Generalized Identity in a Successive Matching-To-Sample Procedure in Rats: Effects of Number

Generalized Identity in a Successive Matching-to-Sample Procedure in Rats: Effects of Number

of Exemplars and a Masking Stimulus

Mark Galizio, Michael Mathews, Ashley Prichard & Katherine E. Bruce

University of North Carolina Wilmington

Author note: This research was supported in part by grant DA029252. Michael Mathews is now at West Virginia University and Ashley Prichard is now at Emory University. The authors thank

Katherine Dyer, Angela Goolsby, Madeleine Mason, Catharine Nealley and Tiffany Phasukkan who assisted in data collection and analysis. 2

Abstract

Two experiments examined the emergence of generalized identity matching in rats using a successive (go, no-go) discrimination procedure with olfactory stimuli. Trials consisted of the presentation of two odors separated by a 1 s inter-stimulus interval. Responses during the second odor presentation were reinforced only if the two odors were identical. In Experiment 1, rats began training with two odors and after the discrimination was mastered, were exposed to sessions during which unreinforced probe trials with novel odors were presented. There was evidence of higher response rates on matching probe trials in some rats, but matching did not approach baseline levels. Then two new odors were added to the baseline and after training with four exemplars, all rats showed clear transfer to novel odors. In most rats, generalized matching to novel odors was equivalent to levels obtained in the baseline. Experiment 2 tested the possibility that detection of stimulus change, rather than generalized identity might have been responsible for the transfer seen in the first study. A masking odor was inserted during the 1-s inter-stimulus interval so that stimulus change occurred on all trials and above chance transfer to novel stimuli was still observed in most animals. These findings support the hypothesis that transfer to novel odors in this successive matching-to-sample paradigm is based on a generalized identity relation. Multiple exemplar enhanced the development of relational responding, although it may not have been necessary to produce it. 3

The ability to learn same-different relations has been viewed by many as fundamental to complex human reasoning (Delius, 2004; Wasserman & Young, 2010). Perhaps for this reason, research on abstract concept learning in non-humans has focused on the same-different relation.

Techniques developed in operant laboratories have provided the main tools for such research including identity matching-to-sample (MTS) and non-matching-to sample (NMTS). In these procedures, responding to the stimulus that is identical to (in MTS) or different from (in NMTS) the sample is reinforced. However, a demonstration of accurate responding with training stimuli is not sufficient to demonstrate acquisition of same-different relations. Rather, early studies found that responding was under the control of stimulus configurations or the directly trained stimulus-stimulus relations (Carter & Werner, 1978; Cumming & Berryman, 1965; see

McIlvane, 2013 for a review). A test with novel stimulus exemplars is generally conducted in order to demonstrate abstract same-different relational responding. If behavior is under the control of the trained configurations or stimulus-stimulus pairings, then responding will occur at chance levels when novel stimuli are substituted for the training stimuli.

Training with just two stimuli is generally not sufficient to bring about same-different learning, but rather, training with many different exemplars is required (Katz & Wright, 2006).

This finding is of more than passing interest as multiple exemplar training (MET) has been proposed to play a critical role in the development of the arbitrarily-applicable relational responding that is the key concept in Relational Frame Theory (Hayes, Barnes-Holmes & Roche,

2001). Although MET is also thought to be important in the development of non-arbitrarily- applicable relational responding, there has been very little research addressing this point (see

Galizio & Bruce, in press, for a review). However, Katz, Wright and their colleagues (Katz, 4

Wright & Bachevalier, 2002; Katz & Wright, 2006; Wright, Rivera, Katz, & Bachevalier, 2003) have developed a technique called the set-size expansion procedure which permits the analysis of

MET in the development of same-different concept learning. In these studies, capuchin monkeys, rhesus monkeys, or pigeons were trained on a same-different procedure in which two different responses were available to the animal; on trials when the sample and comparison stimuli were identical, responding on the “same” lever or response key was reinforced, but on trials when they differed, responding on the “different” lever or response key resulted in reinforcement. When accurate responding developed to the first set of stimuli, the stimulus set was expanded, i.e., new stimuli (exemplars) were added to the mix.

Using these set-size expansion procedures, Katz, Wright and colleagues have found that monkeys required exposure to at least 32 different exemplars before showing above chance same-different responding to novel stimuli. Accuracy continued to improve with additional exemplars and accuracy to novel stimuli matched baseline levels of 80% correct or higher after exposure to 128 exemplars (Katz, et al., 2002; Wright, et al., 2003). Pigeons required more MET with 64 or more exemplars needed to reach above chance (70% correct) performance on novel stimuli, and 256 exemplars required before performance on novel stimuli was equivalent to baseline levels of accuracy (Katz & Wright, 2006). The set expansion procedure has also been used with simultaneous MTS and NMTS procedures in pigeons (Daniel, Wright & Katz, 2015;

Bodily, Katz & Wright, 2008). The outcomes were similar in that accuracy on novel trials improved with the number of trained exemplars, although with MTS/NMTS tasks above chance accuracy was seen with fewer exemplars relative to same-different procedures.

The set-size expansion studies with monkeys and pigeons lead to the conclusion that training with many exemplars is required to produce same-different concept learning, but there 5 have been exceptions. Oden, Thompson & Premack (1988) trained MTS with only two stimuli to young chimpanzees and yet found very high levels of transfer to novel stimuli. Oden et al. suggested that transfer with such limited experience might be unique to apes and humans, but the emergence of generalized MTS/NMTS responding after training with few exemplars has been observed in other species as well. For example, in a study from our laboratory Prichard, Panoz-

Brown, Bruce & Galizio (2015) trained rats on a successive (go, no-go) MTS task with four different odor exemplars. After high levels of accuracy were reached with the four trained odors, accuracy on probe tests with four novel odors matched the baseline performances. In a more recent study, these surprising results were replicated and extended to a successive NMTS procedure (Bruce, et al., in press). The present experiments represent an effort to follow-up these previous studies with olfactory stimuli in rats. The first question raised is whether, given that rats can learn abstract same-different concepts after training with four exemplars, might they, like chimpanzees, also show generalized matching after training with only two exemplars? In

Experiment 1 we present data on transfer of MTS to novel stimuli after training with two and four exemplars.

Experiment 1: Generalized identity matching-to-sample following training with two and four exemplars.

Methods

Subjects

Subjects were five naive male Sprague-Dawley albino rats approximately 90 days old at the beginning of training. All rats were individually housed on a reversed 12-hour light-dark cycle, and were maintained at 85 percent of their free feeding weight with ad libitum access to water. Animals were maintained and experiments conducted consistent with the NIH Guide for 6 the Care and Use of Laboratory Animals were approved by the UNCW Animal Care and Use

Committee.

Apparatus

Sessions were conducted in Med Associates operant chambers (30.5 cm long by 24 cm wide by 21 cm high) with three response ports located across the front panel; however, only the center port was active during the experiment. A stimulus light could be illuminated inside the center port which also contained a photo beam for nose-poke response detection, and openings for scents to be pumped in and vacuumed out. A pellet dispenser was located at the rear of the chamber opposite the response ports. Chambers were housed in sound attenuating cubicles and interfaced to a computer equipped with MED-PC software. Three five-channel Med Associates flow olfactometer systems (ENV-275-5) were added to each chamber permitting delivery of 15 different odors. An input pump (Linear AC0102, 2.84 pound per square inch with an airflow of .

177 cubic feet per min) delivered air through glass jars containing odorant solution that forced scented air through Teflon tubing and a manifold into the center nose port of the chamber.

Computer-controlled solenoids gated the delivery of odorants. A vacuum pump (Linear VP0125,

-9.84 Hg vacuum and air displacement of .247 cubic feet/min) operated throughout the session removing air from a tube located at the bottom of the center port [see Prichard, et al. (2015) for an illustration].

Odorants

Liquid odorants purchased from The Great American Spice Company, Nature’s Garden, and local stores were used. Odorants were diluted to a solution of 6.7 ml oil per 100 ml distilled water. Sixteen odors were grouped into sets of four stimuli: Set A (cinnamon, apricot, bubblegum, root beer), Set B (brandy, vanilla butternut, almond, licorice), Set C (clove, honey, 7 blueberry, geraniol), and Set D (apple, grass, coconut, sandalwood).

Procedure

Shaping phase. After initial magazine training, rats were trained to make nose-pokes to the center port. The beginning of the session was signaled by the onset of the center port light and house light. A nose-poke turned off both lights, and a 45 mg sugar pellet (Bioserve) was delivered accompanied by illumination of a hopper light located at the rear of the chamber. After a 5-s period, the hopper light terminated; the house and center-port lights came on and the nose- pokes continued to be reinforced on a FR1 schedule. When rats were responding consistently, the reinforcement schedule was thinned to FI-5s over several sessions.

Two Exemplar Training. MTS training involved presentation of a pair of odors presented through the center port. Four rats (N25, N35, O2, O4) were trained with cinnamon and apricot as odorants, and the fifth (O1) was trained with brandy and vanilla butternut (see Table

1). Each trial began with the onset of the house light and center port light. An initial observing nose-poke response was required, after which a sample odor was presented. The first nose poke after 5 s terminated delivery of the sample odor and produced a 1s termination of the house and center port lights, which was followed by the onset of the comparison odor and both lights. On positive trials the comparison odor was identical to the sample and responding was reinforced on an FI-5s schedule. The first response after 5 s resulted in termination of the comparison odor, the house light, and the center port light and a 5 s onset of the hopper light along with delivery of a sugar pellet. On negative trials, the comparison was different from the sample and was presented for 5 s and then terminated, along with the house and center port lights. A 30 s inter-trial interval

(ITI) separated the termination of the comparison stimuli from the onset of the next trial. Nose- pokes occurring during the 5 s when the S+ or S- was presented were used to calculate 8 discrimination ratios (S+ responses/S+ responses + S-responses) and response rates.

Sessions were conducted five days/week and contained 48 trials with four different trial types, two positive and two negative. For example, for Set A scents, positive trial types were cinnamonsample-cinnamoncomparison and apricotsample-apricotcomparison, and negative trial types were cinnamonsample-apricotcomparsion and apricotsample-cinnamoncomparison. Trial types were distributed randomly with the constraint that no more than 4 consecutive positive (reinforced) or negative

(non-reinforced) trial types were permitted and each odorant occurred equally often as a sample and comparison. Two-exemplar MTS training continued until a mastery criterion was met such that an average discrimination ratio (DR: responses to S+ divided by responses to both S+ and

S-) of .8 with a minimum DR of .75 on each set of trial types was met on two consecutive sessions. One rat (O1) failed to meet this criterion and after 80 sessions of training, the criterion was relaxed to a DR of .79 overall with minimum of .75 on all trials which was met on Session

Novel Stimulus Probe Sessions. When criterion was met, a probe was conducted on the next scheduled session. Probe sessions mixed 28 baseline MTS trials using the two originally trained odors (16 reinforced matching and 12 unreinforced non-matching trials), with 8 unreinforced probe trials using novel odors drawn from a different stimulus set (see Table 1 for odors used for each rat). Each of the four novel odors was presented once in a matching trial and once in a non-matching trial. The purpose of the imbalanced matching and non-matching baseline trials programmed on probe sessions was to keep the overall reinforcement density similar to that used in regular baseline sessions. After each probe session, rats were returned to the original baseline schedule until the mastery criteria were met again for two consecutive sessions and a probe session was conducted on the next scheduled session. This phase of the 9 experiment continued until 8 probe sessions had been completed.

Four Exemplar Training and Probes. For four of the rats (O1 was not tested in this phase) MTS training continued with the set size expanded from two to four odorants by adding trials with bubblegum and rootbeer as odorants along with the previously trained apricot and cinnamon. Training sessions were 48 with each odorant serving as a sample 12 times: 6 times on reinforced matching trials (e.g., bubblegumsample—bubblegumcomparison) and 6 times on non- reinforced non-matching trials (e.g., bubblegumsample—rootbeercomparison). Mastery criteria were as above with a minimum of ten sessions of four-exemplar training required before probe testing.

Probe sessions were conducted as above, but the probe stimuli were drawn from a scent set different from the one used in the first round of probe sessions (see Table 1).

Results and Discussion

Number of sessions required to meet acquisition criterion for the two-exemplar training ranged from 13 to 82 sessions with a mean of 34.8 sessions (three additional animals began two- exemplar training but were dropped from the study after 50 or more sessions with performance remaining at chance levels). The main results are shown in Figure 1 which presents mean response rates from the 8 probe sessions after two-exemplar training. White circles show performances on baseline trials and, as would be expected, response rates were high on matching trials and much lower on non-matching trials for each rat. Black circles show comparable performances with novel stimuli on probe trials and response rates were relatively low on both matching and non-matching trials. Still, for some rats (N25, O1, O2) response rates were consistently somewhat higher on matching than on non-matching trials. Statistical analyses were conducted by comparing response rates on matching vs. non-matching trials for each rat as the repeated measure using a matched pair t-test with the 8 probe sessions. The tendencies for higher 10 probe response rates for matching trials relative to non-matching trials was statistically significant for Rat N25 (p = .017), and approached significance for O1 (p = .051) and O2

(p=.058). Thus, for these three rats there was some evidence of generalized matching-to-sample with novel stimuli even after training with only two stimuli. Significant differences between matching and non-matching conditions were evident for each rat on baseline trials (p < .01 for all five rats).

Figure 2 shows comparable response rate data after four-exemplar training. On baseline trials with the four trained stimuli (white circles), high rates of responding were evident on matching trials with much lower rates on non-matching trials. Responding was much better differentiated on novel probe trials as well. Indeed, for two rats (N35 and O4) matching on probe trials was virtually indistinguishable from baseline. The other two rats also showed higher response rates on matching than on non-matching trials, but not to the degree observed on the trained baseline trials. All four rats showed significantly higher response rates on matching than non-matching trials for both baseline and probes (p < .01).

Figure 3 allows comparison of matching after training with 2 vs. 4 exemplars by showing mean DR for each rat on baseline and probe conditions. The horizontal line marks a DR of 0.5 which indicates chance performance (equivalent response rates on matching and non-matching trials). Note that baseline DRs were nearly 0.8 or higher for all rats with both two- and four- exemplar training. After two-exemplar training, probe DRs were near or slightly above chance levels (with the exception of rat N25), but after four-exemplar training all four of the rats tested showed DRs that were well above 0.5. For three of rats (N35, O2 and O4), training with two exemplars did not transfer to novel stimuli, but the additional training with four-exemplars led to generalized matching to novel stimuli at levels that were comparable to those achieved on the 11 trained baseline trials with DRs approaching 0.8. Rat N25 was unusual in two ways. First, he was the only rat to show significant transfer to novel stimuli after exposure to only two exemplars.

Second, four-exemplar training produced a much smaller effect for Rat N25. In both two- and four-exemplar probe conditions N25 showed significant transfer, but in neither case did it approach baseline levels.

One important finding in the set expansion studies reviewed above was that there appeared to be an intermediate pattern of responding between absence of transfer to novel stimuli and levels of transfer that were equal to baseline levels. Katz and colleagues refer to this final level as “full concept learning” and the intermediate pattern as “partial concept learning” (Daniel et al., 2016; Katz & Wright, 2006). Viewed from this perspective, Rat 25, and to a lesser extent rats O1 and O2, showed partial concept learning after training with only two exemplars. Rats

N35 and O4 showed little evidence of transfer after two exemplar training, but both these two and O2 approached full transfer after four exemplars. Overall, these results replicated the high levels of transfer obtained by Prichard et al. (2015) and Bruce et al. (in press) after training with four exemplars. Transfer of any kind after so few training exemplars is noteworthy, because it is in marked contrast to the much larger number required to produce even partial transfer in pigeons and monkeys (e.g., Katz & Wright, 2006).

Although generalized MTS has been observed after only two exemplars in chimpanzees

(Oden et al., 1988), interestingly, a few studies have observed the emergence of generalized

MTS responding after training with few exemplars in pigeons. For example, Cook, Kelly & Katz

(2003) found transfer of same-different responding to novel stimuli in pigeons after training with only two exemplars, and above chance generalized identity matching was also observed by

Urcuioli (2011) in pigeons after training with only two stimuli. Both of these studies used 12 successive (go, no-go) discrimination procedures, as did the present study and Bruce et al. (in press) and Prichard et al. (2015) with rats. Might some features of this procedure accelerate the development of relational responding? Alternatively, is it possible that non-relational cues rather than the identity relation may have come to control behavior in these studies? One possible cue that might be particularly salient in the present study is control by stimulus change. In the present study (as well as Bruce et al. and Prichard et al.), the sample odor was presented for 5 s followed by a 1-s inter-stimulus interval (ISI). On matching trials the ISI is followed by a 5-s presentation of the same odor, but given that there may be a lingering trace of the sample odor, this may functionally serve as an 11-s (or longer) presentation of a single odor which is terminated by a reinforced response. In contrast, the odor duration is shorter on non-matching trials as a different stimulus is presented after the ISI and responding never results in reinforcement. Perhaps rats learn to respond after long odor presentations and not after short ones. If so, it might be that this form of stimulus change/temporal control is what transfers to novel odors, rather than a generalized identity relation.

One way to test this possibility is to create a stimulus change on both positive and negative trials by inserting a masking odor during the ISI. Such a procedure was used by Lu,

Slotnick & Silberberg (1993) who observed accurate odor matching in rats when a masking odor was presented for up to 10 s between sample and comparison stimuli. The purpose of Experiment

2 was to determine whether rats would show transfer of matching to novel stimuli under these conditions. If rats learn a generalized identity relation then a masking odor during the ITI should have little effect. On the other hand, if discrimination is based on stimulus change, then the masking odor should prevent the transfer of matching because a stimulus change will occur on every trial. 13

Experiment 2: Effects of a masking odor between sample and comparison stimuli on transfer of matching.

Subjects and Apparatus

Four rats from Experiment 1 (N25, N35, O2, O4) and one additional rat (O10) served as subjects. Apparatus and odorants were as in Experiment 1.

Procedures

General procedures were the same as in Experiment 1 except for the addition of a masking odor (coffee for rats N25, N35, O2 and O4; marshmallow for O10) during the 1-s ISI on each MTS trial. Rat O10 received initial MTS training to criterion with four exemplars from the

Set C odorants before the start of Experiment 2 (see Table 1) while the rats from Experiment 1 continued with MTS with the Set A stimuli. All rats received a minimum of 10 sessions under the Mask MTS conditions. Novel probe sessions with the masking stimulus could then begin once the rat had met the DR criterion of .8 or higher overall with no less than .75 on any trial type for two consecutive sessions. Eight probe sessions were conducted for each rat as in

Experiment 1 with the requirement that criterion on two consecutive baseline sessions had to be met before the next probe session was scheduled. Novel probe stimuli were used for all rats: from Set A for O10, Set C for N25 and N35, and Set D for O2 and O4. The masking odor used on novel probe trials was the same as in baseline trials (marshmallow for O10 and coffee for the other four).

Results and Discussion

Figure 4 shows mean DR for the last two baseline sessions without the masking odor for each rat (left bars), the mean DR for the first two of the baseline masking sessions (middle bars), and the last two masking session before probe sessions (right bars). The introduction of the 14 masking odor produced some disruption of accuracy in all six rats. This disruption was quite dramatic in some rats (note N35 and O10 in particular), but recovery was rapid as most rats met criterion to move to probes in the minimum of 10 sessions, and three required just a few sessions longer (see Table 1). Generally, discrimination accuracy reached levels with the masking stimulus that were comparable to the initial baseline as the rightmost bars show; however, DRs in the masking condition were somewhat lower for some rats (e.g., N35, O4).

The main focus of Experiment 2 was whether rats would show transfer to novel stimuli with the masking odor creating a stimulus change on every trial. Figure 5 shows response rates for baseline and probe trials with matching trials on the left and non-matching trials on the right.

All rats show accurate discrimination on baseline trials, with higher response rates on matching trials (t-tests at p<.01 in all cases). Overall response rates were lower on probe trials, but all five rats showed somewhat higher rates on matching trials relative to non-matching. These differences were statistically significant for N25, N35, O4 and O10 (t-tests at p<.01) and approached significance for O2 (p = .058). Note that although response rates were lower on matching probe trials relative to baseline, they also lower on the non-matching probe trials for most rats.

Analysis of DRs provides another way of looking at these results and as Figure 6 shows, both baseline and probe DRs are well above chance for all rats. Indeed, for rat N35, DR on probe trials was somewhat higher than baseline DR. This type of outcome might be interpreted as full concept learning (cf. Daniel et al., 2016) in that matching with novel stimuli was as strong as with the directly trained odors. From that perspective, the other four rats showed partial concept learning as probe DRs, while above chance, were somewhat lower than baseline DRs. However, the overall lower response rates on probe trials suggest that the masking stimuli disrupted probe 15 responding to some degree. Perhaps the combination of the masking odor and the novel stimuli created a new context with some attendant generalization decrement. Still, the differentiation between matching and non-matching trials is consistent with the hypothesis that transfer in these successive odor matching studies is based on generalized identity matching, rather than stimulus change per se.

General Discussion

One purpose of the present study was to determine whether training with two exemplars would be sufficient to produce generalized identity matching. Response rates on matching novel probe trials were much lower than on baseline trials in all rats after training with just two exemplars. Still, three of the five rats showed higher rates of responding on matching relative to non-matching probe trials that reached or approached statistical significance. After training was expanded to four exemplars, all four of the animals tested showed above chance matching and three out four had response rates and DRs on probe trials that were nearly equivalent to baseline.

In sum, Experiment 1 showed evidence for what Katz, Wright and colleagues (Daniel et al.,

2016; Katz & Wright, 2006) have called partial concept learning after training with two exemplars, and for full concept learning after four-exemplar training.

In Experiment 2, a masking odor was inserted between the sample and comparison stimuli and above chance transfer to novel stimuli was still observed in four of the five animals, although response rates on probe trials were reduced. This finding supports a generalized identity interpretation of transfer on probe trials because stimulus change occurred after 5 s on both matching and non-matching trials. If transfer of matching to novel stimuli were based solely on discrimination of odor duration or stimulus change, then the masking stimulus would have prevented transfer. 16

As noted above, generalized same-different responding is not usually observed until training with many exemplars is provided, so it is worth considering factors that may be responsible for the present findings. Use of olfactory stimuli in the present study is clearly one important feature. Rats generally perform poorly in MTS tasks using visual stimuli (Iverson, refs) as control by stimulus location can interfere with relational learning. This seems to be less of a problem when olfactory stimuli are used and same-different concept learning has been demonstrated (April, Bruce & Galizio, 2011; Pena, Pitts & Galizio, 2006). These studies were not designed to assess the number of exemplars necessary to produce concept learning, but April et al. showed generalized identity and oddity matching after training with only 10 exemplars.

Both April et al. and Pena et al. used a simultaneous MTS procedure which indicates that transfer after few exemplars is not limited to the successive MTS procedure use in the present study.

However, it would be interesting to determine transfer after two and four exemplar training in a simultaneous MTS procedure as it may be that training with fewer exemplars is needed to obtain transfer using successive procedures in pigeons (e.g., Cook et al., 2003; Urcuioli, 2011).

Training with four exemplars in Experiment 1 resulted in increased matching in all rats and yielded response rates and DRs on novel probes that approached those of baseline trials in three of the four rats tested. This would be described as full concept learning (cf., Katz &

Wright, 2006) and is what would be expected if responding has come under the control of the identity relation. On the other hand, the finding of above chance matching that was still well below baseline levels (i.e., partial concept learning) observed in some animals in Experiment 1 and most in Experiment 2 seems more problematic. From a behavior analytic perspective, the patterns of behavior termed partial concept learning might better be understood in terms of a mixture of different types of stimulus control—termed stimulus control topographies by Dube 17 and McIlvane (1996). The stimulus control topography hypothesis posits that different features of the stimulus environment compete to control behavior with the likelihood of any particular feature exerting control depending on the reinforcement history associated with that specific feature (see also McIlvane & Dube, 2003).

In the present example, after training with two exemplars, responding to novel stimuli might be controlled primarily by generalization from specific properties of the training stimuli with the sample-comparison relation being less likely to control responding. Exposure to additional exemplars would result in a reduction of control by specific features of the first two training stimuli as reinforcement for these is reduced, and an increased likelihood of control by the identity relation which is consistently associated with reinforcement. As multiple exemplar training continues, relational responding should become applicable to both novel and familiar stimuli and accuracy would eventually become equivalent across baseline and novel stimuli— full concept learning. Such an analysis might help to account for the reduction in transfer observed in some rats in Experiment 2 (O2 and O4). Perhaps during baseline training, stimulus compounds formed by mixing the masking odor with the training stimuli became part of the stimulus control topography. If so, then on probe trials with novel stimuli, the formation of different compounds may have resulted in some generalization decrement or otherwise disrupted responding. This hypothesis could be tested by providing additional multiple exemplar training with the masking conditions as control by compounds involving different training odors and the masking odor should drop out and relation responding become stronger. Indeed, such research and other variations on the set size expansion paradigm would seem an ideal strategy for future assessment of how multiple exemplar training produces relational responding. 18

References

April, L.B., Bruce, K., & Galizio, M. (2011). Matching- and nonmatching-to-sample concept

learning in rats using olfactory stimuli. Journal of the Experimental Analysis of Behavior,

96(2), 139-54.

Bodily, K. D., Katz, J. S., & Wright, A. A. (2008). Matching-to-sample abstract-concept learning

by pigeons. Journal of Experimental Psychology: Animal Behavior Processes, 34(1),

178-184.

Carter, D. E., & Werner, T. J. (1978). Complex learning and information processing by pigeons:

A critical analysis. Journal of the Experimental Analysis of Behavior, 29, 565-601.

Cook, R. G., Kelly, D. M., & Katz, J. S. (2003). Successive two-item same-different

discrimination and concept learning by pigeons. Behavioural Processes, 62, 125-144.

Cumming, W., & Berryman, R. (1965). The complex discriminated operant: Studies of

matching-to-sample and related problems. In D. I. Mostovsky (Ed.) Stimulus

generalization (pp.284-330), Stanford, CA: Stanford University Press.

Daniel, T. A., Goodman, A. M., Thompkins, A. M., Forloines, M. R., Lazarowski, L. & Katz, J.

S. (2016). Generalization cannot predict abstract-concept learning. In M. C. Olmstead

(Ed.). Animal cognition: Principles, evolution and development (pp. 131-145). New

York: Nova Science Publishers.

Delius, J. D. (1994). Comparative cognition of identity. In P. Bertelson, P. Eelen, & G.’Ydewalle

(Eds.), International perspectives on psychological science, 1, 25-40. Hillsdale, NJ:

Erlbaum.

Dube, W. V. & McIlvane, W. J. (1996). Implications of stimulus control topography analysis for

emergent behavior and stimulus classes. In T. R. Zentall & P. M. Smeets (Eds.), 19

Advances in psychology: Stimulus class formation in humans and animals, vol. 117 (pp.

197-220). Amsterdam: Elsevier.

Hayes, S. C., Barnes-Holmes, D., & Roche, B. (2001). Relational frame theory: A post-

Skinnerian account of human language and cognition. New York: Plenum Press.

Iversen, I. H. (1993). Acquisition of matching-to-sample performance in rats using visual stimuli

on nose keys. Journal of the Experimental Analysis of Behavior, 59, 471-482.

Iversen, I.H. (1997). Matching-to-sample performance in rats: A case of mistaken identity?

Journal of the Experimental Analysis of Behavior, 68, 27-45.

Katz, J. S., & Wright, A. A. (2006). Same/different abstract-concept learning by pigeons.

Journal of Experimental Psychology: Animal Behavior Processes, 32, 80-86.

Katz, J. S., Wright, A. A., & Bachevalier, J. (2002). Mechanisms of same/different abstract-

concept learning by rhesus monkeys (Macaca mulatta). Journal of Experimental

Psychology: Animal Behavior Processes, 28, 358-368.

Katz, J.S., Wright, A.A., & Bodily, K.D. (2007). Issues in the comparative cognition of abstract-

concept learning. Comparative Cognition & Behavior Reviews, 2, 79-92.

Lu, X. C. M., Slotnick, B. M., & Silberberg, A. M. (1993). Odor matching and odor memory in

the rat. Physiology & Behavior, 53, 795-804.

McIlvane, W. J. (2013). Simple and complex discrimination learning. In G. J. Madden, W.V.

Dube, T. D. Hackenberg, G. P. Hanley, & K. A. Lattal (Eds.) APA handbook of behavior

analysis, Vol. 2: Translating principles into practice (pp. 129-163). Washington, DC:

American Psychological Association.

McIlvane, W. J., & Dube, W. V. (2003). Stimulus control topography coherence theory: 20

Foundations and extensions. The Behavior Analyst, 26, 195-213.

Mackintosh, N. J. (2000). Abstraction and discrimination. In C. Heyes & L. Huber (Eds.), The

evolution of cognition (pp. 123-141). Cambridge, MA: MIT Press.

Oden, D. L., Thompson, R. K., & Premack, D. (1988). Spontaneous transfer of matching by

infant chimpanzees (Pan troglodytes). Journal of Experimental Psychology: Animal

Behavior Processes, 14, 140-145.

Peña, T., Pitts, R.C., & Galizio, M. (2006). Identity matching-to-sample with olfactory stimuli in

rats. Journal of the Experimental Analysis of Behavior, 85(2), 203-221.

Penn, D. C., Holyoak, K. J., & Povinelli, D. J. (2008). Darwin's mistake: Explaining the

discontinuity between human and nonhuman minds. Behavioral and Brain Sciences, 31,

109-130.

Prichard, A., Panoz‐Brown, D., Bruce, K., & Galizio, M. (2015). Emergent identity but not

symmetry following successive olfactory discrimination training in rats. Journal of the

Experimental Analysis of Behavior, 104(2), 133-145.

Urcuioli, P. J. (2011). Emergent identity matching after successive matching training, I:

Reflexivity or generalized identity? Journal of the Experimental Analysis of Behavior,

96, 329-341.

Wasserman, E. A., & Young, M. E. (2010). Same–different discrimination: The keel and

backbone of thought and reasoning. Journal of Experimental Psychology: Animal 21

Behavior Processes, 36, 3.

Wright, A. A., Magnotti, J. F., Katz, J. S., Leonard, K., & Kelly, D. M. (2016). Concept learning

set-size functions for Clark’s nutcrackers. Journal of the Experimental Analysis of

Behavior, 105(1), 76-84.

Wright, A. A., Rivera, J. J., Katz, J. S., & Bachevalier, J. (2003). Abstract-concept learning and

list-memory processing by capuchin and rhesus monkeys. Journal of Experimental

Psychology: Animal Behavior Processes, 29(3), 184-198. 22

Table 1. Stimulus sets/number of sessions to complete the phases of Experiments 1 and 2.

Rat 2-Ex train Probe 4-Ex train Probe Mask train Mask Probe

N25 A/38 D/35 A/10 B/43 A/10 C/19

N35 A/23 D/16 A/10 B/20 A/14 C/41

O1 B/82 A/37 ------

O2 A/13 C/27 A/10 B/14 A/10 D/20

O4 A/18 C/16 A/10 B/14 A/10 D/18

O10 -- -- C/33 -- C/14 A/14 23

Figure Captions

Figure 1. Mean response rates from identity probe sessions following training with two exemplars in Experiment 1. Response rates on baseline trials are represented by open circles and rates on probe trials by closed circles.

Figure 2. Mean response rates from identity probe sessions following training with four exemplars in Experiment 1. Response rates on baseline trials are represented by open circles and rates on probe trials by closed circles.

Figure 3. Mean discrimination ratio (DR) for each rat on identity probe sessions of Experiment 1.

DRs on baseline trials are represented by black bars and rates on probe trials by white bars. The left panel shows DRs after two-exemplar training and the right panel after four-exemplar training.

Figure 4. Mean DR for the last two baseline sessions without the masking odor for each rat in

Experiment 2 (left bars-white). Middle bars (striped) show the mean DR for the first two sessions after the masking sessions were introduced, and the right bars(cross-hatched) show DRs on the final two baseline masking sessions before probes began.

Figure 5. Mean response rates from identity probe sessions with masking odors in Experiment 2.

Response rates on baseline trials are represented by open circles and rates on probe trials by closed circles.

Figure 6. Mean discrimination ratio (DR) for each rat on identity probe sessions with masking odors in Experiment 2. DRs on baseline trials are represented by black bars and rates on probe trials by white bars.

Figure 1. 24

Figure 2

Figure 3

Figure 4 25

Figure 5

Figure 6.