One of the seemingly simple questions about vocabulary that has stub- bornly resisted a simple answer is, Does wide reading increase stu- dents’ vocabulary size? A slightly more general version of this question is, How much vocabulary do students pick up simply while reading, without any deliberate attempt to learn word meanings?
The Urbana team started with a predisposition to believe that indi- viduals with large vocabularies had gained them largely through read- ing. There was reason to believe that not much vocabulary instruction was being done in schools (Durkin, 1978; Jenkins & Dixon, 1983) and that traditional vocabulary instruction was not especially effective (Jen- kins, Pany, & Schreck, 1978; Mezynski, 1983). Although some research- ers had examined students’ ability to infer word meanings from context, in 1980 there did not seem to be a definitive answer to the question of to what extent vocabulary acquisition could be attributed to reading.
Our first study on learning from context during normal reading (Nagy, Herman, & Anderson, 1985) revealed a modest but statistically significant level of learning from context. Furthermore, the amount of learning from context did not seem to vary significantly much by the nature or leniency of the measure of word knowledge used. In subse- quent studies, we addressed a variety of questions, such as: Is there any evidence of learning words incidentally a week after reading? Are the results different for different kinds of words and different kinds of texts (Nagy, Anderson, & Herman, 1987)? What types of text modifi- cations increase learning (Herman, Anderson, Pearson, & Nagy, 1987;
Diakidoy, 1998)? Does asking children to pay attention to the words increase their rate of learning (Kilian, Nagy, Pearson, Anderson, &
García, 1995)?
Most important, we found that children do learn the meanings of words while reading, even when they are not reading with the purpose of learning these words. A meta-analysis by Swanborn and de Glopper (1999) later confirmed the picture that emerged from our work—that children learn a modest percentage (about 15%) of the unknown words that they encounter while reading.
We interpreted our results as supporting a positive answer to the question of whether wide reading could be a means of increasing stu- dents’ vocabularies. One might estimate, for example, that (1) if a stu- dent read 500,000 words a year (25 minutes a day, 100 words per minute, 200 days a year), (2) if 2% of these words were unknown, and (3) if the student learned 15% of the unknown words, then the vocabulary gain would amount to 1,500 words a year. This estimate admittedly depends on some extrapolation, but the numbers that go into them seem plau- sible. (Carver, 1994, presents evidence that 2% is an overestimate of the words not known by students, but it could also be argued that his method of identifying unknown words—self-report by students—is an underestimate.) If our extrapolations are anywhere close to accurate, the modestly ambitious goal of increasing student reading to an hour a day could produce gains matching even generous estimates of the aver- age students’ annual vocabulary growth (Beck & McKeown, 1991)
In their 1991 review of vocabulary research, Beck and McKeown (1991) did not challenge our findings, but they certainly did question our interpretation, on several grounds. Beck, McKeown, and McCaslin (1983) had already entered the context arena with the assertion that “all contexts are not created equal.” Their examination of texts from two widely used basal reading series revealed that the contexts of words to be learned by students varied greatly in the amount of information they provided, some being nonsupportive and others even misleading.
This finding is, of course, not inconsistent with our claim that a small percentage of words learned from context can have large cumulative effects; the point of that article is more to address unrealistic expec- tations that teachers may have about the effectiveness of context as a strategy for figuring out the meanings of new words that students encounter while reading. We would agree that in this regard it is impor- tant not to overestimate the power of context.
However, Beck and McKeown (1991) also called into question our optimism about the effectiveness of context as a contributor to long- term vocabulary growth. Deftly using a quote from other researchers to voice the sharpest part of this criticism, they pointed out that the effect of context in our studies might be characterized as “statistically significant—but minute” (Schatz & Baldwin, 1986, p. 448). Schatz and Baldwin’s (1986) interpretation of our results is understandable, given that it comes in an article reporting null results; the students in their study learned nothing at all about the meanings of low-frequency words they encountered in context.
The Word Games 77 We would, of course, want to make a case for considering Schatz and Baldwin’s (1986) findings to be an underestimate. For example, in their study they used relatively brief contexts and low-frequency words, that is, words that were likely to be quite unfamiliar to students.
If words are learned incrementally, one small step at a time, more learn- ing would be found in a study such as ours, which used natural texts in which a number of words would be partially known to students.
When readers encounter partially known words in context, it is likely that even a modest level of contextual support could nudge a few of these words over whatever threshold of word knowledge is adopted for the test.
The most serious challenge to our optimism about learning from context as a source of vocabulary growth is the fact that those who need most to learn more words are the least likely to be able to benefit from it. McKeown (1985) had found (as had others; e.g., Shefelbine, 1990;
Sternberg, 1987; van Daalen-Kapteijns & Elshout-Mohr, 1981) that the ability to infer the meanings of new words from context is strongly related to reading ability. It might be true that good readers pick up large amounts of vocabulary from context while reading. However, for less proficient readers, there are a number of factors that make reading a less promising source of vocabulary knowledge: They read less, they learn a smaller proportion of the unknown words they do encounter, and they may spend less of their reading time in texts that are intel- lectually challenging. So to advocate reading as the main channel of vocabulary growth would be to leave some students on the short end of the “Matthew effect” phenomenon.
Finally, those who were suspicious of our extrapolations suggested that the argument for reading as a major source of vocabulary growth was primarily a default argument: Vocabulary must be learned through reading because it wasn’t clear where else it was learned. It seems rea- sonable to believe that persons with large vocabularies acquired them through wide reading because of the lack of other possible sources. In particular, it has been argued that vocabulary instruction in schools covers only a small number of words, perhaps 300 per year under good circumstances (Jenkins & Dixon, 1983), and that oral language—at least normal conversation, even among college-educated adults—is rela- tively impoverished in its vocabulary (Hayes & Ahrens, 1988).
Admittedly, the default argument is not a strong one, because, as Beck and McKeown (1991) pointed out, we know relatively little about how much vocabulary might be acquired through oral language in the
classroom. Though normal conversation may not be rich in rare words (Hayes & Ahrens, 1988), it is quite likely that classroom talk, both lec- tures and discussions, is much richer than out-of-school conversations.
And the number of words that teachers explicitly teach or explain during a school day undoubtedly far exceeds the number covered in
“vocabulary instruction,” narrowly construed.
Where are We now?
Swanborn and de Glopper’s (1999) meta-analysis could be said to have resolved the question of the rate of learning word meanings from context while reading. However, the question of the efficacy of inferring word meanings from context while reading as a source of vocabulary growth remains unresolved. In the National Reading Panel report (National Institute of Child Health and Human Development [NICHD], 2000), for example, it is acknowledged that students learn words from con- text while reading. However, the National Reading Panel also found no evidence that increasing students’ volume of reading had a significant impact on their vocabulary size. This is one reason that I would charac- terize vocabulary as a stubborn area to research: We know that children can learn words from context while reading, and we have a fairly good idea of the rate at which they do so. And yet the jury is still out on the question of whether having children read more is an effective way of increasing their vocabulary size.
One of the interesting differences between the Urbana and Pitts- burgh lines of research on context was the relationship between learning from context and reading ability. McKeown, in line with the majority of researchers who had investigated this question, found there to be a strong relationship between reading ability and success at inferring the meanings of unfamiliar words. In my research with colleagues at the Center for the Study of Reading, on the other hand, ability effects were relatively weak and, in one study, not even sta- tistically significant. For example, Pat Herman’s 1985 dissertation—
published as Herman and colleagues (1987)—showed a significant interaction of learning from context with ability, but students in the 99th percentile would have learned only a little over twice as much as students in the 3rd percentile. If the amount of learning from context reported in our studies could be described as “statistically significant but minute,” ability effects of this magnitude might warrant the same description.
The Word Games 79 This discrepancy in results—the fact that Urbana studies on learn- ing from context tended to find small or no ability effects, whereas studies done in Pittsburgh and elsewhere found a strong relationship between reading ability and learning from context—might be attributed to differences in methodology. One of the key distinctions is between incidental learning of word meanings and deriving word meanings.
The Urbana studies focused on incidental learning: Students were given texts to read, in most cases without being told that word learning was the purpose of the task. Students were tested on their knowledge of word meanings from the text after an interval (of from 15 minutes to a week) without the text present. In McKeown’s (1985) study, and in many others that found a strong relationship between reading ability and learning words from context, the task was deriving word mean- ings: Students were asked to infer the meanings of words with the texts present.
I would speculate that incidental learning from context and delib- erately deriving meanings from context rely to a large extent on dif- ferent learning mechanisms. Truly incidental learning might, for example, rely more heavily on associative learning, as exemplified by Landauer’s latent semantic analysis model (Landauer & Dumais, 1997).
A simulation based on this model succeeded in “learning” the mean- ings of words from context with no constraints, prior knowledge, or metacognitive abilities. Roughly equivalent to a connectionist network, this model simply gathers information about which words occur in the proximity of which other words. Such associative learning might be only weakly related to reading ability.
Deriving word meanings from context, on the other hand, is explic- itly metacognitive: Students make hypotheses about the meanings of unfamiliar words they encounter and evaluate these hypotheses on the basis of the information the context provides. As Sternberg and Pow- ell (1983) point out, the processes required to make and test inferences about the meanings of new words overlap substantially with those required for reading comprehension in general, so one would expect a strong relationship between reading ability and the ability to derive word meanings from context.
As far as I know, the hypothesis that truly incidental learning from context shows smaller ability effects than deliberate inferring of word meanings from context is consistent with the available research.
Another reason I find this hypothesis appealing is the success that most
children experience in oral language acquisition, which presumably relies substantially on their ability to learn word meanings incidentally from context. Studies that show large ability effects for learning from context tend to portray less able readers (McKeown, 1985) or younger learners (Werner & Kaplan, 1952) as having such difficulty with using context that one wonders how they have managed to acquire any lan- guage at all.
VoCabuLary instruCtion as a Way to inCrease reading CoMprehension
A second stubborn question about the relationship between vocabu- lary and reading comprehension is, Does vocabulary instruction increase students’ understanding of text? This question was addressed in a series of studies by Isabel and her colleagues on rich and intensive vocabulary instruction (Beck, Perfetti, & McKeown, 1982; McKeown, Beck, Omanson, & Perfetti, 1983; McKeown, Beck, Omanson, & Pople, 1985).
This line of research represents the most thorough examination of the impact of vocabulary instruction on reading comprehension that has been conducted. In addition to the high quality of the instruction used in the interventions, the work was exemplary in its use of multiple dependent measures (including accuracy of word knowledge, fluency of lexical access, and comprehension of text) and in the comparison of different levels of instructional intensity. This series of studies and additional publications that have sprung from it (e.g., Beck, McKeown,
& Kucan, 2002) have been extremely influential.
The main finding from this series of studies was a confirmation of the instrumentalist hypothesis (Anderson & Freebody, 1981)—that is, that vocabulary instruction can in fact increase reading comprehension.
Some might take these results as having put the final nail in the coffin of Urbana talk about the “futility” of vocabulary instruction. However, in the world of research, it is the sworn duty of colleagues to disagree with each other, and we in Urbana did not shrink from our duty.
As I mentioned earlier, in the debate over learning vocabulary from context during reading, Urbana researchers came up with find- ings that were challenged as being statistically significant but educa- tionally irrelevant. In the debate over the effects of vocabulary instruc-
The Word Games 81 tion on reading comprehension, it could be said that the roles were reversed.
One of the critiques of the vocabulary studies of Beck and her col- leagues was that the intensity of instruction was higher than could (or at least would) be maintained in classrooms. In the first two studies (Beck et al., 1982; McKeown et al., 1983), there were approximately 22 minutes of instruction for each of the 104 words taught (averaging across words in the “some” and “many” conditions, the latter taking part in at least twice as many instructional encounters).
In a third study—perhaps in response to this problem—the amount of time per word was less: 15 minutes per word averaged across the high- and low-encounter conditions (12 and 4 instructional encounters, respectively). It is likely that the high-encounter words received about 20 minutes of instructional time per word and that the low-encounter words received less than 10. However, the low-encounter condition did not produce significant gains in reading comprehension, despite the high quality of instruction. Thus these studies demonstrated that vocabulary instruction can improve reading comprehension, but there is no clear indication that this can be done with less than 20 minutes of instruction per word.
A related criticism is the small number of words that could be cov- ered by instruction of this intensity. The first two studies covered 104 words. Furthermore, most of these words were relatively low in fre- quency or else (as in the case of fast meaning “to abstain from food”) represented a rather low-frequency meaning of a common word. If one adds up the frequencies of all these words and includes all their mor- phological relatives, the total is less than 1,000 occurrences per million words of text. That means that after more than 37 hours of high-quality vocabulary instruction spread out over 6 months, students would know one more word per thousand words of text that they read than would comparable students who had not received this instruction.
It seemed unlikely to us in Urbana that knowing one additional word out of a thousand words of text would result in a measurable gain in reading comprehension. If vocabulary instruction were to result in gains on standardized measures of reading comprehension, the effect might be due to generalization of the instruction to knowledge of unin- structed words or to gains in metalinguistic awareness produced by the instruction (Nagy, 2007).
In fact, the initial study (Beck et al., 1982) did show generalized effects, with the treatment group experiencing greater gains on stan- dardized measures of both vocabulary and reading comprehension.
However, such generalization was apparently not examined in two sub- sequent studies (McKeown, Beck, Omanson, & Perfetti, 1983; McKeown, Beck, Omanson, & Pople, 1985).
Another possible critique of the Pittsburgh vocabulary instruction studies concerned the type of texts used. Reading comprehension was measured in terms of recall, and the texts being recalled were written specifically to incorporate the words that had been targeted in instruc- tion. It is likely that these texts involved a higher density of low-frequency words and the use of low-frequency words in more crucial positions than would texts written for other purposes. These studies thus do not demonstrate that high-quality vocabulary instruction would produce comprehension gains for normal text.
Where are We now?
The Pittsburgh vocabulary instruction studies are an example of high- quality vocabulary instruction and a powerful demonstration that the instrumentalist hypothesis is, at least under some circumstances, true.
However, maintaining my stance of collegial skepticism, I would argue that the impact of vocabulary instruction on reading comprehension is another case in which vocabulary has proved to be a very stubborn field to research. There have certainly been some very promising stud- ies, and the Stahl and Fairbanks (1986) meta-analysis certainly comes down on the instrumentalist side. But published studies since then showing an impact of vocabulary instruction on reading comprehen- sion are rare. Mary Beth Curtis reports vocabulary interventions that have led to substantial gains on standardized measures of reading comprehension (Curtis & Longo, 2001), but the recent large-scale pub- lished studies I am aware of either did not use standardized measures of comprehension (Carlo et al., 2004) or did not find significant effects for such measures (Foorman, Seals, Anthony, & Pollard-Durodola, 2003).
Furthermore, even when there is some evidence of the impact of teaching words on generalized measures of comprehension, I don’t think we have any clear idea of what the mechanism is. Is it because students became familiar with key high-utility academic vocabulary? Is it because they have become more familiar with how