Documenti di Didattica
Documenti di Professioni
Documenti di Cultura
Abstract
The study of word-length effect concentrated mostly on the accuracy of recalling short
and long words in both pure and mixed lists. Previous studies showed that pure long lists were
much poorly remembered. Hulme et al. (2004) found that word-length effect could be abolished
in mixed lists when the short and long words are alternated. We investigated distinctiveness and
found it to be a salient cue for improved correct recall when the list of words has a single
distinctive transition. Lists contained three short words following by three long words and vice
versa. Surprisingly, in the short-long condition, there was also an improvement in position 3
recall. One of the possible explanations could be the strategic shift of working memory resource.
saliency, processing, activation levels, and attention, are all different facets that explain the same
effect.
List Effect
Baddeley, Thomson & Buchanan (1975) theorized that the word list is rehearsed within a
fixed size phonological buffer that is temporally based on word duration and not structurally
based on the number of syllables. All elements that can fit in the buffer suffer similar time based
decay with rehearsal reactivating the associated memories. It would predict that the serial
position curve for an alternating list of short and long words would average out and lie between
that of the pure long word list and the pure short word list curves (figure 1). Unfortunately, the
word list effect was abolished, and an intermediate effect was not seen by Hulme et al. (2004).
List Effect on Accuracy
100
% Correct
80
60
40
SSSSSS
LLLLLL
SLSLSL
20
0
1
Word Position
Figure 1. List effect prediction on accuracy (data for illustrative purposes only)
Word Effect
Neath & Nairne (1995) explained the primacy-recency curve by considering each word
item as a sublist of time based segments that must be assembled based on their features, with
longer segments being less likely to be assembled correctly due to increased interference of
features. If the length of the word item caused the effect, then an alternating list of words curve
should have a sawtooth shape with the short and long items segregated as shown in figure 2. This
too was not seen by Hulme et al. (2004) as the alternating word list was as smooth as the pure
short word list.
It should be pointed out that the item and word accounts differ in another dimension.
Baddeley, Thomson & Buchanans account is based on an activation decay mechanism, while
Neath & Nairnes (1995) account is based on feature interference. These two mechanisms will
become entangled in related theories that follow.
Item Effect on Accuracy
100
% Correct
80
60
40
SSSSSS
LLLLLL
SLSLSL
20
0
1
Word Position
Figure 2. Item effect prediction on accuracy (data for illustrative purposes only)
Lag-Recency Effect
The list and item explanations differ in scope of influence, global or local respectively.
Although salience is an item feature, distinctiveness is a relation between the current and
previous item and this could lead to an inter-item scope effect. The concept of distinctiveness is
often asymmetrical. The second item is considered more distinct than the first because it is only
when processing the second that distinction can be recognized. With rehearsal, of course,
distinction of the first item can be subsequently processed but to a lesser amount overall. Howard
& Kahana (2002) have shown that the conditional probability of picking an item close to the
specifically rehearsed and distinctive item follows the Lag-Recency curve as shown in figure 3.
The previous retroactively interfering items are overall less likely than the following proactively
interfering items.
The possible implication to our hypothesis is that if indeed the fourth item is recalled
better, than this could affect the third and fifth items as well.
Retroactive Interference, Patterns and Structural Effects
The Lag-Recency effect shows both proactive and retroactive interference. When dealing
with patterned information, there is yet another account that seems to show all effects combined,
and adds a hierarchical structure effect as well. Our experiment made use of only four patterns of
word lists. Moreover, the pattern of three short words followed by three longs words has
similarities to music where there are strong and weak beats. Palmer (2005) describes how
Figure 4. Predictions of memory retrieval strength for past, present and future sequence events
during performance of a melody. Bars indicate amount of memory strength for the different
sequence events at the time at which the circled note is performed. Xs indicate metrical accent
strength (the more Xs, the more accent).
The aim of the present study was to investigate the time course and impact on accuracy
for a single distinctive transition. We used lists of six words where the first three were short and
the last three were long (and vice versa). We hypothesized that the accuracy of the distinctive
fourth word would improve, but how this would affect the accuracy on nearby words was left as
a subject of this investigation.
Method
Hulme et al. (2004), in their study have shown that the effect of world length on recall is
abolished in lists containing alternating long and short words. Neither the standard item-based
nor list-based models adequately explain this finding. Hulme et al. (2004) proposed an
explanation based on item complexity and distinctiveness. The purpose of this experiment was to
investigate these factors separately and in greater detail. Thus, the present study included the
long-short lists (LLLSSS, where L stands for long and S stands for short) and short-long lists
(SSSLLL) of words as well as pure long lists and pure short lists of words. The transition from
short-to-long word and long-to-short words in mixed lists can indicate distinctiveness, which
should improve recall of nearby words but would separate the entire list into two sublists.
Participants
A total of 16 graduate students from Carleton University voluntarily took part in this
study. The experiment was performed in a large classroom with all participants present. The
participants sat at tables set around the perimeter of the room. The participants were able to see
the projector screen. Each participant signed the first page of their answer booklet for
participation in the experiment as per the guidelines and approval from the Carleton University
Ethics Committee for Psychological Research. Of the 16 participants, 14 were male and 2 were
female, and their ages ranged from 24 to 50 years.
Materials
The words used in this experiment came from experiment 1 of Hulme et al. (2004) with
one exception. The original word stoat was deemed to be less familiar due to the cultural
differences and was replaced by the word mink. The list consisted of eight 1-syllable words
(zinc, mink, scroll, math, switch, mumps, school, Greece) and eight 5-syllable words (aluminium,
Results
Correct Recall
Correct recall is defined as the correct word recalled in the correct position, correct
order having been the instruction given to the participants in both this experiment, and in Hulme
et al. (2004). Proportion correct recall was the dependent variable used to measure differences
between the four stimulus presentation types as defined in the previous section (pure short, pure
long, short-long, long-short).
A within-participants analysis of variance was performed using list type and word order
(six positions). List type was significant, F(3,45) = 12.40, MSE = .043, p < .001. Participants
recalled fewer words from the pure long list than either the pure short or the mixed lists. As can
be seen in Table 1, proportion correct for pure long was 0.584, while proportion correct for the
pure short was 0.702, a difference of 0.118. This recall difference between pure lists replicates
Table 1
Proportion of Items Recalled Correctly and Omitted as a Function of List Type
---------------------------------------------------------------------------------------------List type
Pure short
Pure long
Long-short
Short-long
---------------------------------------------------------------------------------------------Proportion Correct 0.702
0.584
0.726
0.754
---------------------------------------------------------------------------------------------Omission Errors
0.128
0.146
0.093
0.078
Of interest as well are the omission errors, wherein participants responded blank or
indicated in another way that they did not know a given item. As can be seen in table 1 these
errors were larger for the pure lists than the mixed lists.
The effect of word order was also significant, F(5,75) = 11.32, MSE = .063, p < .001.
Recall performance varied with serial position for all four list types. Both pure short and pure
long recall showed classic primacy-recency effects with position one having the highest
proportion correct (primacy). Participants recall declined through the serial presentation, but
improved on the last word in the order (recency) as seen in figure 5. The two mixed presentations
had patterns that differed from those of the pure lists.
Proportion Correct
0.8
0.7
0.6
0.5
0.4
0.3
pure long
pure short
0.2
one
two
three
four
five
six
Word Position
Figure 5. Mean proportion of words recalled in the correct position as a function of list type and
serial position
There was also a significant List x Order interaction F(15,225) = 5.55, MSE = .019, p <
.001. Both mixed lists (short-long and long-short) acted as if they have been divided into two
separate pure word sublists each with primacy-recency effects (see figures 6 and 7). Participants
did not know until the fourth word whether they were being presented a pure or a mixed list.
When participants were presented with long words first, the results are the same for the first
three words in both pure long and long-short presentations. For word presentation four through
six, the results diverge. Participants recall for the short half of the long-short list improved
significantly compared to the pure long list.
Proportion Correct
0.8
0.7
0.6
0.5
0.4
0.3
Pure long
Long-short
0.2
one
two
three
four
five
six
Word Position
Figure 6. Mean proportion of words recalled in the correct position as a function of list type and
serial position
When short words were presented first participants again split the mixed list (short-long)
in two with position three (the end of the short word presentation) and position four (the
beginning of the long words) being significantly different from the corresponding pure short list
(figure 7). Position four recall improvement was expected. What was a surprise was the
significant result and improvement for the third word in this mixed short-long list. This might be
due to distinctiveness. When rehearsing the Participant has a chance to repeat the list and as such
may be able to cue the shorter word by referencing the longer improving both position three and
four recall. Or it might be due to complexity, or rather the ease with which the one syllable word
can be retained compared to the five syllable word.
Proportion Correct
0.9
0.8
0.7
0.6
0.5
short-long
pure short
0.4
one
two
three
four
five
six
Word Position
Figure 7, Mean proportion of words recalled in the correct position as a function of list type and
serial position
This implies a strategy. In a task to remember as many words as possible in the correct
order the Participant may favour the shorter word over the longer. Position six for pure short,
pure long, and pure short-long were not significantly different (see both figure 5 and 7).
Participants responding to the questions at the end of the experiment wrote that the most
difficult recall was the long word list whereas the easiest were the pure short and mixed. They
had difficulties due to time limits, length of the word, and remembering how to spell the word.
At least five participants did speak English as a second language
Discussion
Hulme et al. (2004) proposed a novel explanation of the word-length effect drawing on
ideas of item complexity and item distinctiveness. They argued that the word-length effect
depends upon the phonological complexity of items and not upon their spoken length per se.
Hulme et al. (2004) argued that in addition to the role of item complexity, how well any given
item in a list can be recalled would depend upon the relative distinctiveness of the item in
relation to other items in the list. In this view, the recall of long items in mixed lists (alternating
short and long items) is boosted by their relatively greater distinctiveness in mixed lists than in
pure lists. When equal numbers of short and long words are presented in the same list the word
length effect is abolished, with both types of words being recalled at levels indistinguishable
from the recall of short words.