Abstract
Students’ ability to read complex texts is emphasized in the Common Core State Standards (CCSS) for English Language Arts and Literacy. The standards propose a three-part model for measuring text complexity. Although the model presents a robust means for determining text complexity based on a variety of features inherent to a text as well as considerations outside the text, the grammar used in a text is not an overt component of the model. In this essay, we argue that the grammar of a text—especially, the syntactic complexity of sentences in a text—should be included as an explicit and distinct component in a text complexity model due to the fact that grammar contributes to the meaning of text and grammatical meaning impacts reading comprehension. We summarize findings from linguistics research on academic English to support this argument.
Keywords
Text complexity refers to the level of sophistication and challenge of a reading selection or other type of text. 1 Depending on how it is defined, text complexity may encompass specific aspects of the language of a text, such as the difficulty level of the vocabulary; features manifested through language, such as organizational structure; and features that are more fully cognitive than linguistic, such as the need for the reader to connect new information in the text to his or her existing knowledge. The Common Core State Standards (CCSS) for English Language Arts (ELA) and Literacy in History/Social Studies, Science, and Technical Subjects (Common Core State Standards Initiative, 2010a) emphasize the importance of students’ ability to read complex texts. CCSS does not specify an operational definition of text complexity, but a three-part model for measuring it is presented in Appendix A of the standards. The model is to be used to determine “how easy or difficult a particular text is to read” (Common Core State Standards Initiative, 2010b, p. 4), thereby equating text complexity with readability. The model recommends that text complexity be determined through a combination of (a) analysis of “qualitative dimensions” of a text, (b) quantitative readability measures, and (c) “reader and task considerations.” In this way, CCSS defines text complexity—and readability—as including a variety of features inherent to a text as well as considerations outside the text.
Other than the measure of syntactic complexity and, in a couple cases, other dimensions of grammar that quantitative readability measures entail, the grammar used in a text is not considered an aspect of the text’s complexity according to the model. 2 CCSS Appendix A does note that “grammatical knowledge can also aid reading comprehension and interpretation” (Common Core State Standards Initiative, 2010b, p. 29). And sample analyses of texts in Appendix A mention the Coh-Metrix (Graesser, McNamara, Louwerse, & Cai, 2004) measure of syntax. Grammar is thus acknowledged as relevant, but its exclusion from the model for measuring text complexity, except as an aspect of a quantitative measure, nevertheless suggests a low priority placed on its consideration.
Traditional readability measures, such as Dale-Chall and Flesch-Kincaid Grade Level, employ a single, indirect measure of the syntax of sentences in a text: sentence length as a proxy for syntactic complexity. Although sentence length is used as a proxy for syntactic complexity (e.g., Ravid, Dromi, & Kotler, 2010), McNamara, Louwerse, and Graesser (2002) explain that a key weakness in so doing is that this approach does not account for textual cohesion, which has been shown to impact comprehension. They present these sentences to exemplify the point: “One part of the cloud develops a downdraft. Rain begins to fall.” and “One part of the cloud develops a downdraft, which causes rain to fall” (McNamara et al., 2002, p. 11). The first example has low cohesion because there are no linguistic cues to indicate the causal relationship between the two sentences. This presents a greater challenge to the reader, who must infer the relationship based on adjacency of the two clauses alone. Yet a readability score based on sentence length will indicate that the first example is easier than the second example, even though comprehension of the second example is aided by the verb causes as a cohesive device that explicitly identifies the relationship between the two ideas. Another reason that sentence length is sometimes a problematic proxy for syntactic complexity is that long sentences need not require more complex syntactic structures; sentences can be increased in length by inclusion of prepositional phrases or conjoining simple sentences, for example.
In their study of seven readability tools, Nelson, Perfetti, Liben, and Liben (2012) note that three more recently developed measures—Coh-Metrix Text Easability Assessor, Reading Maturity Metric, and SourceRater, which is now called TextEvaluator 3 —continue the traditional use of sentence length as a proxy for syntactic complexity but add a broader range of linguistic and text features than traditional measures include. Coh-Metrix and TextEvaluator, in particular, take account of grammar in more robust ways. For example, both look at grammatical features of individual words in a sentence, such as part of speech and verb tense, which may relate to syntactic structure of sentences. In addition, TextEvaluator looks at average number of clauses in a sentence, which is a more direct gauge of sentence structure than sentence length, as well as average frequency of prepositions, which indicates number of prepositional phrases per sentence (Sheehan, Kostin, Futagi, & Flor, 2010). The Coh-Metrix construct of syntactic simplicity is based on the average size, or density, of noun phrases in sentences; number of words before the main verb of the main clause in sentences; words that signal logical relationships within a sentence (e.g., if-then conditional); number of constituents in sentences; frequency of passive voice constructions; and similarity of syntactic structure in adjoining sentences (Graesser et al., 2004; Graesser, McNamara, & Kulikowich, 2011). In comparing correlations between the measures provided by the seven studied text analysis tools and independent second estimates of the complexity of texts used in the study, Nelson and colleagues found that the readability tools “that included the broader range of linguistic and text measures produced higher correlations than the measures that [relied exclusively on] word difficulty and sentence length measures” (Nelson et al., 2012, p. 4). 4 This suggests that a more robust measure of the syntactic and other linguistic features of a text contributes to a better measure of the text’s complexity.
Nelson et al. (2012) also found that for texts that were identified as either informational or literary narrative, the readability measures in their study usually produced higher correlations with the independent estimates of text complexity for the informational texts than for the literary narrative texts. This finding points to general differences in the characteristics that typify these two types of text, including their grammatical features. Coh-Metrix and TextEvaluator are in fact designed to account for these differences. Coh-Metrix includes a narrativity measure that is based on grammatical features, such as length of noun phrases, prevalence of pronouns, number of main verbs and adverbs, and syntactic complexity (Graesser et al., 2011). TextEvaluator uses two distinct prediction models depending on whether a text is identified as informational or narrative based on features of vocabulary difficulty, syntactic complexity, academic orientation, and topic development (Sheehan et al., 2010).
In short, Coh-Metrix and TextEvaluator exploit current technology in order to look more broadly and more directly at the impact of grammatical features on text complexity than traditional readability measures. Advances in quantitative measures notwithstanding, we argue that it is also important to consider the grammar of a text from a qualitative perspective because of the fact that grammar contributes to the meaning of text, and grammatical meaning impacts reading comprehension. Linguistics research on academic language—that is, the style of language that is typical of the school context—shows that grammar plays a fundamental meaning-making role in academic texts, with variation by subject area according to discipline-specific requirements (e.g., Fang, 2010; Schleppegrell, 2004, 2014; Veel, 1999; Zwiers, 2014). We propose, then, that academic language research can serve as a source of information to address a gap in the CCSS text complexity model.
In the next section of the paper, we discuss some findings from academic English research that highlight the meaning-making function of grammar.
Academic English Research Findings
There is general consensus that there are different styles or varieties of language use—a business communication has a style and a format that differ from a personal communication; an interaction with a supervisor may have a more formal style than a discussion with a coworker; a novel uses a different style of writing than a technical manual. In the same way, the language of schooling has a distinct style, and in fact, each academic discipline exhibits its own variety of language use, or register, a linguistics term that Schleppegrell (2004) defines as “the configuration of lexical and grammatical resources which realizes a particular set of meanings” (pp. 45–46). Schleppegrell explains the reason for this: Language differs in the discourses of different subject areas due to differences in the epistemologies of the disciplines as well as differences in methodologies and pedagogies. Each subject area of schooling has its own expectations in terms of the genres that students will read and write, and each genre is constructed through grammatical resources that construe the disciplinary meanings. Developing facility with new genres involves learning new lexical and grammatical strategies to fit new tasks and contexts. While each genre has its own register characteristics, each discipline as a whole can also be characterized in terms of the linguistic choices that are typical and pervasive. (Schleppegrell, 2004, pp. 113–114)
Academic language and academic English are the terms commonly used to refer to the features of language that are typical of textbooks, teacher talk, and assessments and that are represented—explicitly and implicitly—in instructional standards. In short, academic English is the register of language that is commonly used in the English-medium scholastic context (Schleppegrell, 2001, 2004). The study of academic English has primarily developed in and informed the field of English learner (EL) instruction and assessment because of the manifest need to identify, support through instruction, and monitor progress in the specific aspects of English that ELs need for success in school (Bailey, 2007, 2012). Describing the relationship between ELs’ understanding and use of academic English and their academic performance, Francis, Rivera, Lesaux, Kieffer, and Rivera (2006) state that “mastery of academic language is arguably the single most important determinant of academic success for individual students” (p. 7). Scarcella (2003) goes further, arguing that “learning academic English is probably one of the surest, most reliable ways of attaining socio-economic success in the United States today” (p. 3).
Academic English theory and research have not traditionally informed subject-area instruction and assessment, perhaps because native speakers of English, who constitute the majority of professionals responsible for subject-area instruction and assessment, are generally not consciously attuned to the language they are using as a distinct variety of English and are thus not fully aware of the particular language demands placed on all students in the school setting (Ernst-Slavit & Mason, 2011; Homza, 2011; Lee, 2011). However, there is increasing acknowledgment that all students can benefit from greater explicitness during instruction about the rules of language (Zwiers, 2014), the styles of language specific to different disciplines (e.g., Bailey, Butler, Stevens, & Lord, 2007; Fang, 2010; Fillmore & Snow, 2000; Quinn, Lee, & Valdés, 2012; Snow & Uccelli, 2009; Van Lier & Walqui, 2012), and explicitness in how language constructs the meanings it conveys (Schleppegrell, 2004, 2014). Explicitness about language forms, meaning, and usage is in fact reflected in the emphasis that the CCSS for ELA and Literacy places on literacy and language (Frantz, Bailey, Starr, & Perea, 2014).
Bailey and Butler (2007) explain that there are “at least two dimensions of potential variation [in academic English]: content-specific subject matter and grade level” (p. 69). Academic English research does not yet provide a comprehensive, systematic picture of the grammatical features that are typical of K–12 academic texts across subject areas and grades, but it does offer insights about variation by subject area and grade. These insights can guide and inform analysis of the impact of grammar on text complexity.
Academic English Variation by Subject Area
Bailey et al. (2007) report on a study of the language used in fifth-grade mathematics, science, and social studies textbooks. The researchers examined 12 text selections per subject area drawn from textbooks by three different publishers (i.e., four text selections from each of three different textbooks per subject area). For math, the researchers chose word problems as the unit of analysis rather than instructional prose in part because the language of word problems in textbooks presumably mirrors the language of word problems encountered on math in-class tasks and assessments. Among the findings of the study in terms of the grammar used in the texts are the following:
Number of instances of both noun phrases and prepositional phrases was comparable across the three subject areas.
There was a higher percentage of simple sentences in the math texts that were part of the study than in the science and social studies texts and, correspondingly, a higher percentage of complex sentences in science and social studies than in math.
Passive voice, although relatively rare, was most commonly used in science and least commonly used in math.
The reported differences in syntactic complexity and use of passive voice across the three subject areas are evidence that different disciplines make use of grammatical resources in different ways in order to construe particular disciplinary meanings. Correspondingly, these findings present a new understanding of how syntax contributes to differential textual challenges in the different disciplines. In terms of a relationship between syntax and text complexity and readability, a higher percentage of complex sentences and the more common use of passive voice constructions suggest that the science texts had the most challenging syntax to read and process, followed by the social studies texts, and then the mathematics texts.
In contrast to Bailey et al.’s (2007) findings, Graesser et al. (2011) found that science texts had the simplest syntax of three subject areas examined. Graesser and colleagues conducted an automated analysis of a corpus that contained 37,520 kindergarten–through–Grade 12 texts, about 85% of which were classified as language arts, science, and social studies/history texts. Using the Coh-Metrix construct of syntactic simplicity, as defined earlier, the researchers found that the language arts texts had the lowest levels of syntactic simplicity (i.e., had the most complex syntax) across all grades, followed by the social studies/history texts, and then the science texts. They propose that writers may use simpler syntax in order to compensate for unfamiliar and challenging subject matter in informational texts, although this explanation does not account for the difference between the science texts and the social studies/history texts, all of which presumably are informational and contain unfamiliar and challenging subject matter.
The differing findings of these two studies highlight at least two issues. First, the data set or corpus that is used to make determinations about subject-area variation in language use could impact findings. Bailey et al.’s (2007) use of word problems to represent math text means that a study that uses math instructional prose may reach different conclusions about language use in math text. Although Graesser et al. (2011) state that the corpus they used “is representative of the texts that a typical senior in high school would have encountered from kindergarten through 12th grade” (p. 228), the sources of the texts in the corpus are not described—so it is not clear, for example, how many came from instructional textbooks versus other sources—and they note that the mean length of the individual text selections in the corpus was only 289 words, which is relatively short. Second, and more importantly to this paper, the focus of the analysis will of course impact findings. Bailey et al. elected to count passive voice verb forms as one of several distinct grammatical features, and they were thus able to report specifically on that feature of the texts in their study. Their finding that passive voice was most common in the science texts in fact accords with other research, usually on adult-level texts, that shows a heavy use of passive voice in science writing (e.g., Leong, 2014). Although Coh-Metrix counts passive constructions, Graesser et al. do not report specifically on frequency of passive voice across the three primary subject areas in the corpus they used because their focus is on generating a measure of syntactic simplicity as defined by Coh-Metrix.
Researchers have been conducting automated analyses of patterns of language use in large corpora of adult-level texts for years (e.g., Biber, Conrad, & Reppen, 1998; Biber, Johansson, Leech, Conrad, & Finegan, 1999). Corpus-based studies of K–12 academic English are rare; the Graesser et al. (2011) study is an exception rather than the rule, and as explained, Graesser and colleagues were not interested in determining patterns of use of individual linguistic features. K–12 academic English analyses tend to be descriptive, and they tend to focus on individual subject areas. Corpus-based research that targets K–12 subject-area variation in use of individual linguistic features is needed. Findings from studies done on college-level corpora could in fact play a role in designing K–12 corpus-based studies, given that a goal of college and career readiness preparation for K–12 students is ultimately the ability to engage with college-level academic language (Frantz et al., 2014). Existing K–12 descriptive studies that focus on the language of individual subject areas are nevertheless useful in showing, as Schleppegrell (2004) explains, the way that particular grammatical resources are used to realize particular disciplinary meanings. Following are some highlights from a select sampling of such work.
The language of mathematics
Research on language use that is characteristic of individual disciplines illuminates discipline-specific differences and suggests reasons for discipline-specific uses of grammatical resources. Two fundamental mathematical operations—measurement and calculation—entail precision. Precision of meaning is achieved through specific, technical meanings of certain words in the math context that differ from their meanings in everyday use. Schleppegrell (2007) explains that conjunctions are often used in precise ways in math. For example, the conjunction if generally indicates possibility of an occurrence or conditions for the occurrence. In math, if is used to set up a problem to be solved, for example, If Allen has three apples and he gives one to Min . . . . 5 There are many examples in addition to conjunctions.
Veel (1999) notes that mathematics commonly uses long, dense noun phrases. For example, an eighth-grade student learning about the Pythagorean theorem might encounter a noun phrase such as the perimeter of a right triangle whose two shorter sides are three inches and four inches long. Veel explains that long noun phrases used in math often have this structure:
Quantifiable mathematical attribute (the perimeter of) of something to be measured or analyzed
Classifying adjective (right) that describes the object to be measured or analyzed
Class of object (triangle)
Additional qualifying information (whose two shorter sides are three inches and four inches) further describing the object
This grammatical structure serves to efficiently encapsulate three aspects of a quantification—mathematical attribute, (mathematical) classificatory information, and descriptive details—that involves measurement or calculation. It also arguably presents a more complex reading processing challenge, although this needs to be empirically verified.
Veel (1999) also lists grammatical metaphor as a prominent feature of math texts. Nagy and Townsend (2012) explain grammatical metaphor as when a part of speech is used with a meaning not prototypical of that part of speech. Typically, nouns represent persons, places, or things; verbs represent actions; and identifiable agents (e.g., people) perform actions. However, in grammatical metaphor, nouns can represent complex processes, and abstract concepts can “perform” actions. (p. 94)
The nouns estimate and height in the phrase an estimate of the building’s height are examples of grammatical metaphor. The noun estimate is used instead of the verb estimate, which would have more congruently expressed the fact that an action is required, and the noun height is used instead of the adjective tall, which would have more congruently expressed the quality to be estimated. This example demonstrates one reason that, according to Veel, grammatical metaphor is used in math—one can create “quantifiable entities for the purposes of calculation” (Veel, 1999, p. 194), similar to the purpose for using long, dense noun phrases. Nagy and Townsend argue that grammatical metaphor “presents the most significant issue for students” (p. 94) when it comes to learning the features of academic language that are new and different from the features of social-interactional language.
The language of science
Schleppegrell (2004) explains that science theorizes about and builds people’s experience of the world. Within this broad purpose, a key feature of the language of science is an impersonal, authoritative tone (Fang, 2010; Schleppegrell, 2004). Extensive use of technical vocabulary contributes to this tone. In addition to technical vocabulary, Fang identifies three grammatical resources that are used to create the tone:
Declarative sentences present information both as factual and in an assertive manner, for example, Earth’s atmosphere causes stars to appear to twinkle.
Passive voice foregrounds concepts, conclusions, phenomena, processes, and so on while suppressing whatever human agency may lie behind those things, thereby presenting information as objective rather than as impacted by human subjectivity, for example, Mass is defined as the amount of matter contained in a physical object. The fact that this definition of mass is a product of a human decision made for the purpose of scientific inquiry is obscured by use of passive voice.
Long, abstract noun phrases that serve to encapsulate complex concepts, phenomena, processes, and so on present information in a way that seems detached or distant from our everyday experience, for example, the consumption of different nutrients depending upon what is available.
Passive voice and long noun phrases arguably present a greater reading processing challenge, although this assumption, like others stated earlier, needs empirical verification.
Grammatical metaphor is used in science texts as well. A particular type of grammatical metaphor common to science allows for logical relationships to be expressed within a clause instead of through the use of a conjunction joining two separate clauses (Fang, 2010; Schleppegrell, 2004). This results in a more efficient presentation of information. For example, the verb require in the single-clause sentence Electric motors require energy to operate stands in for the conjunctive phrase only if in the two-clause sentence Electric motors operate only if they receive electricity and obviates the need for the pronoun they and the verb receive, both of which are artifacts of the use of a second clause.
The language of history/social studies
Schleppegrell (2004) states that the broad purpose of history is to interpret experience. Within this purpose, events, institutions, groups, ideas, artifacts, and places are prominent focuses. Although individual human actors obviously play critical roles in historical events, Schleppegrell explains that human actors are often subsumed in history texts within representations of larger historical entities expressed as noun phrases—events (the Vietnam War), institutions (the Nixon administration), groups (antiwar activists), ideas (the spread of communism), artifacts (the Paris Peace Accords), and places (Hanoi, Washington)—which become the actors in the texts, for example, Toward the end of the 1960s, the Nixon administration’s continuation of the previous administration’s efforts to stop the spread of communism in Indochina met with increasing opposition from antiwar activists. In this way, history is similar to science in the way that it encapsulates complex concepts, phenomena, processes, and so on in noun phrases. Understanding that noun phrases often represent active entities within historical scenarios may present a comprehension challenge to children who are accustomed to thinking about the world in terms of individuals acting autonomously.
Cause-effect relationships and temporal relationships are central to history texts. As in science texts, logical relationships may be achieved within clauses rather than between clauses through conjunctions (Schleppegrell, 2004). In the example sentence in the previous paragraph, the verb phrase met with increasing opposition from antiwar activists presents the effect caused by the Nixon administration’s policy in Vietnam in a way that is more abstract and arguably more challenging to comprehend than if the relationship were expressed more explicitly by using a conjunction, such as because, as in the sentence Antiwar activists increased their opposition because the Nixon administration continued the previous administration’s efforts to stop the spread of communism in Indochina.
Academic English Variation by Grade Level
Regarding grade-level variation, Anstrom et al. (2010) explain that academic English “is developmental with trajectories of increased sophistication in language use from grade to grade” (p. v). The academic-English literature includes no studies designed specifically to identify grammatical developmental trajectories at different grade levels, so this information must generally be pulled from separate studies focused on different grades. As Nagy and Townsend (2012) point out, work in this area suggests that textbooks become more academic—that is, use increasingly complex academic language—as grade level increases, “but little data are available from empirical studies on the specific language demands of the various grade levels” (p. 103). This is a gap in the literature. Studies that target grade-level developmental trajectories are needed to complement the research that describes academic-English variation by subject area.
Research in psycholinguistics and in developmental psychology that looks at child language acquisition presents findings on syntactic development that is typical by age. For example, Nippold, Hesketh, Duthie, and Mansfield (2005) examined syntactic development in native-English-speaking 7-year-olds to 49-year-olds in terms of both conversational and expository discourse. They explain that a key marker of syntactic development among school-age children is increasing use of dependent clauses to combine into complex sentences ideas that young children tend to express in strings of independent clauses. In their study, they found that production of relative clauses in expository discourse, in particular, was one of the best indicators of age-related growth in syntax. Dick, Wulfeck, Krupa-Kwiatkowski, and Bates (2004) studied interpretation of active and passive constructions and subject and object clefts with and without subject-verb agreement cues by native-English-speaking 5- to 17-year-old typically developing children, language-impaired children, and children who had experienced early brain injuries. Among their findings, Dick and colleagues found that typically developing children were slower and less accurate in responding to the more complex and less familiar (i.e., noncanonical) sentence structures, which was as expected. They also found very gradual increases in mean responses across age, which was consistent with the authors’ theoretical standpoint as expressed by the competition model, a connectionist model for information processing that suggests the importance of frequency at which children encounter sentence structure types and availability of linguistic cues for sentence structure types for acquisition.
These studies and others in this vein look at age-related processing and production of syntax rather than age-related exposure to syntax in academic texts.
Implications of the Research for Instruction and Assessment
Empirical research aimed at describing academic English confirms it as a distinct register with register-specific lexical, grammatical, and discourse features. Further, some linguistic features may be shared across academic subject areas, whereas others are discipline specific. Discipline-specific linguistic features support the particular ways in which a discipline construes knowledge. In addition, the research points to variation of linguistic features across grades, suggesting developmental trajectories. More research is required to elaborate both subject-area variation and the developmental trajectories of particular features, especially, grammatical and discourse features.
In terms of the specific focuses of this paper—syntactic complexity and text complexity—academic-English research clearly indicates that grammatical complexity, and syntactic complexity specifically, is an important aspect of the language of a text, and we argue, on the basis of the existing body of research, that syntactic complexity should be an explicit and distinct component of a text complexity model. This has implications for both instruction and assessment.
Teachers are a critical conduit through which students can learn academic language, including how syntax shapes meaning in texts. A teacher’s lack of knowledge about the language learning needs of ELs or lack of understanding of the different meaning-making resources of a vernacular dialect of English spoken by other language-minority students can negatively impact academic success for these students (Delpit, 1995; Philips, 1993; Schleppegrell, 2004). Conversely, a teacher’s understanding that some students face a greater challenge in learning the language typical of school (i.e., academic language), including syntactic forms conventionally used in discipline-specific ways in texts, can positively impact academic success for these students (Achugar & Schleppegrell, 2005; Bunch, 2014; Fang, 2010; Fillmore & Snow, 2000; Zwiers, 2014). Snow and Uccelli (2009) argue that promotion of academic language development is in fact “a crucial task for educators of all students” (p. 114). However, teachers must themselves understand what academic English is. This is a responsibility for teacher education and professional development programs.
In terms of assessment, we have argued elsewhere (Frantz et al., 2014) that using item and test specifications that are informed by academic-English research for the development of a K–12 English language proficiency test that includes or focuses on academic-English measurement can contribute to a validity argument for that test. We extend that argument in this paper to propose that academic-English research should inform passage specifications in future test development in other subject areas. Specifically, what the academic-English research says about discipline-specific and grade-appropriate syntactic complexity should inform development of test passages that students will read in order that these passages be aligned with the types of linguistic features that students encounter during instruction. Attempting to mirror instructional text language in assessment text language will contribute to a validity argument for a test.
Finally, we would like to stress the need for additional research in K–12 academic English. Although existing work that describes the ways in which particular grammatical resources are used to realize particular disciplinary meanings in K–12 texts is useful, there is clearly a need for more robust research—ideally, corpus based—that targets K–12 subject-area variation and grammatical developmental trajectories at different grade levels.