the code activated by the sample is no longer active.  In the next section, we consider whether outcome expectancies need necessarily involve affectively potent events in order to enhance acquisition and retention in delayed matching.  

Differential Color Outcomes  

Kelly and Grant (1998b) performed an experiment similar to that described in the preceding section.  As in the previous experiment, pigeons were trained on an identity matching task using horizontal and vertical stimuli, and correct responses were followed equally often by one of two outcomes.  However, the outcomes employed were a 1.5-s presentation of blue or yellow light.  Following a correct choice, the choice stimuli were terminated and a color was projected onto the key that the bird had chosen.  Termination of the color was followed by 3-s access to food.  As illustrated in Figure 8, sample type (and hence correct choice) was correlated with outcome in the differential outcome group.  In the nondifferential outcome group, sample type (and hence correct choice) and outcome were not correlated. 
   
Figure 9 shows matching accuracy in group differential and nondifferential during the first 16 sessions of training.  As was the case when the outcomes were food and no food, a differential outcome effect is apparent in that accuracy was higher in the differential than in the nondifferential outcome group.  Sessions to meet an acquisition criterion of two consecutive sessions in which accuracy was at least 80% correct with both types of samples also suggested an enhancement in acquisition in the differential outcome group.  Specifically, the nondifferential outcome group required a mean of 13.0 sessions to met the acquisition criterion, whereas the differential outcome group required a mean of only 8.3 sessions. 

Figure 10 shows the results of two retention tests.  The first test, using delays of 0, .3, and 1 s, was conducted immediately following acquisition of 0-s delayed matching.  The second test, employing delays of 1, 5, and 9 s was conducted after extensive training with a fixed delay of 1 s on all trials.  In neither case was there evidence that differential outcomes enhanced retention.  In fact, there was a slight tendency in the opposite direction.

In an effort to increase the salience of the color outcomes we increased their duration from 1.5 s to 5.0 s, and reduced the probability of food following correct choices from 1.0 to 0.5.  Moreover, retention tests were conducted after training at various nonzero fixed delays.  In spite of these manipulations, we have generally observed overlapping retention functions in the two groups.  To date, therefore, we have obtained no evidence that differential color outcomes enhance short-term retention.  

Summary

Research using differential outcome procedures has provided evidence that pigeons learn to anticipate affective outcomes.  If the outcomes are differential with regard to the combination of sample and correct choice, then sample presentation triggers an expectation of the outcome associated with that sample.  Data from retention testing reveal that these expectations are particularly effective in mediating short-term retention.  To date, however, we have not obtained evidence that pigeons learn to anticipate more affectively neutral outcomes (i.e., different colors).  The finding that differential color outcomes enhance acquisition could be explained without postulating that birds in the differential outcome group learn to anticipate the color outcome.  Specifically, differential color outcomes may enhance the discriminability of the line sample stimuli, a phenomenon known as acquired distinctiveness of cues.  Early in training, at which point the ability to distinguish between the lines is still developing, enhancing the discriminability of the lines by differentially associating them with color outcomes would facilitate performance and increase rate of acquisition.  After extensive training, the ability to discriminate between the lines based on their visual properties may be sufficient to allow accurate identification of the sample on all trials.  At this point, enhancing the discriminability of the lines by associating them with different colors would no longer enhance the likelihood of correct sample identification.

Although our data do not provide evidence that pigeons can anticipate color outcomes, they also do not, of course, rule out that possibility.  In particular, it is possible that birds in the differential outcome group did learn to anticipate the color outcomes, but that those anticipations failed to facilitate retention.  It may be that the anticipation of a color, in contrast to the anticipation of an affective event, is not more memorable than the code activated by the sample.  If so, no enhancement in retention would occur in spite of the fact that birds in a differential outcome group learn to anticipate the color outcome.  Research is continuing in our laboratory to determine whether there are conditions that generate expectancies for more affectively neutral events that function as effective mediators of short-term retention.

V. "Simple" Response Intentions and Short-Term Retention

Simple vs. Conditional Discrimination Procedures

As discussed previously in this article, matching-to-sample is a conditional discrimination task because the reinforcement contingencies vary from trial to trial.  In a simple discrimination task, in contrast, the reinforcement contingencies remain constant from trial to trial.  For example, pecking a horizontal line might always produce reinforcement whereas pecking a vertical line never produces reinforcement.

Honig and Wasserman (1981) modified the delayed matching procedure and a simple discrimination procedure in order to compare retention in the two tasks.  As developed in more detail below, performance in a delayed simple discrimination can be mediated by a "simple" response intention; that is, to peck or not peck the test stimulus.  Hence, if pigeons are capable of anticipating a future action, and if such an anticipation is an effective mediator of short-term retention, then retention should be higher in a delayed simple discrimination than in a delayed conditional discrimination.  

In Honig and Wasserman's (1981) procedure, trials began with presentation of an initial stimulus that, after a delay interval, was followed by presentation of a single test stimulus.  On some trials the test stimulus was positive and the first peck after 5 s in the presence of the test stimulus produced reinforcement (i.e., fixed interval, FI, 5 s schedule).  On other trials the test stimulus was negative and terminated in nonreinforcement after 5 s (i.e., fixed time, FT, 5 s extinction schedule).  

As pigeons learn to discriminate positive and negative test stimuli, rate of keypecking becomes increasingly differential:  high in the presence of positive test stimuli and low in the presence of negative test stimuli.  The extent of differential responding is indexed by a discrimination ratio that represents the proportion of responses to all test stimuli that are emitted to positive test stimuli.  Thus, a ratio of .5 indicates equivalent responding to, and hence no discrimination between, positive and negative test stimuli.  Ratios between .5 and 1.0 indicate increasingly higher levels of discrimination.  

In Honig and Wasserman's (1981) experiment, the simple and conditional procedures differed in terms of the basis for determining the positive or negative status of a test stimulus.  In the delayed  simple discrimination, illustrated in the left half of Figure 11, the initial stimulus provided all the information necessary to determine whether pecking the test stimulus would or would not be reinforced.  In the case illustrated, a red initial stimulus signalled that the test stimulus was positive, regardless of whether that test stimulus was a vertical or horizontal line.  A blue initial stimulus, in contrast, signalled that the test stimulus was negative, regardless of whether that test stimulus was a vertical or horizontal line. Click for an animated illustration of the delayed simple discrimination procedure 

In the delayed conditional discrimination, illustrated in the right half of Figure 11, the combination of initial and test stimulus determined whether pecking the test stimulus would or would not be reinforced.  As illustrated, vertical was positive after violet but negative after yellow, whereas horizontal was positive after yellow and negative after violet.  Two sets of initial stimuli were employed to allow a within-subjects comparison of retention in the two tasks (i.e., all subjects were trained and tested on both tasks). Click for an animated illustration of the delayed conditional discrimination procedure 
  
     Notice that because the identity of the test stimulus is irrelevant in the simple discrimination task, retention could be mediated by a "simple" response intention.  That is, birds could remember to peck on trials initiated by red, and not to peck on trials initiated by blue.  However, because the identity of the test stimulus is relevant in the conditional discrimination task, mediation by a simple response intention is precluded.  Hence, if pigeons can anticipate a future action (i.e., pecking or not pecking), then one might expect better retention in the simple than in the conditional task.  This prediction is based on the notion that a simple response intention (i.e., "peck" or "do not peck") is more memorable than is either a prospective code (e.g., "peck vertical" or "peck horizontal") or a retrospective code (e.g., "violet" and "yellow").  Tests of this prediction are considered in the next section.  

Retention in Simple and Conditional Procedures  

Following acquisition, Honig and Wasserman (1981) assessed retention in the two tasks.  As shown in Figure 12, retention in the simple task was considerably more robust than that in the conditional task (the data points in Figure 12 were estimated from Figure 4 in Honig & Wasserman, 1981).  This finding is consistent with the notion that in the simple task, pigeons remembered an anticipated action (to peck or not peck the test stimulus).

Although Honig and Wasserman's (1981) findings are consistent with the view that simple response intentions mediate performance in a delayed simple discrimination, their findings do not demand such an account.  As they noted, enhanced retention in the simple task could be an instance of the differential outcome effect.  Notice that in the simple task, the initial stimuli are correlated with different outcomes.  In the illustration shown in Figure 11, red is associated with a food outcome and blue with a no-food outcome.  Hence, red should come to evoke an expectation of food and blue should come to evoke an expectation of no food.  It may be that these expectations persisted throughout the retention interval and controlled responding to the test stimuli.  Specifically, an expectation for food evoked pecking of the test stimulus, and an expectation for no food failed to evoke pecking.  In the conditional task, the initial stimuli are not correlated with different outcomes and hence should evoke the same expectancy (i.e., an expectancy for food).  As discussed in Section IVb, anticipated affective outcomes are particularly effective mediators of short-term retention in pigeons.  Hence, enhanced retention in the simple task could reflect mediation of retention in that task by differential outcome expectancies.  

Urcuioli and Zentall (1990) devised a clever design which allowed them to assess whether the enhanced retention in the simple task obtained by Honig and Wasserman (1981) reflected mediation by simple response intentions or mediation by outcome expectancies.  They modified the simple task such that all trials, regardless of which initial stimulus was presented, ended in reinforcement.  The effect of this modification, of course, is to remove the correlation between initial stimulus and outcome.  As shown in the left half of Figure 13, the initial stimuli in the simple task signalled not whether the trial would end in food or no food, but rather which behavior, pecking or not pecking the test stimulus, was required to procure reinforcement.  Following a red initial stimulus, the first key peck after 5 s in the presence of the test stimulus produced reinforcement (i.e., FI 5 s schedule), regardless of whether the test stimulus was vertical or horizontal.  Following a green initial stimulus, in contrast, the bird had to refrain from pecking for 5 s to produce reinforcement (i.e., differential reinforcement of other behavior, DRO, 5 s schedule).  As in the simple task employed by  Honig and Wasserman (1981), the initial stimuli in Urcuioli and Zentall's symmetrically-reinforced simple task supplied all the information necessary to determine the behavior required at testing.  Importantly however, and in contrast to Honig and Wasserman's simple task, the initial stimuli were not correlated with differential outcomes in Urcuioli and Zentall's simple task.  Hence, retention in this task could be mediated by simple response intentions (i.e., peck, do not peck), but could not be mediated by differential outcome expectancies. Click for an animated illustration of a symmetrically-reinforced delayed simple discrimination 
   
To maintain comparability between the simple and conditional tasks, the conditional task was modified so that all trials ended in reinforcement, as was the case in the simple task.  Also as in the simple task, the response required to produce reinforcement varied from trial-to-trial (see right half of Figure 13).  Notice, however, that in the symmetrically-reinforced conditional task it was the combination of initial and test stimulus that determined whether pecking or not pecking was reinforced on the current trial.  Hence, performance in the conditional task could not be mediated by a simple response intention.  Click for an animated illustration of a symmetrically-reinforced delayed conditional discrimination 

Thus, if simple response intentions mediate retention in the simple task, then Urcuioli and Zentall's (1990) procedure should reveal enhanced retention in the simple task, as did Honig and Wasserman's (1981) procedure.  If, however, the facilitation in retention observed in the simple task by Honig and Wasserman was mediated exclusively by differential outcome expectancies, then Urcuioli and Zentall's procedure should fail to reveal enhanced retention in the simple task.  

Following acquisition, two retention tests were conducted.  The first involved retention intervals of 0, 5, and 10 s, and the second involved retention intervals of 0, 10, and 20 s.  The results of the retention tests are shown in Figure 14 (the data points plotted in Figure 14 were estimated from Figure 3 in Urcuioli & Zentall, 1990).  Although overall accuracy was somewhat higher in the simple than in the conditional task, this difference was not statistically significant in either retention test.  More importantly, and in contrast to the findings of Honig and Wasserman (1981), there was no evidence of enhanced retention in the simple task.  In fact, rate of forgetting was slightly faster in the simple than conditional task, although this difference was not statistically significant. 

Urcuioli and Zentall's (1990) results suggest that the enhanced retention in the simple task, compared to that in the conditional task, obtained by Honig and Wasserman (1981) was due entirely to mediation of retention by differential outcome expectancies in the simple task.  However, as Urcuioli and Zentall (1992) noted, the results of their 1990 study (illustrated in Figure 14) do not rule out the possibility that response intentions did mediate performance in the delayed simple discrimination.  To account for equivalent rates of forgetting in simple and conditional tasks, one would be required to assume that a simple response intention is no more (and no less) memorable than the code that mediates retention in a delayed conditional discrimination (i.e., a retrospective or prospective code).  

To further pursue the possibility that response intentions mediate performance in delayed simple discriminations, Urcuioli and Zentall (1992) employed transfer tests. In one test, illustrated in Figure 15, pigeons were trained sequentially on two simple discriminations involving different initial and test stimuli.  Following acquisition, the initial stimuli from the first task were followed by the test stimuli from the second task.  If performance is mediated by simple response intentions, then considerable transfer would be expected.  Specifically, if red evokes the anticipation "peck", then a high rate of pecking the blue and white test stimuli should occur on transfer trials initiated by red.  Further, if green evokes the anticipation "do not peck", then a low rate of pecking the blue and white test stimuli should occur on transfer trials initiated by green.  In contrast to this prediction, red and green failed to control different rates of pecking the blue and white test stimuli.  Thus, results of transfer testing also failed to provide evidence of mediation of short-term retention by response intentions.  
 
The failure of Urcuioli and Zentall (1990, 1992) to obtain evidence of mediation by simple response intentions does not, of course, rule out the possibility that under some conditions such mediation might be observed.  Research in our laboratory has recently addressed this possibility.  

As noted previously, the identity of the test stimulus is irrelevant in a simple discrimination.  Hence, only a single test stimulus could be used and, for example, trials might involve reinforced pecking of a white test stimulus after a red initial stimulus and reinforced nonpecking of white light after green.  Presumably, Honig and Wasserman (1981) and Urcuioli and Zentall (1990, 1992) used two test stimuli in the simple task to maintain comparability with the conditional task which requires use of at least two different test stimuli.  Notice, however, that the possibility of mediation of retention by simple response intentions is predicated on pigeons being sensitive to the irrelevance of test stimulus identity.  If, however, pigeons are insensitive to the irrelevance of test stimulus identity in a simple task using two test stimuli, then mediation by simple response intentions would not be expected.  

Recent research in our laboratory has addressed the possibility that pigeons fail to detect the irrelevance of test stimulus identity when two test stimuli are employed in the simple task (Grant & Kelly, 1998; Grant, Kelly, & Steinbring, 1997).  In particular, we reasoned that the irrelevance of test stimulus identity might be more readily detected if either (a) only a single test stimulus was employed or (b) many test stimuli were employed.  Therefore, we implemented a modification of the design employed by Urcuioli and Zentall (1990).  As illustrated in Figure 16, in the two simple discrimination groups, the initial stimulus supplied all the information necessary to determine which behavior would be reinforced at the time of testing.  In the one test stimulus group, the test stimulus was always horizontal.  In the many test stimuli group, the test stimulus was equally often one of nine possible stimuli.  Finally, the delayed conditional group received training equivalent to that received by Urcuioli and Zentall's (1990) conditional discrimination group.  Grant and Kelly (1998) replicated the procedure shown in Figure 16 except that (a) horizontal replaced red and vertical replaced green as initial stimuli and (b) red replaced horizontal and green replaced vertical as test stimuli. 
  
Following acquisition, birds received a retention test involving delays of 0, 5 and 15 s.  As shown in Figure 17, accuracy was slightly higher in the simple task employing one test stimulus than in both the simple task employing nine test stimuli and the conditional task, although this difference was not statistically significant.  More importantly, neither experiment produced evidence that rate of forgetting was reduced in either simple task relative to that in the conditional task.  Hence, neither experiment produced evidence that simple response intentions can mediate short-term retention in pigeons. 

Summary  

Recent research has shown that an early report of enhanced retention in a delayed simple discrimination, relative to that in a delayed conditional discrimination, was an instance of the differential outcome effect.  When the operation of a differential outcome effect in the simple task is precluded, rate of forgetting in simple and conditional tasks is equivalent.  Hence, research has not revealed convincing evidence that short-term retention in pigeons can be mediated by simple response intentions (i.e., peck, do not peck).  

  VI. Conclusion  

The question of interest in this chapter is whether an anticipation of a future action and/or event can function as an effective mediator of short-term retention in pigeons.  Although the research reviewed above permits that question to be answered in the affirmative, mediation of retention by anticipated events may occur only in a limited set of circumstances.  Perhaps the strongest evidence for anticipatory memory-mediation comes from Zentall et al.'s (1990) study investigating memory for spatial locations.  Recall that they found that a delay interpolated after four of five locations had been chosen produced less disruption in performance than a delay interpolated after two or three locations had been chosen.  This finding provides compelling evidence that, as the trial progressed, pigeons recoded from remembering locations chosen to remembering locations not yet chosen.

Research using differential outcomes also appears to provide strong evidence for anticipatory memory-mediation.  Specifically, retention is enhanced when correct responses following each sample result in different outcomes.  This finding suggests that the samples in such differential outcome procedures evoke different outcome expectancies that persist during the delay and control choice performance.  However, one might ask about the nature of these expectancies and, in particular, whether they are truly anticipatory.  To elaborate, consider the possibility that the expectancies take the form of a visual image of food (e.g., an illuminated feeder with grain accessible) and a visual image of no food (e.g., darkness or illuminated feeder with grain inaccessible).  Likely most readers would agree that expectancies of this sort are truly anticipatory.  However, consider that the occurrence of food and no food elicit radically different emotional reactions in a hungry pigeon.  Therefore, for example, a red sample should evoke a positive emotional reaction if correct responses on red-sample trials always result in food.  In contrast, a green sample should evoke a negative emotional reaction if correct responses on green-sample trials always result in no food.  Perhaps it is these emotional reactions, unaccompanied by a representation of the outcome, which persist during a delay and guide choice behavior.  If so, is this a genuine instance of anticipatory memory?  In our view, it is not.  

It may be argued that an account of the differential outcome effect in terms of persisting emotions is flawed because the different outcomes employed have not always involved one favorable and one unfavorable outcome.  For example, enhanced retention as a function of differential outcomes has been observed when the outcomes have been different probabilities of food or different types of appetitive reinforcers (e.g., different types of food).  Even in these instances, however, the different outcomes are likely to evoke, at the least, different magnitudes of the same affective state.  In our view, it is reasonable that a bird could learn to choose between two stimuli based upon its current level of "excitement".  

Research in our laboratory using different colors as outcomes is consistent with an interpretation of the differential outcome effect in terms of persisting emotional reactions.  Because the color outcomes are followed equally often by food, they should not elicit different types or amounts of affect.  Hence, samples associated differentially with those colors should also not evoke emotional states which are different in either type or magnitude.  Hence, if facilitation of retention in a differential outcome procedure requires that the samples evoke qualitatively or quantitatively different emotional reactions, then such facilitation should not occur when those outcomes are colors with an equivalent historical relation to food.  Consistent with this supposition, we have not observed enhanced retention by differential outcomes when those outcomes have been colored lights.  It should be noted, however, that other explanations of the failure for differential color outcomes to enhance retention are possible.  For example, pigeons may be capable of anticipating outcomes only if those outcomes have strong affective properties or, alternatively, perhaps pigeons do learn to anticipate color outcomes, but those anticipations are no more memorable than retrospective or prospective codes.  

Also reviewed in this article were data from retention and transfer testing that failed to provide evidence that simple response intentions mediate short-term retention in pigeons.  Hence, pigeons appear incapable of using an anticipated action (i.e., peck or do not peck) to mediate short-term retention.  Moreover, data from these experiments may have implications for the view that standard matching tasks typically involve prospective mediation.  Recall that retention in our simple discrimination involving many test stimuli was equivalent to that in the conditional discrimination.  This finding suggests that pigeons in the former group were not coding prospectively: that is, they were not remembering the nine test stimuli to which responding should or should not be directed on each trial.  Had this been the case, retention would be expected to be markedly lower in the simple discrimination with nine test stimuli than in the conditional discrimination because the latter required only a single prospective code on each trial.  One could argue, of course, that birds in the many test stimuli group coded retrospectively and remembered which color initiated each trial.  Although this was likely the case, one might then ask why retention was equivalent in this group and in the conditional discrimination group, which, on the basis of Roitblat's (1980; Roitblat & Scopatz, 1983) research, would be expected to code prospectively.  Although one could argue that retrospective codes of color stimuli and prospective codes of line stimuli are equally memorable, such a notion strains credibility.  

In our view, the most parsimonious explanation of research comparing retention in simple and conditional discriminations is that pigeons in all groups remembered the initial stimulus during the delay and then made a decision about responding at the time of testing based on that retrospective code.  But if this explanation is correct, then how does one account for Roitblat's (1980; Roitblat & Scopatz, 1983) data implicating prospective coding in choice matching?  One possibility is that conditional discriminations involving colors, lines and forms are mediated by prospective codes when choice responding is involved, but are mediated by retrospective codes when go/no go responding is involved (but see Grant & Spetch, 1991; Spetch & Grant, 1993; Spetch, Grant, & Kelly, 1996).  

Alternatively, perhaps prospective coding was induced by presentation of three choice stimuli on each trial, which was the case in both Roitblat (1980) and Roitblat and Scopatz (1983).  It may be that the utility of prospective coding, relative to that of retrospective coding, increases as the number of choice stimuli that need to be identified and evaluated on each trial increases.  To illustrate, consider a situation involving five samples and five choice stimuli, each of the latter being presented simultaneously at the time of choice.  If the animal codes retrospectively, it may be necessary to repeatedly retrieve the response rule(s) associated with that code as each choice stimulus is successively evaluated.  The response decision process may be simplified if the animal remembers prospectively, and hence need only determine whether a particular choice stimulus matches the active code.  Thus, there is some basis for suggesting that the occurrence of prospective coding may be restricted to tasks, such as that employed by Roitblat (1980) and Roitblat and Scopatz (1983), in which color or line samples are followed by the simultaneous presentation of more than two choice stimuli.  

In summary, the literature provides only moderate evidence of a role for anticipation in mediation of short-term retention in pigeons.  The strongest evidence comes from a spatial task in which pigeons appear able to remember locations not yet chosen.  The differential outcome literature is consistent with mediation by anticipated affective outcomes, but may reflect instead persisting emotional or behavioral reactions elicited by samples correlated with outcomes that differ affectively.  Research using many-to-one sample-to-choice mapping procedures is similarly not definitive with regard to whether the codes mediating retention have anticipatory content.  Moreover, research has failed to obtain evidence that short-term retention can be mediated by either (a) anticipated outcomes which are affectively equivalent or (b) simple response intentions.  Finally, although analysis of confusion errors and interference suggests prospective coding in choice matching with lines or colors (at least in procedures involving the simultaneous presentation of three choice stimuli), the small sample size and mixed statistical evidence suggests caution in evaluating the theoretical significance of those results.  Continued research is necessary to clarify the role of anticipation as a mediator of short-term retention in pigeons.  

   VII. References  

     Brodigan, D. L., & Peterson, G. B. (1976).  Two-choice conditional discrimination performance of pigeons as a function of reward expectancy.  Animal Learning & Behavior, 4, 121-124.

     Colwill, R. M., & Rescorla, R. A. (1985).  Postconditioning devaluation of a reinforcer affects instrumental responding.  Journal of Experimental Psychology:  Animal Behavior Processes, 11, 120-132.

     Conrad, R. (1964).  Acoustic confusions in immediate memory.  British Journal of Psychology, 55, 75-83.

     Cook, R. G., Brown, M. F., & Riley, D. A. (1985).  Flexible memory processing by rats:  Use of prospective and retrospective information in the radial maze.  Journal of Experimental Psychology:  Animal Behavior Processes, 11, 453-469.

     Dickinson, A., Campos, J., Varga, Z. L., & Balleine, B. (1996).  Bidirectional instrumental conditioning.  Quarterly Journal of Experimental Psychology, 48B, 289-306.

     Edwards, C. A., Jagielo, J. A., Zentall, T. R., & Hogan, D. E. (1982).  Acquired eqivalence and distinctiveness in matching to sample by pigeons:  Mediation by reinforcer-specific expectancies.  Journal of Experimental Psychology:  Animal Behavior Processes, 8, 244-259.

     Grant, D. S. (1981).  Short-term memory in the pigeon.  In N. E. Spear & R. R. Miller (Eds.), Information processing in animals: Memory mechanisms (pp. 227-256).  Hillsdale, NJ:  Erlbaum.

    Grant, D. S. (1982a).  Prospective versus retrospective coding of samples of stimuli, responses, and reinforcers in delayed matching with pigeons. Learning and Motivation, 13, 265-280.

     Grant, D. S. (1982b).  Samples of stimuli, responses, and reinforcers:  Effect of incongruent sample type, serial position, and mode of presentation. Animal Learning & Behavior, 10, 7-14.

     Grant, D. S. (1991).  Symmetrical and asymmetrical coding of food and no-food samples in delayed matching in pigeons. Journal of Experimental Psychology:  Animal Behavior Processes, 17, 186-193.

     Grant, D. S., & Kelly, R. (1998).  [Failure to find evidence that response intentions mediate retention in delayed simple discriminations with pigeons.]  Unpublished raw data.

    Grant, D. S., Kelly, R., & Steinbring, K. M. (1997, June). The role of response intentions in delayed simple discriminations in pigeons.  Presented at the meeting of the Canadian Society for Brain, Behaviour, and Cognitive Science, University of Manitoba.

     Grant, D. S., & Spetch, M. L. (1991).  Pigeons' memory for event duration:  Differences between choice and successive matching tasks. Learning and Motivation, 22, 180-199.

     Grant, D. S., & Spetch, M. L. (1993).  Analogical and nonanalogical coding of samples differing in duration in a choice-matching task in pigeons. Journal of Experimental Psychology:  Animal Behavior Processes, 19, 15-25.

     Grant, D. S., & Spetch, M. L. (1994).  Mediated transfer testing provides evidence for common coding of duration and line samples in many-to-one matching in pigeons. Animal Learning & Behavior, 22, 84-89.

     Holland, P. C., & Rescorla, R. A. (1975). The effect of two ways of devaluing the unconditioned stimulus after first- and second-order appetitive conditioning.  Journal of Experimental Psychology: Animal Behavior Processes, 1, 355-363.

     Holland, P. C., & Straub, J. J. (1979). Differential effect of two ways of devaluing the unconditioned stimulus after Pavlovian appetitive conditioning.  Journal of Experimental Psychology: Animal Behavior Processes, 5, 65-78.

     Honig, W. K., & Thompson, R. K. R. (1982).  Retrospective and prospective processing in animal working memory.  In G. H. Bower (Ed.), The psychology of learning and motivation: Advances in research and theory (Vol. 16, pp. 239-283).  New York, NY:  Academic Press.

      Honig, W. K., & Wasserman, E. A. (1981).  Performance of pigeons on delayed simple and conditional discrimination under equivalent training procedures.  Learning and Motivation, 12, 149-170.

     Hull, C. L. (1943). Principles of behavior. New York, NY: Appleton-Century-Crofts. 

     Hull, C. L. (1952). A Behavior system. New Haven, CT: Yale University Press. 

     Kamin, L. J. (1968). "Attention-like" processes in classical conditioning. In M. R. Jones (Ed.), Miami symposium on the prediction of behavior: Aversive stimulation (pp. 9-31). Miami, FL: University of Miami Press. 

     Kamin, L. J. (1969). Predictability, surprise, attention, and conditioning. In B. A. Campbell & R. M. Church (Eds.), Punishment and aversive behavior (pp. 279-296). New York, NY: Appleton-Century-Crofts. 

     Kelly, R., & Grant, D. S. (1998a).  Coding of duration and hedonic samples in many-to-one matching-to-sample with pigeons.  Manuscript submitted for publication.

    Kelly, R., & Grant, D. S. (1998b).  [The effect of differential hedonic and nonhedonic outcome procedures in delayed matching in pigeons].  Unpublished raw data.

     Peterson, G. B., Wheeler, R. L., & Armstrong, G. D. (1978).  Expectancies as mediators in the differential-reward conditional discrimination performance of pigeons.  Animal Learning & Behavior, 6, 279-283.

     Peterson, G. B., Wheeler, R. L., & Trapold, M. A. (1980).  Enhancement of pigeons' conditional discrimination performance by expectancies of reinforcement and nonreinforcement.  Animal Learning & Behavior, 8, 22-30.

     Rescorla, R. A. (1966). Predictability and number of pairings in Pavlovian fear conditioning. Psychonomic Science, 4, 383-384. 

     Rescorla, R. A. (1968). Probability of shock in the presence and absence of CS in fear conditioning. Journal of Comparative and Physiological Psychology, 66, 1-5. 

     Riley, D. A., Cook, R. G., & Lamb, M. R. (1981).  A classification and analysis of short-term retention codes.  In G. H. Bower (Ed.), The psychology of learning and motivation: Advances in research and theory (Vol. 15, pp. 51-79).  New York, NY:  Academic Press.

     Roitblat, H. L. (1980).  Codes and coding processes in pigeon short-term memory.  Animal Learning & Behavior, 8, 341-351.

     Roitblat, H. L. (1993).  Representations and processes in working memory.  In T. R. Zentall (Ed.), Animal cognition: A tribute to Donald A. Riley (pp. 175-192).  Hillsdale, NJ: Erlbaum.

     Roitblat, H. L., & Scopatz, R. A. (1983).  Sequential effects in pigeon delayed matching-to-sample performance.  Journal of Experimental Psychology:  Animal Behavior Processes, 9, 202-221.

     Santi, A., & Roberts, W. A. (1985).  Prospective representation:  The effects of varied mapping of sample stimuli to comparison stimuli and differential trial outcomes on pigeons' working memory.  Animal Learning & Behavior, 13, 103-108.

     Spear, N. E., & Riccio, D. C. (1994).  Memory: Phenomena and principles.  Boston, MA: Allyn and Bacon.

     Spence, K. W. (1956). Behavior theory and conditioning. New Haven, CT: Yale University Press. 

     Spence, K. W. (1960). Behavior theory and learning. Englewood Cliffs, NJ: Prentice-Hall. 

     Spetch, M. L., & Grant, D. S. (1993).  Pigeons' memory for event duration in choice and successive matching-to-sample tasks.  Learning and Motivation, 24, 156-174.

     Spetch, M. L., Grant, D. S., & Kelly, R. (1996).  Procedural determinants of coding processes n pigeons' memory for duration.  Learning and Motivation, 27, 179-199.

     Thorndike, E. L. (1898). Animal intelligence: An experimental study of the associative processes in animals.  Psychological Review Monographs, 2(8).

     Tolman, E. C. (1932). Purposive behavior in animals and men. New York, NY: Century. 

     Tolman, E. C. (1948).  Cognitive maps in rats and men.  Psychological Review, 55, 189-208.

     Tolman, E. C., & Honzik, C. H. (1930a).  "Insight" in rats.  University of California Publications in Psychology, 4, 215-232.

     Tolman, E. C., & Honzik, C. H. (1930b).  Introduction and removal of reward and maze performance in rats. University of California Publications in Psychology, 4, 257-275. 

     Tolman, E. C., Ritchie, B. F., & Kalish, D. (1946a). Studies in spatial learning I. Orientation and the short-cut. Journal of Experimental Psychology, 36, 13-24. 

     Tolman, E. C., Ritchie, B. F., & Kalish, D. (1946b). Studies in spatial learning II. Place learning versus response learning.  Journal of Experimental Psychology, 36, 221-229. 

     Trapold, M. A. (1970).  Are expectancies based upon different positive reinforcing events discriminably different?  Learning and Motivation, 1, 129-140.

     Urcuioli, P. J. (1990).  Differential outcomes and many-to-one matching:  Effects of correlation with correct choice.  Animal Learning & Behavior, 18, 410-422.

     Urcuioli, P. J., DeMarse, T., & Zentall, T. R. (1994).  Some properties of many-to-one matching with hue, response, and food samples: Retention and mediated transfer.  Learning and Motivation, 25, 175-200.

     Urcuioli, P. J., & Zentall, T. R. (1986).  Retrospective coding in pigeons' delayed matching-to-sample.  Journal of Experimental Psychology:  Animal Behavior Processes, 12, 69-77.

     Urcuioli, P. J., & Zentall, T. R. (1990).  On the role of trial outcomes in delayed discriminations.  Animal Learning & Behavior, 18, 141-150.

     Urcuioli, P. J., & Zentall, T. R. (1992).  Transfer across delayed discriminations:  Evidence regarding the nature of prospective working memory.  Journal of Experimental Psychology:  Animal Behavior Processes, 18, 154-173.

     Urcuioli, P. J., Zentall, T. R., Jackson-Smith, P., & Steirn, J. N. (1989).  Evidence for common coding in many-to-one matching:  Retention, intertrial interference, and transfer.  Journal of Experimental Psychology:  Animal Behavior Processes, 15, 264-273.

     Wagner, A. R. (1969). Stimulus validity and stimulus selection in associative learning.  In N. J. Mackintosh & W. K. Honig (Eds.), Fundamental issues in associative learning (pp. 90-122). Halifax, Nova Scotia: Dalhousie University Press. 

     Wagner, A. R., Logan, F. A., Haberlandt, K., & Price, T. (1968). Stimulus selection in animal discrimination learning.  Journal of Experimental Psychology, 76, 171-180. 

     Zentall, T. R., Jagielo, J. A., Jackson-Smith, P., & Urcuioli, P. J. (1987).  Memory codes in pigeon short-term memory:  Effects of varying the number of samples and comparison stimuli.  Learning and Motivation, 18, 21-33.

     Zentall, T. R., & Sherburne, L. M. (1994).  Role of differential sample responding in the differential outcomes effect involving delayed matching by pigeons.  Journal of Experimental Psychology: Animal Behavior Processes, 20, 390-401.

     Zentall, T. R., Steirn, J. N., & Jackson-Smith, P. (1990).  Memory strategies in pigeons' performance of a radial-arm-maze analog task.  Journal of Experimental Psychology:  Animal Behavior Processes, 16, 358-371.

     Zentall, T. R., Urcuioli, P. J., Jagielo, J. A., & Jackson-Smith, P. (1989).  Interaction of sample dimension and sample-comparison mapping on pigeons' performance of delayed conditional discriminations.  Animal Learning & Behavior, 17, 172-178.

Acknowledgement   Preparation of this chapter was supported by a grant from the Natural Sciences and Engineering Research Council of Canada awarded to the first author.  Correspondence concerning this article should be sent to: Douglas S. Grant, Department of Psychology, University of Alberta, Edmonton, Alberta, Canada  T6G 2E9.  Electronic mail should be sent to:  dgrant@psych.ualberta.ca.