Background knowledge is a reader's understanding of the specific concepts, situations and problems associated with the words encountered in the text. Knowledge of the topic provides readers enough understanding to make meaning and build onto what they currently know. Background knowledge may also be referred to as content knowledge. It is closely tied to vocabulary, an understanding of word meanings, however, it goes beyond to mean understanding the circumstances, situations, or ideas that impact comprehension of the reading.
Prior knowledge differs slightly from background knowledge. Prior knowledge includes the experiences, understandings, skills, and abilities children bring to the learning process, including cultural and language knowledge [Hennessy, 2021]. These understandings are essential foundations for future learning as learners connect those pieces of information to new learning. Essentially, children build upon prior knowledge to develop a depth of background knowledge in specific areas.
Authors of text assume readers have enough background knowledge to access the meaning to work with the text. Therefore, as children read increasingly more complex text, their background knowledge will help them comprehend and develop further understanding. This is because "...knowledge is not only cumulative, it grows exponentially. Those with a rich base of factual knowledge find it easier to learn more - the rich get richer." [Willingham, 2006]. In some research studies, having some background knowledge of the text topic was more valuable to comprehension than strategies such as visualizing, summarizing, predicting, or making inferences. In fact, having some background knowledge helps readers use strategies to make sense of their reading.
Classroom teachers contribute towards building children's content knowledge through read-alouds, first hand experiences, classroom conversation, and interactions with all types and genres of text across the curriculum. Meaningful learning experiences in all curricular areas contribute towards children gaining knowledge about their world to apply to unfamiliar texts.
While prior knowledge of a passage topic is known to facilitate comprehension, little is known about how it affects word identification. We examined oral reading errors in good and poor readers when reading a passage where they either had prior knowledge of the passage topic or did not. Children who had prior knowledge of the topic were matched on decoding skill to children who did not know the topic so that the groups differed only on knowledge of the passage topic. Prior knowledge of the passage topic was found to significantly increase fluency and reduce reading errors, especially errors based on graphic information, in poor readers. Two possible mechanisms of how prior knowledge might operate to facilitate word identification were evaluated using the pattern of error types, as was the relationship of errors to comprehension. Implications of knowledge effects for assessment and educational policy are discussed.
Beginning with Bransford and Johnson's [1972] seminal study, many researchers have shown that having some prior knowledge about the topic of a passage enables both greater comprehension of the text and better memory for it [McNamara & Kintsch, 1996; Rawson & Van Overschelde, 2008; Recht & Leslie, 1988; Spilich, Vesonder, Chiesi, & Voss, 1979]. The effects of prior knowledge on comprehension are so pronounced that researchers have advocated including assessments of prior knowledge in comprehension batteries [Johnston, 1984; Pearson & Hamm, 2005], and Hirsch [2006] designed a reading curriculum that aims to improve comprehension scores by establishing a core of common knowledge for all students.
In contrast, there has been very little research investigating whether prior domain knowledge might also improve word identification. Logically, it is possible that part of the benefit of prior knowledge on comprehension may be because when one has topic knowledge, then the words can be identified more readily. Accurate word identification is a strong predictor of comprehension [Fuchs, Fuchs, & Compton, 2004; Pinnell et al., 1995], and deficits in fluency are thought to be at least partly responsible for comprehension problems for many students [Duke, Pressley, & Hilden, 2004]. Indeed, interventions targeted at improving fluent word identification in struggling readers can improve comprehension [Chard, Vaughn, & Tyler, 2002; National Institute of Child Health and Human Development, 2000; Nicholson & Tan, 1999; Stahl & Heubach, 2006; Tan & Nicholson, 1997]. So, if it can be shown that differences in prior knowledge lead to differences in how words are identified within a passage, then it would provide further insights into the processes of word reading and comprehension and how one might improve both. We know that attentional resources are limited [LaBerge & Samuels, 1974] and therefore there can be tradeoffs between allocating resources to word decoding and comprehension [Perfetti, 1985]. If prior knowledge facilitates word reading, then more resources could be allocated to comprehension processes rather than to word identification, thus facilitating comprehension.
But is there any evidence that having prior topic knowledge does in fact help word identification? There is plenty of evidence that context facilitates word identification, and undoubtedly context taps into prior knowledge to have its effects, but there can be knowledge differences even when there are no context differences and there is almost no research on how topic knowledge affects word identification.
Beginning with Tulving and Gold's [1963] classic study, it has often been shown that words can be read more accurately and quickly in context [e.g., Nation & Snowling, 1998; Perfetti, Goldman, & Hogaboam, 1979; Stanovich, 1984]. This is especially true for poor readers, who seem to benefit more from the additional semantic activation provided by the context to compensate for their poor decoding skills [Gough, 1983; Nation & Snowling, 1998; West & Stanovich, 1978]. Semantic effects on word identification have even been shown when words are presented in just a single word context or even in isolation [e.g. Nation & Cocksey, 2009; Rodd, 2004; see Keenan & Betjemann, 2007, for a review]. However, even though context must exert its effects by evoking prior knowledge, having context is not the same as having prior topic knowledge. This is evident from the fact that it is possible for readers to differ in the amount of background knowledge they have when reading the same text and thus having the same amount of context available. The question we raise in the current study is whether such differences in the amount of prior knowledge matter for word identification. When reading words in the context of a text passage, is it possible that a reader who has prior knowledge of the passage topic will identify the words in the passage better than another reader who has the same level of word decoding skill but does not have prior knowledge about the topic?
The only study that directly assessed the effects of prior topic knowledge on reading errors in developing readers was done by Taft and Leslie [1985]. They examined oral reading accuracy in third graders by comparing readings of a passage about the food chain from children who attended different schools, where only one of the schools had provided instruction in the food chain as part of their curriculum. They found that the children with prior knowledge of the subject made few graphically similar errors relative to their counterparts who had no knowledge of the topic, and their errors tended to preserve meaning. This was a promising finding about the potential of topic knowledge to facilitate word identification, but to our knowledge, it has not been followed up in subsequent research. However, a study by Malik [1990] of adults learning English as a foreign language provides some corroborating support in that errors were more likely to be semantically acceptable when reading familiar than unfamiliar passages.
Children with better word decoding skill are also often those who have a wide range of domain knowledge. For example, in our own sample, we find that it is quite easy to find poor readers who do not have prior knowledge of assorted passage topics; however, it is much harder to find good readers who do not have prior knowledge. Better word decoding skill appears to go hand in hand with an increased chance of having prior knowledge. This raises the question of whether the effects that Taft and Leslie [1985] observed were due to better word reading skills or more prior knowledge of the passage. Because they did not have an independent assessment of word decoding skill, we thought it important to determine whether knowledge differences would have an effect even when children are matched on word decoding skill across knowledge groups, defined here as performance on reading words in a list.
In the present study we unconfound word decoding skill and knowledge by matching the children in our knowledge groups on word decoding skills and by including both good and poor readers in both the knowledge group and the no knowledge group. The logic behind matching our knowledge groups on decoding skill is as follows. If we know that two children perform equally well at reading words in a list, when we then assess their skill at reading a passage for which they have different levels of domain knowledge, we would expect those children to perform equally on reading that passage if domain knowledge has no effect. If, however, we do find differences in oral reading accuracy on the passage, we can be confident that those differences are not due to decoding skill differences and that it is likely domain knowledge differences that underlie the difference in performance on the passage.
This study examines oral reading accuracy in both typically developing readers and poor readers in order to determine whether the effects of prior topic knowledge depend on the reader's skill level in word decoding. Stanovich's [1984] interactive compensatory model of reading suggests that, just as context has been found to have a greater influence on word identification in poor readers than in those with typical reading skill, so too differences in knowledge about the passage topic may have a greater influence on decoding accuracy in poor readers. In fact, a recent study by Miller and Keenan [2009] showed that prior knowledge effects on comprehension are strongest for poor readers. In this paper we therefore examine knowledge effects both in typical readers and in children who are poor readers.
Using Substitution Errors to Identify Processing Mechanisms Underlying Knowledge Effects
This paper not only examines whether prior knowledge facilitates word recognition, but also how. We assume that word identification consists of a set of neural networks that map orthographic input onto sounds and meaning, as shown in Figure 1 [adapted from Plaut, McClelland, Seidenberg, & Patterson, 1996]. We examine two possible mechanisms that might operate on this model to yield knowledge effects. One is whether prior knowledge facilitates reading by allowing the reader to skip mapping the orthographic onto the phonological units because the word is so predictable based on the child's knowledge, e.g., “hill” in “Jack and Jill climbed up the…” A child familiar with the nursery rhyme would not even need to look at the word, much less process the orthographic information, because their prior knowledge of the rhyme would allow them to generate the correct word. We refer to this process of bypassing orthographic information as the bypass process.
Open in a separate window
Figure 1
Illustration of two hypothesized mechanisms underlying knowledge effects: a] bypass hypothesis, and b] constraint satisfaction hypothesis. Figure adapted from Plaut et al., [1996].
Support for the bypass process would be evident if the child makes a substitution that is semantically similar, but not graphically related; for example, if the text stated “Jack and Jill climbed up the mountain,” and the child said “hill” instead of “mountain.” Although it may seem unlikely that semantic information can overwhelm the graphemic input so as to produce such errors, even highly skilled readers have shown evidence for just this in a phenomenon known as the Moses Illusion. Studies of semantic illusions like the Moses Illusion have participants answer questions like “How many animals of each kind did Moses take on the ark?” Even after reading the sentence out loud, and being told to look out for distorted questions, participants tend to answer “two”, as if “Noah” instead of “Moses” had been included in the question [Erickson & Mattson, 1981; Reder & Kusbit, 1991].
An alternative mechanism by which prior knowledge may facilitate word identification is that it gives the network that maps orthographic to phonological and semantic units a boost, rather than bypassing the mapping, so that it has more information to use in the constraint satisfaction process and can more readily and more accurately settle on a solution. We refer to this process as the constraint satisfaction use of semantic activation from prior knowledge. Although this semantic boost would generally keep one from making errors, evidence for this process would be reflected by a decrease in substitutions that are graphically similar and semantically dissimilar to the original word. Whereas a bypass mechanism leads to substitutions that may not be at all related graphically, like “hill” is to “mountain”, the constraint satisfaction process would lead to fewer substitutions that are only graphically similar, and not semantically similar, for example something like “maintain” for “mountain.” Just as context constrains possible choices for a word, so might having prior knowledge, resulting in fewer miscues that are based solely on graphic information.
It is important to note that both models predict more semantically similar miscues as an outcome of having prior knowledge, due to the added contribution of semantics. The difference in the two models lies in the relative contributions of graphic and semantic information, where the constraint satisfaction model predicts fewer errors based on only graphic information, and the bypass model predicts more errors based on only semantic information.
In order to determine the relative occurrence of these two processing mechanisms, we examined oral reading substitution errors for their graphic and semantic similarity to the original word. This approach is labeled miscue analysis [Goodman, 1969], and has a long history of use, with demonstrated validity and reliability [Murphy, 1999; Sadoski, Carey, & Page, 1999]. It was used by Taft and Leslie [1985] when they explored the effects of prior knowledge; however, because they averaged over semantic similarity when examining graphic similarity, and vice versa, it is not possible to determine from their results the mechanism by which prior knowledge is operating. To do that, we need to do more than just examine overall graphic similarity and overall semantic similarity; we need to classify a substitution on both dimensions simultaneously, as we do in the current study. Table 1 shows the types of substitutions and examples of each kind.
Table 1
Examples of the Types of Graphic and Semantic Errors, Using the Example Sentence: “Jack and Jill climbed up the mountain.”
Type of errorExample
[target word: mountain]Graphically and semantically similar [G+S+]moundSemantically similar but graphically dissimilar [G−S+]hillGraphically similar but semantically dissimilar [G+S−]maintainNeither graphically nor semantically similar [G−S−]apple
Open in a separate window
If prior knowledge allows readers to bypass word identification and “read” what they think the word should be based on their knowledge, then we expect to find that they will show this by having more G−S+ substitutions than readers with no knowledge. Alternatively, if prior knowledge gives a boost to the constraint satisfaction process of mapping, then we may see fewer substitutions that are G+S−. It is common to find that when children who are better readers do make errors, their oral reading errors are more semantically similar than those of poorer readers [e.g., Goodman & Goodman, 1977; Leslie, 1980]. More semantically similar errors is also what Taft and Leslie [1985] reported when children had prior domain knowledge. Again, to dissociate any effects that might be due to better word decoding instead of prior knowledge, we matched our knowledge groups on decoding.
Because we expect that having prior knowledge leads to more automatic word recognition, we will also examine the fluency of good and poor readers with and without prior knowledge. We expect that having prior knowledge will lead to more accurate and faster word identification, particularly for poor readers. In addition, because previous research has demonstrated the importance of studying self-corrections [Beebe, 1980; Sadoski et al., 1999], and because previous research has failed to find a relationship between prior knowledge and self-corrections [Malik, 1990; Taft & Leslie, 1985], we will also examine the rate of self-corrections as a function of prior knowledge and reading ability.
A final aim of this study is to examine the relationship between oral reading accuracy and comprehension to explore whether certain types of errors predict reading comprehension more strongly than the overall number of errors. In particular, based on previous research that shows that poor readers make more miscues that change the meaning of the text [Goodman & Goodman 1977; Leslie, 1980], and given previous findings that semantically dissimilar miscues are more related to comprehension than other types of errors [Pinnell et al., 1995], we expect that semantically dissimilar substitutions may be the most strongly related to comprehension.
In sum, the purpose of the present study is to investigate the mechanisms by which prior knowledge might facilitate word identification. By being able to match our participants on age and word decoding skill so that they only differ in their prior knowledge of the passage topic and by examining both fluency and the types of oral reading errors, we can determine whether and how knowledge facilitates word identification. By examining knowledge effects in both typical readers and in poor readers, we can also determine whether there is differential facilitation on oral reading for poor readers with prior knowledge. Finally, we examine the relationship between oral reading errors and comprehension to determine if the type of oral reading error, or the total number of errors, is more predictive of reading comprehension; if a specific type of error is more predictive, then it would suggest that the extra effort required to classify error types is warranted.
Method
Participants
The data analyzed for this study were previously collected as part of a language comprehension assessment battery in a larger ongoing research project conducted by the Colorado Learning Disabilities Research Center [c.f. Keenan et al, 2006; Olson, 2006]. The sample for the present study was much smaller than the full sample, which consisted of over 130 children in the 4th grade, because we had to match the participants in the prior knowledge group with children in the no prior knowledge group on single word reading. Because we were doing this matching for both poor readers and for controls and because there is a lower rate of poor readers who had topic knowledge and controls who had no prior knowledge of the topic, we were fortunate to have a large number of children to select from so that we could closely match our knowledge groups on decoding skill. For this study the participants consisted of 60 4th graders [33 female], with a mean age of 9.7, originally recruited to the larger study either on referral for a reading disability [RD] or as controls.
Thirty participants comprised the prior knowledge group, with 15 being poor readers and 15 good readers. The no prior knowledge groups also had 30 participants who were matched to the individuals in the knowledge group on word decoding skill. The method for defining knowledge groups and reading abilities groups is given below.
Table 2 presents descriptive statistics on age and word decoding skill for the four groups, as well as vocabulary skill and listening comprehension. Decoding skill was measured with two tests, the Timed Oral Reading of Single Words [Olson, Forsberg, Wise, & Rack, 1994] and the Peabody Individual Achievement Test [PIAT] word recognition subtest [Dunn & Markwardt, 1970]. Each raw decoding score was standardized across the larger sample, and these were combined into a composite in order to have a more reliable measure. The poor readers had an age-adjusted composite decoding z-score of −1 or below. The controls all had word decoding composite z-scores above zero, indicating above average word decoding skills. The matching of individuals in the prior and no prior knowledge groups on word decoding was evaluated by t-tests that revealed no significant differences in decoding skill between the knowledge groups either for poor readers [t [28] =. 12, p =. 91] or for good readers [t[28] =. 46, p =. 65].
Table 2
Average Age, Word Decoding, Listening Comprehension, and Vocabulary Skill for Participants
Poor Readers
No Prior
Knowledge
[n = 15]Poor Readers
Prior
Knowledge
[n = 15]Controls
No Prior
Knowledge
[n = 15]Controls
Prior
Knowledge
[n = 15]Age [years]9.62 [.41]9.72 [.53]9.80 [.53]9.60 [.44]Word Decoding
[composite z-score]−1.54 [.56]−1.52 [.32].49 [.36].55 [.34]Piat Word
Recognition Raw
Score36.60 [6.14]36.87 [3.52]50.27 [3.97]51.60 [4.55]Timed Oral Reading
of Single Words67.87 [22.71]67.80 [16.40]119.47 [17.36]118.67 [17.97]Woodcock-Johnson
Oral Comprehension
Raw Score19.27 [4.23]19.13 [4.67]21.40 [3.02]22.47 [2.13]WISC Vocabulary
Raw Score26.27 [7.25]27.87 [4.98]32.20 [3.97]33.80 [4.35]
Open in a separate window
Vocabulary was measured using the vocabulary subtest of the Wechsler Intelligence Scale for Children, 3rd. ed. [WISC-III; Wechsler, 1991], and listening comprehension was measured using the oral comprehension subtest of the Woodcock-Johnson Tests of Achievement [WJ-III; Woodcock, McGrew, & Mather, 2001]. Interestingly, although readers were not explicitly matched on either listening comprehension or vocabulary, t-tests revealed no significant differences on these variables between those who had topic knowledge and those who did not. For listening comprehension there were no significant differences between the knowledge groups either among the poor readers [t [28] =.08, p =.94] or among the good readers [t [28] =1.12, p =.23], and likewise for vocabulary, among poor readers [t [28] =.71, p =.49] or among good readers [t [28] =1.1, p =.30].
Materials and Background Knowledge Assessment
This study used a 263-word expository passage, titled “Amelia Earhart” from the Qualitative Reading Inventory [QRI; Leslie & Caldwell, 2001]. Although the children were tested on other QRI passages as well, they were not suitable to testing our hypotheses because either all children knew about the topic [e.g., octopus] or none knew anything [e.g., Andrew Carnegie]. The Amelia Earhart passage provided a sufficient variety in children's background knowledge because some children knew about her and some children had never heard of her. Topic familiarity was assessed by a concept question that accompanies the passage in the QRI manual; it always preceded the child's reading. It was scored either as a 0 [no prior knowledge of the topic] or 1 [some prior knowledge of the topic]. Reliability of assessing knowledge was assessed by having multiple examiners score a subset of concept questions, and there were no discrepancies among the ratings of knowledge. Previous research using this method of knowledge assessment on this same passage [Miller & Keenan, 2009] showed it to have sufficient sensitivity to detect knowledge differences that significantly impact comprehension. Given that it has established sensitivity, it seemed preferable to assess knowledge with it rather than more extensive questioning because it avoids the potential problem of telling the child, by way of multiple knowledge assessment questions, what the text is about before they even read it.
Procedure
Before the child read the passage, the tester asked the child a background knowledge question, “Who is Amelia Earhart?” to gauge how much the child knew about the topic of the passage. Answers such as “A pilot” or “The first woman to fly around the world” would be scored as demonstrating prior knowledge, whereas incorrect responses or “I don't know” would be scored as demonstrating no prior knowledge. The child then read the passage orally, while the examiner scored the passage online for rate and accuracy. The child then provided a free recall of the passage. After the testing session, the recalls were scored according to the number of idea units recalled from the idea checklist that accompanies the test. A subset of recalls were scored by multiple raters, and inter-rater reliability of this subset was very high [Cronbach's α = .97]. The readings were transcribed and coded according to the taxonomy described in the next section.
Substitution Error Scoring
Any substitution that was fully sounded out was counted as a substitution error. The taxonomy outlined here is based on the one used by Taft and Leslie [1985]. Table 1 gives examples of each type of substitution.
Graphically Similar [G+]
Substitutions were classified to the degree to which they were graphically similar to the target word by assigning 1 point for each letter shared by the substitution and the target word, and in the same relative order in both words. After totaling the points, they were divided by the number of letters in the target word. Results greater than .5 were judged to be graphically similar, and results less than .5 were judged graphically dissimilar1.
For example, if a child is reading the word mechanical and produces michinical, the substitution would be assigned a graphic similarity score of .80 [8 shared letters, divided by the total number of letters in the word]. A substitution of machine, in contrast, would receive a score of .40 and would not be judged graphically similar. Using Visual Basic and Microsoft Excel, a macro was created which automatically calculates the graphic similarity of each substitution [Priebe & Keenan, 2009].
Semantically Similar [S+]
While we would have preferred to use a quantitative measure of semantic similarity, previous attempts to do so by using latent semantic analysis [LSA; Landauer, 2002] resulted in very low agreement with human ratings of semantic similarity [Priebe & Keenan, 2009]. Therefore, we relied on human ratings of semantic similarity for this study. Substitutions were categorized according to the degree to which they changed the intended meaning of the author of the passage, by reading the sentence up to the substitution and determining if the resulting sentence retained the author's intended meaning for the passage. For example, if a child substituted “inventor” in the sentence “Amelia Earhart was an adventurer”, the substitution would be judged as semantically dissimilar since it changed the intended meaning. A substitution of “adventuresome”, in contrast, would be judged semantically similar.
Other Types of Errors
The number of substitutions, omissions, insertions, repetitions, and skipped lines were summed to created an overall number of errors for each participant. Self-corrections were defined as any error that was subsequently successfully corrected by the child, and these were also summed to create an overall number of self-corrections for each participant.
Reliability of Error Scoring
Intra-rater reliability was assessed through use of BBEDIT, a text-editing software that allows multi-file searches. Each coded word in a transcript was checked against all other occurrences of that word in the remaining transcripts, ensuring that all miscues were categorized in the same way. Inter-rater reliability was obtained by having a trained second observer code 10% of the transcripts for the above categories. Reliability between observers was quite high for all categories, with Cronbach's alphas of .9 for all categories except for substitutions that were semantically dissimilar [Cronbach's α =. 80].
Results
Comprehension
A 2 × 2 analysis of variance [ANOVA] was conducted with prior knowledge [0,1] and reading ability [Poor Readers, Controls] as the between-participants independent variables and the total amount of idea units recalled as the dependent variable. Controls recalled significantly more of the text [M=19.47] than poor readers [M =16.10; F [1, 56] = 5.19, p