Professional Documents
Culture Documents
The Modern Language Journal, 78, iii (1994) cabulary instruction (47).The hypothesis is also
0026-7902/94/285-299 $1.50/0
01994 The M o h n Langwgq Journal
bolstered by L1 and L2 studies showing that stu-
dents can correctly guess the meaning of un-
286 The Modern Language Journal 78 (1994)
known words while reading (2; 11; 12; 31; 41; 60; words in context and also have access to a
61), especially students with higher verbal ability dictionary.
(17; 42; 52; 65). Other studies, however, contra- Dictionary use has also been examined with
dict these findings by showing that readers are respect to L2 reading. A recent study by
often unable to glean the meaning of an un- Hulstijn found no significant difference in the
known word from a text (4; 27; 35; 39; 55). number of words looked up by students with
Part of the difficulty in interpreting the con- high inferencing ability (i.e., those able to guess
flicting findings of these studies is that re- word meaning from context) and students with
searchers have used very different texts to test low inferencing ability-suggesting that some
the same construct. For example, some have students use dictionaries when it may not be
used artificial words or cloze-type blanks in- necessary. Likewise, the results of the seminal
stead of real, unknown words to test word guess- Bensoussan, Sim, and Weiss study question the
ability (2; 12; 43; 52; 65). This practice, however, benefit of dictionary use for reading compre-
removes clues at the morphemic level. Others hension. Their study included three investiga-
have included different amounts of contextual tions in which advanced English as a foreign
support, either increasing contextual clues by language students read texts in three different
artificially constructing paragraphs to provide dictionary conditions: monolingual, bilingual,
rich support (11; 15; 23) or decreasing clues by and no dictionary. Subjects with dictionaries
providing the word with only a single sentence were asked to underline the words they looked
(55). Research on deriving meaning using natu- up while reading. With the text still in front of
ral texts is notably lacking. them, the subjects answered multiple-choice
Dictionary use. In addition to using contextual questions to check comprehension. No signifi-
cues to guess the meaning of an unfamiliar cant correlations appeared between dictionary
word, the L2 reader often chooses another strat- use and reading comprehension scores in any
egy: looking the word up in a dictionary. Al- of the three studies. Thus, while using a diction-
though in recent years, many researchers, ary to increase comprehension has rational ap-
teachers, and textbook authors have encour- peal, research evidence supporting the claim is
aged students to guess, to use inference as the still lacking.
strategy of first choice (30; 48; 49; 64), this ad- Contextual learning. Students may look up an
vice appears to be based more on conjecture unknown word in a dictionary or guess its
than on empirical findings (13). The relevant meaning from context while reading, but later,
literature regarding dictionary use falls into two may not be able to remember the word or recall
major categories: vocabulary learning and read- its meaning. According to Pressley, Levin, and
ing comprehension. McDaniel, guessing from context is only the
In general, the practice of relying on the dic- first step in learning from context (51). Four
tionary as the primary method of increasing recent L1 studies have dealt with incidental
students’ vocabulary has not been substantiated learning of vocabulary while reading (32; 36;
by L1 research (14; 15; 20; 23). Many of the L1 46; 47). The specific results differ, but the gen-
investigations were comparison studies be- eral findings of these studies show small but
tween dictionary definitions and contextual significant gains in word knowledge after read-
guessing to see which condition best facilitated ing. These results, however, have been ques-
the learning of new vocabulary (14; 15; 23). Al- tioned by other researchers whose studies indi-
though the results showed contextual guessing cate that students are often unable to gather
to be superior to dictionary definition, the word meaning from context (35; 55; 70).
findings are confounded by the fact that many Several learning-from-context studies have
of the texts used for the context-only condition been also been conducted in L2, but the find-
provided definitions or gave examples for the ings regarding word learning vary greatly.
targeted words-much the same as the diction- White sought to duplicate an L1 study by Saragi,
ary condition (57; 58). In order to clarify these Nation, and Meister in which subjects learned
findings, Stahl and Fairbanks conducted a sixty-eight percent of the unknown nadsat
meta-analysis of L1 vocabulary instruction words (i.e., neologisms of Russian origin) in
studies and concluded that a combination of Clockwork Orange. White conducted two experi-
definitional and contextual approaches is more ments with students of English as a second lan-
effective than either approach in isolation. The guage (ESL) reading two chapters of the text.
L2 implication is that students would learn The multiple-choice test results showed that six
more new word meanings if they could see the percent of the thirty targeted words were
Susan Knight 287
learned in the first experiment and eight per- electronically recorded. Four texts were used.
cent in the second, percentages that were much Half of the students read the first set of two
lower than for L1 students. White attributed the articles; the other half read the second set. All
low scores to text difficulty and to lack of back- subjects had been instructed to read for mean-
ground knowledge. ing and were told that they would be asked to
Much higher rates for word learning were write an immediate recall protocol. After exit-
found by Haynes and Baker, even though sub- ing the readings, subjects were first tested for
jects had to supply instead of merely select the reading comprehension and then given an un-
definition. Taiwanese college students, who expected vocabulary test on the targeted un-
had studied English for at least six years, first known words (incidental learning). Although
read a passage on the treatment of skin wounds subjects had read only one text-set, they were
and then wrote definitions for the selected tested on the targeted vocabulary of both text-
words. The freshmen learned an average of 6.9 sets to provide a comparison for words learned
out of thirty-six words, while seniors learned in context.
13.6 words. The majority of these targeted
words, however, had rich textual support that RESEARCH QUESTIONS
included definitions, illustrations, and exam-
The present study addressed the following re-
ples. The differences between the findings of
search questions:
the White study and the Haynes and Baker
Incidental learning of vocabulary.
study can perhaps be explained by the types of
1) Is there a significant difference in sub-
words targeted (i.e., neologisms versus real
jects’ vocabulary test scores after they
words), level of difficulty of text, and amount of
have seen words in context (exposure to
contextual support provided.
text containing these words) and when
Conclusion. In summarizing the studies dis-
they have not seen the words in context
cussed thus far, a gap exists between what is
(no exposure to text containing these
known about incidental vocabulary learning
words)?
and dictionary usage and what is advocated in
2) Is there a significant difference between
the classroom. The gap exists because past re-
vocabulary learning measures for low
search has not provided definitive answers to
and high verbal ability students?
issues of guessing from context, dictionary use,
3) Is there a significant difference between
or learning new words through reading. Contra-
the vocabulary learning scores of stu-
dictory findings have resulted because many of
dents who use a dictionary and those
the past studies have used artificially con-
who do not?
structed texts, texts that were too simple or too
Reading comprehension.
difficult, or neologisms and blanks for target
4) Is there a significant difference between
words. In addition, the relationship between
reading comprehension scores for stu-
reading comprehension and vocabulary acquisi-
dents who use a dictionary and those
tion has not been sufficiently examined.
who do not?
*Ability
y4)
Dic tonary
1 Text-Set I
(n = 28)
Text-Set I1 Text-Set I1
Levels
\
No Dictonary
TS
:;I; Text-Set I Text-Set I
(n = 51)
Text-Set I Text-Set I1 Text-Set I1
(n = 26)
(* 50% of the students in the high verbal-ability group were assigned to the Dictionary Condition and
the other 50% to the No-Dictionary Condition. The same procedure was followed for the low verbal
ability group.)
perts first divided the original articles into ac- text-sets were indeed unknown. Later, they were
ceptable pausal units and then ranked the units given the no-exposure vocabulary test (i.e, a
from one to four depending on their salience to supply- and select-definition vocabulary test
the message of the text. Four points were given over the 24 targeted words in the text-set that
to the units having the greatest semantic signifi- students had not been assigned to read).
cance, and one point to those with the least. Reading and testing via the computer. Two
The student protocols were examined for accu- weeks later, subjects met in the computer lab.
rate provision of each valid proposition and As each subject entered, he or she picked up a
awarded points accordingly. The scores from diskette with his or her name on it. Each disk-
each article were combined to give each subject ette had been programmed for either diction-
one comprehension score for his/her assigned ary or no dictionary access and included arti-
text-set (i.e., 298 points possible in Text-Set I; cles and vocabulary tests from either Text-Set I
295, in Text-Set 11.) The scorers’ interrater re- or 11.
liability for the four texts was .95. Subjects were told that the purpose of the
experiment was to test for reading comprehen-
sion using authentic texts. They were directed
PROCEDURES
to read each article for content and then to
Preparation for the study included randomly write, in English, everything they could remem-
assigning subjects within each verbal ability ber of what they had read. No mention was
group (i.e., high or low) to the dictionary or no- made of the vocabulary tests nor were the tar-
dictionary condition. Each of these conditions geted words marked or distinguished in any
was further divided according to reading manner.
passages-either Text-Set I or Text-Set 11. (See To show the students how to use the computer
Table I ) program, the lab director reviewed the function-
The actual study was conducted on three sep- key codes using an overhead projector attached
arate days during the students’ scheduled class to a computer. In addition, these explanations
hour. The following procedures are labeled ac- also appeared on each student’s computer
cording to whether they occurred before, dur- screen during the entire experiment.
ing, or after the actual reading of the articles in The basic procedure on all diskettes was the
the computer laboratory. same. The subjects began by reading on the
Two weeks prior to reading. All subjects were computer the first of the two articles in their
first given the checklist vocabulary test in order text-set, taking as much time as desired. Al-
to validate that all of the targeted words in both though movement throughout the text was con-
290 The Modern Language Journal 78 (1994)
trolled by each student, only those in the diction- ber of words looked up). Both of these measures
ary condition had access to the computerized were designed as descriptive measures and were
dictionary. The computer-dictionary process not considered part of the experimental design
simulated actual dictionary use in two ways: per se.
manner of lookup and type of definition pro- A two-between, one-within subject design was
vided. In order to look up any word, the sub- used to analyze incidental vocabulary learning.
jects in the access condition first pressed the The between-factor variables included two
designated lookup key on the computer and levels of verbal ability (high and low) and two
then typed the root form of the Spanish word in levels of reading condition (dictionary and no
the box. If they entered an incorrect form, the dictionary). The within factor was the presence
three words that would alphabetically precede of exposure (exposure and no exposure) to un-
the false entry in the dictionary as well as the known words in written context. The depen-
three words that would succeed the entry were dent measures were scores on immediate and
displayed on the screen along with instructions: delayed supply- and select-definition vocabu-
“Word not found. Try one of these . . .” Once lary tests.
the correct root was supplied,. the dictionary To assess reading comprehension, a two-
definitions appeared immediately.6 All diction- between subject design was used: verbal ability
ary displays appeared in the center of the (high and low) and reading condition (diction-
screen and covered the majority of the text, ary and no dictionary). The dependent meas-
making contextualizations of the various defini- ure was the combined scores of the recall proto-
tions as difficult as it might be in a paper- cols for each text-set.
dictionary situation. In addition, Pearson correlation coefficients
The computer was programmed to tabulate were found for the number of words looked up
each word a subject looked up, thus providing and vocabulary and recall scores.
an actual dictionary-use count for all the words
in each passage. In addition, the program re-
RESULTS
corded the amount of time each student spent
reading an article. Incidental learning: vocabulary acquisition (Hy-
After exiting each reading, subjects wrote re- potheses one to three). Means and standard devia-
calls in English of everything they could remem- tions for both types of vocabulary tests-supply
ber.’ Following the last recall protocol, students definition (i.e., write the English equivalent)
were administered an unexpected supply- and and select definition (i.e., choose the correct
select-definition vocabulary test over the English equivalent) are presented in Tables I1
twenty-four targeted words in the text-set that and 111, respectively. Each test was given imme-
they had just read. diately after reading and also two weeks later.
As students completed the program, the in- The maximum possible score on each measure
formation stored on their diskettes (i.e., test re- was twenty-four points, one for each targeted
sults, reading time, words looked up) was auto- word.
matically printed out on the mainframe printer. Hol: There will be no significant difference between the
Two weeks afterreading. Subjects took a delayed exposure and the no-exposure conditions on the results of
supply- and select-definition test over the same vocabulary learning measures. Although the aver-
targeted words encountered in their text-set in age supply-definition score was only 3.38 words
order to measure long-term retention. Al- for immediate testing and 2.86 for delayed test-
though this was a paper and pencil test, the ing, these means were significantly different
content and the order were identical to that from the no-exposure mean of .15. Likewise, the
taken in the computer lab two weeks earlier. average select-definition scores for the immedi-
ate test (11.73 words) and for the delayed test
(10.20 words) were also significantly different
DATA ANALYSIS from the no-exposure mean of 1.8 words (p <
Appropriate analyses of variance (ANOVA) .001).
and post-hoc tests were conducted on both de- There were also significant first-order inter-
pendent measures: vocabulary test scores (im- actions on each of the four ANOVAs. On the
mediate and delayed) and reading comprehen- supply-definition scores, exposure interacted
sion scores. Additional data were provided by with dictionary on the immediate results (p <
simultaneous measures taken by the computer .001) and on the delayed results (p = .012). A
during reading (i.e., reading time and the num- comparison of pairwise means indicated that
Susan Knight 291
TABLE I1
Supply-Definition Tests. Means and Standard Deviations of Supply-Definition Vocabulary Scores as a
Function of Exposure, Ability Level, and Dictionary Condition
~ ~~ ~ ~ ~ ~
TABLE I11
Select-Definition Tests. Means and Standard Deviations of Select-Definition Vocabulary Scores as a
Function of Exposure, Ability Level, and Dictionary Condition
although all subjects learned more words when posure scores significant on every measure.
they were exposed to them in context than The null hypothesis was thus rejected.
when they were not, those who had dictionary Ho2: There will be no significant differences between
access learned the most. A Least Squares Means low verbal ability and high verbal ability subjects on the
post-hoc test indicated that the no-exposure vocabulary learning measures. Means differed sig-
means were not significantly different from one nificantly for high and low verbal ability levels
another, yet all other immediate and delayed on the immediate supply-definition test (3.80
supply-definition comparisons were signifi- and 2.98, respectively; p = .049); on the delayed
cant. Thus, all subjects, regardless of verbal abil- supply-definition test (3.29 and 2.44; p = .023);
ity level, were unable to correctly supply many on the immediate select-definition test (12.86
word meanings without first being able to see and 10.67; p = .009); and the delayed select-
the targeted words in context. definition test (11.80 and 8.70; p <.001). The null
ANOVA results for select-definition scores, hypothesis was rejected.
both immediate and delayed, revealed two first- Ho3: There will be no significant differences between
order interactions: exposure*dictionary and the dictionary access and no-dictionary access condi-
exposure*ability. Although subjects identified tions on the scores of vocabulary learning measures.
the meaning of more words after reading them Immediately after reading, those subjects with
in context than they did before, those who had dictionary access achieved a supply-definition
dictionary access identified the most. Similarly, mean score of 4.95 words while those without
high verbal ability students learned more words access obtained a mean score of 1.72. Two weeks
from context than low ability students reading later, the dictionary group mean was 3.37 and
the same articles. All post- hoc test comparisons the no-dictionary group mean, 2.3. Both of
found the difference in no-exposure scores to these comparisons were significant ( p < .001
be nonsignificant and the difference in the ex- and p = .024, respectively.)
292 The Modern Language Journal 78 (1994)
TABLE IV
Means and Standard Deviations of Recall Scores as a Function of Ability Level, and Dictionary Condition
On the select-definition measures, subjects posure mean score and then divided by the to-
having access to the dictionary had a mean tal number of words per test (i.e., twenty four).
score of 14.6 on the immediate test and 12.2 on Table V summarizes the percentage results.
the delayed test compared to the no-dictionary The select-definition test reflected the learn-
group with respective means of 8.8 and 8.1. The ing of a greater number of words than the
immediate and delayed comparisons were both supply-definition test. This difference was ex-
significant at p < ,001. No significant interaction pected, however, because each test taps a differ-
appeared on any of the four tests between dic- ent type or degree of learning. Normally, when
tionary condition and verbal ability level. a word is encountered once in context, the
Reading com$n-ehension. amount of learning that occurs is small-
Ho4: There will be no significant dqference between the allowing only for recognition of a similar defini-
dictionary access/no-dictionary access conditions on tion rather than production (18; 48; 59). In this
reading comprehension scores. As seen in Table W, study, forty-three of the forty-eight words ap-
the dictionary group had an overall reading peared only once in the readings; four ap-
comprehension mean of 74.01, while the no- peared twice; and one appeared five times.
dictionary group attained an overall mean of An unusual finding surfaced in the no-
56.65. The ANOVA indicated a significant dif- dictionary condition between immediate and
ference between the two means (p < .001). The delayed supply-definition scores-after two
null hypothesis is therefore rejected. weeks both ability groups somehow learned ad-
Another ANOVA was performed on these ditional words. Possible explanations are that
reading comprehension scores, but it included subjects may have learned the meanings after
text-set as an additional independent variable leaving the test setting or perhaps confirmed
in order to verify that comprehension did not suspicions during the succeeding select-
vary according to text. Text type was not signifi- definition test. Although the word increase is
cant (p = .739). small, it is interesting that this effect did not
occur in the dictionary condition, suggesting
perhaps that the no-dictionary group processed
DISCUSSION AND ADDITIONAL FINDINGS
the information more deeply (i.e., learned the
Word karning: The above results indicate that words better), perhaps because more analysis
subjects indeed learn a significant number of was needed initially to discover the meaning.
new words while reading for meaning; however, In order to compare the results of this study
high verbal ability students learn more words with those of past studies, it is necessary to dis-
than low verbal ability students, and students tinguish between the findings of those who had
who use a dictionary learn more than those who dictionary access and those who did not.
do not. These results become more meaningful 1) Words learningfor those without dictionary access.
when converted into learning percentages. A The learning from context percentages are con-
method similar to one proposed by Nagy, Her- sistent with results found in similar L1 studies
man, and Anderson (47) was used to determine (e.g., 32, 36, 46, 47). The Nagy, Herman, and
the percentage of words learned while reading Anderson (47) study-which was most similar
with or without a dictionary. The no-exposure to the present study in terms of using authentic
mean score (which represents how many words texts, manner of selecting unknown words, and
the subject could guess without seeing the types of vocabulary measures-found that the
words in context) was subtracted from the ex- likelihood of a subject’s learning a word was
Susan Knight 293
TABLE V
Percentage of Words Learned Incidentally
SUPPLY SELECT
Immediate Delayed Immediate Delayed
NO DICTIONARY
High Verbal Ability 7% 11% 35% 33%
Low Verbal Ability 5% 6% 23% 20%
DICTIONARY
High Verbal Ability 21% 13% 55% 48%
Low Verbal Ability 19% 14% 51% 39%
(% = percentage of words learned after corrected for guessing)
# correct on Exposure test - # correct on No-Exposure test
%=
24 possible words
between .15 and .22 for the select-definition found for the dictionary condition seem to con-
test and between .I1 and .19 for the supply-defi- tradict the study by Bensoussan, Sim, and Weiss,
nition test depending on the level of question which found no significant difference in com-
difficulty. prehension scores for those who used diction-
The two previously mentioned L2 studies also aries and those who did not. Although there
provide comparative percentages for new word were several major differences between the
learning. In White’s study of ESL students read- studies, one of the most important is the profi-
ing parts of Clockwork Orange, select-definition ciency level of the subjects. Even the low profi-
findings were much lower than those found in ciency subjects in the Bensoussan et al.’s re-
the current study (i.e., ESL subjects learned 6% search had studied English for seven years and,
of the words for the first experiment and 8% on in comparison to the present study, would be
the second; compared to 29% in the present considered advanced. When the reading com-
study). As White concluded, however, these low prehension scores were divided according to
results may have been related to passage diffi- verbal ability level, no significant difference was
culty or to insufficient time to finish reading found for the high verbal ability groups in the
the passage. dictionary and the no-dictionary condition-a
Considerably higher percentages of word finding similar to that of Bensoussan et al. This
learning were found in the Haynes and Baker was not true, however, for the low verbal ability
study; for example, on immediate supply- group whose dictionary/no-dictionary scores
definition measures, Taiwanese college fresh- were significantly different (p = .002). (See Ta-
men learned nineteen percent of the targeted ble IV for mean scores.) In addition, it was
words while college seniors learned thirty-eight found that although there were significant dif-
percent. Their study, however, focused on lexi- ferences between scores of the low and high
cal familiarizations and thus provided either a verbal ability groups in the no-dictionary condi-
direct definition, synonym, or illustration of tion, these differences were not significant in
the new concept. In the present study, targeted the dictionary condition. Partial explanation
words were selected only because they were un- for this finding may come from studies showing
known, not because they had strong contextual that beginning readers are much more depend-
support. ent on vocabulary for comprehension than are
2) Word learning for those with dictionary access. more proficient readers (6; 9).
Although the dictionary condition increased Word lookup. It was initially assumed that sub-
the percentage of words learned for both verbal jects who had access to the dictionary would use
ability levels, it appeared to give the low verbal the dictionary frequently. To check this hypoth-
ability group a special advantage. For example, esis, the computer was programmed to tabulate
on the immediate-select-definition test, the dic- words actually looked up. The low verbal ability
tionary condition enabled the low verbal ability subjects in the dictionary condition accessed an
students to learn almost as many words as the average of 35.62 words per text-set while the
high verbal students in the same condition (51% high verbal ability subjects averaged 43.41
and 55%, respectively). words.
Reading comprehension. The main effects Table VI presents Pearson correlation coeffi-
294 The Modern Language Journal 78 (1994)
TABLE VI
Number of Words Looked Up Correlated with Vocabulary and Recall Scores for High and Low Verbal
Ability Groups
VOCABULARY RECALL
Immediate Delayed
Supply Select Supply Select
Low: .24 .54 -.01 .54 .68
p = .224 p = .003 p = .945 p = .004 p < .001
High: .10 .40 -.01 .48 .17
p = ,603 p = .041 p = ,984 p = .011 p = .397
TABLE VII
Means and Standard Deviations for Reading Time in Minutes According to Ability and Dictionary
Condition
~
cients between the number of words looked up bination of both. Because the subjects in the
and scores on the vocabulary and recall meas- dictionary condition also scored higher on the
ures for subjects according to ability level. On vocabulary and recall measures than those is
the vocabulary measures, higher correlations the no-dictionary condition, the question be-
were found for select-definition test scores than comes: Can the increase in learning be attrib-
for supply-definition, for the low verbal ability uted simply to the increase in time? In order to
students than for the high, for the recall meas- determine this, the increase in time between
ure than for the vocabulary measures. Correla- the no-dictionary and the dictionary condition
tions were higher for low verbal ability subjects was calculated for each ability group. These per-
on both vocabulary scores and reading compre- centages appear in Table VIII along with the
hension scores. The high correlation between increase in scores on the dependent measures
recall scores and number of words looked up that occurred between the no-dictionary and
for the low verbal ability group (.68) seems to dictionary condition.
indicate that dictionary lookup does not dis- High verbal ability subjects with dictionary
rupt the short-term memory, but rather en- access spent forty-one percent more time read-
hances comprehension. On the other hand, the ing and looking up words than did their no-
low correlation (.17) for the high verbal ability dictionary counterparts. Only two of the de-
group makes the efficacy of dictionary lookup pendent measures, however, exhibited an equal
questionable, a finding that corroborates the or greater increase than did the time measure.
results of Bensoussan, Sim, and Weiss. Reading comprehension scores increased only
Reading time. In addition to recording the eighteen percent.
words solicited from the dictionary, the com- Low verbal ability subjects with dictionary ac-
puter also recorded the reading time for each cess spent forty-four percent more time in the
passage. These data, when combined with the reading mode than did their no-dictionary
mean vocabulary and recall scores, were used to counterparts. Yet, every dependent measure
produce time-learning comparisons for the dif- shows a similar or greater increase in learning.
ferent conditions. Overall means and standard Thus, although reading comprehension in-
deviations for time are presented in Table VII. creases proportionally with time under this con-
As noted in Table VII, the subjects in the dic- dition, the amount of vocabulary learned in-
tionary condition spent more time on the pas- creases in greater proportion. This analysis of
sages than those in the no-dictionary condition. time on task demonstrates that the low verbal
This may have been time spent comprehending ability subjects benefited from the dictionary
the passage, accessing the dictionary, or a com- more than high verbal ability subjects.
Susan Knight 295
TABLE VIII
Percentage Increases in Time and Dependent Variable Scores from the No-Dictionary Condition to the
Dictionary Condition for High and Low Verbal Ability Groups