Reading in a Foreign Language April 2008, Volume 20, No. 1 ISSN 1539-0578 pp. 92–122 Vocabulary recycling in children’s authentic reading materials: A corpus-based investigation of narrow reading Dee Gardner Brigham Young University United States Abstract Fourteen collections of children’s reading materials were used to investigate the claim that collections of authentic texts with a common theme, or written by one author, afford readers with more repeated exposures to new words than unrelated materials. The collections, distinguished by relative thematic tightness, authorship (1 vs. 4 authors), and register (narrative vs. expository), were analyzed to determine how often, and under what conditions, specialized vocabulary recycles within the materials. Findings indicated that thematic relationships impacted specialized vocabulary recycling within expository collections (primarily content words), whereas authorship impacted recycling within narrative collections (primarily names of characters, places, etc.). Theme-based expository collections also contained much higher percentages of theme-related words than their theme-based narrative counterparts. The findings were used to give nuance to the vocabulary-recycling claims of narrow reading and to more general theories and practices involving wide and extensive reading. Keywords: narrow reading, vocabulary, themes, registers, authorship Over the past 30 years, a large body of literature has touted reading as the major source of students’ vocabulary development (e.g., Cunningham & Stanovich, 2003; Krashen, 1989, 1993a, 1993b; Nagy & Anderson, 1984; Nagy & Herman, 1985, 1987). This claim has also received some empirical support from studies that have found small, incremental gains in word knowledge through contextual exposure during reading (reviewed in Swanborn & de Glopper, 1999), as well as studies that have correlated amount of print exposure with large vocabulary differences among school-aged children (reviewed in Cunningham & Stanovich, 2003). As a result, wide reading (reading large amounts of “authentic” material) and its more robust conceptualization, extensive reading, have been advocated for expanding the vocabularies of various learners in first-language (L1), second-language (L2), and foreign-language instructional settings (Cunningham & Stanovich, 2003; Day & Bamford, 1998, 2002; Graves, 2006; Krashen, 1989, 1993a, 1993b). At the heart of this issue is the assumption that readers will encounter new (unfamiliar) words multiple times in multiple and varied contexts during extensive reading experiences, eventually http://nflrc.hawaii.edu/rfl Gardner: Vocabulary recycling in children’s authentic reading materials 93 resulting in the “incidental acquisition” of those words (Nagy, 1997; Nagy, Anderson, & Herman, 1987; Nagy & Herman, 1987; Shu, Anderson, & Zhang, 1995). Proponents have also put forth this hypothesis as being the best explanation of how young L1 learners acquire the bulk of their th large vocabularies through the 12 grade, with estimates ranging somewhere between 40,000 (Nagy & Herman, 1987) and 80,000 words (Anderson, 1996; Anderson & Nagy, 1992), depending on what is counted as a word. As appealing as this hypothesis has been in reading research and pedagogy, there remains a relative dearth of research studies that have carefully considered the vocabulary input of children’s authentic reading materials to determine how well, and under what conditions, they do recycle vocabulary, particularly those words that are not from the relatively small pool of high- frequency forms found in most texts (the, of, and, a, take, get, mother, play, etc.). The much larger group of more specialized vocabulary items—to which the words of this study belong— constitutes the bulk of the English word stock (Nation, 1990), thus effectively representing the large-scale vocabulary (e.g., 40,000 to 80,000 words) that can potentially be acquired during the school years and beyond. With this background in mind, the aim of the current study is to extend the earlier work of Gardner (2004), in which he analyzed the vocabulary input of a 1.5 million-word extensive reading corpus, consisting of seven children’s narrative collections (four texts each) and seven grade-equivalent expository collections (four texts each). One of his major findings was that the words children are exposed to during narrative reading are vastly different than those they are exposed to during expository reading, particularly at the more specialized, content-rich levels of vocabulary (i.e., beyond the high-frequency words of the language) where 17,921 of 23,857 word types (72.5%) were either found in narrative texts only or grade-equivalent expository texts only (i.e., zero overlap). Additionally, this lack of sharing of critical word types occurred even though many of the narrative and expository collections were related by common themes—one of two conditions proposed by advocates of narrow reading for improving vocabulary recycling in reading curricula of English as an L2 or a foreign language (e.g., Cho, Ahn, & Krashen, 2005; Day, 1994; Krashen, 1981, 1985, 2004; Schmitt & Carter, 2000), the other being authorship (i.e., using texts written by the same author). Krashen (1985) has articulated these two conditions as follows: If the Input Hypothesis is correct . . . it suggests that narrow input is more efficient for L2 acquisition, that early specialization rather than late specialization is better, that students should be encouraged to read on only one topic at a time, or several books by the same author, in the intermediate stage, and that [L2] students stay on somewhat familiar ground when they first enter the mainstream. . . . In addition, each topic has its own vocabulary, and to some extent its own style; the same can be said for each author. Narrow input provides many exposures to these new items in a comprehensible context and built-in review. (p. 73) The current study examines this assertion from the standpoint of authentic vocabulary input from the children’s reading corpus (Gardner, 2004), considering theme (topic) and authorship, in addition to register, as primary variables of interest in order to tease apart nuances of specialized vocabulary recycling in authentic reading collections. Reading in a Foreign Language 20(1) Gardner: Vocabulary recycling in children’s authentic reading materials 94 At the outset, the potential benefits of narrow reading are recognized to extend beyond vocabulary recycling only (e.g., exposing L2 readers to consistent stylistic and discourse moves of certain authors). However, because vocabulary recycling is a central tenet of this position, it deserves more careful examination. A clearer understanding of the impact of text relationships on vocabulary recycling will serve as a guide for theories and practices in English language education in general, particularly in the areas of wide and extensive reading and vocabulary development. The findings may also prove informative in L1 settings, where the assumed language benefits of theme-based instruction (e.g., Walmsley, 1994) and the known challenges with content-area, nonfiction reading materials (e.g., Bamford, Kristo, & Lyon, 2002; Vacca & Vacca, 1996) have also received a great deal of attention. Why a Focus on Authentic Reading Materials? Before proceeding, it is important to note that the existence of narrow reading and similar approaches (e.g., Dubin, 1986) is largely a result of the linguistic characteristics of authentic reading materials. Such materials, unlike graded readers (e.g., Waring, 2003; Wodinsky & Nation, 1988), basal readers (e.g., Bello, Fajet, Shaver, Toombs, & Schumm, 2003), decodable texts (e.g., Mesmer, 2001), or other linguistically engineered materials, do not intentionally control for the presentation of vocabulary and other language structures. By their nature, authentic reading materials are fairly unpredictable in terms of the language demands they place on readers, as well as the language-learning opportunities they afford. While authentic oral communication is often simplified and repeated in order to achieve the conditions of comprehensible input, the same is not true for most authentic written language, which is made permanent in print, thus removing the author from the reader in terms of both time and space. While modern technology may hold the key to making written text more flexible as a language learning tool (Cobb, 2007; Huang & Liou, 2007), and while such technology has also introduced e-mailing, on-line chatting, and text-messaging with their real-time, two-way communication capabilities, these modes of written communication are vastly different from the linguistically frozen materials of printed school English (novels, trade books, textbooks, etc.). By extension, narrow reading is simply one attempt to deal with this challenge of authentic written input by suggesting that collections of authentic texts written on similar topics or by one author will improve the chances that essential linguistic redundancy will actually take place, or in other words, that readers, especially L2 readers, will be exposed to necessary levels of repetitive, comprehensible input as they move from one text to the next. Essentially, vocabulary learning from extensive reading is very fragile. If the small amount of learning of a word is not soon reinforced by another meeting, then that learning will be lost. It is thus critically important in an extensive reading program that learners have the opportunity to keep meeting words they have met before. (Nation, 1997, p. 15) A clearer understanding of how relationships between authentic reading materials might affect such crucial vocabulary recycling is at the heart of the current study. Reading in a Foreign Language 20(1) Gardner: Vocabulary recycling in children’s authentic reading materials 95 Why a Focus on Specialized Vocabulary? The work of Paul Nation and his colleagues has been instrumental in showing the distributions of vocabulary in authentic written and spoken materials. Table 1 is a repurposing of Nation’s (2001) analysis of the distribution of vocabulary in the American Heritage Intermediate (AHI) corpus (Carroll, Davies, & Richman, 1971), which consists of 5 million running words taken from a random selection of third- through ninth-grade texts. This corpus is particularly important to the current study because it was the primary source for the landmark claims associated with the incidental hypothesis (Nagy & Anderson, 1984) and the call for wide reading in reading instruction (Nagy, Herman, & Anderson, 1985). Table 1. Vocabulary coverage in the American Heritage Intermediate corpus Number of word families Cumulative % of text coverage 10 23.7 100 49 1,0 74.1 2,0 81.3 3,0 85.2 4,0 87.6 5,0 89.4 12,448 95 43,831 99 86,741 100 Note. Adapted from Nation (2001, p. 15). Table 1 shows clearly that a small subset of high-frequency word families (i.e., base forms plus their inflections and transparent derivations, e.g., climb, climbs, climbing, climbed, climber, climbers) covers most of the running words of the AHI corpus. For instance, the top 100 word families cover nearly half (49%) of the running words, and the top 1,000 word families cover nearly three-fourths (74.1%) of the running words. Examples of these high-frequency words include function words (the, of, and, a, to, in, etc.) and high-frequency content words (take, get, said, people, find, water, words, know, etc.), many of which can be found in authentic children’s texts. However, the remaining 85,741 word families in the AHI corpus (86,741 minus 1,000) cover only slightly more than one-fourth (25.9%) of the running words. This means that they repeat much less frequently in general than the 1,000 high-frequency word families. In most cases, however, these less frequent word families characterize a particular text or content area. They are also the words that children are less likely to know, and for which the long-term vocabulary learning benefits of extensive reading are most likely to be realized (nourishment, saturated, tomb, mineral, topographic, prohibition, tomahawk, etc.). Determining how often, and under what conditions, these words actually repeat in collections of authentic reading materials is the focus of this study. Reading in a Foreign Language 20(1) Gardner: Vocabulary recycling in children’s authentic reading materials 96 Linguistic Studies of Vocabulary Recycling in Narrow-Reading Materials Most of the linguistic studies that consider the impact of text-level variables such as theme or authorship on vocabulary recycling have focused on adult-level materials. The findings are nonetheless important to the current study. For instance, Hwang and Nation (1989) performed an analysis of the vocabulary load in running stories from newspapers versus the vocabulary load in unrelated stories, concluding that [A] higher proportion of word families outside the 2,000 [most frequent] words will recur in stories from the same series, thus reading running stories reduces the vocabulary load to a greater extent than reading unrelated stories … [and] running stories provide more repetitions of more words outside the first 2,000 words [italics added] than unrelated stories, and thus provide more favorable conditions for learning vocabulary [italics added] than unrelated stories. (p. 332) The authors also suggested that their findings have implications for other texts besides newspapers, especially in settings of English as a foreign language, where several disparate topics often comprised textbooks. Sutarsyah, Nation, and Kennedy (1994) also found substantial differences in the distribution of vocabulary between a single content text (economics), consisting of approximately 300,000 words, and a corpus of 160 shorter academic texts (from over 15 subject areas), consisting of approximately the same number of words. While the diverse corpus contained a much larger vocabulary base than the single text, the words were mostly of lower frequency. In contrast, “a small number of words that were closely related to the topic of the text occurred with very high frequency in the economics text” (p. 34). Additionally, with the exception of higher general frequency words (from the 2,000 word family list) and a few subtechnical terms common to many disciplines, there was little overlap in vocabulary between the narrower textbook and the broader corpus, leading the researchers to conclude the following: Most English courses make use of a series of unrelated texts. This can increase the vocabulary load of the course enormously. If teachers or course designers wish to avoid this, it is worth considering making the course consist of a few themes so that the texts within a theme bear more relationship to each other and thus make use of a smaller vocabulary. (p. 49) It is important to note that the single text in this particular study was an expository textbook, consisting of a tight theme (macroeconomics) written by one author. Finally, Schmitt and Carter (2000) compared the vocabulary of a series of nine theme-related newspaper stories (the tragic death of Princess Diana) to the vocabulary of nine unrelated stories from the same newspapers, containing the same number of total running words (7,843). The findings indicated that the theme-related Diana stories contained 156 fewer types (different words) for L2 readers to deal with, and repeated those types more often in general than the unrelated stories. This overall trend was also true when content words and proper nouns were examined, leading the researchers to the general conclusion that narrow reading may facilitate Reading in a Foreign Language 20(1) Gardner: Vocabulary recycling in children’s authentic reading materials 97 earlier access to authentic L2 reading materials “by lowering the lexical load required of the learner” (p. 8). The important point for purposes of the current study is that the theme-related texts were tightly related to each other (i.e., death of Princess Diana), whereas the unrelated stories had no connections beyond the fact that they belonged to the newspaper register in general. It is also crucial to note that only five of the content words actually listed in the study (crash, palace, photographers, police, princess—all occurring in the Diana stories) would be considered as specialized vocabulary in the current investigation, as the rest would have been identified as general high-frequency forms (e.g., said, car, pay, school, people, work, time, one, year); in other words, they come from the relatively small pool of general high-frequency forms and are therefore likely to occur in many texts, regardless of the relationships between those texts. While there is no question that continued exposure to such high-frequency forms is essential for building general reading fluency and text comprehension, it is equally clear that they do not represent the types of topic- and content-related words upon which a reader can build an extensive vocabulary. Taxonomy of Textual Relationships To date, very little has been done to formalize the potential relationships between authentic reading materials in terms of how such relationships might affect language sharing and recycling. While popular book-leveling schemes in elementary education (e.g., Fountas & Pinnell, 1996, 1999, 2005) have provided important guidelines for grouping texts according to general linguistic and print characteristics (percentages of higher-frequency vs. lower-frequency words, numbers of morphologically and conceptually complex words, font size, words per line, etc.), they do not address specific vocabulary redundancy that may occur as a result of thematic, authorship, or similar relationships between those texts (e.g., genre and register). Viewed another way, traditional leveling schemes tend to relate two or more texts based on the linguistic demands they place on young readers (i.e., how well such readers will be able to comprehend those texts), not on the potential redundancy of the textual content. Therefore, a book about plants and a book about outer space could both be rated at the same difficulty level, even though there is likely to be very little overlap in the topic-related words of the two texts (e.g., blossom and root vs. star and comet). Likewise, a children’s adventure novel and a children’s trade book about magnets could both be rated as at the same level, depending on their general linguistic and print characteristics. Figure 1 depicts a proposed taxonomy for classifying relationships between authentic texts that could directly impact specialized (topic-related) vocabulary recycling within such materials. The taxonomy is in essence a classification scheme that could be used to more accurately predict the chances that blossom and root or, alternatively, star and comet, will appear in Text 1, Text 2, and so forth. Three primary text relationships are considered in the taxonomy: themes, authorship, and registers. In the case of themes, the primary considerations are twofold: (a) the general presence or absence of thematic relationships between texts and (b) the relative tightness of a given theme. For instance, mummy is a tighter theme than mystery in this study and might therefore be expected to recycle specialized vocabulary more efficiently. In general, Gold Rush Reading in a Foreign Language 20(1) Gardner: Vocabulary recycling in children’s authentic reading materials 98 is a tighter theme than Westward Movement, which is a tighter theme than American History; bees is a tighter theme than insects, and so forth. Specialized Vocabulary Recycling No No themes themes Less vocab Loose Loose themes themes More vocab a Semi-tight Semi-tight NA themes themes vocab Tight Tight themes themes vocab vocab Narrative Expository vocab Multiauthor Multiauthor vocab Narrative Expository Uniauthor Uniauthor Figure 1. Proposed taxonomy of textual relationships for specialized vocabulary recycling in collections of authentic reading materials. a There are no collection possibilities for the two cells on this particular row (Expository Uniauthor and Narrative Uniauthor) because uniauthor creates a potential relationship between the texts, even though they are not related by a content theme. With regard to authorship issues, it has been broadly accepted that text collections written by one author (uniauthor) are more efficient in recycling vocabulary than text collections written by multiple authors (multiauthor). Finally, regarding register issues, the primary consideration has been the differences between the culturally- and socially-oriented vocabulary of narrative fiction (storybooks) and the informationally-oriented vocabulary of expository nonfiction. Reading in a Foreign Language 20(1) Gardner: Vocabulary recycling in children’s authentic reading materials 99 The possible combinations of these variables are depicted in Figure 1 and will subsequently be referred to as the taxonomy of textual relationships. A fourth dimension, content-area, may also have a bearing on vocabulary recycling in text collections. For instance, science-based materials may exhibit different vocabulary characteristics than history-based materials. However, because of the practical constraints of using an existing corpus, this dimension will only be addressed loosely in the current study by analyzing possible differences between the history-based collections under Westward Movement and the science-based collections under Mummy and Mystery. It is clear that relatively little is known about vocabulary recycling as a function of text relationships, especially with regard to authentic children’s reading materials. The current study will more carefully examine this issue by analyzing the specialized vocabulary of several collections of children’s texts (Gardner, 2004) written at approximately the fifth- to sixth-grade level. The following question will be used to focus the analyses: To what extent do specialized words recycle within various collections of authentic children’s reading materials that are related by (a) theme (Mystery, Westward Movement, Mummy), (b) authorship (texts written by different authors vs. texts written by one author), (c) register (narrative fiction vs. expository nonfiction), and (d) the various combinations of (a–c) above? Method and Procedure Constructs of Word, Vocabulary, and Type The terms word, vocabulary, and type are used broadly and interchangeably in this study, and all three are defined conservatively as “unique spellings.” While it is realized that some children may be able to make connections during reading between the morphologically related words of English (e.g., climb, climbs, climbing, climbed, climber), there is growing evidence of disparities in this ability based on children’s individual reading skills (Carlisle, 2000; Mahony, Singson, & Mann, 2000; Singson, Mahony, & Mann, 2000) and the amount of direct instruction they receive in raising their morphological awareness (Carlo et al., 2004; Cunningham, 1998; Stahl & Shiel, 1992). Furthermore, differences in children’s awareness of morphological relationships have been isolated as one of several significant variables predicting early vocabulary acquisition (McBride-Chang, Wagner, Muse, Chow, & Shu, 2005). The fact that many of the studies cited above deal with native English-speaking children or bilinguals suggests that this morphological- awareness problem may be even more pronounced for nonnative children trying to negotiate the complex morphological system of English. In fact, Schmitt and Zimmerman (2002) found that even adult learners of English (university students) struggle to make many morphological connections without explicit help, particularly when derivation is involved. It should also be noted that the definition of word, vocabulary, and type used in this study does not account for multiword items (phrasal verbs, idioms, etc.) or variant meaning for the same Reading in a Foreign Language 20(1) Gardner: Vocabulary recycling in children’s authentic reading materials 100 word forms (homonymy and polysemy). However, the more specialized nature of the words in this study suggests that there will be fewer chances for form-meaning error than if high- frequency words were being analyzed (Ravin & Leacock, 2000). Children’s Thematic Corpus The actual corpus of children’s extensive reading materials comes from Gardner’s (2004) earlier study. Hereafter the corpus will be referred to as the Children’s Thematic Corpus. With the aid of an experienced fifth-grade teacher and a children’s librarian, Gardner established four collections of four texts each for each of three popular themes used in upper elementary education (fifth and sixth grades): Mummy (tight, science-based theme), Westward Movement (semitight, history- based theme), and Mystery (loose, science-based theme). This collaboration resulted in the 12 text collections outlined in Appendix A. A total of 48 texts were used to establish the four collections in each of the three themes: 3 × 4 × 4 (Themes × Collections × Texts). Of the 48 texts in the 12 collections, 27 are from documented (published) thematic units (see the theme unit source key in Appendix A), and 21 were chosen with the expert assistance of the fifth-grade teacher and children’s librarian, based on (a) subjective grade-level readability assessments or readability scores printed on the back covers of several books, (b) thematic fit, and (c) popularity of texts. A narrative and an expository control collection (no thematic or authorship relationships between the texts) were also established with the assistance of the teacher and children’s librarian (see Appendix A). The control narrative collection consisted of four popular Newbery Medal books from four different genres of fiction (science, mystery, adventure, and romance), and the control expository collection consisted of four grade- equivalent informational books from four different content-areas (earth science, political science, life science, and geography-culture). From the perspective of the current investigation, the two control collections could alternatively be viewed as examples of wide reading, whereas the thematic collections would be more appropriately labeled as narrow reading. Preliminary Procedure for Analyzing Vocabulary Scanning. Each of the 56 texts (48 thematic and 8 control) was scanned into the computer using Omnipage text scanning software. Words not able to be scanned because of font and background problems were entered into the computer by keyboard. Each electronic document was then carefully edited to correct the relatively few scanning errors that occurred. Equalization of word counts. For comparative purposes in the current study, each of the electronic texts was reduced to equal chunks of running words as follows: the first 5,000 running words of each text, beginning with the first word on page one. This was done for two reasons: to account for differences in text length, especially between the lengthy narrative texts and the relatively short expository texts at the same grade level, and to allow comparisons of vocabulary repetition within a consistent number of running words that a child could encounter in a normal reading experience. Identification of specialized vocabulary. The texts in each of the 14 collections (12 thematic and Reading in a Foreign Language 20(1) Gardner: Vocabulary recycling in children’s authentic reading materials 101 2 control) were run, by collection, through the Range vocabulary program (Heatley, Nation, & Coxhead, 2002) and sorted into lists of High-Frequency Words and Other Words. The predetermined high-frequency list consisted of words from the first 1,000 word families of the General Service List (GSL; West, 1953), which accompanies the Range program, and which, unlike the second 1,000 GSL word families, have been found to be fairly stable over time (Nation & Hwang, 1995). An additional 108 function words and numerical terms that were not found in the first 1,000 GSL list were also added (e.g., ahead, amid, billion, eighths, during). The Other Words (i.e., not in the high-frequency list) were subsequently identified as being specialized if they appeared in at least three texts of a four-text collection. Hereafter these words are referred to as specialized words, specialized vocabulary, or specialized types interchangeably. These are the words of interest in the current study, because they tend to characterize the content of the various collections of extensive reading materials (e.g., mummy, pyramids, museum, archeologist, buffalo, prairie, investigation). It is crucial to reiterate that these are shared, specialized words, occurring in several different texts of a collection instead of one text only (e.g., Hirsh & Nation, 1992). They are thus more representative of the types of words that children could encounter in an extensive reading program that uses themes, authors, and registers to organize instruction. They also fulfill the well established assumption for successful incidental word acquisition, namely, that children will encounter new words multiple times in multiple and varied contexts within a reasonable time frame. Data Analysis Once the specialized words were identified, three measures of vocabulary recycling were selected for comparison purposes: 1. Total number of specialized types (number of different words occurring in several texts). For instance, the words mummy and prairie would be counted as one type each, even though they might repeat 100 times and 4 times respectively. 2. Total number of specialized tokens (raw frequency counts). For instance, the word mummy would have a token count of 100 in the example above, and the word prairie would have a token count of 4. 3. Total number of specialized types that repeat at least six times (6+). For instance, the word mummy would be counted as one 6+ type in the scenario above (i.e., it repeats at least six times), whereas prairie would not be counted (i.e., it does not occur at least six times). The 6+ figure is a fairly conservative estimate of the number of incidental encounters that is generally necessary for new vocabulary to be acquired during extensive reading (see Zahar, Cobb, & Spada, 2001, for review.) Each of these three measures addresses a different aspect of vocabulary recycling in authentic extensive reading collections. The first gives a general sense of how many different specialized words are drawn together by the relationships between texts or collections. The second provides an indication of how often these different words repeat in general, and the third provides information about specific specialized words that reach repetition levels conducive to incidental Reading in a Foreign Language 20(1)