Computation in the Learning System of Cephalopods

9 Pages·1991·3.6 MB·English

by  J Z Young

Reference: Bio/. Bull. 180: 200-208. (April,
Computation in the Learning System of Cephalopods
YOUNG
J. Z.
Department ofExperimentalPsychology, University ofOxford,
South Parks Road, Oxford OX1 3UD, UnitedKingdom
Abstract. The memory mechanisms of cephalopods to identify objects or situations, say a rock or a tree or a
consist ofa series ofmatrices ofintersecting axes, which fish, but learn theclassification ofparticularsetsofstimuli
find associations between the signals ofinput events and byvirtueofthelarge numberoftheirneurons. It hasbeen
their consequences. The tactile memory is distributed claimedthat thisinvolvessimply "thespontaneousemer-
amongeight such matrices, and there isalso some suboe- gence ofnew computational capabilities from the collec-
sophageal learning capacity. The visual memory lies in tive behaviouroflarge numbers ofsimple processingele-
the optic lobe and four matrices, with some re-exciting ments" (Hopfield. 1982). Biologists will probably suspect
pathways. In both systems, damage to any part reduces that agenetic component is involved in the organization.
proportionally the effectiveness of the whole memory. Cephalopodshavesuchnervoussystemswith numerous
These matrices are somewhat like those in mammals, for neurons, and there is sufficient information about their
instance those in the hippocampus. arrangement to suggest how they function. Formerly, I
The first matrix in both visual and tactile systems re- have emphasized that the circuits in their brains must
ceives signals ofvision and taste, and its output serves to allow for the outputs from feature detectors to produce
increase the tendency to attack or to take with the arms. alternative effects after learning (Fig. 1). I proposed that
The second matrix provides forthe correlation ofgroups the feature detectors must become restricted during
ofsignals on its neurons, which pass signals to the third learning to establish units ofmemory or mnemons. This
matrix. Here large cells find clusters in the sets ofsignals. view is correct in that it emphasizes the possibility ofal-
Their output re-excites those ofthe first lobe, unless pain ternative outputs from feature detectors, but it is much
occurs. In that case, thisset ofcellsprovidesa recordthat too restrictive. Emphasis on units obscures the essential
ensures retreat. fact that these are systems with numerous parallel, inter-
There is experimental evidence that these distributed acting channels. It is now evident that the various lobes
memory systems allow forthe identification ofcategories ofthe brain provide sequences ofmatrices ofintersecting
of visual and tactile inputs, for generalization, and for axes, with feedback. Theyenabletheidentification ofcat-
decisiononappropriatebehaviorinthelightofexperience. egories of input and storage of records of the probable
The evidence suggests that learning in cephalopods is value ofeach, in the form ofbias to particulardirections
not localizedto certain layers or "grandmothercells" but ofaction to each set ofinput signals.
is distributed with high redundance in serial networks, Theprinciple ofthe matricesistoprovide forselection
with recurrent circuits. ofpaths that are used and inhibition ofthose that are not
used. This is accomplished by various means that allow
Introduction
interaction between pathways. One ofthe best analyzed
Responding appropriately in a complex environment systemsisin the mammalian hippocampus(Fig. 2)(Rolls,
depends upon thecategorization ofeventsand adecision 1990). Inacompetitivelearningmatrixsuchasthedentate
ofwhat to do. Animals with good brains have the ability gyrus, "different input patterns on the horizontal axons
tolearn theuseful responsestoparticulareventsthatthey will tend to activate different output neurons. The ten-
encounter. They may not be born with receptors tuned dency for each pattern to select different neurons can be
enhanced by providing inhibition between the output
Received 7 August 1990; accepted 18 January 1991. neurons. . . . Synaptic modification then occurs . . . and
200
COMPUTATION IN CEPHALOPOD LEARNING SYSTEM 201
Attack becomes linked by strengthened synapses with any other
Basal Taste Medsupfr conjunctively activated cell or set ofcells" (Rolls, 1990).
Feature lobes Vertical Asaresult, duringrecall "presentation ofeven part ofthe
detectors original pattern . . . comes to elicit the firing of the
whole set of cells that were originally conjunctively
activated."
In the hippocampus, this result is achieved by the col-
Subvertical laterals ofthe CA3 cells, which reactivate thedendrites of
their own and a large number ofother CA, cells. It may
bethat in cephalopodsasimilareffect isachievedbypass-
ing the signals through a series oflobes, each servingas a
Magno matrix whose output may be returned to a previous
cellular
memberoftheseries(Fig. 3). The functioningofanysuch
Retreat matrix system depends on the particular anatomical ar-
Figure 1. Schemetoshowalternativepathwaysfromthe visual fea- rangements and details of synaptic functioning and its
ture detectors ofan octopus. There are output pathways for attack or alteration with use. We do not know enough about such
sreytsrteeamt.. HAertehirpdartpiactuhlawraypatlteeardnsstooftvhiesuaflousrigmnaaltsriacreescoofmbtihneevderwtiictahltlhoobsee factors in cephalopods to beableto specify precisely how
oftastetoincreasethefuturetendencytoattack,orwithsignalsofpain they operate. However, the system is simple enough to
to reduce it. allow usto followthewhole sequencethrough these lobes,
from sense organs to motor output, and at least to spec-
ulate about its functioning.
the response of the system as a categorizer climbs over Matrix systems ofthis sort have the properties that we
repeated iterations" (Rolls, 1990). associate with complex animal behavior. Theyensurethat
In autocorrelation networks, such as the CAi cells there isgeneralization: presentation ofeven a part ofthe
(Fig. 2), the preferred pathways are reinforced by mutu- original figure, or one like it, activates the firing of the
ally strengthening each other. In this case, the cells of whole set of cells that were conjunctively activated
the matrix have collaterals that feed back to their own ("completion""). Moreover,thesystemcontinuestooperate
inputs. These recurrent synapses follow the Hebb rule even if some ofthe cells fail to operate or are removed
so that "any strongly activated cell or set of cells ("fault tolerance"). We can show that these essential
FDgc CA1
Neocortex Parahigpyproucsarripal
to septum Parahippccampal
mammillary bodies gyrus
1 1
Neocortex
Figure 2. Schematic representation ofthescheme ofmatricesandconnectionswithin the primate hip-
pocampusandwiththeneocortex.Thecompetitivematrixinthedentategyrusleadstoanautoassociation
matrix formed bytheCA3cells, which in turn lead toacompetitive matrix on theCA, cells(From Rolls,
1990, with permission).
YOUNG
202 J. Z.
visual discrimination. Many ofthe experiments are done
after the whole supraoesophageal lobe has been bisected.
The arms ofthe two sides then learn independently and
median
can even be trained in opposite directions.
inferior frontal
>2 Afferent fibersfrom thearmsandalsotaste fibersfrom
sub trauma the lips cross the dendrite systems ofthe first tactile lobe,
fro3nta,lv the lateral inferior frontal (Fig. 6, 7). The axons of the
cells of this first lobe pass partly to the fourth lobe, the
trauma posterior buccal, and partly to the lateral superiorfrontal
and sotothevertical lobesystem(below). The fibersfrom
the arms and lips then pass on to the second matrix, in
the median inferior frontal, where they interweave and
cross the trunks of a large sample of the 10" cells, this
allows maximum opportunity for any cell ofthe lobe to
receivesignals from a varietyofinput fibers(Fig. 7). These
'sucker median inferior frontal cells then send their axons to the
third matrix, thesubfrontal lobe, whichcontains relatively
Figure3. Diagramoftheconnectionsofthetactilememorysystem few large cells with twisted trunks and many bushy den-
ofOclopus.Thesuccessivematricesarelabelled 1 to8.Inaddition,there drites and, in addition, a great number (5 X 10b) ofvery
issomelearningcapacity in thesuboesophageal centers. small amacrine cells. The subfrontal also receives nu-
merous fibers from below, presumed to signal trauma.
The large subfrontal cells send their axons to the fourth
properties ofmemory systems are present in the nervous lobe,theposteriorbuccal, fromwhich, inturn, largeaxons
system ofoctopuses. pass directly to the arms and cause them either to draw
The nervous system of an octopus provides for cate- in or reject the object touched (Budelmann and Young,
gorizationandsettingupofmemories both forvisionand 1985). These cells must be oftwo sets, some causing the
touch. There are two distinct systems, each made up of object touched to be drawn in, the others to reject it.
fourlobes, withelementswhose arrangement can nowbe
seen to constitute sequences ofmatrices(Fig. 3, 4). Itwas
subv med^upfr medinf.fr
thought at one time that these were independent visual
andtactilesystems, butitisnowclearthatintactilelearn-
ing all eight lobes are involved (Young, 1983). This,
therefore, constitutes a remarkable example ofa distrib-
uted memory system using a series of networks. This
model is rather similarto that suggested by Wells(1978).
It does not depend on detailed preformed connections
and so avoids problems ofcomplex morphogenesis.
The Chemo-Tactile Memory System
Octopusesreadilyrecognizedifferencesinthechemical
natureandtextureofobjectsbytouch, althoughtheycan-
not discriminate between shapes (Wells, 1978). The re-
ceptors for touch are in the rims ofthe suckers (see Gra-
ziadei in Young, 1971). Their axons proceed through ant has sub.fr sup.buc
synapsesinthearm,butnodetailsareknownofthecoding Figure 4. Sagittal section ofthe supraoesophageal lobe ofOctopus
signals that are sent to the brain. mlgarisstainedwithCajal'ssilvermethod.Abbreviationsforallfigures:
In ourexperiments. Wellsand I train octopusesto dis- ant.bas.,anteriorbasal;b.med.,medianbasal;buc.p.,posteriorbuccal;
tinguish between plastic balls, either smooth or with up cer. br. con., cerebrobrachial connective;cer. tr.. cerebral tract; lat. int.
to thirteen incised rings (Wells, 1978; Young, 1983). We fr, lateral inferior frontal; lat. sup. fr., lateral superior frontal; mag.,
train an animal by giving it food when it takes one ball msuapgenroicoerllfurloanrt;alm;eodp..,inofp.tifrc.;.mpeedd.i,apneidnufnecrlieor;fproosntt,alb;umc.e,d.possutpe.rifro.r,mbeucdciaaln;
(say a smooth one) and by giving it no reward ora small plex., plexiform layer, prec., precommissural: pv., palliovisceral; ret.,
electricshock fortakingtheother(rough) (Fig. 5). In crit- retina; subfr.. subfrontal; sup. buc.. superiorbuccal; subv.. subvertical;
ical experiments, the optic nervesarecut toavoid possible vert., vertical.
COMPUTATION IN CEPHALOPOD LEARNING SYSTEM 203
4 tests tests -,
204 J. Z. YOUNG
topuses could probably make some discrimination even
between a difference ofone ring.
By such tests we can compare the discrimination by
animals after various lesions. Without the median inferior
frontal there is still discrimination, but it is much less
accurate than in control animals (Fig. 9). Removal ofthe
vertical lobe also reduces accuracy, although to a lesser
extent. Clearly each of the lobes through which the in-
formation passes adds something to the effectiveness of
the representationsthat are formed,aswould beexpected
from a system ofmatrices.
Animalswithout vertical lobesshowerrorslargelywhen
they take the negative ball, showing again that this lobe
serves to increase the effectiveness ofshocks. In normal
octopuses, learningispossibleeven ifrewardsaredelayed
for upto 30 safterthe ball has been removed. In animals
without vertical lobes, such delay is no longer possible
(Wells and Young, 1968). The re-excitation within the
vertical lobe system serves to maintain the necessary ex-
Figure7. TransversesectionoftheinferiorfrontalsystemofOctopus citability for a Hebb type oflearning.
cvri.i/tgra.r.iscesrteabirnaeldtwriactth(Cfarjaolm'ssusbivleverrtimcealthloodb.e):Abpb.retrvaicattioofnpsroasbaFbilgyurpeai4n. We have used this technique to make a large number
fibersfrom hind end ofbodv. ofexperiments, leaving one side as a control. With this
technique it is possible to remove the subfrontal lobe,
which cannot be approached laterally. The effect is to
a complex plexus to the vertical lobe. Here there are rel- produce a complete inability to learn on that side: there
atively few large cells (65,000), with complex dendrites, isastrongandirreversible preferencefortherougherballs
exactly like those ofthe subfrontal, and no less than 25 (Young, 1983). The lobeevidently hassome specificeffect
million amacrine cells. on the coding and discrimination process.
The largecellssendtheiraxonsdowntothesubvertical After cutting the cerebro-brachial tract (Fig. 6), the
lobe and so to the posterior buccal, but also back to the whole influenceofthe inferiorfrontal and vertical systems
lateral superior frontal(Fig. 3). Thiscircuitevidently plays isremoved. Nevertheless, there isstill aslight capacity for
some part in re-enforcing the conjunctions, possibly by learned discrimination (Fig. 10), which must lie in the
maintaining particular patterns by re-excitation. suboesophageal ganglia, or in the arms themselves. This
residual learning is difficult to study. Animals with these
The Distributed Tactile Learning System
The system for touch learning thus includes no less CONTROLHALF-BRAINS
than eight distinct lobes with matrix structure (Fig. 3).
Therelative partsplayed by the various lobeswasstudied
overa numberofyearsin a large numberofanimalswith
divided brains. Lesions were made on one side, and the
other was left as a control. In many ofthe experiments,
discrimination was between completely smooth balls (0
rings) and those with 13 incisions. The sequence oftrain-
ing for 129 normal sides is shown in Figure 5.
A useful measure of the extent and reliability of dis-
crimination is to give a series of24 extinction tests with
balls ofdiffering roughness, shown at short intervals (1-
3 min) without any reward. Such tests are arduous to
give, but theyshowthat habituation proceeds more slowly 4 6 9 13
in proportion to similarity of each ball to the one for numberofrings
which reward was previously given (Fig. 8). The capacity Figure 8. Tests after training in three different directions. Means
for discrimination was also tested by using more nearly andstandarderrorsofratiosoftakesofeachballtototal takes,with 24
similar balls, with 4 and 7 rings. With long training, oc- trialswith each ball. Thebarsshowstandarderrors.
COMPUTATION IN CEPHALOPOD LEARNING SYSTEM 205
~\NV
"0469 0469 0469
13 13 13
numbernfrings
Figure 9. Tests after training. Comparison ofresults (a): 11 animals after cutting the tract between
superior frontal and vertical lobes on one side (NSFV) and removing the vertical lobe (NV) on the other
side;36controlsforcomparison(b): 36animalsafterremovingthemedianinteriorfrontal(NMIF);controls
andNV added forcomparison. In a& b, all trainingwaswith smooth ballspositiveand 13 rings negative.
In (c) balls with six rings were positive and smooth balls negative. The figure shows the results for nine
animalswith the median inferior frontal removed on one side three animalswith no vertical lobe on one
side added for comparison. Note that here (and Fig. 8) the ballswith 9 and 13 rings were seldom taken,
although they had not been associated with shock.
large operations do not feed well and tend to hold pieces component ofthe visual learning system. The existence
offoodand otherobjectsclosetothemouth. Nevertheless, ofthis double capacity is a striking demonstration ofthe
the differences in numbers of takes of the balls during powerofsuchadistributed matrixsystem tostoreavariety
tests are significant and show that some learning has oc- ofinputs. Itwill be very interestingto investigate whether
curred. individual cells ofthe vertical lobe system play a part in
Some measure oftheaccuracy ofthe memory afterthe both systems. In a study ofcombined visual and tactile
various lesions isgiven bythedifference between the takes training, no mutual interactions between the two modal-
ofthe rough andsmooth ballsin thetests(Table I). Using ities was seen (Allen ct at., 1986).
the difference in controls as a standard (100%), we can The study of visual memory is made difficult by the
judge that animals without vertical lobes are rather less complexity ofthe connections in the optic lobes, which
than halfas efficient, without the median inferior frontal are not fully understood. Cells with large tangential den-
are one third as efficient, and that the suboesophageal drites in the plexiform layer probably act as feature de-
contribution isaboutonesixth. Damagetothesubfrontal tectors (Fig. 1 1). Their axons form columns proceeding
produces a perverse effect. to the center ofthe lobe, where they interact in an inter-
The Visual Memory System weaving matrix ofcells and fibers. Second or third order
The vertical lobe system, which plays a part in tactile visual neurons then send axons to the central nervous
learning, forms, with the optic lobes, the main and only
control Table I
control trained13-t-_,
trnai=n3e6d0-*- n~L6 Meantakesinfinaltestsaftervariouslesions
NCB.13+
I 20
04
6 9 13
numberofrings
Figure 10. Tests after training. Comparison ofcontrol sides with
thosewith thecerebrobrachial tractscut(NCB).
206 J. Z. YOUNG
vert interruption of the circuit, an octopus will continue to
med supfr
make attacks, even at crabs, in spite ofreceiving shocks,
Iatsupfr unless these shocks aregiven at intervals offive minutes
or less. "The setting-up ofa memory representing asso-
ciation of a given situation with a shock is therefore a
buc property ofthe optic and basal lobes but persistence of
the representation depends upon the presence ofthe ver-
tical lobe" (Boycott and Young, 1955).
Many other experiments have confirmed that learning
ofvisual discrimination is impaired by lesionsofthe ver-
tical lobe system (Young, 1961. 1965). Ifpart ofthe ver-
tical lobe is removed, the accuracy ofthe memory ispro-
portionately reduced. This "graceful degradation" is a
property to be expected in such a distributed system. In-
cidentally, Boycott and I were able to show that the same
is true ofthe optic lobes. Memories are retained after re-
moval ofat least 50% ofthe lobe or after making lesions
op. in several places with a cataract knife.
Discussion
ret V The two memory systems ofan octopus thus work on
precisely similar principles. The input signals are passed
Figure 11. Diagram of the connections of the visual and tactile through a series ofmatrices ofintersecting axes allowing
learning system ofan octopus. 1, retina; 2, second order visual cells for particular groupings of signals to interact and to be
(feature detectors): 3, centrifugal cells; 4, amacrine cells; 5, tangential directedtothepathwaysforattackorretreat. Thesystems
cells; 6, optic-peduncle; 7. optic-magnocellular; 8, optic-anterior basal; are tuned to produce exploratory investigation ofnovel
9.optic-median basal; 10.optic-lateralsuperiorfrontal; 11, optic-median situations. Ifthe results are favorable, the particular set
superior frontal; 12. taste fibers-lateral inferior frontal; 13, taste fibers
median inferiorfrontal; 14, lateralinferiorfrontal-lateralsuperiorfrontal; ofconnections in the lateral frontal lobes are re-enforced
15. lateral inferiorfrontal-mediansuperiorfrontal; 16, median superior by signals oftaste, and this set later produces more rapid
frontal-vertical; 17, vertical-lateral superiorfrontal; 18, lateral superior attacksortakes by thearms. The inputsaregiven further
frontal-subvertical; 19,painfibers-vertical:20,painfibers-subfrontal;21, opportunity forinteraction in the matrices ofthe median
vertical-subvertical:22,subvertical-optic;23,subvertical-precommissural: frontal lobes. In the vertical and subfrontal lobes, partic-
24, precommissural-palliovisceral; 25, precommissural-magnocellular;
26, subvertical-posteriorbuccal; 27, chemo-tactile libers-lateral inferior ularsets are then concentrated into rather few large cells.
frontal; 28,chemo-tactile fibers-median inferiorfrontal; 29, median in- The recurrent output from these to the lateral superior
ferior frontal-subfrontal; 30. subfrontal-postenor buccal; 31, lateral in- frontal lobe presumably re-enforces the tendency to pos-
feriorfrontal-posteriorbuccal; 32.motorfibersfromposteriorbuccal to itive action, unless pain occurs. In that case, the other
arms.
outputs from these large cells ofthe vertical orsubfrontal
prevent further investigation ofthat configuration ofin-
system (Fig. 1 1). Some pass to the magnocellular lobe, puts. The numerous amacrine cells in these lobes are ev-
and this is probably a pathway for rapid escape reactions. idently concerned with establishing the conjunction be-
Other fibers pass to the peduncle and basal lobes, which tween particular sets ofinput signalsand the pathwaysof
together regulate movement, including attack. A third retreat.
pathway leadstothesuperiorfrontal, andsotothevertical The organization ofthese lobes, and the effects ofre-
lobe, and is responsible for learned behavior. moving them, suggests that learning in these animals is
The system is organized exactly as we have seen for not localized to one or two "hidden layers" or to a few
tactile learning. In the lateral superior frontal, the visual essential "grandmothercells,"butisdistributedwith high
fibers interact with those oftaste, and this is a pathway redundancy in a series ofmatrices networks, with recur-
that promotes attack. After removing this lobe from one rent circuitry, up to a late stage where funneling to a few
side, an octopus will no longer attack when that eye has cells occurs.
been used toseeacrab, forinstance, atadistance (Boycott We can gain some insight into how this process has
and Young, 1955). The median superior frontal and ver- evolved by considering the differences between octopods
tical lobes provide a system that prevents visual attack and decapods. Cuttlefishes and squids have a system of
when trauma occurs. After removal ofthese lobes or an matrices for visual learning similar to that of octopods
COMPUTATION IN CEPHALOPOD LEARNING SYSTEM 207
B
vert
Figure13. Anearlysuggestionofre-excitationasthebasisofmemory.
Diagram ofScpiu toshow howcirculation between the lateral superior
frontal (lat. sup. fr.) and vertical (vert.) might facilitate the firing ofa
motorneuron(M)byconjunctiveexcitationfromtheopticlobe(O)and
tastefibers(V)(Young.
croneurons have processes restricted to a limited field,
wherethey mayservetorepressactivity inthelargercells.
In this context, it is especially interesting that we found
Figure 12. Sagittal section ofthe brain ofSepia. Note that there is some simple capacity for tactile memory in the suboe-
nomedianinteriorfrontalorsubfrontal.Thesuperiorfrontalhasamatrix sophageal lobes.
structure like that ofOctopus. The vertical lobe has a rather different It is suggested that the amacrine cells ofthe subfrontal
structure. Cajal silverstain, b.a., anteriorbasal; h. med., median basal; and vertical lobes ofoctopods have evolved from inhib-
fr. ]..interiorfrontal;fr.s.,superiorfrontal;prec.,precommissural;subv., itorsofthereciprocal feedingreflexes.Theinferiorfrontal
subvertical; v.. vertical.
and vertical lobe systems are backward extensions ofthe
superior buccal lobes (see Fig. 4). The matrices that are
(Fig. 12). In an early experiment it was shown that inter- responsible forlearninghaveevolvedbythe modification
ruption of the vertical lobe circuit damages the visual ofthesesimplerreflexcenters. Theincomingafferentfibers
memory system of Sepia (Young, 1938; Sanders and have become marshalled into rows crossing the axons of
Young. 1940). This was the first suggestion that the cir- cells of the lobe, allowing the formation ofconjunctive
culation ofimpulses around a circuit provides a basis for response to the incoming patterns ofstimuli. The ama-
memory (Fig. 13). There has been little further progress crine cells became collected together in distinct lobes,
because the experiments are more difficult than in octo- servingto prolongtheeffects, perhapsespecially ofinhib-
pods. In decapods, the inferior frontal system is much
simpler than in octopods: there is no median inferior
frontal or subfrontal lobe. These animals detect the prey
visually and often seize by ejection of the tentacles. It
seems likely that they have, at best, only a small capacity
for learned tactile discrimination; the operations of ma-
nipulatingand eatingthe preyarecomplex, butare prob-
ably largely reflex. Nevertheless, there must be a mecha-
nism for release ofany object that gives pain when it is
held. Probably all reflex systems have some method of
inhibition, especially ifthey involve musclesactingrecip-
rocally, such as flexorsand extensorsin mammals, where
the inhibition is produced by Golgi type II cells in the
spinalcord. In cephalopods, reciprocal inhibition isprob-
ably produced by the smaller amacrine neurons that are
common amongthe larger motorneurons ofthe superior Figure14. DrawingofasinglelargecellfromthepedallobeofLoligo.
buccal and suboesophageal centers (Fig. 14). These mi- accompanied by verysmall cellswith branches in the neighborhood.
YOUNG
208 J. Z.
itory inputs. The details are far from clear, but this pro- Budelmann, B.-l., andJ. Z. Young. 1985. Central nervouspathways
vides a possible scenario for the evolution of memory for the arms and mantle ofOctopus. Phil. Trans. R Soc. B 310:
109-122.
mechanisms, at least among cephalopods. Hopfield.J.J. 1982. Neuralnetworksandphysicalsystemswithemer-
gent collective computational abilities. Proc Natl. Acad. Sci. USA
Acknowledgments 79:2554-2558.
Rolls, E. T. 1990. The representation and storage ofinformation in
The work reported here has been helped by many col- neuralnetworksintheprimatecerebralcortexandhippocampus. In
leagues. I am especiallygrateful to Brian Boycott, Martin TheComputingNeuron. R.Durbin.C. Miall,andG. Mitchison,eds.
Wells, Marion Nixon, and Pamela Stephens. TheStazione Addison-Wesley, Wokingham.
Zoologica at Naples providedexcellent conditions forex- Sanders. F. K., and J. Z. Young. 1940. Learningand otherfunctions
periment over many years. Recently, work has been Wellosf,thMe.hiJ.gh1e9r7n8e.rvoOucstocpeunstr.esInofPhSeypsiiao.loJg.yNeaunrdopBheyhsaiovli.ou3r:o5f01a-n52A6d.-
helpedbygrantsfromTheWellcomeTrustandUniversity vancedInvertebrate. Chapman and Hall. London.
College, London. I am grateful to ProfessorL. Weiskrantz, Wells, M. J., and J. Z. Young. 1968. Learningwith delayed rewards
F.R.S., foraccommodation inthePsychology Department in Octopus. Z I'ergl. Physiol. 61: 103-128.
at Oxford and to my wife Raye for her secretarial help Young, J. Z. 1938. The evolution ofthe nervous system and ofthe
and typing. Dr. M. J. Wells kindly commented on an erseslaatyisonpsrheispeonfteodrgtaoniEs.mS.anGdooednrviircohn,meGn.t.R.Ppd.e1B7e9e-r2.04ed.inCElvaorluetnidoonn.
early draft ofthe paper. Press, Oxford.
Young,J. Z. 1961. Learninganddiscrimination in theoctopus. Biol
Literature Cited Rev 36: 32-96.
Young,J.Z.1965. Theorganizationofamemorysystem.TheCrooman
Allen, A., J. Michels, and J. Z. Young. 1986. Possible interactions Lecture. Proc. R Soc. B 163: 285-320.
betweenvisualandtactilememoriesinoctopus.Mar. Behav.Physiol. Young, J. Z. 1971. TheAnatomy oftheNervous System ofOctopus
12:81-97. vulgaris. Clarendon Press. Oxford.
Boycott, B. B., and J. Z. Young. 1955. A memory system in Octopus Young,J.Z. 1983. ThedistributedtactilememorysystemofOctopus.
vulgarisLamark. Prof. R Soc. B 143:449-480. Proc. R Soc. B 218: 135-176.