Anim Cogn DOI 10.1007/s10071-009-0251-y
ORIGINAL PAPER
Serial list combination by monkeys (Macaca mulatta): test cues and linking F. Robert Treichler · Mary Ann Raghanti
Received: 18 February 2009 / Revised: 2 June 2009 / Accepted: 4 June 2009 © Springer-Verlag 2009
Abstract This investigation assessed prospective bases of non-human primate cognitive operations that support serial list memory. Four macaques learned 3-, 5-item ordered lists of objects (as two-choice problems) and then either did or did not (in a within-subject design) receive training on pairs that linked the three original lists into a 15-item serial order. Next, subjects experienced selective exposure trials on object pairs that either maintained or contrasted to the serial position relationships seen during original learning. Subsequent comprehensive tests assessed the interactive eVects of linking and exposure conditions on choosing in accord with a combined 15-item serial order. Linking readily induced monkeys to merge lists into a 15-item order, but restricting early exposure to pairs with the same positional relationships as original training slowed, but did not prevent, list combination. Exposure to positional relationships congruent with the combined (15-item) list and diVerent from those of original 5-item training aided both expression of the linking eVect and acquisition after no link training. Thus, list linking facilitated serial reorganization by inducing release from error derived from memory for prior learned positional relationships. The task was recommended as a prospective evaluator of continuity of cognitive processes among species.
F. R. Treichler (&) Department of Psychology, Kent State University, 331 Kent Hall, Kent, OH 44242, USA e-mail:
[email protected] M. A. Raghanti Department of Anthropology and School of Biomedical Sciences, Kent State University, Kent, OH 44242, USA
Keywords Macaques
Serial memory · List linking · Transitivity ·
Comparative study of serial list memory as an indicant of information processing and integration was enhanced by McGonigle and Chalmers’ (1977) pioneering study of transitive inference in squirrel monkeys. Subsequently, analogous tests have been conducted with a wide variety of species ranging from pigeons (von Fersen et al. 1991) to chimpanzees (Gillan 1981). Although the interpretation of results from these tests has generated controversy (see Penn et al. 2008, and its responses), comparative investigation of serial memory in various non-human primates continues to reveal unique processing characteristics (Merritt et al. 2007). One especially salient reason for interest in nonhuman primate performance derives from Terrace’s (1993) Wnding that monkeys and pigeons employ distinctively diVerent memory processes when treating serially ordered information (see also Terrace and McGonigle 1994). In these demonstrations, monkeys remembered the positions of objects in a serial list, while pigeons seemed to choose on the basis of conditioned reinforcing value derived from serial proximity to items at or near the always-rewarded end of a list. Similarly, items associated with the neverrewarded end were avoided yielding an “end-anchor” eVect. As a consequence of these species-speciWc performance characteristics, there has been increased interest in the processes that monkeys use when asked to remember relationships among arbitrary stimuli arranged as an ordered series. Several prospective mechanisms have been suggested. One that has been nearly universally implicated is retention of information about the relative positions of the list items (e.g., Terrace et al. 2003). An especially convincing demonstration of this property was reported by
123
Anim Cogn
Chen et al. (1997) when their monkeys, after learning several diVerent independent lists, were asked to make choices on test pairs comprising items from diVerent lists. In such pairs, subjects reliably chose the item that had appeared at an ordinal position associated with reward during initial learning. For example, if two ascending ordered lists were trained (A < B < C < D < E and F < G < H < I < J), items from a later position were selected when test pairs contained stimuli from the two diVerent lists. Thus, monkeys seemed to assign a value or rank to each item based on its position in the series and subsequently use that property to guide selections on between-list pairings. Another integrative operation that has thus far been demonstrated only in humans (Woocher et al. 1978) and nonhuman primates (Treichler et al. 2003) is the linking of lists. In the Wrst demonstration of linking by macaques (Treichler and Van Tilburg 1996), two independently trained 5-item serial lists were joined by isolated training on only a single pair made up of the always-rewarded (high end) item from one list and the never-rewarded (low end) item from another, but now the latter (formerly low) item was the correct one. Within the Wrst block of test trials that provided all possible object pairings, monkeys treated the items as a single 10-item list arranged according to the serial order deWned by the link. A follow-up investigation (Treichler et al. 2003) expanded the number of lists to three and found further conWrmation of the linking eVect. Perhaps more importantly, the latter study also revealed that linking represented a capability diVerent from, but acting in concert with, memory for list positions. In the absence of linking information, it was possible to provide training suYcient to combine three 5-item lists; however, several hundred trials were necessary to attain the level of accuracy seen immediately after link training. The course of eventual adoption of the 15-item series without linking indicated that subjects gradually overcame a speciWc error factor, their proclivity to choose on the basis of the originally learned list positions. The contrast of outcomes suggested that monkeys could use revised information about only a few list-end items to support rapid reorganization of an extensive ordinal series. In this investigation, we attempted to explore the manner in which test trials that provided selective reward experience might inXuence choices when serial list combination was tested either in the presence or absence of link training. In the tactical approach of this within-subject investigation, macaques Wrst learned three 5-item serial lists. After acquiring a set of 5-item lists, the subjects either did or did not receive training on pairs that linked between initial lists. Subsequently, tests to determine whether the three lists were incorporated into an inclusive 15-item serial array were administered. During these tests, novel object pairings were initially restricted to ones that either conformed or did
123
not conform to the ordinal relationship between objects that had been experienced during original training. After administration of these selective exposure conditions, a further 6-day test presented pairs that conformed to the single 15-item series and included all possible contrasts of original list positions. Several of the original training pairs also appeared in these tests. Thus, this investigation assessed the possible biasing eVect of restricting speciWc order-related characteristics during early test experience. That inXuence was measured when monkeys were subsequently required to merge three, previously learned, ordered lists into a single inclusive list. The purpose of the various measures was to obtain information about learning mechanisms and integrative operations that support the list linking process and to evaluate the prospective utility of linking as an indicant of cognitive processing. Prior research on list linking has suggested that serial memory in monkeys may, in some instances, be interpreted as a process involving internal organization (Treichler 2007), with the corollary that serial memory of at least some animals is not exclusively determined by simple conditioning (but see Wynne 1998 for a diVerent view). Interpreting monkey serial memory as an organized process accords well with Marcel Kinsbourne’s (2005) contention that cognitive processes vary continuously among phyla rather than being unique to each. Kinsbourne buttressed his premise of continuity by claiming that some animals may demonstrate a distinctively cognitive behavior, episodic memory. His criterion for that performance was demonstration that retained information could be automatically updated so that, in his words, “superseded external reality is replaced by a present but internalized reality.” Interpreting his quote in terms of our prior list linking outcomes, we believe that isolated link training allows the development of new, internally organized relationships among objects that comprised earlierlearned 5-item lists. Accordingly, in this research, we explored some properties of monkey list linking procedure that might support or impair such memory reorganization. Adding empirical support to Kinsbourne’s contention, Sherwood et al. (2008) provide further biologically based justiWcation for cognitive continuity in their recent report of anatomical and physiological mechanisms that contribute to cognitive function in both human and non-human primates After reviewing a substantial variety of physiological evidence, their concluding statement was “we consider it inescapable to recognize the continuity in mentality between the LCA (last common [primate] ancestor) and us, despite the signiWcant disparity in phenotypes.” One purpose for the development of cognitive indices such as list linking is to aid in the search for behaviors that allow further understanding of the issue of cognitive continuity.
Anim Cogn
Methods and materials Subjects Four mature (one 15 years old and three 21 years old), laboratory-reared female rhesus monkeys (Macaca mulatta) served as subjects. All had been maintained for over ten years in Kent State’s AALAC-approved primate facility that conformed to all US Department of Agriculture guidelines. Rewards were always incentives (fruit, nut, or cereal segments), and no nutritional deprivations were ever imposed. All animals had extensive experience with concurrent conditional discrimination tests and had served as subjects in both the Treichler and Van Tilburg (1999) study and the Treichler et al. (2003) study.
gation, the four subjects were each trained on three diVerent 15-item lists with list orders varied across subjects and the several lists allowing various combinations of linkage and initial test experience conditions In order to meet the requirements of another investigation (Treichler et al. 2007), only three 15-item tasks were presented, and consequently, complete counterbalancing of conditions over the four subjects was not possible. The design of the experiment was a 2 £ 2 factorial with linkage versus non-linkage as one factor and the imposition of diVerent kinds of test pairs during the Wrst two test blocks (see test procedure section below) as the other. All four animals were tested under the linked condition combined with both the initial test experience conditions. Under the non-linked (control) condition, only two subjects were exposed to the diVerent initial test conditions.
Apparatus All tests were carried out in a WGTA (Wisconsin General Test Apparatus; see Meyer et al. 1965) with a retractable tray that allowed lateral placement of two three-dimensional wood, metal, or plastic objects attached to thin black plastic bases. When an opaque screen concealing the tray was raised, subjects were allowed to distally displace one object (in tracks) to reveal an underlying food well. Upon displacement, the tray was retracted. The tracks contained switches that allowed automatic recording of display durations intended as measures of choice latency. However, these did not provide suYciently reliable data for analysis because of substantial within and between subject variability. In order to provide three diVerent 15-item lists, a total of 45 objects were chosen from a pool of over 1500. Choice of speciWc objects was designed to assure that distinctive items comprised the eventual 15-item lists that resulted when three 5-item lists were merged into a longer series. Procedure In an eVort to minimize experimenter bias, test administrators were not informed of the implications of correct and incorrect choices (until a debrieWng session), and trial characteristics were speciWed by a daily protocol sheet. Testing was conducted on 5 or 6 days per week with some exceptions for holidays. Design Figure 1 depicts the sequence of training and testing that was followed to impose the variety of treatments included in this investigation. Roman numerals on the left margin indicate the progression through the various phases of experimental manipulation. Over the course of the investi-
Preliminary list training (indicated by Roman numeral I in Fig. 1) Each list was introduced as a 5-item ordered series derived from the four, two-choice premise pairs, e.g., A < B, B < C, C < D, and D < E. Two further 5-item lists comprising diVerent objects were designated by the letters F–J and K– O, and orders of training on the diVerent lists were varied among animals. For each 5-item list, daily 40-trial sessions always provided 10 trials on each premise pair with the sequence of appearance of the pairs intermixed and counterbalanced in terms of correct right and left tray locations. For each list, trial-rerun correctional training (errors yielded trial repetition until correct choice) was conducted for 15 daily sessions. On the 16th day, a new set of Wve objects was introduced and trained as a list for 15 days in the same manner as the Wrst list, followed by 15 days of training on a third 5-item list. In the next session after completion of training on this third list, retention of all the three 5-item lists was tested. On the Wrst day, either the Wrst- or second-learned list was presented using non-correctional (no trial repetitions) procedures, and, subsequently, each of the other lists was presented on the ensuing two days of a 3-consecutive-day block. Four such 3-day blocks were conducted with the order of appearance of the three lists (A–E, F–J, or K–O) varied systematically within each block. Retention on each set of three 5-item lists was tested in similar manner at the completion of its training in an eVort to allow all constituent lists to be well retained before attempting their merger in later tests. Link training (indicated by Roman numeral II in Fig. 1) Since each set of three 5-item lists appeared to be well retained at the completion of the 12-day tests, it was
123
Anim Cogn Fig. 1 Sequence of training procedures and treatments imposed to yield test measures
I
TRAIN AND RETAIN THREE 5-ITEM LISTS
II
LINK
III
POSITIVE GAP EXPOSURE
IV
NO LINK
NEGATIVE GAP EXPOSURE
ALL-INCLUSIVE 15-ITEM TEST
Test Blocks 1 & 2
Test Blocks 3 & 4
Treatment combinations yield the four major test conditions: LINK POSITIVE – displayed in Figure 3A LINK NEGATIVE – “ “ Figure 3B NO LINK POSITIVE – “ “ Figure 3C NO LINK NEGATIVE – “ “ Figure 3D
deemed appropriate to administer subsequent treatments involving either presence or absence of link training. When the predetermined sequence of treatments called for link training, it consisted of exposure to two novel pairings that bridged between the Wrst and second (E–F+) and the second and third (J–K+) of the previously learned 5-item ordered lists (speculatively, generating a 15-item series). These two pairs were presented in 32-trial daily sessions where each appeared 16 times in a systematically intermixed sequence balanced for tray placement of the correct alternative. Link training was conducted under the trial-rerun correctional procedure and was continued until both pairs yielded 14 of 16 correct initial choices within the same session. In the session following attainment of that criterion (or after no linkage when that condition was imposed), tests trials on 15-item lists were initiated. Early test exposure conditions (Roman numeral III in Fig. 1) Tests of whether or not the three 5-item lists were combined into one 15-item serial list were provided in daily sessions of 48 non-correctional trials. When testing was begun, 36 of the trials were made up of inter-list pairings
123
representing one of the two early exposure conditions, and the remaining 12 trials comprised original training pairs. Under one exposure condition, all inter-list pairings were ones where the objects bore the same ordinal position relationship that had been in eVect during original learning, i.e., the correct object (for the 15-item list) always came from a higher original ordinal position in a 5-item list than the object with which it was paired. Pairs representing this condition were termed “positive serial gaps” (Fig. 2 shows the basis of the disparities termed “serial gaps”). The 36 between-list trials contained nine trials from each possible disparity (1, 2, 3, or 4) between positions that comprised this category (e.g., pair B < N of Fig. 2). Under the other early test condition, objects that conformed to the 15-item series (the “correct” ones) were always ones that had appeared at a lower ordinal position in original learning and such pairings were termed “negative serial gaps” (e.g., pair D < K of Fig. 2). As in the positive serial gap condition, there were nine pairs representing each of the four possible disparities. Under these exposure conditions, only those between-list pairs appropriate to a speciWc treatment were administered in six consecutive daily sessions (as the Wrst 2, 3-day blocks).
Anim Cogn
Tests of all possible positional combinations (Roman numeral IV in Fig. 1) After completion of the two blocks that provided an early exposure condition, daily pairings provided a test that presented all possible combinations of objects from the diVerent original list positions. Once again, daily 48 non-correctional trial sessions with reward contingencies arranged to conform to a 15-item ordinal series (A–O) were presented. However, now, each 3-day (144-trial) test block was made up of 12 appearances of the original training pairs, 108 inter-list pairs, and 24 novel pairings of objects that had originally been learned within the same list (e.g., B vs. E or G vs. I). These test pairings were designed to allow comparison of choice proportions as a function of disparity between original list locations and the requirements of the new merged 15-item list. Test pairs assessed the consequence of presenting an object from positions 1–5 in original training with an object from another list at a diVerent or the same position in the later test (as depicted in Fig. 2). For example, if an object from the Wfth position of the initially trained list (e.g., E) was paired with the Wrst item of the list designated to comprise items 11–15 (e.g., K), a correct choice would require selecting an object from an originally learned position #1 over one that had initially appeared at position #5. That magnitude of position contrast would be designated as a serial gap distance of ¡4. If all possible location contrasts for 5-item lists are considered, there are nine prospective levels of serial gap distance varying from ¡4 to 0 (where both objects are from same list positions), to +4 (where a position 5 item from a later list is paired with a position 1 from an earlier list). In this investigation, the
Serial List Position 1st
2nd
3rd
4th
5th
A
B
C
D
E
F
G
H
I
J
K
L
M
N
O
Fig. 2 Depiction of a 15-item serial list showing list positions during original 5-item acquisition and the basis for generating pairs representing the various serial gaps of subsequent 15-item merged list tests where a later letter is always correct. Test pair B–N provides an example of a serial gap of +2 where the correct alternative is at a location two positions ahead of its original training location, the pairing D–K represents a serial gap of ¡3 (three positions behind its training location), and E–J represents pairings of objects at the same original list positions
108 inter-list test pairings that appeared in each 3-day block represented one of the three categories: (a) negative serial gaps, i.e., where the correct object was from a lower original list position (combining gaps ¡4 to ¡1 and presenting 48 such trials/block), (b) positive serial gaps, where the incorrect object was from a lower position (combining gaps +1 to +4, with 42 such trials/block), and (c) same, where the two paired objects had each appeared at the same serial position during learning of original lists (18 such trials/ block). Each of the serial gap distances (1–4) was represented in each daily session with slightly more pairings devoted to the shorter distances. Examples of pairings that provide the various levels of the “serial gap distance” conditions may be seen in Fig. 2, and a more extensive array with designation of all possible appropriate pairings among objects in our 15-item-list tests appears in Fig. 1 of Treichler et al. 2003.
Results Initial training of 5-item lists and multiple list retention (Roman numeral I, Fig. 1) The purpose of providing results from this phase of training was to assure that both acquisition and retention of information required for our later assessment of linking characteristics was accomplished. Since the four premise pairs that comprised each 5-item serial list appeared to be readily acquired, performance on the last 3 days of each 15-day training session was chosen as an appropriate measure of terminal acquisition. DiVerences among acquisition scores on the diVerent pairs within 5-item lists were tested in a repeated-measures 4 £ 3 factorial ANOVA. This analysis compared, as one factor, object pair location within each 5-item list expressed as the constituent pairs, A–B+ (1st), B–C+ (2nd), C–D+ (3rd), and D–E+ (4th). These designations served merely as indicants of serial location within the list; therefore, there were corresponding F–G+ , G–H+ , K–L+ , etc., pairs that similarly represented each of the four possible pair locations. The other factor in this analysis was the order of presentation (1st, 2nd, or 3rd) of the sets of lists. Acquisition did not diVer as a consequence of list presentation order (F2,6 = 3.14, P = 0.117), but terminal performances at the various pair locations did yield signiWcant diVerences (F3,9 = 18.67, P < 0.001, partial 2 = 0.862) with no signiWcant interaction of these two factors (F6,18 = 1.37, P = 0.297). Retention was evaluated in a factorial ANOVA that compared performances as a function of within-list pair location (the signiWcant factor in acquisition) and the succession of 4, 3-day blocks that served as tests for any three concurrently trained lists. A recording error resulted in loss
123
Anim Cogn Table 1 Mean, standard deviation and conWdence intervals for scores based on proportion of choices conforming to a 15-item combined serial list on four successive, 3-day test blocks Test condition
Block
Link positive
Link negative
No link positive
No link negative
Negative gaps
Positive gaps
Training pairs
1
87.9 § 5.9 (84.2 ¡ 91.6)
62.5 § 11.5 (55.2 ¡ 69.8)
2
92.2 § 7.9 (87.2 ¡ 97.2)
72.9 § 15.0 (63.4 ¡ 82.4)
3
67.3 § 17.3 (56.4 ¡ 78.3)
92.4 § 8.2 (87.2 ¡ 97.6)
75.0 § 15.1 (65.4 ¡ 84.6)
88.9 § 14.7 (79.6 ¡ 98.3)
4
74.7 § 14.1 (65.7 ¡ 83.6)
95.3 § 8.1 (90.2 ¡ 100.5)
56.3 § 21.7 (42.5 ¡ 70.0)
88.8 § 10.9 (81.8 ¡ 95.7)
1
74.4 § 16.6 (63.9 ¡ 85.0)
51.5 § 9.8 (45.3 ¡ 57.7)
2
81.4 § 9.9 (75.2 ¡ 87.7)
51.4 § 13.6 (42.8 ¡ 60.0)
3
86.3 § 11.4 (79.1 ¡ 93.5)
84.2 § 10.6 (77.4 ¡ 90.9)
50.0 § 18.5 (38.3 ¡ 61.7)
84.7 § 20.7 (71.5 ¡ 97.8)
4
84.0 § 15.0 (74.5 ¡ 93.5)
88.6 § 8.5 (83.2 ¡ 94.0)
52.1 § 22.5 (37.8 ¡ 66.4)
90.2 § 11.2 (83.1 ¡ 97.3)
1
95.2 § 4.4 (90.6 ¡ 99.7)
87.5 § 11.5 (75.4 ¡ 99.6)
2
98.2 § 3.3 (94.8 ¡ 101.6)
82.0 § 8.3 (73.2 ¡ 90.8)
3
47.0 § 11.8 (34.7 ¡ 59.3)
93.0 § 7.7 (85.0 ¡ 101.0)
75.0 § 22.4 (51.5 ¡ 98.5)
75.0 § 17.4 (56.8 ¡ 93.2)
4
64. 7 § 14.0 (50.0 ¡ 79.4)
97.7 § 3.6 (93.9 ¡ 101.5)
66.7 § 25.8 (39.6 ¡ 93.8)
75.0 § 17.4 (56.8 ¡ 93.2)
1
37.0 § 7.0 (29.7 ¡ 44.3)
2
51.7 § 19.7 (31.0 ¡ 72.3)
3
77.2 § 18.8 (57.5 ¡ 96.9)
94.2 § 9.3 (84.4 ¡ 104.0)
54.2 § 10.2 (43.5 ¡ 64.9)
86.5 § 12.4 (73.5 ¡ 99.5)
4
80.3 § 13.2 (66.4 ¡ 94.2)
93.0 § 7.7 (85.0 ¡ 101.0)
62.5 § 21.0 (40.5 ¡ 84.5)
89.0 § 17.0 (71.1 ¡ 106.9)
64.0 § 11.4 (52.0 ¡ 76.0) 65.3 § 9.8 (55.0 ¡ 75.6)
of retention measures for one of the sets of lists, so that this analysis was limited to two of the three sets of 5-item lists that would eventually comprise 15-item lists. However, the two available sets were analyzed as one factor in a 2 (list presentation order, earlier vs. later) £ 4 (within-list pair location, A–B, B–C, etc.) £ 4 (successive test blocks) ANOVA. Neither successive blocks (F3,9 = 0.25, P = 0.857) nor presentation order (F1,3 = 5.69, P = 0.097) yielded signiWcant main eVects, although within-list pair location did (F3,9 = 25.94, P < 0.001, partial 2 = 0.896). There were no signiWcant interactions among these factors (all Ps > 0.10), so that performances in acquisition and retention appeared quite similar. This was conWrmed in a further 2 £ 4 factorial analysis that compared acquisition and retention for the four diVerent within-list locations. That analysis yielded a main eVect of within-list pair location (F3,9 = 29.39, P < 0.001, partial 2 = 0.907), but no diVerence between acquisition and retention scores (F1,3 = 5.45, P = 0.102) and no interaction of these factors (F3,9 = 2.49, P = 0.126). The signiWcant list location eVect was further evaluated in a Tukey's HSD post-hoc test that revealed signiWcantly better performances on A–B list locations than on B–C or C–D locations (both Ps < 0.01), while D–E pairs exceeded only the B–C pair location (P = 0.02). Thus, serial list learning and memory in the present instance showed a pattern of diVerences among successive pairs that has been suYciently characteristic of 5-item ordered lists to be labeled a “serial position eVect” by von Fersen et al. (1991). When terminal acquisition and retention scores were graphically plotted, near identical overlapping U-shaped functions were revealed. In summary,
123
Same
although merely a prerequisite to further testing, our initial acquisition and retention measures (acquisition 80.4% and retention 77.3% correct) indicated that performance characteristics (especially, error distributions) were nearly identical to those of previous investigations, and accordingly, provided information appropriate for further use in this study. Link training (Roman numeral II in Fig. 1) Concurrent training to criterion when two pairs served as links between 5-item lists proceeded rapidly. Eight instances of link acquisition (four animals on two linking opportunities) were provided, and subjects met the acquisition criterion in as few as two and no more than four 32-trial sessions. 15-item list performance (Roman numeral III in Fig. 1) Some measures from the 15-item tests were combined to facilitate graphic presentation and statistical analysis via scores representing proportional choices that conformed to the 15-item combined serial list. Table 1 shows means, standard deviations, and 95% conWdence intervals at each 3-day block of testing under each of the four conditions derived from the factorial combination of linking or no linking and early (Wrst two blocks) exposure to either conforming (positive gap distance) or non-conforming (negative gap distance) pairs. Mean scores on each of four kinds of object pairing (if provided during that phase of testing) at each successive 3-day test block are shown in Fig. 3a–d
Anim Cogn
under each of the four above noted treatment conditions. As a consequence of diVerences in the number of animals tested under the 15-item list conditions (four on linked vs. two on non-linked), it was not possible to provide an orthogonal design appropriate for application of a factorial ANOVA as a statistical test. However, with concern and recognition of possibilities for Type I error, we chose to conduct Student’s t tests of diVerences from the speciWc alternative of chance performance (applying a Bonferroni adjustment that modiWed = 0.05 to = 0.0031) for each measure at each block of testing. Support for this decision was based on the generally robust eVects of the treatment variables, their pattern of correspondence to previous tests of linking and distance eVects and the large numbers of trials that contributed to each score. Note that on Blocks 1 and 2 (selective early exposure), each subject received 108 treatment trials and 36 training pair trials per block. In Blocks 3 and 4, for each subject at each block, there were 48 trials representing negative gap distances, 42 with positive gap distances, 18 with objects from the same original location, and 36 training pairs. Results of the Student’s t tests can be found in Table 2, with measures signiWcantly diVerent from chance indicated in bold type. When linking was provided, the pattern of outcomes for the several measures displayed in Fig. 3a and b indicated that such training resulted in reorganization of the lists into one 15-item list. Under the condition where only pairs with negative gap distances (Fig. 3b, Block 1) appeared in the initial sessions, choices almost immediately indicated that a revised, inclusive 15-item list was guiding selections. That interpretation seemed warranted because 75.5% of the Wrst 108 such choices were for objects diVerent from the ones predicted on the basis of memory for list location. Similarly, reorganization was indicated by scores on original training pairs that did not diVer from chance at any point during test, i.e., unlike expectations from a simple conditioning view, original premise pairs provided the most diYcult problems in retention, a property treated more fully in our discussion section. Further indication of list reorganization as a consequence of link training may be seen in Fig. 3a where only original-order-conforming pairs (positive gaps) appeared in the Wrst two blocks. By Block 2, training pairs were at greater than chance level (perhaps aided by selective reward of many order-conforming pairs). However, the introduction of tests with all types of pairings indicated that some memory for the linking experience might still be operative. That prospect was suggested because, when negative gap pairs appeared in Block 3 tests, these were chosen at slightly, but not signiWcantly, above chance levels. However, with further exposure to the 15-item contingencies provided by Block 4, selections improved to above chance levels. When test pairs were comprised of objects that had
been at the same original list locations (e.g., both had been at position #2 in 5-item lists), Blocks 3 and 4 both yielded selections that were signiWcantly above chance, a further indication of treating objects as a merged 15-item list. While there is no ready explanation, pairs involving objects from equivalent serial positions appear to be especially sensitive indicators of list combination, and that characteristic was encountered across the various conditions of this study. When tested in the absence of link training, subjects based choices on retained (and unmodiWed) information about list positions. Accordingly, Fig. 3d shows that monkeys initially given selective exposure to negative gap distances were below chance (on the conformity to longlist measure) on these pairings. That would be the expected outcome if selections were guided by retained information about original list positions; however, that characteristic changed with further experience. By the second block of exposure to only negative gaps, choices were near chance. With the subsequent introduction of allpairs testing (Block 3), choices began to improve and by Block 4 were signiWcantly above chance, reXecting organization of these objects into a 15-item list. Of special note is that Block 3 pairings of objects from the same original list positions were above chance and improved even more with additional training. This provides a further example of the sensitivity of these pairings, although perhaps in this case, scores were aided by transfer eVects from training on the newly imposed 15-item serial order. The Wnding of chance levels of choice on original training (premise) pairs on Blocks 3 and 4 also lends support to the contention that the list was organized as a 15-item series. Under conditions of no linking and pre-exposure to only positive gap distances (Fig. 3a), introduction of all pairing types at Block 3 revealed chance performance on pairs with negative gap distances, but there was some improvement with continued exposure to the testing/training regimen. However, under these conditions, selections on pairs comprising items from “same” original positions did not diVer from chance, while choices on training pairs declined from the levels seen during pre-exposure. Perhaps, this was an indication of the beginning of list reorganization. Since it was necessary to attain proWcient acquisition of all lists for a subsequent study (Treichler et al. 2007), two additional training blocks were administered after the present tests, and this training did yield eventual overall performance of over 80% correct on all lists. The general consequence of restriction of preliminary test pairs to only those with positive serial gaps may be seen in Fig. 3a and c. Seemingly, both in the presence and absence of link training, experiencing only positive gaps retarded transition to an inclusive 15-item list. Even with linking, the introduction of test trials with negative gaps
123
Anim Cogn
123
Anim Cogn 䉳 Fig. 3 Mean proportion choice scores (P) conforming to a 15-item (combined) list for each of four test-pair categories during four successive 3-day blocks of testing. Graphs a–d display results for the four diVerent treatment groups. Scores with asterisks were signiWcantly above chance (Bonferroni adjusted P < .0038). a With link training and selective initial exposure to positive gap distances. b With link training and selective initial exposure to negative gap distances. c No link training and selective initial exposure to positive gap distances. d No link training and selective initial exposure to negative gap distances
yielded only about 62% conformity to the combined list, and without linking there was even less conformity (48%).
Discussion The present research attempted to provide information about mechanisms of serial list memory related to the phenomenon termed “list linking.” During the course of the investigation, several previously observed list memory characteristics were conWrmed. Among these were the ease of acquisition of concurrent conditional discriminations by sophisticated macaques, the pattern of errors associated with premise pair order (the “serial position” eVect) and memory of an item’s serial location as a “default” mechanism for guiding choices when tests presented pairs of objects from diVerent previously learned lists. The tests administered after linking revealed that limited (but speciWc and isolated) new information about inter-list relationships could induce monkeys to make choices
indicating that they readily merged the several previously learned lists into a comprehensive serially ordered list. As indicated in earlier reports, when conditions were appropriate, these choices reXected near spontaneous reorganization of the original list elements. Especially sensitive indicators of the transition from three short lists to a single, inclusive list were provided by two measured characteristics in the present tests, and those properties were evident in both the presence and absence of link training. Predictably, one property associated with transition to the combined list was suppression of choice based on the relative positions that objects occupied during original learning. In our graphs (Fig. 3a–d), this measure is indicated by the rapid gain in proWciency (or error reduction) on pairs that represented negative gap distances. In addition, a second, less obvious indicator of list combination was the appearance of chance (or near chance) levels of choice on the original training pairs when they appeared during test. Whether initially, as after linking, or eventually with continued training, such an outcome revealed that subjects treated these pairings as especially diYcult discriminations. We have contended that this characteristic is present because monkeys generate an internalized memory array that yields progressively better performance on pairs separated by more intervening items. In contrast, the minimal separation aVorded by immediately adjacent pairs (i.e., the training pairs) renders them diYcult to discriminate, and they remain so even with extended testing. This property has been shown in both our own (Treichler and Van Tilburg 1999, see especially Fig. 5) and other (Rapp et al. 1996) list memory studies. This outcome
Table 2 Results of t tests comparing scores from Table 1 to chance performance Test condition Link positive
Link negative
No link positive
No link negative
Block
Negative gaps
Positive gaps
Training pairs
1
t(11) 22.34
t(11) 3.76, P = 0.004
2
t(11) 18.53
t(11) 5.31
Same
3
t(11) 3.48, P = 0.01
t(11) 18.03
t(11) 5.75
t(11) 9.16
4
t(11) 6.04
t(11) 19.43
t(11) 1.00, P = 0.34
t(11) 12.35
1
t(11) 5.09
2
t(11) 11.05
3
t(11) 11.09
t(11) 11.14
t(11) 0.00, P = 1.00
t(11) 5.80
4
t(11) 7.86
t(11) 15.76
t(11) 0.32, P = 0.75
t(11) 12.45
1
t(5) 25.40
t(5) 7.96
2
t(5) 36.30
t(5) 9.40
t(11) 0.53, P = 0.61 t(11) 0.36, P = 0.73
3
t(5) ¡ 0.63, P = 0.56
t(5) 13.74
t(5) 2.74, p = 0.04
t(5) 3.53, P = 0.02
4
t(5) 2.56, P = 0.05
t(5) 32.30
t(5) 1.58, P = 0.18
t(5) 3.53, P = 0.02
1
t(5) ¡ 4.58, P = 0.01
t(5) 3.00, P = 0.03
2
t(5) 0.21, P = 0.84
t(5) 3.83, P = 0.01
3
t(5) 3.54, P = 0.02
t(5) 11.63
t(5) 1.00, P = 0.36
t(5) 7.24
4
t(5) 5.62
t(5) 13.74
t(5) 1.46, P = 0.20
t(5) 5.61
Tests were conducted on each pairing category measure under four test conditions at successive test blocks. Results shown in bold were signiWcantly diVerent from chance at < 0.05 with a Bonferroni adjustment to < 0.0031
123
Anim Cogn
also accords well with D’Amato and Colombo’s (1990) postulate of “symbolic distance” between paired items as an inXuence on list memory. Perhaps most meaningfully, it is reemphasized that optimal diYculty in discriminating highly trained pairs provides a direct contrast to the contention that serial list retention is attributable to simple conditioning processes. The consequences of restricting preliminary test pairs to ones with either positive or negative gap distances also seemed consonant with an organizational view of list memory in macaques. If link training had been provided prior to experience with only positive gap distances, adoption of choices consonant with the 15-item combined list was impaired. The extent of that impairment may be compared to outcomes with the present negative gap pre-exposure or to our earlier tests with immediate testing on all possible pairings (about 74% conforming in Treichler et al. 2003). In the absence of link training and with initial experience on only positive gap distances, transition to the 15-item list was quite slow and required training (i.e., continued exposure to 15-item test pairs) beyond that of the present test phases. However, selective preliminary exposure to negative gap distances operated as reasonably eVective instruction and provided some support for transition to the new inclusive list. In overview, this series of tests demonstrated that combining three lists into one required monkeys to be informed that the pairs at test entailed choices diVerent from those derived from old memories of an object’s original list position. List combination could be accomplished without linking, but to do so required much training, especially if immediate postlinking trials continued to provide only list position information consonant with initial training. However, if these early trials countermanded original positional information, acquiring the new list was enhanced, although some further training was required. In contrast, providing isolated link training was suYcient to inform subjects of the change in rules, although once again, selective exposure to pairs that continued to conWrm to the positional properties of the initial premise pairs slowed transition to the inclusive list. Seemingly, the limited information provided in linking pairs allowed reorganization of serial information that resulted in emergence of an inclusive, new serial order. The essential function provided by linking was to deliver the message that new reward relationships were operative, even for objects that had never before appeared together. Further, that information was projected into a reorganized representation that was internal, near immediate, and predictive of choices on any object pairing. While those characteristics have been demonstrated in our previous reports, present outcomes provide new information indicating that common learning phenomena may inXuence tests of linking, and the inXuence of link training appears to
123
operate by reducing speciWc error factors related to list position memory. Although some early demonstrations of serial list linking (Treichler and Van Tilburg 1996) could be questioned because of possible measurement artifacts, the use of three linked lists (rather than 2) in Treichler et al. 2003 and in the present tests allows improved conWrming evidence for the reliability and utility of the procedure. However, we caution that the present performances are speciWc to sophisticated subjects capable of tolerating reversal and conditionality of reinforcement regarding the objects that make up serial lists. That macaques require an extensive training history to overcome the proactive interference inherent in such tasks has been demonstrated by Treichler (2005). A recent report by Vigo and Allen (2009) emphasized that facility in making conditional choices is a requirement for transitive performance, and they have characterized list linking as a “higher-order modal similarity judgment.” Due to the sophistication requirement, we encourage long-term use of the same animals as a research tactic that may yield unique information as well as conserve animal resources. In addition, the requirement of preliminary sophistication when testing cognitive behaviors provides a condition analogous to that frequently encountered in human research eVorts. Thus, we believe linking represents processing that is diVerent from and more complex than retention of information about location in a serial list. Although memory for an object’s location in a series provides an appropriate indicant of ordinal memory, an even more advanced level is indicated by the internal and rapid reorganization attendant upon list linking. With the prospect that diVerent hierarchical levels of integrative operation may exist in the serial memory of monkeys, further generalities among the cognitive operations of diVerent animals are suggested. In accord with that precept, we believe that list linking and other demonstrations of transitive ordering lend support to Kinsbourne’s (2005) and Sherwood et al.’s (2008) views advocating cognitive continuity among species.
References Chen S, Swartz KB, Terrace HS (1997) Knowledge of the ordinal position of list items in rhesus monkeys. Psychol Sci 8:80–86 D’Amato MR, Colombo M (1990) The symbolic distance eVect in monkeys (cebus apella). Anim Learn Behav 18:133–140 Gillan DJ (1981) Reasoning in the chimpanzee: II. Transitive inference. J Exp Psychol Anim Behav Process 7:150–164 Kinsbourne M (2005) A continuum of self-consciousness that emerges in phylogeny and ontogeny. In: Terrace HS, Metcalfe J (eds) The missing link in cognition. Oxford University Press, New York, pp 142–173 McGonigle BO, Chalmers M (1977, June 23) Are monkeys logical? Nature 267:694–696
Anim Cogn Merritt D, MacLean EL, JaVe S, Brannon EM (2007) A comparative analysis of serial-ordering in ring-tailed lemurs (Lemur catta). J Comp Psychol 121:363–371 Meyer DR, Treichler FR, Meyer PM (1965) Discrete-trial training techniques and stimulus variables. In: Schrier AM, Harlow HF, Stollnitz F (eds) Behavior of nonhuman primates. Academic Press, New York, pp 1–49 Penn DC, Holyoak KJ, Povinelli DJ (2008) Darwin’s mistake: explaining the discontinuity between human and nonhuman minds. Behav Brain Sci 31:109–178 Rapp PR, Kansky MT, Eichenbaum H (1996) Learning and memory for hierarchical relationships in the monkey: eVects of aging. Behav Neurosci 110:887–897 Sherwood CC, Subiaul F, Zawidzki TW (2008) A natural history of the human mind: tracing evolutionary changes in brain and cognition. J Anat 212:426–454 Terrace HS (1993) The phylogeny and ontogeny of serial memory: list learning by pigeons and monkeys. Psychol Sci 4:162–169 Terrace HS, McGonigle BM (1994) Memory and representation of serial order by children, monkeys and pigeons. Current Dir Psychol Sci 3:180–189 Terrace HS, Son LK, Brannon EM (2003) Serial expertise of rhesus macaques. Psychol Sci 14:66–73 Treichler FR (2005) Successive reversal of concurrent discriminations by macaques (Macaca mulatta). Anim Cogn 8:75–83 Treichler FR (2007) Monkeys making a list: checking it twice? In: Washburn DA (ed) Primate perspectives on behavior and
cognition. American Psychological Association, Washington, DC, pp 143–160 Treichler FR, Van Tilburg D (1996) Concurrent conditional discrimination tests of transitive inference by macaque monkeys: list linking. J Exp Psychol Anim Behav Process 22:105–117 Treichler FR, Van Tilburg D (1999) Training requirements and retention characteristics of serial list organization by macaque monkeys. Anim Cogn 2:235–244 Treichler FR, Raghanti MA, Van Tilburg D (2003) Linking of serially ordered lists by macaque monkeys: list position inXuences. J Exp Psychol Anim Behav Process 29:211–221 Treichler FR, Raghanti MA, Van Tilburg D (2007) Serial list linking by macaque monkeys: list property limitations. J Comp Psychol 121:250–259 Vigo R, Allen C (2009) How to reason without words: inference as categorization. Cognit Process 10:77–88 von Fersen L, Wynne CDL, Delius JD, Staddon JER (1991) Transitive inference formation in pigeons. J Exp Psychol Anim Behav Process 17:334–341 Woocher FD, Glass AL, Holyoak KJ (1978) Positional discriminability in linear orderings. Mem Cognit 6:165–173 Wynne CDL (1998) A minimal model of transitive inference. In: Wynne CDL, Staddon JER (eds) Models for action. Earlbaum, Hillsdale, pp 269–307
123
