Paired Associates Learning

The subject's task in this experiment is to form the association between a stimulus item and a response. Available presentation modes are the method of anticipation with test and study within each single trial, separate study and test blocks or sutdy and test blocks only. Replacement of unlearned response items is possible also.

Program Features

Presentation Method

The method used by default is called method of anticipation. On each trial we present a single CVC on the screen. The subject tries to give the correct response if it is known. If the correct answer is not yet known, the subject may guess and give any of the two possible responses. After the response the correct answer is presented on the screen. This enables the subject to learn the correct correspondence. Each subject has to learn a certain number of items. These items are presented one after the other until the whole list of items has been cycled through. The presentation order of the items is random. If the subject has given only correct responses for a certain number of consecutive presentations of the whole list of items, we assume that all pairs have been learned and stop the experiment.

If the flag T is set then study and test trials are separated. This means that first the complete list of items and responses is shown for study and then all items are tested. In this case there should be no feedback which is prevented by clearing the flag F. Special messages may be given to introduce the different phases: studymsg, testmsg, and testendmsg. It is also possible to completely suppress the test trials by setting flag S.

A complete presentation of the item list is considered to be a block. Each block has at least one parameter, its state. The built in variable blockstate holds the current state of the block. The block's state is the number of consecutive error free blocks until now. So the initial value of blockstate is 0. The blockstate remains 0 while any false response is given during any trial. After each trial the block controlling routine looks at the response. False responses are logically or'ed into a flag such that the flag is set if there is any false response during a presentation of the item list. The blockstate is reset to 0 if the flag is set at the end of the block. If the flag is not set the blockstate is incremented by one. The integer parameter learned defines the learning criterion. This is the number of error free consecutive presentations of the item list which is required before the experiment is stopped.

Items

The stimulus item is given by the string item. The correct response ist contained in link. If the flag M is not set then it is expected that link is a single character string. If M is set then multiple character string responses are accepted and have to be delimited by the RETURN-key. Note that in this case the parameter alength has to give the number of characters in the response string. If the flag W is set then items with false responses are replaced by new items for the next block. The strings for items and responses are taken from the file whose name is given by itemfile. This file is expected to contain two strings per line: the first one is the stimulus string and the second one is the response string. Thus if the subject gives a false response in one block the next block will no longer contain the respective stimulus but will contain a new item instead. This method has been used by Rock to test the effect of repeted presentations.

Simulations

The parameter model may be used to simulate data according to a learning model. If model is 0 data are collected from a subject. If model is nonzero a learning model is used to simulate the data. Currently only the one-element model of Bower (1961) is available (model = 1). The model's parameters are alpha and guess.

Program Parameters

alpha (float)
Learning parameter when simulating the one-element model.

alength (int)
Expected length of response string.

bfactor (int)
The block multiplication factor. At the start of the experiment PXL creates bfactor copies of the item list given in each block of the parameter file. These blocks are then executed until the criterion learned is reached. Thus bfactor gives the maximum number of item list repetitions that are possible.

blockstate (int)
The number of consecutive error free blocks until now. If blockstate is larger than learned, then the experiment is stopped.

feedbackpos (int)
Vertical screen position for the feedback message.

flags (string)

B
Give an attention beep at start of a test trial.

E
Echo responses during test trials.

F
Give feedback during test.

M
Use multiple character responses.

P
Show pair instead of item and link during the study period.

R
Do replace error responses from the file itemfile.

S
Do not run test trials, study only.

T
Do not use the method of anticipation but run study and test phase separately.

W
Do replace error items and responses from the file itemfile.

X
Do not insert blanks between characters for item display.

guess (float)
Guessing parameter for simulating a learning model.

item (string)
The stimulus item.

itemfile (string)
File name for a file of replacement items. This file should either be in the local or in the pxl/exp directory.

learned (int)
The number of error free blocks that is required before all items are considered to be learned.

link (string)
The correct response string.

model (int)
Type of learning model for simulations.

nokeybreak (int)
Since we use the keyboard for response input it might be a good idea to disable the keyboard break function by setting nokeybreak to ON. Thus an accidental press of CTRL-C will not break the experimental session. During testing of the program, however, we can set nokeybreak to OFF in order to allow breaks.

pair (string)
If the flag P is set then this string is shown during the study phase instead of the automatically created combination of item and link.

responsepos (int)
Vertical screen position of the response echo.

studymsg (string)
Message at start of study block.

studypos (int)
Vertical screen position of study message.

studytime (int)
Time to study each item.

studytm (int)
Display duration of studymsg.

testendmsg (string)
Message at end of test block.

testmsg (string)
Message at start of test block.

testpos (int)
Vertical screen position of test start/end message.

testendtm (int)
Display duration of testendmsg.

testtm (int)
Display duration of testmsg.

Example Experiments

Method of Anticipation

The first example is paired associates learning with the method of anticipation. We set bfactor to the maximum number of repetitions that we expect. The first block is for training only, so it is shorter. The second block is the actual data collection block.


Parameter file pala.x from directory \pxl\app\pal

Rock's Method of Item Replacement

The next example is a replication of Rock's (1957) experiment. Two groups of subjects are run: the experimental group has separate study and test trials and if there is a wrong response then the corresponding item is replaced from a file of items ("rock.txt"). For the control group wrong items are not replaced. The parameter file also allows simulation by defining the string MODEL to 1. Given the parameter file is called palr.x then a data run for a subject in the control group is started by the following command line:\labelpalDEF
   pal -f palr.x -DKGRUPPE subj
To run the experiment for a subject in the experimental group the string -DKGRUPPE has to be left out.


Parameter file palr.x from directory \pxl\app\pal

Here is a short section of the file rock.txt:

   G 26
   SS 28
   K 13
   CC 18
   TT 4
   ...
   ...
   ...
   Y 33
   C 5
   HH 42

Back to table of contents


Author: Hans Irtel

irtel@psychologie.uni-mannheim.de