stability had no effect when training was conducted using an array of two ..... Post-training transfer tests were conducted to measure the control ... free-feeding weight. ...... How honey bees use landmarks to guide their return to a food source ..... claves contextuales presentadas por sı¬ solas tuvieron una infl uencia mı¬ nima.
T HE QUA RTE RLY JO U R NA L OF EX P ER IM E N TA L PSYCH OL OG Y, 1996, 49B (4), 307± 345
Landm ark S tability: Fu rther Studies Pointin g to a Role in Sp atial L earning R. Biegler and R.G.M . M orris Centre for N euroscience a nd Depa rtment of Pha rma cology, U niversity of E dinburgh, E dinburgh, S cotla nd
Two exp erim en ts w ere co n d ucted to inve stigate the p os sible ro le of land m ark stab ility in sp atial lear n ing. R ats w ere train ed to search in a large aren a for foo d h idden at a con sisten t distan ce and d irection fro m either a sin gle ra dially sy m m etric land m ark or an ar ray of tw o land m ark s. We varied th e relative de gree to w hich th e landm ark ar ray an d /or th e cu es of th e tr aining co ntext p red icted th e locatio n of food, w ithou t vary ing th e con d ition al p ro bability of food being available given either cu e. E xp erim en t 1 use d vestibu lar d isorientation to en su re con tro l of s earch location by exp erim en ter-c o ntro lled cu es. T h e resu lts sh ow ed th at m aking either a sin gle land m ark or a cluster of tw o ad jacent land m ark s th e sole sp atial p red ictor of rew ard location red u ced th e accu racy of search com p ared to a con d ition w h ere both th e land m ark ar ra y an d con te xt cu es w ere reliable spa tial p red ictor s. Vary ing global land m ark stab ility h ad n o effect w h en tr aining w as con d u cted u sin g an ar ray of tw o land m ark s located so m e d istance from each oth er. C on text cu es, w hen tested alo ne, trig gered very little search ing in ap p rop r iate locatio ns, and th e abso lute m agn itud e of con trol ove r search w as insu f® cien t to accou n t for th e su p eriority of stab le landm arks. T h e better lear n ing w ith a stable land m ark , an d th e d ep end en ce of this effect on th e geom etrical ar ra ng em en t of land m ark s, po ints to th e co n ditio n s of spa tial lear n ing involving add ition al p rin ciples to tho se of sim p le asso ciative con d ition ing. E xp erim ent 2 exam ined land m ark stab ility usin g a s ingle land m ark an d ® xed d ire ction al cues in th e absen ce of vestibu lar d isorien tation . T h is also revealed a relative adv an tage of land m ark stab ility, bu t an im als w ith a land m ark th at m oved fro m trial to trial d id sh ow som e evidence of lear n ing. C on text cues w h en tested alone h ad m inim al in¯ u en ce. P aram etric m an ipu latio n of land m ark stab ility offers a no ve l w ay of in¯ u en cing sp atial lear n ing an d th u s u n d erstan d ing b etter th e p ro cess th ro u gh w h ich ego centric rep resen tation s of p erceived sp ace are tra nsfo r m ed into allocen tric rep resentation s of th e real w orld .
Requ ests for rep rints shou ld be sent to R. Biegler, Cen tre for Ne uroscie nce an d D epartm ent of Ph ar m acology, U niversit y of E din burgh , E dinburgh EH 8 9L E , U.K . E m ail: bie gle r@ ed.ac.u k; r.g.m .m orris@ ed.ac.u k This work w as sup ported by g rants from the Wellcom e Tru st an d the M edical Rese arc h Coun cil. We are also gratefu l to the M cD on nell-Pew C entre for Cogniti ve N euroscien ce (U niversity of Oxford ), to R oge r S poon er an d Patrick S poon er for software an d technic al assistan ce, an d to M ark Good and K atalin Gothard for help ful disc ussio ns. q
1996 The E xp erim ental Psych olog y S ociety
308
B IE G L E R A N D M O R R IS
T he purpose of this p ap er is to describe two experiments co ncer ned w ith identifying the psychological processes of allocentric spatial lear nin g. T he results p oint to the idea that la ndma rk sta bility is an im portant factor in spatial lear ning. Speci® cally, they reveal th at w hether or not a landm ark w ill be used for the purpose of representing the location of an other object (includ ing hidden objects) is in¯ u enced by w hether it is perceived as geom etrically stable w ith re spect to at least one other landm ark an d/or certain geom etric features of the environm ent. T his ph enom en on is relevan t to the ap plication of associative lear ning principles to the spatial dom ain.
Th e P sycho lo g ical P ro cesses o f S patia l Lea rn in g Spatial lear ning has been studied in laboratory an imals for m uch of the century. E arly work used com plex m azes to investigate w hether an imals w ou ld learn the spatial layout of a m az e or w ould lear n by acquiring response habits (Car r & Watso n, 1908; Restle, 1957; Tolm an , 1948). M ore recent research has used a variety of spatial tasks to explore such issues as the cap acity of spatial w ork ing-m em o ry (O lton & S am uelso n, 1976; Roberts, 1979 ), the representatio n of spatial infor m ation (Collett, C artwright, & Sm ith, 1986), an d the role of the h ippo cam pal for m ation an d related structures (G affan & H arriso n, 1989; M or ris, G arrud, Raw lins, & O ’ K eefe, 1982; O ’ K eefe & N ad el, 1978 ). H owever, despite ex tensive study, the processes eng aged by an imals to represent the location of landm arks an d to navig ate effectively throug h the environm ent rem ain a m atter of dispute. W ith respect to the content of w hat is lear ned, som e argue that spatial lear ning involves the for m ation of a cognitive or spatial m ap of space (O ’ K eefe & N ad el, 1978) an d /or fragm ents of m ap s (Worden, 1992), w hereas others have proposed that an imals lear n vectors sp ecifying explicitly the distance an d direction between landm arks (C heng, 1988; C ollett et al., 1986). W ith respect to the co nditions of spatial lear ning, som e ho ld that it can be understood using established principles of associative lear ning (D ie z-Ch am izo, Sterio, & M ackintosh, 1985; Sutherland & Ru dy, 1989), w hereas others suggest that it requires distinct com putational principles (G allistel, 1990). A fund am ental issu e is w hether spatial lear ning differs q ualitatively from lear nin g cau sal relationship s by virtue of certain unique features of spac e that d o not ap ply to the tem poral dom ain in w hich virtually all studies of associativ e conditioning have so far been conducted. For exam ple, in a o ne-dimensional dom ain (such as tim e), th ere can be no such thing as novel shortcuts or detours. Success in taking these m ay depend u pon distinct w ays of representin g infor m ation or upon oper ation s such as v ector add ition . S im ilarly, an imals can them selves deter m ine the rate an d direction of th eir passage through space, but not throug h tim e. E xp loration of this issue is com plicated by th e fact that th ere is a con tinuum of possible accounts from th e m ost parsimon ious associative m odel to the m o st com plex theory of co gn itive m ap ping. T h ere app ear to be at least two different classes of m odel of n avig atio n through associative lear ning. T h e ® rst is the id ea of chains of stimulu s± respo nse associations that cover the entire area an an im al has previously traversed (D eutsch, 1960). A ccord ing to this id ea, an an im al lear ns to navig ate throu gh sp ace by for m ing associations between its actions in ap proach ing a goal an d th e stimulus (or stimulus con® g ur ation s) present at the tim e a goal-directed action that eventually culminates in reach ing rew ard is perfor m ed. In the second, the an im al lear ns a retinal sn ap shot of the enviro nm ent as seen
L A N D M AR K S T A B IL IT Y
309
from the goal an d then m oves so as to decrease th e discrepan c y between the rem em b ered sn ap shot an d its cur rent view of the environm ent. For exam ple, C artwright an d C ollett (1982, 1983) describ e an algorithm that can com pute the direction an an imal shou ld m ove to get closer to the goal. A shared feature of these associative accounts is th at the an imal does not represent w here the goal is located, it only lear ns how to get there. Spatial param eters (such as distance, direction, or even proxim ity) are not explicitly represented. A t the other end of the continuum of accounts of spatial lear ning is cognitive m ap ping theory, w hich supposes that an imals can construct quasi-C artesian representation s o f the environm ent encoding the identity an d locatio n of landm arks. T he m inimal req uirem ents of such a spatial representatio n are th at it should h ave intern al geom etric co nsistency an d co nfer the ability to p lan rou tes in advan ce, including novel detours an d shortcuts. A lthough such a representatio n does not, in itself, preclude associative p rocesses, O ’ K eefe an d N ad el (1978) proposed that the lear ning r ules of cognitive m ap for m atio n are rad ically different from those of associative lear ning. Som ew here between these two extrem es are associative accoun ts of potential relevan ce to spatial lear ning such as C ollett et al.’s (1986) vector theo ry (in w hich sp atial param eters are explicitly represented, but lear ning could be associative), an d co n® gural± associatio n theory (e.g. Sutherland & Rudy, 1989). T hough th e latter is not a th eory of navig ation per se, its relevan ce derives from the possibility that individual cues in an enviro nm ent can en ter into a num ber of distinct con® gural association s, o ne o r m ore of w hich is best associated w ith the goal. Both the standard associative m odels an d O ’ K eefe an d N ad el’ s (1978) cognitive m ap theory assert links between w hat D ickin son (1980) refers to as the content an d the co nditions of learn ing. T he associative m odels start w ith study of th e co nditions of lear ning, dr aw inferences about the underlying processes (associative links), an d postulate representations an d com putatio ns that ® t these processes. C ognitive m ap theory derives the putative properties of such m ap s by an alogy w ith physical m ap s an d infers from this type of representation the likely co nditions of learn ing. In ou r view, the proposed links between content an d conditions in both types of theory are unnecessary. In particular, sp atial lear ning could dep end upon relatively simple associative lear ning processes, b ut also use som e special-purp ose m ach inery in the brain not required in oth er lear ning situations. By w ay of exam ple, a neurobiological m odel recently described by M cN au ghto n et al. (1996) su ggests th at hippocam pal place cells encode location w ithin distinct m ultiple reference fram es rather than a unitary glob al m ap (see G othard, Skag gs, & M cN au ghton , 1996, for evidence supporting this proposal). T he representation of locations w ithin any given reference fram e is ``prew ired’ ’ to b e inter nally consisten t even in the ab sence of experience, w ith landm ark infor m atio n added by a simple associativ e m echan ism . N avig ation is primarily a m atter of dead -recko ning, w ith landm ark inform ation providing occasion al updates to correct accu m ulated er rors.
C an S p atial Le arn in g B e U nd ersto od in A sso ciative Term s? O ur experim ents w ere con ducted to deter m ine w hether ``variability m akes it dif® cult, if not impossible, to build a useful m ap ’ ’ (O ’ K eefe & N ad el, 1978, p. 95), or w heth er, as is usually found in associative lear nin g, an imals concentrate on the best predictor of a
310
B IE G L E R A N D M O R R IS
stimulus, to the d etriment of lear ning abou t other signals. T he classic exem plars of this latter principle are the experim ents by Rescorla (1968) an d by Wag ner, L og an , H ab erlandt, an d Price (1968). Rescorla varied how w ell a target stimulus pred icted that reinforcem ent w ould occur. Wag ner et al. varied how w ell other stimuli predicted reinforcem ent w hile keeping co nstant the conditional relatio nship b etween the target stimulus an d reinforcem ent. In that respect, our experiments are analo gous to those of Wag ner et al., except that we explored the effects of system atic variation of spatial, not co nditional, relationships. A s in Wag ner et al.’s experim en t, the relatio nship between the target cu e an d rew ard, in our case the distance to lan dm ark(s), an d the direction relative to land m ark(s) an d p olarizing cues, w ere th e sam e in both groups. T he difference between g roups lay in how infor m ative the non -target cu es w ere. In on e g roup, rew ard w as only found in a sm all area, as speci® ed by distance from curtains an d the w alls an d cor ners of the arena. In th e other g roup that area w as large; these cues w ere, therefore, less infor m ative an d should be less able to com pete w ith the landm arks. A further difference w ith Wag ner et al.’ s experiment lies in the fact that our non-target cues m ust be th e larger fram e of reference, rather than an other landm ark. O nly if there is that larger fram e of reference is there an y b asis for classifying the target landm ark(s) as stable or unstable. In other w ords, w e explored a spatial an alog ue of relative validity an d not, strictly speaking, relative validity itself. In a previous experim ent (Biegler & M or ris, 1993), w e com pared the p erfor m an ce of rats trained w ith food at a set distance an d d irection from one of two landm arks that w ere alw ays placed in ® xed location s in an open-® eld arena w ith a condition in w hich the food w as similarly placed near o ne of two landm arks that w ere m oved independently from trial to trial. We found good spatial lear nin g, co ntrolled by the near landm ark, in th e ® xed co ndition, non e in the varied con dition, but good discrimination between th e landm arks in both. We interpreted these results in ter m s of a req uirem ent for landm ark stability an d argued that this w as in co nsistent w ith a sim ple associative account of spatial lear ning. T he idea is that, over su ccessive trials, the an imals of group ® xed develop a representation of the positions of the landm arks relative to each other an d relative to the geom etry of the arena. A lthough they ap p roach these landm arks from m any different directions on successiv e trials and so have different views of them , their ® x ed positions across trials enables a stable landm ark-centred spatial representatio n to be for m ed. H aving acquired th is stable representation of the im m ediately perceptible features of the environm en t, the an imals can then also encode the locatio n of a h idden ob ject, such as the food. C onver sely, if the landm ark s m ove around w ith resp ect to each other an d w ith respect to the geom etry of the arena, no stable spatial representatio n can be constructed, an d, for this reason, there is no representatio nal fram ew ork into w hich the hidden food’ s locatio n can be incorporated. T his ap paren t disabling of the spatial representatio nal system by geom etric instability does not, how ever, prevent the an im als from detecting that accessible food is closer to one landm ark th an the oth er, an d thus d iscrim ination lear ning, in the sense of lear ning the differential value of each landm ark, is unaffected. T his concept of land m ark stab ility w ill be elab or ated later. For the present, several am biguities in the design of this exp eriment lim it the necessity to ap peal to such a co ncep t. First, w e h ad assum ed that a w hite curtain, extend ing along the full length of one side of the arena, functioned as an effective d irectional cue, bu t w e did not test this.
L A N D M AR K S T A B IL IT Y
311
Seco nd, it is possible that the ® x ed g roup had access to add itional direction al cues, by virtu e of being trained w ith a co nsistent spatial relatio nship between the two landm arks as well as th e con sistent asy m m etric placem ent of the landm ark± feeder arrays in the arena. T hese two factors could account for th e difference b etween g roups. In the presen t study, these problem s were ad dressed by including a co ndition w ith a single rad ially sym m etric landm ark . T his ensured that the only source of direction al infor m atio n w as the explicit polarizing cues. In ad dition, third, an account of Biegler an d M orris’ s (1993) study w ithin the fram ew ork of associative lear ning theory m ay b e possible if it is assum ed that spatial representation involves con ® gural associations as proposed by S utherland an d Ru dy (1989). In each landm ark location in the arena, the view of the near landm ark from the goal m ight for m a unique con® guration w ith background cues. Slower lear ning by g roup varied w ould follow becau se th is group would have to learn as m any different con® gurations as there are landm ark locatio ns. H ow ever, the num ber of con® gural associatio ns to be for m ed should n ot depend o n the spatial arran gem ent of landm arks. G roup varied sh ou ld h ave m ore to lear n an d shou ld show poorer perfor m an ce than g roup ® xed, reg ardless of landm ark ar ran gem ent. A n in terpretatio n in ter m s of context-speci® c associations (H all & H oney, 1989) m akes the sam e pred ictions. In contrast, an interpretatio n of cognitive m ap pin g that allows the for m ation of sm all-scale local m ap s (O’ K eefe, 1991; Word en, 1992) w ould reg ard such a ``geom etric’ ’ param eter as critical. If an imals can co nstruct local m ap s w ith a certain m inimum num b er of suitably spaced landm arks, landm ark stability need o nly be local, rath er than global, an d such rep resentations m ight be used independently of th eir position w ithin a larger fr am e of referen ce. T he m in imum num ber of landm arks is expected to be eith er two or three [Worden, personal com m unication; also O ’ K eefe & N ad el, 1978, p. 73: ``A ny cue, or g roup o f cu es, can b e rem oved from the total ar ray w ithout preventing the rem ainder (so long as at least two or three rem ain) from u niquely specifying, by their relations, a particular p lace’ ’ ].
D isting u ishing th e C o n tro l b y L and m ark C u es an d C o ntext C u es A desirable feature of any experiment ad dressing the issue of relative validity is independen t m easurem ent of the ``asso ciative strength’ ’ of each cue m easu red separately. W hile w e co uld do this for the context cues, w e could n ot for th e landm ark cues. T here are two reasons. F irst, w e had only o ne large training roo m in w hich to conduct the experiments an d, therefore, cou ld o nly test th e am ou nt of lear ning of the context± food an d landm ark± food relationships in that training arena. T hus, tests conducted w ith landm arks present are tests of the control ex erted by landm ark + context cues. M ore importantly, seco nd, our use of b oth single r ad ially sym m etric landm arks an d explicit polarizing cues creates a situation in w h ich the behavioural expression of the spatial relationship between a landm ark an d food m ight only occur if both types of cues are present. A simple w ay to th ink about this is to recognize that context cues could serve a num ber of functions. T hey m ight provide d irection al infor m ation, w hich w ou ld b e equivalent across g roups. T hat an imals extract a location-indep end ent reference direction from polarizing cues is supported by neurobiological data (Taube, M uller, & Ran ck, 1990), an d the idea that directional infor m atio n is used in this w ay is w idely discussed in literature on an im al
312
B IE G L E R A N D M O R R IS
navig ation (E tienne, Joris-L am bert, D ah n-H ur ni, & Rev erdin, 1995). A lternatively, contex t cu es m ight provide distan ce infor m ation, either to deter m in e the geom etric stability of a landm ark an d/or w here the an imal should search for the food. F inally, context cues m ight provide the backdrop for a ``snap shot’ ’ from the food locatio n, this cue acquiring associativ e strength in its ow n right. In the latter cases, effective search in the ® xed groups sh ou ld be observed under the con trol of con text cues alone. E xp eriment 1 sought to add ress both th e issue of w h ether the w hite cur tain provides directional infor m ation an d the question of local versus global land m ark stability. We did so by assessing the effectiveness of search using a single rad ially sym m etric landm ark that co uld not provide d irectional infor m ation on its ow n, an d com paring it w ith training co nditions using a closely or distally sp aced second landm ark. T hese three co nditions were co nducted usin g land m arks w hose location in the arena w as varied or landm arks that were relatively ® xed. Post-training tran sfer tests were conducted to m easu re the control by landm ark + context cues, an d by context cues alone, focusing upon w hether variatio n of landm ark stability an d landm ark ar ran gem ent had similar or distinct effects in these different types of test.
E X PE R IM E N T 1 M eth o d Subjects T h e su bjects w ere 42 exp erim en tally n aive m ale L ister-h ood ed rats, m aintained at 85% of th eir free-feed ing w eigh t. T h ese w ere selected fro m a total of 64 an im als r u n for th e ® rst tw o d ays of train ing. T h e rats co n tinu ing in th e exp erim en t were tho se tha t search ed reliably for food . A n im als w ere assig ned at r and om to gro u p s before th is selection w as m ad e.
App aratus T h e exp erim en ts w ere con d u cted in a 3.3-m 3 3.3-m sq uare aren a, w ith 40-cm -hig h w h ite w alls an d ® lled to a h eigh t of 5 cm w ith w ood sh avings. T h is aren a w as set w ithin a larger squ are ro om (w h ose w alls w ere p ainted black) an d su r ro un d ed by m oveable ceiling-to -¯ oor cu rtain s h an ging betw een th e w alls of th e ro om and the aren a. T h ese cu r tains w ere black o n th ree sid es an d w h ite on th e rem aining sid e, w ith th e w h ite cu rtain serving as a p oten tial p olariz ing cu e p rov iding inform atio n abou t d irection . A lthou gh w e refer to th is referen ce d irection as ``n or th ’ ’ , the orien tatio n of the w h ite cur tain w as varied betw een trials in a r an d om sequ en ce, w ith th e con strain t th at all orien tatio ns occu rr ed equ ally often an d n on e tw ice in a row. A loud sp eaker w as p laced beh ind th e centre of each w all. D u ring tr aining, th e sp eaker beh ind th e w all w h erever th e w h ite cu r tain w as located (i.e. to th e ``n or th’ ’ ) em itted w h ite n oise. T h is serv ed as an add ition al d irectio n al cu e an d m ask ed p oten tial acoustic cu es. T h ere w ere tw o land m ark s (F igu re 1A ). L 1 w as a steep p yram id w ith a 15-cm -w ide base an d 29.2 cm h igh , an d con taining tw o p ro tr u din g ``te r races’ ’ of 2.5-cm an d 2.0-cm w idth located 11.6 an d 19.4 cm , resp ectively, above th e base. B etw een th e tw o ter races, th e land m ark w as covered w ith rou gh sa n dp ap er w ith a black stripe of 2.2-cm w idth below th e lower ter r ace. T h e rest of the lan dm ark w as p ainted ``sig n al red ’ ’ . Insid e th e tip of L 1 w as som e carp et to w h ich 1 m l of alm on d esse n ce w as ad ded before each sessio n . H oles in th e top allowed th e od ou r to escap e. L an d m ark L 2
L A N D M AR K S T A B IL IT Y
313
FIG . 1. A. L an dm ark arran gem ents, arena, and areas in clud ed in the quan titative an alysis. In this and all followin g ® gures of E xperim ent 1, the diagram s are rotated so that the polariz in g cues are at the top of the picture, re gard less of the actu al orien tatio n in a particu lar trial. In the sin gle -lan dm ark arran gem ent, the id entity of the land mark w as cou nterbalanced within grou ps. In c lu stered and sp ace d arrang em ents, the iden tity of the land m ark s in Test 3 was coun terbalanced. Time a t F+ was record ed in all group s. Directiona l B ia s com pared cumulative time in the three light-g rey c ells in the direction of F+, w ith the cum ulative tim e in the corre sp ond in g cells in the three othe r cardina l direc tio ns. Dista nce R a tio com pared the average time in the F+ an d E cells with the average time in the ``backg roun d’’ Ð this bein g de® ned as inclu ding all cells in a 5 3 5 g rid centre d on the land m ark, excep tin g those of the lan dm ark locatio n itse lf (dark g rey) and the F+ an d E cells. The analysis of dire ctional bias an d distan ce ratio in sin gle an d c lu stered arran gem ents is id entical. In the spaced arrang em ent,
only 2 cells are at equivalent direction s from both lan dmarks, and the chan ce level of direction al bias is there fore .5. (Exp erim ent 1.)
con sis ted of a stack of fou r layer s of fou r golf balls on a w ood en base, w ith a green ten nis ball on top. T h e lower ha lf of th e tenn is ball h ad 1 m l of straw ber ry esse n ce ad d ed . L 2 w as 25.5 cm h igh, w ith a m axim um d iam eter of 9.8 cm . T hu s, th e tw o land m ark s d iffered in th eir visu al, tactile, an d olfactor y p rop er ties. T h ere w ere tw o cop ies of each land m ark , u sed in alter n atio n th ro u gh ou t tr aining in an ticipation of a po st-training p ro be test w ith tw o iden tical land m ark s. T h e feed er (F + ) w as m ad e of ch ro m e tu be, 5 cm lo n g an d 19 m m in diam eter, m ou nted on a p lyw ood base; it con tained th e c ereal K ellogg’ s Ricicles as rew ard . W h en in p lac e, all bu t its top su rface w as ben eath th e w ood shav ings. T h e locatio n of th is feede r relative to th e land m ark (s) is d escr ibed later. T h e an im als’ m ovem en ts th ro ug h th e aren a w ere tracked by an overh ead video cam er a co n nected to a fram e-g rab bing digitizer cap tu rin g im ages at 10 H z. T h e s oftw are sy stem ru n n ing on an A cor n A 5000 com p u ter su btr acts, in r eal tim e, an im age of th e aren a w ithou t th e anim al p resen t from each su ccessiv e im age w ith th e an im al p resent, en abling u s to tra ck th e m oving an im al irresp ective of
314
B IE G L E R A N D M O R R IS
FIG . 1 a . B. M oveme nt of land m ark-fee der arrays in the varied co ndition s. Du ring train in g, 15 positio ns in the 4 3 4 g rid were used , w ith 50-cm spacin g be tween adjac ent positions . One position (broken lin es on tw o edge s) was only used in Test 2 (as a novel locatio n). Single an d c lu stered land marks w ere placed directly on a grid in tersection (show n for varie d± cluster ed). For the spaced arran ge me nt, the centre of the lan dmark array coin cid e d with a g rid in tersection (sh own for varied± sp aced ). D urin g the exp erim ent, there was on ly ever one
array in side the m aze. C. In the ® xed con dition, the arrays were not literally imm ob ile, but they were moved between position s in a grid with 10-cm spacin g. Otherw ise procedu re s were the sam e as in the varied cond ition. D. Possib le loc ation s of the 25-cm 3 25-cm F+ squ are s in g roup s varied± single and varied± clustered (areas of ligh t grey sh ad in g). The sam e set of 16 areas existed for grou p varied ± sp ac ed, but these were shifte d by 25 cm to the north and west. E . Possible locatio ns of the 25 cm 3 25 cm F+ squ are in g roup s ® xe d (sin gle area of dark g rey sh adin g). This was also shifted for group ® xed± sp aced . The g re y areas of D an d E were used to c alcu late the N orm aliz ed Tim e at F+ (no land m arks). (E xperim ent 1.)
w h eth er it is da rk er or ligh ter th an th e land m arks placed w ithin th e arena (S p oon er, T h om p so n , H all, M or ris, & S alter, 1994).
Procedure P retra ining. T h e rats h ad 10 d ays of h an d ling an d train ing to eat fro m th e feeder in a sep ar ate cage in th e an im al h ou se. T h is w as followed by 3 d ays of h abituation in the m az e. In th e ® r st 2 d ays, they sp ent 2 h alf-h ou r sessio n s p er d ay in the em p ty aren a, in gro u p s of fou r. O n th e th ird d ay of h abituation , ind ividu al an im als ha d o ne 10-m in ha bituatio n sessio n d u rin g w h ich 4 feede rs w ere in
L A N D M AR K S T A B IL IT Y
315
the aren a, ar ran ged in a sq uare of 1 m each sid e. We recorde d w h ich feed er w as visited ® r st and th e laten cy un til th e cor rect feed er w as fou nd . Tra ining a nd A ssignment to Groups. Training con d ition s differed alo ng tw o ind ep en d ent d im en sion s: landm ark stab ility an d land m ark arr an ge m en t. In th e stab ility d im ension , g ro u p s w ere d ivided into ® xed an d varied . For th e th ree varied gro u p s, th e land m ark s w ere m oved r an d om ly betw een 15 p ossible location s in a 4 3 4 g rid w ith 50 -c m s pa cing (i.e. aro u n d a 1.5-m 3 1.5-m area w ithin th e 3.3-m 3 3.3-m aren a; see Figu re 1B ). For th e th ree ® x ed g rou p s, th e land m ar ks were n ot literally im m obile, but m oved in a sim ilar g rid w ith 10-cm sp acing betw een ad jacent po sitio n s (i.e. in a 30-cm 3 30-cm area; Figu re 1C ). In th e other d im e nsio n , g ro u p s were train ed w ith on e of th ree landm ark ar ran gem en ts : sin gle, clustered , an d sp ac ed. T he sing le land m ark ar ran gem en t con sisted of o n e lan dm ark always p laced 50 cm n orth of F + , clustere d of tw o land m ark s side by side rep lacing th e on e land m ark of the sin gle ar ran gem en t, an d th e s paced arr an gem ent of land m ark s an d feed er for m ing a right-an gled trian gle w ith both land m ark s p laced 50 cm fro m th e feede r (F igure 1A ). O n e ach train ing trial, o n e of th e 15 location s of th e g rid w as occu pied by th e land m ark(s ) for g ro u p s sing le an d clustered , an d by th e cen tre of th e land m ark ar ray for g rou p sp aced . T h u s, th ere w ere six gr ou p s: ® x ed± sin gle (F igu re 1), varied± sin gle, ® xed ± clustered , varied ± clustered (F igu re 1B, low er righ t), ® xed ± s p aced, an d varied± sp aced (F igur e 1B , up p er left). W ith N = 42 for th e exp erim en t, each g rou p con tained N = 7 sub jects. T h e land m ar ks and feed er w ere introd u ced on the ® r st d ay of tr aining. T he ar ra y of lan dm ark s an d feed er w as m oved before each trial, in a r and om sequ en ce, to a d ifferen t g rid p osition , w ith th e con strain t th at th e ar r ay w as n ot in o ne of th e tw o p osition s closest to th e sta rt location . T h e orien tatio n of th e p olarizing cue s an d ar ray s w as c han ged betw een trials in a ra nd om sequen ce. Tra ining P rotocol. T h e su bjects w ere tr an sp or ted to th e sta rt location s in an op aqu e, slow ly rota ting box. T hey star ted a trial facing on e of the w alls in a ra nd om s eq uen ce, w ith th e con strain t that each of the fou r star t p osition s w as use d o nce a da y. T h ey w ere ru n in g rou p s of 6± 8 an im als, so the IT I for each p articular rat d epe nd ed on the latenc y of th e oth er s. IT I varied fro m abo ut 80 m in at the be ginn ing of tra ining to abou t 60 m in at th e en d . T h e rats w ere train ed for 25 d ays. D u rin g the ® rst 2 da ys, th ere w ere 2 trials p er d ay, both con d ucted w ith the land m ark s p res ent an d rew ard ed . T h ereafter, th ere were 4 trials per d ay, 3 w ith the land m ark s an d rew ard ed , 1 w ithou t land m ark s an d no n -rew ard ed . T h u s, the co n d ition al p rob ability of rew ard given th e land m ark (s), p(R, L ) = 1.0; an d th e co n d itio n al p ro bability of r eward given the po larizing (an d oth er con text) cu es, p(R, PC ) = 0.75. T h ese co n dition al p ro babilities w ere iden tical in all g ro u p s. A train ing trial beg an w ith th e rat being p laced into th e aren a fro m th e op aqu e box and en de d w h en th e r at fou n d th e food , or after 300 sec, w h ichever w as the sh or ter. A r at th at failed to ® n d th e food w ithin 300 sec w as p icked u p an d p laced in fro n t of th e feeder u n til it ate. T h e feed er w as th en re® lled an d th e bed din g th oro ug h ly shu f¯ ed befor e th e n ext r at w as p u t into th e m az e. D u rin g the ® rst 5 da ys, th e top of th e feed er w as ap p rox im ately 2 cm above th e bedd ing. O n D ays 6 to 8, th is m arg in w as red u ced to abou t 1 c m . O n all sub sequ en t d ays, th e bedd ing w as level w ith th e feed er. Rats w ere rem oved from th e m aze be fore th ey s top p ed eating. T he d aily n on rew ard ed trial w as 60 sec in d u ratio n. Tra nsfer Tests. A fter 60 rew arde d trials, 4 in e ach of th e 15 po sition s use d in tra ining, a ser ies of tran sfer tests w as con d u cted to exam ine w ha t th e an im als h ad lear n ed abou t th e location of th e food . T h ere w ere 4 typ es of test, all 60 sec in d ur atio n. Test 1 (D ay 21) w as co nd u cted w ith p olarizing cu es an d w ith the land m ark arr ay in a po sitio n p reviou sly u sed in train ing (fam iliar location ). In Test 2 (D ay 23), th e land m ark ar ray w as in a no vel po sitio n, th e 16th location of th e grid s n ot u sed d u rin g
316
B IE G L E R A N D M O R R IS
train ing. Test 3 (D ay 25) w as con d ucted w ithou t p olarizing cu es. T h ere w ere black cur tains o n all sides, all fou r sp eakers em itted n oise, an d two iden tical land m arks w ere u sed for the sp aced an d clustered g rou p s (L 1 an d L 2 w ere cou n terbalanced w ithin gro u p s). Test N L (n o land m ark s) w as th e ave rag e of 3 tests on D ays 20, 22, an d 24 an d w as p roced u ra lly iden tical to th e d aily no n-re w ard ed trial. T h is serv ed to m easu re the exten t of c on tro l over search ing by con text cu es alon e.
Data Co llection and An alyses La tency to Find the H idden Food during A cquisition. T h e tim es taken fro m placem en t in th e arena u ntil th e an im al fou nd th e food w ere ave ra ge d across trials in 2-d ay blocks (sec 6 1 S.E.M .). A s th ere w as an od d nu m ber of train ing d ays, D ay 25 w as excluded fro m th e an alysis. Tra nsfer Tests. T h e tim e sp en t searchin g in each of 29 cells aro u nd th e landm arks w as com pu ted (tha t is, 25 cells in a 5 3 5 g rid w h ose cen tre coincided w ith th ose of th e respe ctive land m ark ar r ays, p lus 1 cell beyon d th is g rid in each of th e fou r card inal d irection s). E ach cell m easu red 25 cm 3 25 cm . Fou r m easu res of perfo r m an ce w ere th en calcu lated, thre e w ith th e land m ark s p resen t an d the fou rth w h en they w ere absen t. First, Time a t F + p ro vided an abso lute m easu re (sec 6 1 S.E .M .) of com m itm en t to search at the cor rect loc ation (i.e. th e 25 3 25 cm squ are tha t w ou ld h ave co n tained th e food h ad it been p rese nt). It alon e can be com p ared w ith the tim e sp ent in th is sam e locatio n in th e absen ce of th e lan dm ark s and th u s pro vides an ind ex of th e relative d eg ree of con tro l by th e land m ark s and th e con text cu es. Seco nd , N orma liz ed Time a t F + (no la ndma rk s) is th e equ ivalent m easu re calculated for Test N L. T h e p rob lem aro se of w h at p os itio n th e F + location sho uld be con sid ered to occu p y in the absen ce of a land m ark . We con sid ered seve ral op tio ns and rep ort h ere the on e indicating th e h ighest d eg ree of con tr ol of searc h by con text cue s in th e ® xed gr ou p s w h en com p ared to varied g rou p s, th at is, th e an alysis th at m ost favou rs an interp retation in ter m s of add itive inter action s betw een land m ark an d con text cu es r ather th an land m ark stability. In th e ® xed gro u p s, th e land m ark ar ray m oved betw een 16 location s sp re ad over a 30-cm 3 30-cm area acro ss trials. T h ese d im en sion s w ere add ed to th e 25 cm 3 25 cm of th e F + cell. T h e tim e spe n t in th at 55- cm 3 55-cm area (F igure 1E ) w as th en n or m alized to be com p ar able to tim e in a 25-cm 3 25-cm cell. In g ro up s varied , th e large r sp acing of th e g rid of ar ray p ositio ns kept po ssible F + p os ition s sep ar ate (Figu re 1D ). T im e in these location s w as ad d ed an d also n or m alized to be com pa rable to a 25-cm 3 25-cm cell. T h ird , Directiona l B ia s p ro vides a m easu re of h ow m uch the an im als ha d lear n ed abou t the d irection from the land m ark in w h ich to search . It took the for m A /(A + B ) an d w as calculated in either of tw o w ays. For the sing le and clustered con d ition s, A w as d e® n ed as th e total tim e sp en t in the th ree cells in th e d irectio n of F+ (i.e. F + itself an d th e cells closer to an d fur th er fro m th e lan dm ark ) and area B as th e su m of th e tim es in the cor resp on din g cells in the othe r th ree card inal d irection s (see Figu re 1A ). T h e cha nce level is 0.25. In the sp aced -lan d m arks con d ition , o n ly tw o cells are at equ ivalent directio ns fro m both land m ark s (F+ an d E in Figu re 1A , bottom righ t). T h erefore, in calcu lating Directiona l B ia s u sing th e r atio A /(A + B ), A w as taken as the tim e at F + an d B as th e tim e at th e op p os ite location (E ). C on sequ ently, th e ch an ce level ch an ges to 0.5, an d sta tistical com p ariso ns w ith sin gle an d clustered ar ra ng em en ts are n ot p ossib le. Directiona l B ia s m ay be less sens itive tha n S pa tia l B ia s u se d in ou r p reviou s stu d y (B iegler & M or ris, 1993), bu t it h as th e adva nta ge of b eing ind epen d en t of d istan ce. We sho w later th at con trol by con text cues alon e w as m easu rab le bu t very sm all. A s taking accou n t of its in¯ u en ce w ou ld h ave involved add ition al an d , in ou r view, u n n ecessary com p lication s, w e m ad e no attem p t to su btract their relative con tribu tio n before calcu lating D irectio nal Bias.
L A N D M AR K S T A B IL IT Y
317
Fou rth , Dista nce B ia s p rov ides a m easu re of ho w m u ch th e an im als h ad lear n ed abou t th e distan ce fro m a land m ark at w h ich to search . It w as also an A /(A + B ) ratio, w h ere A is th e ave ra ge of tim e at F + an d at th e equ ivalen t locatio ns (E in F igure 1A ) an d B th e avera ge tim e in th e ``b ackg rou nd ’ ’ (i.e. all th e rem aining cells of th e 5 3 5 g rid aro u n d the land m ark , except th e cells con taining land m arks, F + or E location s. C h an ce levels are equ al in all gr ou p s (0.5). T h e p ossible m inim al co ntrib u tion of con text cu es w as ag ain disre g ard ed.
R esu lts Acqu isition Qua lita tive Description. A lthough initially reluctan t to leave the w alls, only 3 rats (out of 42) failed to ® nd rew ard unaided during the ® rst trial (D ay 1). Perform an ce im proved stead ily b efore (Blocks 1± 4) an d after (B locks 5± 1 2) th e feeder w as ¯ ush w ith the surface (F igure 2). System atic search patter ns app eared w ith in a few days in g roup s ® x ed± single an d ® x ed ± clustered. M ost frequently, the rats of these groups w ou ld check a location at the cor rect distance from the landm ark(s), retur n to the landm ark(s), an d then tu rn an d visit the next locatio n until ® nding food at th e correct place. Such patter ns were m uch less frequent in the varied± single an d varied± clustered groups, for w hich the an imals ad opted ap p arently ran dom searching near the landm ark(s). H ow ever, it w as also ap parent as th e exp erim ent proceeded that searching w as less well focused to the F+ regio n in each of these 4 g roups than in our previous study. In the spaced ± landm arks g roup s, preferen tial searching o n the ap prop riate side of the 2 landm arks em erged relatively early, w ith the accuracy im proving during training. V isits to the sym m etric location opposite to the feeder occurred o nly rarely. T he absolu te level of perfor m an ce w as broad ly com p arable to that ob served in B ie gler an d M or ris (1993).
Qua ntita tive A na lysis. D uring the initial training period w hen the feeder w as visible, an alysis of latencies showed neith er an effect of arran gem ent nor stability, Fs < 1. T he improvem ent across blocks w as highly signi® can t, F(3, 108) = 84.19, p < .0001, as w as the L an dm ark A rran gem ent 3 Blocks in teraction, F(6, 108) = 3.51, p < .01. D ifferences between ar ran gem ents existed o nly during the ® rst block, w hen the spaced± landm arks g roups’ latencies were longer. W hen the feeder w as hidden, effects of ar ran gem ent, F(2, 36) = 6.44, p < .005 , an d stability, F(1, 36) = 16.54, p < .001, ap peared. Single landm ark g roups had longer latencies than did clustered an d sp aced g roup s, p < .05, t-test, an d varied groups took longer th an ® x ed g rou ps. T he rats im proved over blocks, F(7, 252) = 24.70, p < .0001, w ith a Stability 3 Block interaction, F(7, 252) = 4.53, p < .0001. In an an alysis of B locks 10± 12 (i.e. the transfer test d ays), the effects of stability, F(1, 36) = 7.39, p < .01, an d arran gem ent, F (2, 36) = 8.96, p < .001, were still signi® can t, bu t the spaced groups found the hidden food faster than either single or clustered g roups, p < .01, t-test, w hich in tur n did not differ from each oth er, p > .05. T hu s, search latency w ith a hidden feeder ap peared to be in¯ u enced by the num ber of landm arks early in training, an d by th eir sp atial ar ran gem ent later o n.
318
B IE G L E R A N D M O R R IS
FIG . 2. L aten cy to ® nd food (se c 6 1 S.E.M .) durin g train ing w ith a visib le feeder (Blocks 1± 4 ) and with a hidd en feeder (Blocks 5± 12). Note faster tim es to ® nd food in group s w ith two lan dm arks during the early stages of train in g with a hid den feeder. (E xp erim ent 1.)
Transfer Tests Qua lita tive Description. T he paths taken by all rats du ring Tests 1 an d 2 are show n in Figure 3A . T hree qualitative features of th e paths are ap parent: First, searching w as not particularly well focused at F+ in the single an d clustered groups, the m ost focused searching at F + being in the spaced g roups (F ig ure 3A , bottom row ). Second, the ® x ed± single an d ® x ed ± clustered g roups showed no obvious directional preference relative to the polarizing cues, b ut focused their search at the ap p ropriate distance an d direction relative to the w alls (the geom etry of the arena de® nes 4 equivalent locations). T hird, the varied±
L A N D M AR K S T A B IL IT Y
319
FIG . 3A . A. Path s taken during the ® rst 30 sec of Tests 1 and 2 (su perim pose d). Only the area im m ediately su rrou ndin g the land m ark s is sh own . The c ell w here F+ w as placed durin g train in g is m arked by a sq uare. Other cells at equ ivalen t distan ce an d directions (relativ e to the walls of the m az e) are m arked by broken sq uares. The cells m easure 25 c m across. Scale bar = 1 me tre. (E xpe rimen t 1.)
single an d varied± clustered g roups searched w ith a m ore d iffuse patter n than the corresponding ® xed g roups an d closer to the lan dm ark(s). T he paths taken during the three tests w ithout landm arks (co nstituting Test N L ) are show n in Figure 3B. A ll an im als sp ent a great deal of tim e near the side w alls. O ther than in g roup ® xed± single, there is relatively little searching in the general area of the arena w here food m igh t have been found. T he quan titative an alyses bear out this qualitative d escription.
320
B IE G L E R A N D M O R R IS
FIG . 3B . B. Path s taken durin g all tests w ith no land m arks. N ote that a great deal of tim e is spen t at the side walls and corn ers by all grou ps. Note change of scale from Pan el A. (Exp erim ent 1.)
Qua ntita tive A na lysis. T here are two preliminary consider atio ns. First, it w as decided to base the an alysis on th e ® rst 30 sec of each tran sfer test becau se searching w as m ore focused during the ® rst half of the tests. Averaged across all g roups an d tests, 61.6 6 2.1 % of search at the F+ an d E locations occurred during the ® rst 30 sec. S econd , w e treated the landm ark cluster (2 landm arks side-by-side) as if it w ere a single landm ark de® ning four, not two, equivalent directions. H ad the an im als used the axis of sym m etry of th e 2 clustered landm arks, they wo uld have b een expected, in Test 3 (w ithout
L A N D M AR K S T A B IL IT Y
321
polarizatio n an d w ith identical landm arks), to have spent m ore tim e at the F + an d opposite locations than at th e E locations to the right or left of the lan dm ark cluster (see F igure 1A ). T he proportion w as 0.43 6 0.07Ð no differen t from chan ce. T he an imals ign ored the breaking of the m az e’ s fourfold sym m etry by the single sym m etry axis of the landm ark cluster, an d it can therefo re be treated as if it w ere a single landm ark. Time a t F+ . W ith the landm arks present, the anim als sp ent virtually all of the ® rst 30 sec of each test searching near them . O ur focus of attention w as on how they distributed their search w ithin th is area, be ginnin g w ith T ime at F + (Figure 4). G roups ® xed± spaced an d varied± spaced spent 5.4 6 0.9 an d 3.7 6 0.5 sec, respectively, in the F+ cell. A nalysis of Tests 1 an d 2 revealed signi® can t effects of stability, F(1, 36) = 5.90, p < .05, an d arran gem ent, F(2, 36) = 8.18, p < .0 05. A subsequent an alysis of the effect of arran gem ent revealed that the spaced land m ark g roups spent m ore tim e at F + than the m ean of th e single an d clustered g roups, F(1, 36) = 16.12, p < .001, w hich, in tur n, did not differ from each other, F < 1. T here w as no difference in perfor m an ce between Test 1 (F + in one of the training locations) an d Test 2 (F+ in a 16th novel position), an d, for sim plicity, the results of these two tests have therefore been aver aged in Figure 4A . T hus, T ime at F+ is in¯ uenced by both landm ark stability an d landm ark arran gem ent. N orma liz ed Time a t F+ ( N o La ndma rks) . A n important step in the an alysis is to establish th e relative deg ree of control over searching by the co nte xt cues an d by the landm ark (s). Figure 4B sh ow s the N or m alized T ime at F+ (no landm arks) for all groups to the sam e scale as in Figure 4 A . A key point is that the absolute values of these tim es are very low, never exceeding 0.65 sec p er square. In ad dition, the scores for the two spaced g roupsÐ 0.1 0 6 0.04 an d 0.08 6 0.02 secÐ w ere 1.5 orders of m agnitude lower than those for T im e at F + in Tests 1 an d 2 (see earlier). A n an alysis of variance did, however, sh ow effects of stab ility, F(1, 36) = 13.42, p < .001, ar ran gem ent, F(2, 36) = 6.98, p < .005, an d a Stability 3 A r ran gem ent interactio n, F(2, 36) = 4.82, p < .025. T he patter n giving rise to this interaction w as different from w h at w e observed w ith the landm ark s present. Signi® can tly g reater searching w as observed in the g roups w ith one landm ark than in those w ith two landm arks, an d an alysis of the interaction show ed that this difference w as on ly ap parent in groups ® x ed , p < .001. T he clustered an d spaced g roups did not d iffer, p > .05. L ikew ise, the effect of landm ark stability occur red only in the single-landm ark g roups, w ith g reater co ntrol by co ntext cues in group ® x ed± single than group varied± single, p < .001. T hus, control over searching by context cues alone depended o n stability an d the num ber of landm arks rather than their arran gem ent. Directiona l B ia s. F igure 5A show s the d irectional bias averaged across Tests 1 an d 2, together w ith the chan ce lev el of perfor m an ce. T he single an d clustered g roup s w ere, as a w hole, not above chan ce, w hile the spaced g roups had clearly lear ned in w hat direction from the two landm arks to search for F + . A nalysis of the single an d clustered groups sh ow ed no signi® can t difference between the ® x ed an d varied g roups, [stability, F(1, 24) = 3.05, p = .094], no difference b etween single an d clustered ar ran gem ents, F < 1, an d no Stability 3 A r ran gem ent interactio n, F(1, 24) = 1.59, p > .2. T h e above chan ce perfor m -
322
B IE G L E R A N D M O R R IS
FIG . 4. A. Tim e at F+ (sec 6 1 S.E.M .). D ata are average d across Tests 1 and 2. F+ (no land marks) plotted to sam e scale as A. (E xpe rim ent 1.)
B. N orm alize d Tim e at
an ce of g roup varied ± single w as un expected, 0.41 6 0.03, t = 5.48, p < .001. E xam in atio n of the paths revealed that this g ro up showed a tendency to search closer to the landm ark than g ro up ® xed± single an d that its search distribution w as displaced away from the polarizing cues. H ow ever, displacem ent alo ne can not account for the above-chan ce perfor m an ce, as directional b ias is insensitive to distance. T he separate an alysis of the two spaced landm ark g roups revealed no effect of stability, F(1, 12) = 1.48, p > .20 , b ut, as show n in Figure 5A , b oth g rou ps had direction al bias scores that were well above ch an ce, ps < .001. T he u se of uncontrolled cues can be excluded o nly if directional bias fell to chan ce for all three landm ark arran gem ents w hen the polarizing cues were rem oved an d, for two of these conditions, the landm ark s w ere identical. U n der th ese co nditions
L A N D M AR K S T A B IL IT Y
323
FIG . 5 . D irection al bias. A. Test co nducted with polarizin g cues and the land m arks used in train in g. D ata are average d across Tests 1 an d 2. B. Test w ith iden tical land m arks but w ithout polarizi ng cues (value 6 1
S.E .M .). (E xperim ent 1.)
(Test 3), no ne of the g roups differed from chan ce (Figure 5B). Further m ore, two separate A N OVAs show ed that the g roups did not d iffer from each other, Fs < 1. T h us, the single an d clustered g roups ap pear not to have acquired any directional infor m ation from the polarizing cues, w hile the spaced g roups, having the b ene® t of a local array, did learn in w hat direction to search.
324
B IE G L E R A N D M O R R IS
Dista nce R a tio. T he results described so far indicate that the polarizing cues (w hite cu rtain an d localiz ed sou nd source) were less salient than we had expected. T he question therefore arises of w hether distance info rm ation can still be processed effectively in the relative ab sence of directional infor m atio n, an d of how such processing is in ¯ uenced by landm ark stability. T his is especially relevan t becau se landm ark arrays w ere tran slated (i.e. ch an ged their distance), but not rotated relative to context cues. Figure 6 sh ow s th e m ean distance ratios of the 6 g roups averaged across Tests 1± 3 (Test 3 can be in cluded becau se the com p utatio n of d istance ratio is independent of directional infor m ation ). T he overall an alysis revealed a high ly signi® can t effect of stability, F(1, 36) = 27.15, p < .0001, arran gem ent, F(2, 36) = 17.13, p < .0001, an d a signi® can t Stability 3 A rran gem ent interactio n, F(2, 36) = 5.58, p < .01. T h e ® xed g roups perfor m ed b etter than the v aried groups in the single an d clustered landm ark co nditions, ps < .05, but there w as no such difference in the sp aced groups, p > .20. T hese contrasts w ere established using separate F tests an d Scheffe ’ s m ethod of ad ju sting er ror rate for m ultiple com p ariso ns (M yers, 1966, p. 333). A com plication in this otherw ise striking patter n of results is that the overall an alysis also revealed a signi® can t ch an ge in distance ratio across successive tran sfe r tests, F(2, 72) = 3.46, p < .05, an d a signi® can t triple interaction b etween tests, stab ility, an d arran gem ent, F(4, 72) = 3.48, p < .025; data not show n. Inspection of the data showed that the p erform an ce of g rou p ® xed± clustered decreased m arkedly from Test 1 (landm arks in a training location) to Test 2 (landm arks in a novel locatio n). H ad this effect ap p eared in the varied g roups, we w ould have taken it as evid ence for separate lear ning of each landm ark location. We have no ex planation for its occurrence only in o ne of the ® xed g roup s but w ill consider this point in m ore detail in the general discussio n.
F IG . 6 .
D istan ce ratio. D ata are ave raged across Tests 1± 3 (value
6
1 S.E .M .). (Expe rim ent 1.)
L A N D M AR K S T A B IL IT Y
325
D iscussio n T he three m ain ® ndings were: (1) the polarizing context cues exerted little control over searching on their ow n, b ut an effect of relative validity w as observed; (2) an effect of landm ark stabilityÐ lear ning the distance b etween food an d a landm ark w as imp aired w hen a single landm ark m oved locations during training; (3) the square geom etry of the arena an d the spatial arran gem ent of a p air of landm arks w ere both im portant determ inan ts of lear nin g. The Im p lications of Poor C ontro l over Searching by C ontext Cues In this experim ent, the ® x ed± single an d ® xed± clustered g rou ps sh ow ed no directional bias. T his ® nding ind icates that the w h ite cur tain an d noise source, d espite their ap parent prom inence, were not very effective polarizing cues. C hen g (1986) an d M argules an d G allistel (1988) have re ported sim ilar observatio ns. N evertheless, context cues did have a sm all but sign i® can t in¯ uence on searching patter ns w hen the landm arks w ere absent. A s ex pected on the b asis of associative lear ning theory, g reater control w as observed in g rou p ® xed± single than varied ± single, an d there w as a decline in con trol as the num ber of landm arks increased. T his patter n is in accord w ith a principle of relative validityÐ co ntext cu es show ing g reatest (albeit rather w eak) contro l in a situatio n in w hich the context is m ost infor m ative an d the landm ark(s) are least salien t. T he an imals in g roup ® x ed± single m ay h ave lear ned the general area of the arena in w hich to search, p erhap s on the basis of a sn ap shot taken w hen cor rectly ap proach in g the F + lo cation du ring train ing. H owever, the co ntrol by context cues is clearly insuf® cient to acco unt for the m uch larger d ifferences between g roups observed w ith the landm arks present. A qu ite separate im plication is th at, in our previous study (B ie gler & M orris, 1993), the co nsistent spatial relationship b etween L + an d L 2 an d/or the asym m etric placem ent of L + in g ro up ® xed probably provided an important directional cue. A s g roup varied did not bene® t equivalently, the conclu sion draw n th en that landm ark stability in ¯ uences sp atial lear ning is insecure on the basis of that study alone. Land m a rk S ta bility In¯ uences Perfo rm an ce T he signi® can t effect of lan dm ark stability in the g roups w ith either a single landm ark or clustered landm arks w as show n in the T ime at F + m easure and , m ost clearly, in the distance ratio m easure of Figure 6. T h e paths taken by the rats in the ® xed± single an d ® xed± clustered g roups indicate that direction (relative to the w alls) an d distan ce (from the landm ark ) had b een learned. T hat is, these two g roups had lear ned the geom etrical arran gem ent between food an d landm ark relative to the w alls of th e aren a. H ad they lear ned only the ap propriate distance, they w ould have searched around th e landm ark in a circle; had they lear ned o nly to w alk tow ards or aw ay from the landm ark(s) parallel to one of the w alls, not know ing at w hat distance to stop, they w ould have spent a sub stantial prop ortion of time too far aw ay from the landm ark. N either patter n w as observed. T hus, landm ark stability can positively in¯ uence perfor m an ce even w hen d irection al infor m ation is am b iguou s or w eak.
326
B IE G L E R A N D M O R R IS
An Effect of Landm ark Arra ngem ent T he co ntrast between perfor m an ce w ith th e spaced -landm ark arran gem ent an d that w ith either the sin gle or clu stered ar rangem ents, both in overall ter m s an d in the interactio n b etween stability an d geo m etric ar ran gem ent, is co nsistent w ith predictions of O ’ K eefe’ s (1991) an d Worden’ s (1992) m ap ping theories that local landm ark stability sh ou ld be suf® cient to support spatial lear ning given a m inimum of two ap propriately sp aced landm arks. H owever, th e striking difference b etween the clustered an d spaced landm ark arran gem en ts co ntrad icts the pre diction of co n® gural associatio n th eory to the effect that m ovin g a landm ark array, so creating m ultiple con® gurations that have to be learned individually, should be independent of the spatial ar rang em ent of the landm arks. N ote, however, that if spatial relations are assum ed to be elem ents of con® gural associations, this theory becom es dif® cult to distinguish from cognitive m app ing theories (Sutherland & Rud y, 1989, p. 130). Processing w ithin the con® gural association system is not described in suf® cient detail, even in its m ost recent presentation (Rudy & Su therland, 1995), to allow a de® nitive interpretation.
E X PE R IM E N T 2 T he primary w eakness of E xperiment 1 is th e lack of effective polarizing cues for the single an d clustered groups. T hese g roups ap paren tly disreg arded the polarizing curtain an d noise source an d so did not have an un am biguous directional referen ce. T he w eakness of the context cu es m ay also have contributed to the very m odest deg ree of co ntrol over searching observed w hen these cues w ere te sted alo ne. We sou gh t to add ress this problem by considering two contrib utory factor s. In a recent study of hippocam pal place-cells an d thalam ic directional cells, K nierim , K ud rim oto, an d M cN au ghto n (1995) fo und th at a disorientation procedu re similar to ours m ad e rats disreg ard a cue card in a sm all circular open ® eld m aze an d frequently led to ran d om rem ap ping of place ® eld s. Vestibular an d visual cues need to be m utually co nsistent for both place- an d head -direction cells to m ain tain stable ® ring ® elds acro ss time (see also E tienne, L am b ert, Reverd in, & Teroni, 1993). In E xperiment 2, w e sought to achieve this by om itting disorientation of the rats’ v estibular system at the start of each training trial an d, consequently, not rotating the polarizing cues between trials. A separate factor is that, in E xperiment 1, the polarizing cues w ere present on the single non-rew arded trial of the day. We therefore decided to in crease their correlatio n w ith rew ard by rem oving them o n this trialÐ i.e. chan ging p (R, PC ) from 0.75 to 1.0. Both ch an ges should increase the deg ree of control by the co ntext cues an d so provide a m ore rigorous test of the dual role of such cues in this kind of task. Follow in g train ing, tran sfer tests w ere ag ain co nducted to exam ine th e possible ro le of landm ark stability under cond ition s that w e h oped w ould be characterized by effective directionality, an d we took the opportunity to exam in e the relative contrib utions of vestibular an d explicit polarizing cues in deter m ining the an imals’ sense of direction.
L A N D M AR K S T A B IL IT Y
327
M eth o ds Subjects T h e sub jects w ere 12 m ale L ister ho od ed rats, selected fro m 18 after D ay 2 of tra ining. T h ey w ere treated exactly as th e anim als in E xp erim en t 1.
App aratus T h e ap p aratu s w as iden tical to th at in E xp erim en t 1.
Procedures T h e train ing pro ced u res w ere iden tical to th ose in th e sin gle land m ark con d ition of E xp erim en t 1 w ith th e follow ing excep tion s: (1) th e an im als w ere car ried to th e sta rtin g p oint w ithout ro tatio n of the tran sp ort b ox; (2) th e p olarizing cu es (cu rtain s and n oise so u rce) always occu p ied th e sam e w all of the laborato ry ro om ; th ey w er e, h ow ever, ro tated by 90 8 for tw o of the tran sfer tests (see later); (3) th e p olarizing cu es w ere absen t d u rin g the n on -re w arde d trials; th e n oise w as sw itch ed off an d th e w h ite cu rtain re placed by a black o ne; (4) th e feeder w as p laced 50 cm ``e ast’ ’ of th e sin gle land m ark (rath er than ``so u th’ ’ , as in E xp erim ent 1), to exclude th e sim p le strate gy of w ithd raw al fro m po larizing cu es (these being to th e ``no r th ’ ’ ); (5) a series of 8 tra nsfe r tests w ith land m ark s (nu m bered co ns ecu tively) w as b egu n on D ay 21 (Tests 1± 8), 1 tran sfe r test w ithou t land m ark s b ut w ith p olarizing cu es (Test P O, for Polarizing cues On ly), an d 6 inter m ixed da ys of testin g w ithout either land m ark s or p olarizing cu es, w h ich, w h en ave rag ed , w ere co n sid ered as Test N LP (N o Lan d m ark s or Polarizing cu es). T h ese 15 tests w ere sch ed u led d aily (D ays 21± 35). Tests w ith land m ark s w ere always 2 d ays ap art. T h e p roc edu res for th ese tran sfer tests w ere as follows: Tests 1 an d 2 w ere iden tical to th e cor respo n d ing tests in E xp erim en t 1, excep t th at vestibu lar cu es w ere available. Test 1 w as w ith th e land m ark s in a fam iliar locatio n, an d Test 2 w ith th e lan dm ark s in a n ovel location n ot u sed d u rin g tr aining. Tests 7 an d 8 w ere a rep eat of th ese tests at th e en d of the test series. T he p u rp ose of th ese 4 tests w as to re-exam ine th e n otion of lan dm ark stab ility an d to test th e effects of fam iliarity of th e landm ark ’ s p ositio n an d of rep eated testing. Tests 3 to 7 w ere design ed to investigate th e effects of po larizing an d vestibular cues w ith th e land m ark in a fam iliar location . In Test 3, po larizing cues w ere absent but vestibular cues p resen t. In Test 4, both p olarizing an d vestibular cues w ere p resen t, but w ith th e p olarizing cues ro tated 90 8 coun terclockw ise relative to their orientation du ring trainin gÐ that is, these two d irection al cues disag reed. In Test 5, the p olarizing cu es w ere ro tated 90 8 clockw ise (relative to tr aining) an d vestibu lar cu es d istu rb ed by at least 10 slow ro tation s of the tran sp ort b ox (w e refer to this as vestibu lar cu es ``ab sent’ ’ an d justify th is d escription in th e discu ssio n ). In Test 6, n either p olarizing n or vestibu lar cu es w ere p resen t. In Test 7, both po larizing an d vestibu lar cu es w ere pr esen t an d in ag reem en t. Test PO (D ay 22) w as con d u cted w ith p olarizing cu es bu t w ithout land m ark s. Its p u rp ose w as to explore w h ether p olarizing an d backg rou n d cu es alon e were su f® cien t for focu sed search . A s in E xp erim ent 1, w e com p u ted N or ma liz ed Time a t F+ (no la ndma rk s) for each of th e tw o g ro u ps. Test N L P (D ays 24 , 26, 28, 30, 32, 34) w as con d u cted w ith n either p olarizing cu es n or land m ark s. It w as th e u su al extinction trial u sed th ro ug h ou t train ing. T h e sam e m easu re w as calcu lated as in Test P O just d escribed .
328
B IE G L E R A N D M O R R IS
R esu lts Acqu isition D uring the ® rst 4 blocks, w hile the feeder w as visible, there w as a signi® can t decline in latency across blocks, F(3, 30) = 37.94, p < .000 1, no difference in latency between the ® xed an d varied g roups, F(1, 10) = 1.42, p > .2, an d no L an d m ark Stab ility 3 Block interaction, F(3, 30) = 1.58, p > .2. T hus, landm ark stab ility had n o effect on task dif® cu lty w hile the feeder w as visible (Figure 7). W hen the feeder w as hidd en, from Block 5 onw ards, the effects of stab ility, F (1, 10) = 12.58, p < .01, blocks, F(12, 120) = 7.14, p < .0001, an d the Stability 3 Blocks interaction, F(12, 120) = 2.32, p < .02, were all signi® can t. Further an alysis show ed that the faster latencies in group ® xed w ere only present on B locks 5, 7, an d 8, ps < .02, an d that the decline in latency across blocks w as signi® can t only in g roup varied, F(12, 120) = 8.0 7, p < .001, but not g roup ® xed, F(12 , 120) = 1.38, p > .1. G roup ® xed ap pears to have lear ned the sp atial relationship between landm ark an d feeder w hile the feeder w as visible, an d there w as, therefore, little opportunity for further im provem ent w hen the feeder w as hidden. Transfer Tests T he paths taken during Tests 1 an d 7 (com b ined) an d Tests 2 an d 8 are show n in Figure 8. T hese show relatively fo cused search at F+ in group ® x ed an d a m ore dispersed patter n in g roup varied. A n alysis of the tim e at F+ revealed a signi® can t effect of stability, F(1, 10) = 7.34, p < .02 5Ð F igure 9. T here w as n o difference between early an d late tests, F < 1, but, surprisingly, there w as a decrem en t in search at F+ if the landm ark w as in a n ovel locatio n, F(1, 10) = 9.42,
FIG . 7. Laten cy to ® nd food (se c 6 1 S.E .M .) during train ing trials with a visib le feeder (Blocks 1± 4) an d a hidd en feeder (Bloc ks 5± 17 ). (E xpe rim ent 2.)
L A N D M AR K S T A B IL IT Y
329
FIG . 8 . Paths during the ® rst 30 sec of tests w ith fam iliar and novel land m ark positions . Paths from early an d late tests have been su perim posed . Only the portio n of the paths close to the lan dm arks is sh own . The cell wh ere F+ w as placed during trainin g is marked by a square. Other cells at e qu ivalen t distance and direc tio ns (relative to
the walls of the m az e) are m arked by broke n squares. Polariz in g cues (locatio n not to scale) are situ ated towards the top of the page. For illu strative purpose s, both lan dm arks are shown . They were cou nterbalanced w ithin group s, but an y ind ividu al rat was on ly trained an d teste d w ith the sam e lan dm ark. (Exp erim ent 2.)
p < .02Ð Figure 9. T he distance ratio m easure followed a similar trend, b ut th e effects of stability an d fam iliarity nar row ly failed to reach signi® can ce, F(1, 10) = 4.05, p = .072, an d F(1, 10) = 4.19, p = .068, respectively, all oth er Fs < 1Ð Figure 9. In the an alysis of directional bias, on ly the effect of fam iliarity ap proach ed sign i® can ce, F(1, 10 ) = 3.85, p = .07 8; all other ps > .2Ð F igure 9. T he absolute valu es for Time a t F+ for Tests 1 an d 7 com b ined w ere g roup ® x ed = 6.9 4 6 0.92 sec an d g roup varied = 4.15 6 0.5 7 sec, a difference of 2.79 sec. T he question arises of w heth er this difference w as, in part, due to co ntrol over searching by context cues alone. T he ab solute values for N orma liz ed Time a t F+ (no la ndma rk s) for Test PO were ® xed = 0.24 6 0.08 an d varied = 0.15 6 0.02, values that are even lower than in E xp eriment 1 (desp ite a better cor relation of context cues w ith reinforcem ent), an d a difference b etween g roups of only 0.09 sec. A nalysis revealed no effect of landm ark stability d uring training upon perfor m an ce during Test PO (F < 1). It follow s that the signi® can t effect of landm ark stab ility in Tests 1 an d 7 re¯ ects som e positive in¯ uence
330
B IE G L E R A N D M O R R IS
FIG . 9. 2.)
Time at F+ (se c
6
1 S.E .M .) D istan ce ratio an d D irection al bias for Tests 1, 2, 7, an d 8. (E xp erime nt
upon spatial representatio n of having a landm ark present that is ® x ed w ith respect to the geom etry of the w orld around. C om parison of Test PO w ith Test N L P (no landm arks or polarizatio n) revealed a further drop in the absolute m ean sco res from 0.24 6 0.8 to 0.05 6 0.02 sec per square for g roup ® xed, an d from 0.15 6 0.02 to 0.07 6 0.02 sec for g roup varied. Rem oval of the polarizing cues cau sed a signi® can t decline in searchin g, F(1, 10) = 9.05, p < .025, b ut landm ark stab ility during training w as w ith out in¯ uence. T hus, the polarizing cues had m easu rable but m inim al associative strength w hen tested alone.
L A N D M AR K S T A B IL IT Y
331
Interactions betw een D irectional C ues T he paths taken during Tests 3± 7 are show n in Figure 10. Test 3 (polarizing cues absent, vestibu lar cues present) show s focused search to the east o f the landm arkÐ as in Tests 1, 2, 7, an d 8 of F igu re 8. In Test 4, rotatio n of the polarizing cu e by 90 8 cou nterclockw ise (i.e. from north to west) failed to sh ift the direction of m axim al search from that sh ow n in Test 3; the rats ap peared, therefore, to determ ine the location of F+ using vestibular in preference to polarizing cues. H owever, in Test 5, w ith vestibular cues absent, rotatio n of the polarizin g cues by 90 8 clockw ise did result in the locatio n of preferential search rotating by an equivalent am ount. T hat is, instead o f searching in predom inantly the sam e direction as in Tests 3 an d 4, the an imals focused w ith in a cell to the south (in room coordinates) of the landm ark. T hu s, tho ugh vestibular cues m ay be u sed in preferen ce to explicit polarizing cues, the polarizing cues are suf® ciently salient on their ow n to direct focu sed search ap p ropriately. In Test 6, w ith bo th polarizing an d vestibular cues ab sent, search w as no longer focused in an y one direction. T h is indicates that the an imals w ere not using an y further uncon trolled cues. In Test 7, in w hich the two sets of cu es w ere in ag reem ent, search w as focused in the co rrect direction to the east. A n A N OVA of Tests 3, 5 , 6, an d 7 for T im e at F+ revealed signi® can t effects of polarizing cu es, F(1, 10) = 25.65, p < .0005, an d of vestibular cues, F(1, 10) = 8.98, p < .02Ð F igu re 11A . D istance ratio w as higher in th e presence of polarizing cues, F(1, 10) = 6.41, p < .05. A nalysis of directional bias revealed a similar patter n to that sh ow n for tim e at F+ : the effect of polarizing cues w as reliable, F(1, 10) = 24.18, p < .001, w hereas that of vestibular cues w as only m arginally signi® can t, F(1, 10) = 3.87, p = .078.
Com pa rison of Experim en ts 1 an d 2 Two com parisons are of interest. T he ® rst concer ns the focus of this pap erÐ landm ark stabilityÐ exam ined as a function of th e effectiveness of directional cues. E xam inin g the data from Tests 1 an d 2 in b oth experim ents revealed that in both m easures w ith a directional com po nen t, T im e at F + an d direction al bias, perfor m an ce increased between experiments, F(1, 22) = 13.80, p < .002, an d F(1, 22) = 10.75, p < .01, an d there were E xperiment 3 S tab ility interaction s, F(1, 22) = 5.26, p < .05, an d F(1, 22) = 6.45, p < .02, du e prim arily to improvem ents in the ® xed co ndition. E ffects of landm ark stability w ere signi® can t in m easures that included a com ponent of distance [Tim e at F + : F(1, 22) = 6.48, p < .02, an d distance ratio: F(1, 22) = 18.05, p < .001], but not signi® can t w ith respect to directional bias. Varying the distance between the landm ark± feeder ar ray an d the w alls w hile keeping direction co nstant affected only the an imals’ know ledge of distance. T he second com parison exam ined w hether th e effectiveness of the polarizin g cues, co nsidered in isolation, chan ged between experiments. T he key differences w ere that these cues w ere differentially reinforced in Experiment 2 an d that vestibular cues w ere also present d uring training. We com pared perfo rm an ce in tests w ithout vestibu lar cues in both experiments. Both T im e at F + , F(1, 22) = 13.06 , p < .002, an d directional bias,
3 32
L A N D M AR K S T A B IL IT Y
333
F(1, 22) = 22.01, p < .0001, w ere higher in E xperiment 2 in tests conducted w ith landm arks presen t. T hus, p olarizing cues w ere m ore effective in E xperiment 2 than in E xperiment 1, even though (a) the simultan eously present vestibular cues took preceden ce w hen both cues w ere available an d in disag reem ent, an d (b ) the co ntrol over searching by co ntext cues alone w as unchan ged b etween the two studies.
D iscussio n T he m ain ® ndings of this study were: (1) landm ark stab ility w as ag ain seen to in¯ uence sp atial learn ing positively, although signi® can t learn ing w as observed in g roup varied; (2) au gm enting the effectiveness of the directional cues im proved the accuracy of se arch, w ithout m ajor effect up on searching in the absence of the landm arks; (3) the sam e polarizing cues as used in Ex periment 1 w ere found to be effective in this experiment, althou gh vestibular cues nor m ally had priority; an d (4) placing the landm ark in a novel location not used in training w as associated w ith lower levels o f perform an ce. Land m a rk S ta bility Still In¯ uences Perform ance w hen Directional C ues A re Effective T hree chan ges in procedure w ere m ad e, com pared to E xperiment 1, to improve the effectiveness of the directional cues. T he result w as that, instead of search ing p referentially in four discrete locations (Figure 3A ), g roup ® x ed now con centrated its search at one single location (Figure 8). We do not know w hich of the three ch an ges w as the m ost important, but w e have n everth eless realized a condition in w hich rats, given a single rad ially sym m etric landm ark an d effective directional cues, w ill focus their search accurately. T his ® nd ing is consistent w ith C ollett et al.’ s (1986) observations o n gerbils, but our co ntrol of th e directional cues an d explicit m an ipulation of landm ark stability have provid ed ad ditional in for m atio n. Perfor m an ce w as better in g roup ® xed than in g roup varied, w ith m inim al control of searching by context cues tested alone. T his ® nding is consistent w ith our p roposed principle of landm ark stab ility. U se of D irectional Cues T he rats’ use of directional cues has som e theore tical implicatio ns. First, if the increase in the cor relation between the polarizing cues an d rew ard is w hat has imp roved th eir effectiveness, an alogous to the results of Rescorla (1968), O ’ K eefe an d N ad el’ s (1 978) FIG . 1 0. Path s during tests examin in g the relative in¯ uence of dire ctional cues. Polariz in g cues (wh ite curtain an d lou dspeaker) are sh own , wh en prese nt, in their actu al orie ntation durin g the test (locatio n not to sc ale). In Test 3, vestib ular cues on ly were present. In Test 4, two cells are m arked as F+ as the polariz in g cues pred icted F+ to be north of the land mark, whe reas ve stib ular c ues pred ic ted it to be to the east. The rats gave priority to vestib ular cues. Test 5 sh ow s that po larizin g cues cou ld de® ne a referen ce directio n in the absenc e of vestib ular cues. In Test 6, the absen ce of e ith er polarizin g or vestib ular cues resu lted in the rats showin g no direction al preference. In Test 7, both polariz in g an d vestib ular c ues were presen t an d in ag reem e nt. Pre fere ntial search retu r ns to the origin al F+ location , ind icatin g that the series of tests has not resu lted in an irreve rsib le disrup tion of perfo rm an ce. Both land m arks are sh own. They w ere c ou nterbalanc ed within g roup s, but any in divid ual rat was on ly trained an d tested with the sam e land m ark. (Exp erim ent 2.)
334
B IE G L E R A N D M O R R IS
interp retatio n of cogn itive m ap theo ry w ou ld ap pear to be incorrect. T heir theory m akes one of two predictions in a circum stance in w hich a cue is only som etimes present: either the subject co nstructs o ne cognitive m ap of the enviro nm ent but excludes this cue becau se it is not a stable featureÐ in w hich case the location of the food re lative to the landm ark can not be encoded w ithin the m ap Ð or two separate m ap s are co nstructed, one w ith the cue an d one w ithoutÐ an d lear nin g w here th e food is w ith the cue present sh ou ld be unaffected. T he theory does not predict that rem oving a cue from the environm ent o n som e trials could ever improve lear ningÐ the outcom e that w e m ay have obtained . O n the other han d , second, Etienne et al. (1993) have suggested that m utual correlation b etween visual cues an d dead recko nin g cau ses them to potentiate each other, w hereas K nierim et al. (1995) found that creating a co n¯ ict between visu al an d vestibular inform ation led rats to ignore visual cues. T he latter argued that disorientation through rotation of a tran sport box resets vestibular cues at ran dom . O ur resu lts are also consistent w ith the idea that visual an d vestibu lar cues should be m utually con sistent to b e effective. T here w as no overshad ow ing of the w eaker by th e stron ger cue. Test 4 of this experim ent sh ow ed th at vestibular cues have a higher priority than do polarizing cues, although in Test 5 polarizing cues alone w ere found to be suf® cient to provide a directional reference. H owever, w hen th e stronger vestibular cues w ere m ad e unreliable in E xperim ent 1, co ntrol of search direction by the polarizing cues w as w eaker. T his is co nsistent w ith the results of Jackso n an d Fritsche (1989), w ho foun d in taste aversion conditioning that, if th e salience of an over shad ow ing cue w as weakened suf® cien tly, it could be potentiated. O ur results do not allow us to determ ine de® nitively w hether there w as m utual p otentiation of vestibular an d visual cues. T hat effect would be expected from an ap plication of a landm ark stability princip le to directional cues an d would go beyon d the results of Jackso n an d Fritsche (1989). O n th e basis of the pre sent d ata, it is impossible to decide w h ether increasing the co ntingency between polarizing cues an d rew ard, or the m utual ag reem ent b etween vestibular an d polarizing cues, or b oth of these factors, w ere responsible for the g reater effectiveness of the polarizing cues in this experiment. H owever, one reason for favouring the latter is th at their effectiveness as directional cu es increased dram atically between E xperiments 1 an d 2, but their ab ility to trig ger ap propriate searching w hen tested alone w as unchan ged (or, if any thing, d eclined).
FIG . 11 . A. Time at F+ (sec 6 1 S.E.M .) for Tests 3± 7. The high est tim es are shown for Test 7, wh en polarizin g cues and vestibu lar cues are both present and in agreeme nt. The black an d white bars in Test 4 show data calculate d usin g a geocentric reference direction (consistent with vestib ular cues). These data are used in all statistical com parison s across tests. The dark- an d light-gre y bars show the same data calcu lated usin g a reference direction given by the polarizin g cues. Note that vestib ular cues dom inate. B. D istance ratio. C . D irectional bias. (E xperime nt 2.)
335
336
B IE G L E R A N D M O R R IS
The Behavioural Effects of V estibu lar D isorienta tion We ag ree w ith K nierim et al. (1995) that vestibular d isorientation can cau se ran dom resetting of rats’ directional system , at least as represented by thalam ic ``head -direction’ ’ cells, but w e b elieve that the resulting con¯ ict between directional cues can not, in our ex periment, be the o nly factor in the devaluation o f local visual cues. Test 4 of E xperim ent 2 established that our su bjects relied m ore on vestibular than o n visual an d au ditory polarizing cu es. E very thing else being the sam e, ran dom resetting of vestibular cues in Test 5 should hav e bro ugh t the g roup m ean s for both T ime at F+ an d directional bias to ch an ce, w hereas increasing variance for b oth m easu res b ecau se ind ividual r ats should have show n idiosyncratic directional preferences as stro ng as in Test 4. Instead , the rats disreg arded vestibular infor m ation (this justi® es ou r d escription of vestibular cues as functionally ``absen t’’ in tests w ith disorientatio n, rather than m erely ``disturbed ’ ’ ). We su ggest that rats can reclassify vestibular infor m ation as unreliable w h en they are diso rien ted, an d that this d oes not require previou s experience (provid ed the disorientin g rotation is above detection threshold). In oth er w ords, both the m agnitude of a variable an d an estimate of its reliability are represented. If this interpretation is correct, it w ould represent a challenge to associative lear ning theory. A de® ning characteristic of associative lear ning is that its output is restricted to a single variableÐ an associative strength. E xplicit representation of the reliability of a variable is not possible w ith existin g m odels. N either have m o dels of cognitive m ap ping yet attem pted to acco unt for the possible representation of reliability. O bviou sly, a single result can not establish this interpretatio n of our data as a general feature of lear ning, an d the q uestion must rem ain open for the m om ent.
W hy W as Perform ance Affected by U se of N o vel Landm ark Lo cations? T he perfor m an ce d ecrem ent in the tests w ith a novel landm ark locatio n confor m s to predictions of a con® gural± associatio n acco unt. T he p uzzling feature is that the ® xed g roup w as as affected as group varied. In the ® x ed g roup, the novel location of L + w as only 10 cm from the two nearest fam iliar locatio ns, w hereas the nearest arena w all w as 1.5 m aw ay. H ow ever, as the search distributions of g roup ® xed g ave no indications that localizatio n w as accurate to w ithin less than 10 cm w hen search w as guid ed by a landm ark only 50 cm aw ay, it is puzzling that a 10-cm shift relative to w alls 1.5 m aw ay w as detected. In g roup varied, the n ovel location w as 50 cm from the nearest fam iliar locatio n an d 95 cm from an arena w all. A ccordingly, if th e rats in b oth groups lear ned each training locatio n separately, even w ith only 4 or 5 training trials at each location, a g reater decline in search accuracy for the novel locatio n m ight have been ex pected in g roup varied , that is, there shou ld have b een a Stability 3 Fam iliarity interaction. T his did not occur. T he decrem ent in the novel lan dm ark condition is also inconsistent w ith the results of E xperim ent 1 w here only g roup ® xed± clustered sh ow ed a decrem ent.
L A N D M AR K S T A B IL IT Y
337
G EN E R A L D IS C U S S IO N T he prim ary purpose of these experiments w as to explore w hether aspects of allocen tric sp atial lear nin g can be u nderstood in ter m s of established prin ciples of associative conditioning. T hese include the idea that lear ning proceeds m ost read ily between a stimulus an d reinforcer w hen that stimulus is th e m ost reliable pred ictor of reinforcem ent (Rescorla, 1968; Wag ner et al., 1968). O ur results show th at the ab ility of co ntext cues to direct ap p ropriate searching w as greatest w hen they w ere m ost infor m ative w ith re gard to th e location of the food. H owever, the absolute m agnitude of this con trol w as very weakÐ too weak to explain th e g reater effect of lan dm ark stability upon p erfor m an ce m easu red in two d ifferent w ays. A ccordingly, our results do not so mu ch call into q uestion w hether the p rinciple of relative validity holds w ithin th e spatial d om ain; rather, they su ggest that varying the relative spatial predictability of a landm ark by variatio n in its geom etric stab ility has unan ticipated ad ditional effects. Speci® cally, spatial lear nin g ap pears to b e im paired under conditions in w hich landm arks are not geom etrically stable. E xperiment 1 con® r m ed the general p rinciple under im proved testing co nditions b ut indicated that local stability w ithin an array of at least two landm arks is suf® cien t to su pp ort spatial lear ning. It also raised the possibility th at the encoding of d istance an d direction w ere differentially affected by intertrial tran slatio n of the landm arks, alth ough not com pletely ind ependent. In E xperim ent 2, directional stability w as achieved by m aintaining a co nsisten t reference direction w ith respect to the room . U nder these conditions, stability of a single landm ark still had a positive in ¯ u ence o n perfor m an ce, but the effect w as m erely q uantitativeÐ an imals trained w ith m oving land m ark s w ere cap able of som e sp atial lear ning.
Th e R ole o f Lan dm ark S tability in S p atial Learning In contrast to the m ost straightforward interpretation of associative learning theory, lear ning about the landm ark’ s spatial relationship to rew ard w as impaired, not aided, by the spatial inconsistency between context cues an d the landm ark± feeder array. Fixed and varied g roups, in both the single and clustered landm ark condition of E xperiment 1, used the sam e directional cues in different w ays. The ® xed g roups focused search in four discrete directions an d at the correct distance, whereas the varied g roups searched with a m ore diffuse distribution an d a weak preference both in one direction an d closer to the landm ark. E xperiment 2 provided suf® ciently salient directional cues that both groups had an unam biguous reference direction. U nder these conditions, the superiority of the ® xed group rem ained, although the varied group showed above-chanc e levels of spatial learning. T his latter result forces us to aband on the notion that inconsistency between sources of spatial infor m ation com pletely disables spatial learning (Biegler & M orris, 1993). H owever, the deleterious effect of instability on spatial lear ning is a replicable phenom enon. W hat our ® nd in gs o n landm ark stability poin t to is an im portant task that an y allocentric spatial lear ning system has to do. Speci® cally, it has to discover the relative locations of stimuli that are perceived as com prising the stable physical environm en t± that is, the locatio ns of landm arks that do not m ove around, such as trees, b ou lders, or rivers. A lthough an im als can ap proach su ch lan dm arks from m an y different directions
338
B IE G L E R A N D M O R R IS
an d so have multiple views of them , their categorical perception of the w orld som ehow en ables scenes to b e com partm entalized into (a) la ndma rksÐ w h ich are best if geom etrically stable, an d (b ) other objectsÐ w hich can m ove an d are not used for navig atio n (su ch as other an im als). A cquiring a stable m ental representatio n o f environm ental landm arks m ay be a prerequisite, or at least a help, in encoding the location of hidden ob jects that can not be perceived. E xposing an imals to a w orld in w hich land m arks inexplicably m ove aroun d in relation to each other w ould be deleterious to the process of tran slating infor m atio n derived from a viewer-cen tred p erceptual representatio n to a landm ark-centred m em ory representation.
Lo ca l, N o t G lob al S tab ility Is S uf® cie n t for S p atial Lea rn in g E xperiment 1 revealed that consistent spatial relatio ns between two landm arks an d a goal, i.e. local landm ark stability, w as suf® cient to support spatial lear ning. G lobal stability (with respect to location w ithin the training arena) is not necessary. We did not exam ine w hether local landm ark stability is necessary for spatial lear ningÐ by training a g roup in w hich food w as in varied locations across trials w ith respect to both the arena an d local landm ark s, an alogous to the truly ran dom control of Rescorla (1968)Ð becau se it seem s obvious to us that such a group could not lear n. In an y event, that local landm ark stability can b e suf® cien t to support sp atial lear ning supports theories that explicitly take spatial co nstraints into account (G allistel, 1990; O ’ K eefe, 1991; Worden, 1992). C on ® gu ral theories (S uth erland & Rudy, 1989) can not account for th e differential perfor m an ce of g roups varied± clu stered an d varied± spaced as th ey o nly differed in the spatial arran gem ent of the two landm arks an d not w ith respect to th e con ® guratio n of lan dm ark ar ray an d co ntext. E xp eriment 1 show ed th at local landm ark stability w as suf® cient to support abovech an ce spatial lear ning in group varied± spaced, at a level indistinguishable from g rou p ® xed± spaced. W ithin this triangular ar ray, six spatial param etersÐ three distances an d three an glesÐ were kept stable. E xp eriment 1 d em onstrated only that this is a suf® cient requirem ent for the develop m ent of a local spatial representation. It is n ot clear w hether it is necessary. E xperiment 2 revealed that o nly two stable spatial relationsÐ the landm arkF+ distance an d the an gle b etween a reference direction an d th e landm ark-F + directionÐ w ere suf® cient to support above-chan ce perfor m an ce, though an im pair m ent due to th e geom etric instability of this m inimal array rem ained.
N o vel Locatio n T ests In E xperiment 1, in g roup ® xed± clustered, distance ratio d ecreased in the test w ith a novel po sition (Test 2). In E xperim ent 2, tim e at the goal decreased in the novel position tests, w hereas in the other two m easures the d ecrem ent ap proached statistical sign i® can ce. H ad this effect occur red only or to a larger extent in the varied condition, it w ould have strongly sug gested that the search for food co nstituted a sep ar ate task for each in dividual landm ark position. N avigatio n m ight hav e proceeded by m atching a rem em b ered view to the currently perceived view of the enviro nm ent, w ith backg round cues an d the landm ark both being essential parts of this rem em bered view. H owever, it is surprising to ® n d it in
L A N D M AR K S T A B IL IT Y
339
® xed g roup s, w here ad jacent landm ark positions w ithin the g rid used in training (F igure 1C ) w ere o nly 10 cm ap art. T h ere is no evidence in th ese experim en ts that the rats can detect such a sm all displacem ent of the landm arks relative to w alls that are 15 times as far aw ay. O verall, the results from these tests do not ® t an y of the possible p redictions w e co uld think of an d w e m ust reg ard them as inconclusive.
R elatio nsh ip to O th er S tud ies Several other studies have investigated the in¯ uence of ch an ges of the con ® guratio n of cu es o n spatial b eh aviour. T hese hav e also revealed effects of landm ark stability, but not alw ays u nder co nditions that allow unam biguous interpretation. Som e experiments have used discrete choice m az es w ith unstable landm arks not clearly separated from stable ones. Suzuki, Augerinos, & B lack (1980) foun d that ran dom rearran gem ent of stimuli hung over the ends of a rad ial m aze disrupted perfor m an ce in a test phase, w h ereas rotations of the w hole stimulus ar ray had no effect com p ared to co ntrol trials. K raem er, G ilbert, & In nis (1983, E xperiment 2) train ed rats w ith only ex tra-m az e cues or o nly in tra-m az e cues on alter nate trials. Rearran gem ent of intram az e cu es durin g training w as found to delay acq uisition. Spetch an d H onig (1988) train ed pigeons in an op en ® eld an alogue of an eight-ar m r ad ial m az e. Four landm arks were placed am on g th e eight feeders, an d there were three pictures o n the w alls. Ran dom ly interchan ging landm arks an d pictures between trials impaired p erfor m an ce in a forced-ch oice test. H ow ever, in none of these experiments w as the location of the goal reliably cued by a subset of landm arks independen t of the rem aining cues. Four studies provide evidence that the siz e of th e discrepan c y b etween two sources of sp atial infor m atio n in¯ uences the w eigh t given to th em after training in a stable environm ent. C h eng (1988) trained p igeons to ® nd food hidden under bedding in a square en closure. W hen th e landm ark (a cardboard sheet on the w all) w as m oved in test sessions, the search distribution shifted proportion ally to land m ark d isplacem ent, bu t only up to an asym ptote. T he w eight given to the landm ark rem ain ed constant until a m aximum displacem ent w as reached an d then decreased. E tienne, Teroni, H ur ni, an d Portenier (1 990) let ham sters collect food from the centre of a circular arena, retur ning to their hom e cage to hoard it. A very dim ligh t source h ad been in a stable locatio n during pre-e xperimental ex po sure to the environm ent. W hen the light w as rotated 90 8 in experimental sessions, the return directions indicated that the light had a relatively g reater in¯ uence o n hom ing than did self-generated (prop rioceptive an d vestibular) cues. W hen the discrepan cy w as increased to 180 8 , the in¯ uence of the ligh t cue decreased. C hittka an d G eiger (1 995) found that th e in¯ uence of a row of landm arks o n the ¯ ight direction of foraging h oney bees decreased as th e discrep an cy between the orientatio n of the landm ark array an d the sun com pass increased. In a study by C heng, C ollett, P ickhard, an d Wehn er (1987), honeybees followed the displacem ent of two sm all nearby landm arks rather than searching at the position in dicated by four larger landm arks, of equal retinal size. H owever, a secondary search peak ap peared w hen the discrepan cy becam e too large. T hese ® ndings su ggest that an imals classify cues according to both the accuracy an d the reliability o f the inform ation they can provide. In the case of visual land m arks, large an d often distant features are less likely to be altered or obscu red by ran dom chan ges in the environm ent
340
B IE G L E R A N D M O R R IS
an d are therefore m ore reliable, but on ly allow less accurate nav ig ation than close sm allscale features. T he am ount of discrepan cy then indicates w h ether the m ore accurate inform ation shou ld b e tru sted. In th ese exam ples accuracy an d reliab ility are neg atively co rrelated. If it could be show n that such a constrain t is b uilt into an im als’ navig ational system s, it w ould be interesting to investigate the sam e issue in co nditioning task s, w here we would expect accurac y an d reliability to be independent. Factors other than discrep an cy m ay also hav e an in¯ uence o n the weight given to landm ark s. A lthough nearby local cu es are som etim es prefer red (Bennett, 1993; C heng, 1989 ; C heng et al., 1987; C ollett et al., 1986; Spetch, 1995; Spetch & E dw ards, 1986; Spetch & W ilkie, 1994), this is not alw ays the case. M argules an d G allistel (1988) foun d that rats w h o ignored cues that uniquely speci® ed a locatio n in a rectangular environm ent co uld u se a larger fram e of reference to orient them selves an d distin gu ish otherwise am biguous locations. T he often rep orted preferen ce for extram aze cues (K raem er et al., 1983; O lton & C ollison, 1979; O lton, C ollison, & Werz, 1977; O lton & S am uelso n, 1976 : see O ’ K eefe & N ad el, 1978, an d Restle, 1957, for reviews) m ay re¯ ect a strategy of using large-scale features to d istinguish locatio ns m arked by similar-looking landm arks or landm ark arrays (Brodbeck, 1994; C ollett & K elb er, 1988; Spetch & W ilkie, 1994). T he co n® guration of hills or the shap e of a clearing is less likely to be duplicated than is the presence of a nearby bush. A ny duplicatio n that d oes occur is likely to be at a su f® ciently large scale that dead reckoning can resolve the am b iguity. In ad dition to the presence or absence of discrete go al locations an d th e relative salience of cues, an ad ditional confounding factor is the difference between deter m ining either distance or direction . M an y stu dies reporting a preference for extram az e cues have been conducted in rad ial m azes, w here the m ultiple goals are distin gu ished by direction, w hereas stu dies that foun d g reater w eightin g o f nearby lan dm arks have m ore often used op en ® eld m azes w ith landm ark arr ays that are usually tran slated but rarely rotated. N earby cues allow m ore accurate determ inatio n of distance (C heng, 1990), an d d istant cues are m ore useful for deter m ining direction. O ur ex periments were designed to explore the effect o f instability throughout lear ning. A recent stu dy by G ould -Beierle an d K am il (in press) introd uced instability only in a later phase of training. T hey trained C lark’ s nutcrackers to ® nd food n ear the edge of a square tray. A fter a ® rst p hase, w ith a single landm ark in a stable position, the peak of search followed landm ark shifts parallel to, but not perpen dicular to, the edge of the tray. It also followed the parallel com ponent of a diagonal shift. In an attem pt to m ake all cu es outside the tray unreliable, th e tr ay w as tran slated an d rotated b etween trials during a ® nal phase of training. T he nutcrack ers now followed perpendicular an d diagonal shifts as well as parallel shifts. T his raises th e intriguing possibility that, after m ovem ent of the tray, the birds g ave less w eight to the edge of the tray as a cue for distance but still used cu es associated w ith the tray to deter m in e the correct direction from the lan dm ark. T he results of our experim ents an d of those discussed above are all co nsistent w ith the prop osition that spatial represe ntation s are organ iz ed in two o pposing hierarchies, w ith proximal landm arks being treated as m ore accurate an d distal large-scale features as m ore reliable. W hen a con¯ ict arises, the weight given to each source of infor m ation depends o n the am ount of discrepan cy an d on its a priori reliability.
L A N D M AR K S T A B IL IT Y
341
S u m m a ry In our opinio n, our results are inconsistent w ith the two ap plications of associative lear ning theory to navig ation that we h ave outlined. T he m ost straightforw ard associative m odel predicts better control by a m oving landm ark than by a ® xed o ne in o ur con ditions. T his w as never found. T he co n® gural interpretation predicts w orse control by a m oving landm ark ar ray un der all circum stances, re gardless of landm ark ar ran gem ent. T his is also inconsistent w ith our results. H ow ever, m ore com plex ap plication s of associative lear ning to n avig atio n m ay w ell be able to account for our data. T he vector navig ation m odels of C ollett et al. (1986) an d C heng (1988, 1989) can b e interp reted as two-stage associative processes, ® rst an association of landm ark distan ces an d directions w ith a place, then association of that place w ith rew ard . T h e m odel of M cL aren (1995) seem s to fall into the sam e category. M cN au ghton et al. (1996) have described a m odel of cognitive m ap ping an d dead reckoning that postu lates that landm ark infor m ation is add ed to the m ap by associativ e processes. T he inter actions between processes in such two-stage m odels have not yet been speci® ed in suf® cient detail to derive predictions reg arding ou r experimental parad igm . A lthough our results do not conclusively distinguish between an associative an d a co gn itive m ap ping theory of sp atial learn ing, th ey constrain both types of theory. If the results are interpreted in ter m s of cognitive m ap ping, E xperiment 1 has revealed th at the m ap ping system can construct indepen dent sm all-scale m ap s an d E xp eriment 2 that the inform ation co ntain ed in such a m ap can b e the theoretical m inimum for u nam biguous localizatio n (a distance an d a direction). O n the other han d, if the results are inte rp reted in ter m s of associative lear ning th eory, we have found evidence for a principle w h ich w e call ``landm ark stab ility’ ’ for w hich there is no obvious an alog in the con ditional dom ain. In two differe nt ex periments, m aking a single landm ark the best spatial predictor of rew ard location parad oxically decrea sed control over search by th at landm ark, com pared to a co ndition w here other cues co uld also con tribute to localization an d therefore com pete for control. F urther m ore, spatial relationships other than m ere spatial proximity (the an alog of contiguity) have b een show n to be deter m inan ts of lear nin g. M ore than a simple association of a snap sh ot w ith rew ard or a respon se is req uired to account for spatial lear ning in rodents.
R E FE R E N C E S Bennett, A.T.D. (199 3). Sp atial m emo ry in a fo od storing co rvid I: Near tall landm arks are primarily used. J ourna l of Compa ra tiv e P hysiology A, 173 , 193± 207. Biegler, R., & M or ris, R.G.M . (1993) . L and m ark stability is a prerequ isite for spatial but not discrim ination lear ning. N a ture, 361 , 631 ± 633. Brodbeck, D.R. (1994). M emory for spatial an d local cues: A com parison of a storing an d a non -storin g species. Anima l Lea rning a nd B eha vior, 22, 119 ± 133. Car r, H., & Watson , J.B. (1908). O rientation of th e white rat. J ourna l of Compa ra tive N eurology a nd P sychology, 18, 27± 44. Car tw righ t, B.A., & Collett, T.S. (19 82). How ho ney bees use lan dm ar ks to gu ide th eir return to a food source. N a ture, 295 , 56 0± 564.
342
B IE G L E R A N D M O R R IS
Car tw righ t, B.A., & Colle tt, T.S. (1983). L and m ark lear ning in be es. J ourna l of Compa ra tiv e P hysiology, 151 , 521± 543 . Cheng, K . (198 6). A purely geom etric mod ule in the rat’ s spatial representation . Cognition, 23, 149± 17 7. Cheng, K . (1988 ). S om e psyc hop hys ics of th e pige on ’ s use of lan dm arks. J ourna l of Compa ra tive P hysiology A, 162 , 81 5± 826. Cheng, K . (1989). T he vector sum mod el of pig eon lan dm ark use. J ourna l of Experimenta l P sychology: Anima l B eha vior P rocesses, 15 , 366± 375. Cheng, K . (199 0). M ore psyc hop hys ics of th e pige o n’ s u se o f lan dm ar ks. J ourna l of Compa ra tive P hysiology A, 166 , 85 7± 863. Cheng, K., Collett, T.S., Pickh ard, A., & Wehner, R. (1987 ). Th e use of vis ual lan dmarks by hon ey bees: Bees weigh t landm arks ac cor ding to the ir distan ce from th e goal. J ourna l of Compa ra tiv e P hysiology A, 161 , 469± 475 . Chittka, L., & G eige r, K . (1995) . H on eybee lo ng-distan ce orientatio n in a con trolled enviro nment. Ethology, 99 , 117± 126. Collett, T.S., Car tw righ t, B.A., & Sm ith , B.A. (1986) . Land m ark lear ning an d visuo- spatial m em or ies in ger bils. J ourna l of Compa ra tive P hysiology A, 15 8, 835± 851. Collett, T.S., & Kelber, A. (1988). T he retrieval of vis uo-spatial mem ories by hon eybe es. J ourna l of Compa ra tiv e P hysiology A, 163 , 145± 150. Deutsch, J.A. (1960). The structura l ba sis of beha viour. Cam bridge: Cam bridge U niversity Press. Dickinson , A. (1980) . Contempora ry a nimal lea rning theor y. Cam br idge: Cam bridge Univer sity Press. Diez-Cham izo, V., Ste rio, D., & M ackin tosh , N.J. (1985) . B lockin g and ove rshad ow ing betw ee n intramaze an d extra- maz e cues: a test of th e independence of loc ale an d guidance lear ning. Qua rterly J ourna l of Experimenta l P sychology, 37B , 23 5± 253. Etien ne, A.S., Joris-Lam ber t, S., D ah n-Hurni, C., & Reverdin, B. (1995 ). Op tim izing visual landmarks: two- an d th ree -dimensional minim al lan dscapes. Animal B eha viour, 49, 165± 179. Etien ne, A.S., L am be rt, S.J., Reverdin, B., & Teron i, E. (1993) . L ear ning to recalibrate th e role of dead reckon ing and visual cues in spatial navig atio n. Anima l Lea rning a nd B eha vior, 21, 266± 280 . Etien ne, A.S., Tero ni, E., Hur ni, C., & Porten ier, V. (199 0). Th e effect of a single ligh t cue on hom ing behaviour of th e golde n ham ster. Anima l B eha viour, 39, 17± 41 . Gaffan , D., & Harrison, S. (1989). P lace-m emo ry an d scene-mem or y: effec ts of for nix transection in the rhesus m on key. Experimenta l B ra in R esea rch, 74, 202± 212. Gallistel, C.R. (1990) . The orga niza tion of lea rning. Cam b ridge, M assachussetts, M IT P ress. G oth ard, K .M ., S kag gs, W.E ., & M cNau gh to n, B .L . (19 96). Bindin g o f hippoc am pal CA1 neural activity to m ultiple refer ence fram es in a lan dm ark- based navigati on task. J ourna l of Neuroscience, 16, 823± 835. Gou ld-B eierle, K., & K am il, A.C. (in press). Th e use of loc al an d global cues by Clark’ s nutcr ac ker s. Anima l B eha viour. Hall, G., & Ho ney, R.C. (1989). Co ntextu al effec ts in con dition ing, laten t inhibition an d habitu atio n: Associativ e and retr ieval fu nction s of contextu al cues. J ourna l of Experimenta l P sychology: Animal B eha v ior P rocesses, 15, 23 2± 241. Jacks on , R.L ., & Fritsche, M .B. (1989) . Poten tiatio n an d ov ershad ow ing in pig eon s. Lea rning a nd Motiva tion, 20, 15± 35. K nierim, J.J., Ku drim oto, H.S., & M cN au gh ton , B.L . (1995) . P lac e cells, head direction ce lls and the lear ning of lan dm ark stability. J ourna l of N euroscience, 15, 1648± 1659. K raemer, P.J., G ilber t, M .E., & Innis, N.K. (198 3). Th e in¯ uence of cue typ e an d con ® gu ration upon radial-m az e perfo rmanc e in th e rat. Anima l Lea rning a nd B eha v ior, 11, 373± 380. M cL ar en, I.P.L . (1995). A better beta m od el of navig atio n. B ritish J ourna l of Ma thema tica l a nd S ta tistica l P sychology, 48, 51± 55. M cN au gh ton , B.L., B ar nes, C.A., G errard, J.L., G oth ard, K .M ., Jung, M .W., Kn ierim , J.J., Ku drimoti, H., Qin, Y., Skaggs , W.E., Su ster, M ., & Weave r, K .L. (19 96). D eciphering th e hippocam pal polyglot: the hippocam pus as a path integ ration system . J ourna l of Experimenta l B iology, 199 , 173± 185. M argu les, J., & G allistel, C.R. (1988) . Head ing in th e rat: Deter m ination by environ m ental shap e. Anima l Lea rning a nd B eha vior, 16, 404± 410.
L A N D M AR K S T A B IL IT Y
343
M or ris, R.G.M ., G ar rud, P., Raw lins, J.N.P., & O ’ Ke efe, J. (1982). Place navigatio n im paired in rats with hippocam pal lesio ns. Na ture, 297 , 681± 683. M yers, J.L. (1966). Funda menta ls of experimenta l design. Bos to n: Allyn an d Bac on. O’ Keefe, J. (1991). An allocen tric spatial m od el for th e hippocam pal cogn itive map. Hippoca mpus, 1, 230± 235. O’ Keefe, J., & N ad el, L . (1978). The hippoca mpus a s a cognitive ma p. O xford : Ox ford Univer sity P ress. Olton , D.S., & Collison , C. (1979). Intram az e cues an d ``odor trails’ ’ fail to direct choice beh aviou r on an elevate d m az e. Anima l Lea rning a nd B eha vior, 7, 221± 223. Olton , D.S., Collison, C., & Werz, M .A. (19 77). S patial mem or y and rad ial ar m maz e perfor m an ce of rats. Lea rning a nd M otiva tion, 8, 289± 314. Olton , D.S., & Samu elson, R.J. (1976). Remembr an ce of places passed: S patial mem ory in rats. J ourna l of Experimenta l P sychology: Anima l B eha vior P rocesses, 2, 97± 116. Rescorla, R.A. (1968). P robabil ity of shock in th e presence an d abse nce of CS in fear con dition ing. J ourna l of Compa ra tive a nd P hysiologica l P sychology, 66, 1± 5. Restle, F. (1957). D iscrimination of cues in mazes : a resolution of th e `place-vs -respon se’ questio n. P sychologica l R eview, 64, 217± 228 . Rober ts, W.A. (1979) . Sp atial mem ory in the rat on a hierarchical maz e. Lea rning a nd M otiva tion, 10, 117± 140. Rod rig o, T., Cham izo, V.D., M cL aren, I.P.L ., & M ackin tos h, N.J. (1996) . B locking in the spa tia l doma in. M anu script subm itted for publicatio n. Rudy, J.W., & Su th erland, R.J. (19 95). Con® gu ral assoc iatio n th eory an d th e hippocam pal for mation : An ap praisal an d recon ® gu ration . H ippoca mpus, 5, 375± 389. S petch , M .L . (1995) . O ver shad ow ing in lan dm ar k lear ning: tou ch-screen stu dies with pigeo ns an d human s. J ourna l of Experimenta l P sychology: Anima l B eha vior P rocesses, 21, 166± 181. Sp etch , M .L ., & Edwards, C.E. (1986). Sp atial memory in pigeon s (Columba livia ) in an op en-® eld feed ing environm ent. J ourna l of Compa ra tiv e P sychology, 100 , 266± 278 . Sp etch , M .L ., & H on ig, W.K . (19 88). Characteristics of pigeo n’ s spatial workin g m emor y in an op en® eld tas k. Anima l Lea rning a nd B eha vior, 16 , 123± 131. Sp etch , M .L., & Wilkie, D.M . (1994). P igeon’ s use of lan dmarks presented in digitized im ages. Lea rning a nd M otiva tion, 25 , 245± 275. Sp oon er, R .I.W., T hom pson, A., Hall, J., M or ris, R .G.M ., & S alter, S.H . (1994). T he Atlan tis platfor m: A new design an d fu rth er dev elop m ents of B uresova’ s o n-dem an d platfor m for th e water maz e. Lea rning a nd M emory, 1, 203± 21 1. S uthe rland, R .J., & Rudy, J.W. (1989) . Con ® gu ral associatio n th eor y: Th e role of th e hippocam pal for mation in lear ning, m emory an d am nesia. P sychobiology, 17, 12 6± 144. Su zuki, S., Auger inos, G., & Black , A.H . (1980 ). Stim ulus co ntrol of spatial be havior on th e eigh t-ar m maze in rats. Lea rning a nd M otiva tion, 11 , 1± 18. Tau be, J.S., M uller, R .U., & Ran ck, J.B. (1990). H ead -directio n cells recor ded from th e postsubicu lum in freely m oving rats. II. Effe cts of environm ental man ipulation s. J ourna l of Neuroscience, 10, 436± 44 7. Tolm an , E.C. (19 48). Cogn itive maps in rats an d men. P sychologica l R eview, 55 , 189± 208. Wagn er, A.R ., L ogan , F.A., Hab erlan dt, K., & P rice, T. (1968) . S timulus selectio n in an im al discrimination lear ning. J ourna l of Experimenta l P sychology, 76, 171± 180 . Worden, R . (199 2). Navigatio n by fragm ent ® ttin g: a theory of hippocam pal fu nctio n. Hippoca mpus, 2, 165± 187. M a nuscript received 21 December 1995 Accepted revision received 9 J une 1996
344
B IE G L E R A N D M O R R IS
S tab ilite de s rep eÁ res visu els: do nn e es co m p le m en taires qu an t aÁ le ur roà le dan s l’ap pren tiss ag e sp atial D eu x exp e rien ces on t con cer n e le roà le d e la stab ilite d es repeÁ res visuels da ns l’ app ren tissa ge sp atial. D es ra ts o nt e te en train e s aÁ rech erch er d an s u n vaste ch am p d e la n our ritu re cache e aÁ un e d istan ce con sta nte p ar r ap p or t aÁ u n repeÁ re visu el un iqu e et rad ialem en t sy m e triqu e, ou pa r rap p or t aÁ un en sem ble d e d eux rep eÁ res. L e de gre d e p re d ictivite r elative d u rep eÁ re un iqu e ou d e l’ en sem ble d ’ ind ic es con textu els d e la con d ition d ’ en tr aõà n em en t q ua nt aÁ la localisation de la n ou r ritu re a e te varie , san s qu e so it m odi® e e la p ro babilite co n dition nelle de d isp on ibilite d e n ou rritu re asso cie e aÁ ch acu n d ’ entre eu x. L’ exp e rien ce 1 a u tilise la de so rien tation ve stibu laire p ou r qu e la localisation de la rech erche so it assu re e p ar les ind ices con troà le s pa r l’ exp e rim en tateu r. L es d on n e es on t m on tre qu e si le rep eÁ re visu el u niqu e ou l’ en sem ble d e d eu x rep eÁ res ad jacents so nt les seu ls p re d icteu rs sp atiaux d e la localisation d e la re com p en se, la p re cisio n d e la recherch e est re d u ite p ar rap p ort aÁ un e con d ition ouÁ l’ en sem ble d es repeÁ res visu els et d es ind ices con textu els s on t d es p re d icteur s spatiau x ® ables. M od i® er la stab ilite globale d es rep eÁ r es est reste san s effet ap reÁ s un ap p rentissage con d uit ave c u n en sem ble de d eu x rep eÁ res visu els situ e s aÁ qu elqu e d istan ce l’u n de l’ au tre. Teste s se uls, les ind ices con textu els on t sus cite treÁ s p eu d e rech erch e au x localisatio n s ap p ro p rie es, et l’ am p litud e abso lue d u con troà le d e la rech erch e p ar les rep eÁ res visue ls stab les s’ est ave re e insu f® sa nte po u r asse oir leu r sup e riorite . L e m eilleu r ap p rentissage obtenu ave c u n rep eÁ re visu el stable, et le fait qu e cet effet d e pen d e d e l’ ar ran gem en t geom e triqu e d e ces rep eÁ res, ten d en t aÁ ind iqu er qu e l’ ap pre n tiss age sp atial im pliqu e d es p rincipes ad d ition n els aÁ ceu x d u con ditio n n em ent ass ociatif sim ple. L a secon d e exp e rien ce a exam ine la stabilite d es rep eÁ res visue ls p ar l’ u tilisation d’ u n seu l re peÁ re et d ’ ind ic es d irection n els ® xes, en l’ absen ce d e de so rien tation vestibu laire. C es con ditio n s on t, elles au ssi, re ve le l’avan tage re latif d e la stab ilite d es rep eÁ res; cep en da nt, les an im au x do nt le repeÁ re visu el est d e p lace d ’ u n essai aÁ l’au tre on t m anifeste qu elqu e ten d an ce aÁ l’ ap p ren tissa ge. L es seu ls ind ices con textu els on t exerce u n e in¯ u en ce m inim ale. L a m an ipu lation p aram e triqu e d e la stab ilite d es ind ices spa tiaux ou vre u n e voie n ouv elle da ns l’e tu d e d e l’ ap p ren tissa ge sp atial, et, de ce fait, con tribu e aÁ u ne m eilleu re com p re h en sion des p ro ce ssu s p ar lesqu els la r epre sen tatio n e gocen triqu e d e l’ esp ace p ercË u e st tr an sfo r m e e en re pre sen tatio ns allocen triqu es d u m on d e re el.
Estab ilida d d e los p un tos d e re fe ren cia: N u evo s e stu dios qu e ind ican su im po rtan cia e n el ap ren d izaje espa cia l E n d os experim en tos se estu d io el p osible p ap el d e la estab ilidad d e los m ojo nes o p u n tos d e re feren cia en el apr end izaje espac ial. S e en tren o a u n as r atas en u n re cinto am p lio p ar a qu e encon tr asen com ida escon d ida a u n a d istan cia y d ireccio n estables resp ecto a u n u nico m ojo n rad ialm ente sim e trico o d e u n con jun to d e do s m ojon es. Se vario el gra do en qu e el con jun to de m ojon es y/o las claves d el c on texto d e entren am iento p red ecõ an la situ acio n de l alim ento, sin variar la p ro babilidad co n dicion al d e qu e e ste e stu viese d isp on ible d ad a u n a u otra clave. E n el E xperim en to 1 se em p leo la d eso rientacio n vestibu lar a ® n d e asegu rar el con tro l d e la bu sq ued a p or las claves con tro ladas p or el exp erim en tad or. L os resultad os m ostr aro n qu e cua nd o u n u n ico m ojo n o la agr u pa cio n de d os m ojo nes ady acen tes eran el u n ico p red ictor d e la loc alizacio n d e la re com p en sa se red u cõ a la p recisio n de la b u squ ed a, en com p aracio n con un a con d icio n en la qu e tan to la ag ru p acio n d e m ojo nes com o las claves con textu ales er an pre d ictores espa ciales ® ables. L a variacio n d e la esta bilidad glob al d e los m ojon es no tuv o
L A N D M AR K S T A B IL IT Y
345
efecto algu n o cua nd o el en trena m ien to se realizo con u n con jun to d e do s m ojon es situ ad os a cier ta d istan cia un o d el otro. L as claves con textu ales p resen tad as p ro sep ar ad o fu ero n p oco e® caces p ar a gen erar u na bu squed a cen trad a en los lug ares ap ro p iados, sien d o la m agn itud abso luta d el con tro l so bre la con d ucta d e bu squ ed a ins u® cien te p ara exp licar la su pe riorid ad de los m ojon es estables. E l ap ren d izaje m a s e® caz con m ojon es estables, asõ com o la d ep en den cia d e este efecto resp ecto a la disp osicio n geom e trica de los m ojon es, ind ica qu e las con dicion es de l ap ren d izaje espa cial incluyen nu evos p rin cipios qu e n o so n equ ivalen tes a los del co n dicion am ien to aso ciativo sim p le. E n el E xp erim en to 2 se ana lizo el p apel d e la esta bilidad d e los m ojo nes utilizand o u n so lo m ojo n y claves d ireccion ales ® jas, sin d eso rientacio n vestibu lar. E ste p roced im ien to revelo tam bie n u n a relativa ventaja d e la estab ilidad d e los m ojo nes, au nq ue los an im ales en tr an ad os con u n m ojo n qu e cam biab a d e p osicio n en tre en sayo s m an ifestaro n cierto ap ren d izaje. L as claves c on textu ales p rese ntad as p or sõ so las tu viero n u n a in¯ u encia m  õ n im a. L a m an ipu lacio n p aram e trica d e la estab ilidad d e los m ojone s ofrece u n a nue va m an era d e m od i® car las con d icion es d el ap ren d izaje esp acial, p er m itien d o asõ un a m ejor com p ren sio n del p ro ceso a tr ave s d el cu al las re pre sen tacio n es e goce ntricas d el esp acio p ercibido s on tr an sfo r m ad as en rep resen tac ion es aloce n tric as d el m un d o real.
