85-411/711 Cognitive Processes and Problem Solving Homework #1-Solutions The goal of this assignment was to give you an opportunity to experience a couple of classic cognitive psychology research tasks, as well as the effects that they demonstrate, and then to work on explaining those effects I. Memory Experiments. A. Immediate Memory Span. 1. What was the effect of acoustic similarity on the per cent of list recalled correctly? A much smaller percentage of words are recalled correctly for acoustically similar word lists than for acoustically dissimilar word lists. In this particular experiment, some subjects correctly recall fewer than 5% of the words in the acoustically similar word list, whereas most subjects recalled 50-100% of words in the acoustically dissimilar word lists. 2. Why does this effect occur? i.e., what aspect(s) of the cognitive architecture causes this effect? The strong effect of acoustic similarity suggests that the primary representation of words in short-term memory is acoustic (rather than visual or semantic). Many students did a great job of naming and describing this phenomenon, but comparatively few bothered to try to account for it in cognitive architectural terms. Here is the kind of response we were looking for: EPAM's explanation for the confusion among similar words is that more EPAM tests are required to distinguish among the words. To elaborate, one can postulate, in keeping with the theory, that during each trial subjects in this experiment are creating pointers to representations in LTM of the five words presented (MOT 1: 2.3, p. 68). Given the high degree of similarity among the acoustically similar words, the subject would actually need 3 pointers (one for each phoneme) for each word presented, in order to disambiguate them in the discrimination net comprising LTM. Keep in mind that information in short-term memory is lost unless it is rehearsed at least once every 2 seconds. At 3 pointers per word, and five words per trial, this means the subject has to try to keep active 15 pointers in STM. That is too much to ask of the system, and so information is lost and we see a corresponding drop in performance for acoustically similar words. B. Memory for Lists. 1. How and why is the serial position curve different for immediate vs. delayed recall? The main effect of immediate vs. delayed recall is on the recency portion of the serial position curve. (There is also a small interaction with presentation time per item, but that is not of primary interest here). Since the items on the end of the list could be recalled easily immediately after presentation, but not after a delay, it is reasonable to assume that these items were being recalled from STM. In the delay condition, the intervening arithmetic task prevented rehearsal of any items during the delay. Hence, the items from the end of the list could no longer be retrieved from STM, as items fade from STM after 2 seconds if they are not rehearsed. 2. Why might recall be better for the beginning of the list than the middle? There are two main explanations for the primacy portion of the serial position curve. The first explanation, congruent with EPAM, is that subjects tend to rehearse items starting from the beginning of the list. As items are presented, the subject rehearses all the items seen so far. So items presented earlier are rehearsed more often than items presented later, and hence are more likely to be added to LTM. Also, when there are fewer items to rehearse, more time may be devoted to memorizing each item. The second main explanation is the "classical" explanation in the memory literature. Researchers postulated that as items are being memorized there is a build up of inhibition which makes it more difficult to remember subsequent items. This process is called the build up of proactive inhibition (PI). PI also explains why when subjects are given many lists to remember, their performance on each successive list gradually decreases. II. The EPAM simulation. 5. Draw a tree diagram of the network that incorporates the information that you obtained from both "Net Structure" and "Net Contents". Net Contents produces: EPAM NET: (N 0) :: Test=TYPE Branches: VISUAL -> (N 1) (N 1) :: Type=VISUAL Image=(i _) Above=(N 0) Test=First Branches: i -> (N 3) w -> (N 2) (N 2) :: Type=VISUAL Image=(w _) Above=(N 1) (N 3) :: Type=VISUAL Image=(i _) Above=(N 1) Net Structure produces: The tree diagram: See attached epam_net.gif file. 6. What would happen if EPAM tried to recognize the word "IT"? How about the word "its"? There was a lot of confusion here regarding the use of the word "recognize" in the question statement. When an EPAM net is used in a recognition task, it is considered fixed, and the structure of the net does not change. Changes are only made to the net when new terms are "studied," and then only if necessary. If EPAM tried to recognize "IT", it would sort the word only as far as (N 1), as EPAM is case sensitive and does not have a test branch for capital 'I'. Note that this is not necessarily a bug. For example, when reading about "Bill" (proper noun) one would not want to think one had read "bill" (common noun). Since (i _) is the image stored at (N 1), EPAM would recognize "IT" as a word that begins with 'i'. In this case, this is almost correct. However, EPAM would have returned the same thing if it had tried to recognize "fix" or "ED" or "WHEN". If EPAM then studied "IT", it would add another test branch to (N 1) for items which have the letter 'I' in the first position. It would also create a new node (N 4) which would be of type VISUAL and have (I _) as an image. If EPAM tried to recognize "its", it would sort the word as far as (N 3). Since (i _) is the image stored at (N 3), EPAM would recognize "its" as a word that begins with 'i'. If EPAM then studied "its", it would update the image at (N 3) by adding a 't'. Therefore, the new image at (N 3) would be (i t _).