The Simulation of Verbal Learning Behavior by Edward A.Feigenbaum(1961)

1. The Simulation of Verbal Learning Behavior by Edward A.Feigenbaum(1961) 지식표현과 추론과정, 1999.5.3. 인지과학 협동과정 99132-503,김수경

2. INTRODUCTION • Purpose of this Report • To describe in detail an information model of elementary human symbolic learning processes • EPAM(Elementary Perceiver and Memorizer)program • precise statement of an information processing theory of verbal learning • plausible mechanism sufficient to account for the rote learning of nonsense syllables • critical evaluation : its ability to explain and predict the phenomena of verbal learning

3. About the class of Information Processing Models • Models of Mental processes, not brain hardware(no physiological or neurological assumptions) • Conception of Brain : an information processor with sense organs as input channels/ effector organs as output devices/ internal programs for testing, comparing, analyzing, rearranging and storing information • The central processing mechanism:assumed to be serial • As a basic unit : use the information symbol • Deterministic, not probabilistic.

4. The Basic Experiment(1) • Rote memorization of nonsense syllables • three-letter words which have no ordinary English meaning (e.g. XUM) 1)paired-associates presentation : stimulus item-response item, changed order on successive trials 2)serial-anticipation presentation : serial list, same order on successive trials 3)etc.

5. The Basic Experiment(2) • The typical behavioral output of a subject • (failures to respond to a stimulus)>(overt errors) • overt errors : generally due to confusion between similar stimuli/similar responses • oscillation at the threshold of recall • retroactive inhibition • The more similar stimulus syllables are, the more trials it takes to learn.

6.  The Information Processing Model:The Processes and structures of EPAM 1.Overview : Performance and Learning 1) Performance system - producing responses to stimulus items 2) Learning system - learning to discriminate and associate items • Discrimination learning process : builds discriminations by growing the net of tests and branches • Association process : builds associations by storing response cues with stimulus images

7. 2. Input to EPAM:Internal Representation of External Data • The internal representation of a nonsense syllable is a list of lists of bits, each sublist of bits being highly redundant code for a letter of the syllable.

8. 3. Discriminating and Memorizing:Growing Trees of Images(1) • Discrimination Net  primary information structure. • At the terminals of the net: • testsstored programs which examine characteristics of an input code and signal branch left or branch right • At the nodes of the net : • imagelist of symbols, a partial or total copy of an input code

9. 3. Discriminating and Memorizing:Growing Trees of Images(2) • Net Interpreter sorts an input code in the net and produces the image list associated with that input code. • process: 1 Finds the test in the topmost node of the tree and executes this program. 2 The resulting signal tells it to branch left or branch right to find the succeeding test. 3 It executes this, tests its branches again, and repeats the cycle until a terminal is found. 4 The name of the image list is produced, and the process terminates.

10. 3. Discriminating and Memorizing:Growing Trees of Images(3) • Discrimination Learning grows the net. • Initially : No discrimination net, but only a set of simple processes for growing nets and storing new images at the terminal. • Fig. 3,4 (test1, 2 : 1st letter test) • Noticing Order  prescribes a letter-scanning sequence for the matching process. So, which letter next? • Initialized in EPAM as observed from the subjects: • 1st letter > 3rd letter > 2nd letter • Fig.5 (test 4 : 3rd letter test)

11. 4. Associating Images:Retrieval using cues(1) • An EPAM internal associationis built by storing with the stimulus image information sufficient to retrieve the response image from the net at the moment of association. That is  storing {stimulus image + thecue} • How much information is sufficient for the cue? • determined by trial and error • If a test seeks information which the cue does not contain, the interpreter branches left or right randomly(with equal probabilities). • The selection of wrong response is immediately detectable because the response input code is available!

12. 4. Associating Images:Retrieval using cues(2) • Two important Possibilities! 1)A really inadequate cue to guarantee retrieval of the response image may give the correct response image selection during association by ‘luck’. 2)The cue is sufficient only at one particular time. • After the association building process is thoroughly done, those association links are ‘dated’ and ever vulnerable to interruption by further learning. • If the responses may be ‘forgotten’ temporarily because the info. has been lost temporarily in a growing network, the trouble can easily be remedied by adding additional response info. to the cue. • If not, they may be lost in the net.

13. 5. Responding: Internal and External • Response Retrieval • stimulus item  find the cue in the image list  retrieve another image list containing the proposed response image • Response Generation • by a fixed discrimination net(decoding net) which transforms letter codes of internal images into output form.

14. 6. The Organization ofthe Learning Task • As EPAM is a serial machine, the individual items must be dealt with in some sequence. This sequence is not arbitrarily prescribed. It is a result of higher order executive processes whose function is to control EPAM’s focus of attention. • Full exposition of these macroprocesses is available in a paper by Feigenbaum and Simon(1962)

15. Stating the Model precisely:Computer Program for EPAM • The EPAM model has been realized s a program in IPL-V(Newell et al. ,1961) and is currently being run both on the Berkely 7090 and the RAND 7090.

16. Empirical Exploration with EPAM(1) • “Experimenter” program - EPAM as a subject • carrying out the simulations of (basic paired-associate / serial-anticipation experiments / etc.) in the rote learning of nonsense syllables. • A. Stimulus and Response Generalization • Generalization is common to the behavior of all subjects, and is found in the behavior of EPAM. • Xs : Yr X’s : ? r / ? r = Yr • Xs : Yr  ? s : Y’r / ? s = Xs

17. Empirical Exploration with EPAM(2) • B. Oscillation and Retroactive Inhibition • Consequences of simple mechanisms for discrimination, discrimination learning, and association • Interference by later learning may lead to response failure. • Oscillation - new items within the single list being learned • Retroactive Inhibition - new items in the second list being learned in the same discrimination net. • C. Forgetting • EPAM provides a mechanism for explaining the forgetting phenomenon in the absence of any information loss.

18. Conclusion: A Look ahead • Further Studies(Feigenbaum & H.A.Simon, 1961,1962) • “sense modes” corresponding to  visual input : written output / auditory : oral / muscular : muscular • To each mode corresponds a perceptual input coding scheme, and a discrimination net. • Associations across the nets, as well as the familiar associations within the nets, are now possible. • Internal transformation between representations in different modes are possible.  mind’s ear, mind’s eye • Q. Does the Association take place between symbols or between tokens of these symbols?