Thursday, 26 January 2012

The Trouble With Vowels

Vowel Trouble: Revisiting the Difference between Consonants and Vowels in a Phonics-Based Approach to Beginning Reading

Abstract: Drawing on a survey of the teaching practices of more than 100 randomly selected reading teachers and specialists from across the U.S., this article considers the implications for reading instruction of the gap between consonant- and vowel-based pronunciation guidelines. We outline some hurdles faced by decoding-based approaches to reading and argue that phonics-based instruction needs to address the well-established discrepancy between vowels and consonants as guideposts for beginning readers. Our findings reveal that, despite the attempt of current approaches to avoid the use of inaccurate phonics generalizations, these remain a staple of classroom instruction, most likely because of the perceived need to provide cues for reading words out of context and because of the unquestioned assumption that if children are taught consonant sounds they must also be taught vowel sounds. We conclude by outlining a phonics-based alternative to decoding, as it is currently understood, and to approaches that either dispense with phonics instruction altogether or attempt to integrate it, as a whole, into hybrid approaches.

Vowel Trouble: Revisiting the Difference between Consonants and Vowels in a Phonics-Based Approach to Beginning Reading

This article starts with an indisputable fact and a potentially controversial challenge to the way in which phonics is currently conceived in contemporary teaching practice. First, the indisputable (if still controversial) fact: that the most recent large-scale initiative to improve reading instruction in the United States has proven unsuccessful. As the Reading Teacher noted in late 2008, “One of the largest and most rigorous reading studies ever undertaken by the U.S. Department of Education, found that the six billion spending for Reading First has helped more students ‘crack the code’ to identify letters and words, but it has not had an effect on reading comprehension among first and second and third graders in participating schools” (Manzo, 2008, p. 1). This report is striking not just in noting the failure of Reading First, but in highlighting an apparent disconnect between decoding skills and reading comprehension. The three-year study noted that despite the fact that first graders in the program scored significantly higher on tests of decoding skills than peers in comparison schools, the program left proficiency levels largely unchanged, at dismal levels: “Fewer than half of the first graders and less than 40 percent of second and third graders showed grade-level proficiency in their understanding of what they read” (Manzo, 2008, p. 1). The report revealed that the program increased both the instructional time and the amount of professional development received by teachers in decoding instruction without, however, affecting the ability of children to comprehend what they read. G. Reid Lyon, who helped draft the Reading First legislation, criticized the study for not looking at “which kids did respond to Reading First style instruction, and why. And which kids did not respond and why” (Manzo, 2008, p.16).

The Department of Education study is not the only one to have indicated potential problems in the form that decoding-based reading instruction has taken. In a study of a special reading intervention that attempted to teach children decoding rules that would help them pronounce words letter-by-letter, McCandliss (2003) noted that, “…gains in Word Identification raw score were minimal and standard scores actually document a slight decrement in performance over time relative to the sample of children used to construct the norms…” (p. 96). He offered as one quite reasonable explanation for this finding the fact that the word identification test used to measure reading achievement, “contains a high proportion of irregular or inconsistent words that cannot necessarily be read more accurately by enhancing grapheme-phoneme decoding skills…approximately 60% of these items were nondecodable using the grapheme-phoneme elements trained in the 77 lessons of the intervention.”  (2003, p. 96). One of the problems of teaching inaccurate generalizations is that they lead to an expectation of regularity that is defeated by an encounter with actual words. Hebert et al. describe the challenges posed by words that do not follow the rules as peculiar to “less skilled readers” who “display a tendency to regularize the words – for example, to pronounce deaf as “deef” or touch as “towch,” (as described in Adams, 1990, p. 172). The difficult (to put it mildly) challenge faced by beginning readers is to know when not to use the generalizations they have been taught without knowing what word they are looking at.
This article proposes an alternative explanation for the significant research finding that even when decoding skills are successfully taught, reading skills do not improve, and it explores some logical challenges to the way decoding is currently conceived. Then, drawing on original research about teaching practices in the classroom, it argues for the importance of distinguishing between the utility of vowel- and consonant-based phonics generalizations as reliable cues for decoding. The first section outlines the problem with vowels as a guide to beginning readers and the second section explores how these problems relate to existing reading teaching practices in the U.S. and, to the extent that phonics instruction is used, in the U.K. 

The Trouble with Vowels
The reading teaching profession is still, in many ways, coming to terms with the legacy of Clymer’s seminal 1963 article on the utility of phonics generalizations. One of Clymer’s main findings – a discovery repeatedly invoked by his successors, is that consonant rules tend to be significantly more reliable than vowel generalizations. As Johnston (2001) puts it, “Consonants…are mastered more easily and are much more regular. Of the 45 generalizations investigated by Clymer, 10 involved consonants and all but one of those had a utility of 95 percent or better” (p. 134). By contrast, she noted that, “vowel generalizations are much less regular and more subject to interpretation” (p. 134). This distinction has generally been used to license attempts to place more emphasis and spend more time on instruction related to vowel generalizations. Thus, for example, the fact that vowels “pose the most difficulty for beginning readers” is used by Johnston to justify the conclusion that, “vowel generalizations may be the ones they need most” (p. 134). 
Although the opposite conclusion may be drawn from the same set of facts (more on this later in the article), phonics-based instruction has largely taken Johnston’s observation to heart and has attempted a range of strategies for dealing with the vagaries of vowel pronunciation. One approach has been to emphasize rules that apply not to words but to what Caldwell, Roth, and Turner (1978) call “higher order units” – that is, consistent subunits of words that are spelled and pronounced the same. This is the direction that has been taken up by approaches that rely on teaching rhyming patterns and focusing on sub-units of words. Thus, Stahl, Duffy-Hester, and Stahl (1998), describe a category of approaches they call “analogy-based,” (see, Cunningham, 1978, 1979 and Gaskins, Gaskins, and Gaskins, 1992) in which, “students learn how to decode words they do not know by using words or word parts they do know” (Stahl, Duffy-Hester, and Stahl, 1998, p. 347). This approach has also been taken up by teaching strategies that combine “onsets” (beginning letters or letter combinations) with “rimes”—phonogram families that are pronounced and spelled consistently (such as “-ack,” “-ice,” “ick”).
Wiley and Durrell (1970) found that nearly 500 primary grade level words are based on a set of only 37 different “rimes” and that 272 rimes account for 1,437 words in, “Murphy’s inventory of the speaking vocabularies of primary grade children” (as summarized in Adams, 1994, p. 321). In the first case, the “rime” families average 13 words each, in the latter only about five. However, even the not insignificant effort of learning 272 rhymes would assist the beginning reader in recognizing only a fraction of the words in the typical elementary school vocabulary, which is estimated to range from between 24,000 to 40,000 words. Much more work would be necessary to read the remaining words that do not fit the pattern. In the end, the difficulty of approaches that focus on “higher order units” is that they generate a dramatic increase in the number and complexity of generalizations that need to be taught, while providing a relatively small payoff.
If one attempt to increase the reliability of reading cues has been to focus on sub-units of words, another is to craft whole-word rules that incorporate exceptions and qualifications. This is perhaps best illustrated by one of the rules proposed by Gates (1983) and cited by Johnston (2001): “her generalization for silent e, which is reliable at 93 percent, is restated as follows: When a word ends in a single vowel +consonant+ e, the e is silent and the vowel is long, or has the short-i sound (except for words ending in ‘some’, ‘ove’, and vowel + re)” (p. 133).  Johnston charitably described such rules as “unwieldy” – several rules like this would be prohibitively difficult to memorize and would still leave most cases uncovered.
An alternative to creating more complex rules and exceptions has been to narrow down or invent the words to which beginning readers are exposed. If meaningful running text is riddled with exceptions to the generalizations, and it is too hard to change the generalizations, we can respond by changing the text. Thus, current so-called systematic, explicit phonics systems rely largely on teaching individual words out of context or in manipulated text. The pendulum has swung so far away the use of normal, meaningful texts in the U.S. that it has become common to rely on the use of so-called pseudo-words – invented words that “follow” vowel pronunciation rules – and on exposing students to carefully crafted nonsense sentences and stories that “follow” the rules. Open Court (Adams et alia, 1995) exercise books, for example, include stories such as, “Tim skims, Tim dips, Tim tips his hat. Tim hits a pit and sits” (a story about an ice-skating bear, pp. 31-32), and Explode the Code (Hall  & Price, 1981) uses sentences such as, “The deer feels safe in the tree” and “Will a whale wipe its wet hands?” (p. 71, p. 46). The strained, unnatural, unmotivated, and nonsensical tone of this type of writing makes it all but impossible for children to use context and storyline as tools for narrowing down word choices. Adams (1994) highlights the difficulty of such approaches: “Research has shown that text that is composed of high proportions of orthographically and phonologically similar words is inordinately difficult to process. Even when read silently by skillful readers, such texts produce the disruptiveness of tongue twisters” (Adams, p. 322).
 The attempt to increase reliability via pseudo-sense and pseudo-words ends up increasing the effectiveness of one tool at the expense of another. The sentences quoted above may follow the generalizations they are chosen to illustrate, but they are difficult to make sense of and do not help children read normal text. A beginning reader using his or her knowledge about the world, for example, would be unlikely to expect a sentence about a whale to refer to the animal’s “hands.” The sacrifice of meaning and context is a grave loss, since it is the entire point of reading. 
We realize that much current reading practice views the process as one that takes place in stages: skills in decoding are seen as precursors to reading for meaning. This is the sequence outlined by Samuels (2007): “one must identify the words on the page and one must construct their meaning.  If all of a reader’s cognitive resources are focused on and consumed by word recognition, as happens with beginning reading, then comprehension cannot occur at the same time. However, once beginning readers have identified the words in the text, they then switch their cognitive resources to constructing meaning” (p. 964).
This description envisions a process wherein meaning and context come into play only after the fact of word recognition. They have no role to play in the recognition process, which relies on the letter-sound correspondence alone. It is worth stopping and thinking about what the claim here really is: that beginning readers can arrive at the correct sound of a word (for only this would constitute recognition) solely on the basis of the letter-sound correspondences they have learned. For this to be possible it would require a system in which every letter in the word – including the vowels – gave an unambiguous indication of how the word is pronounced. It would mean, for example, that the letter sequence “h-o-w” somehow indicates -- even prior to any recognition of the word (for what is being explained is how the word is recognized in the first place) -- that the letter “o” is to be pronounced to rhyme with “cow” and not with “tow.” Otherwise, of course, the child would “recognize” a very different word (and wonder what a garden implement is doing in the middle of the sentence). 

The Trouble with (Omitting) Context
The attempt to separate out a process of decoding as distinct from comprehension and thus meaning and context is one of the legacies of the vowel-consonant distinction – the fact that, as Johnston (2001) puts it, “Consonants...are mastered more easily and are much more regular” (p. 134). The problem with so-called “authentic” text, from the perspective of phonics instruction, is that meaningful, comprehensible, natural-language sentences are composed of many words that do not follow generalizable pronunciation rules.  Since phonics generalizations that include the vowels do not apply to many of the words that appear in these texts, phonics is taught in controlled contexts such as nonsense or rhyming sentences or lists of pseudo-words where compliance with the taught generalizations can be ensured.
Decoding is, consequently, taken to be an antecedent skill to comprehension – one that can be taught as a distinct process in which pronunciation precedes word identification and word identification, apparently precedes comprehension at the level of the sentence. This distinction is reproduced by Stanovich’s (1984) claim that attempts to use meaning to decode words slow down beginning readers, who should be freed up from the requirement to understand context in order to decode words as rapidly as possible. However, it is not clear that Stanovich’s own findings justify either the distinction between context and comprehension or the importance of dispensing with the cues provided by context. For example in every case, the readers he tested exhibited so-called context effects: that is, context made a difference in the speed of recognition of a written word at the end of a sentence, although he concludes that, “Contextual effects diminish with development” (p. 14).
The resulting assumption is that more advanced readers are less reliant upon context and that therefore the developmental goal of reading instruction should be to facilitate automatic decoding that minimizes the cognitive drain associated with attending to context. This argument suggests that what experienced readers are doing amounts to a speedy version of letter-by-letter decoding. However, it is not clear from the evidence he provides that advanced readers engage in ultra-rapid decoding. Cattell’s classic research (detailed in Adams, 1994, p. 95), still generally accepted in the field, demonstrates that experienced readers recognize whole words as rapidly as they do individual letters. In the face of these and other findings, Adams points out that, “the idea that skilled readers recognize words by translating their letters, one by one, into sounds and then blending the sounds together seems preposterous” (p. 95). “Automaticity,” at least in the case of experienced readers, appears to refer not to ultra-rapid decoding, but to rapid word recognition (these are clearly different processes: it is possible to recognize a word without decoding it.
The fact that Stanovich found that readers at all levels could recognize so-called congruous words   context-based difference between the speed at which experienced readers can recognize congruous and incongruous words bears further examination. It suggests what every experienced reader knows: that the alleged difference between context and comprehension is a vexed one. Experienced readers do not wait until the end of a phrase or sentence (or paragraph, or chapter, for that matter) to attempt to make sense of it. Rather readers build up a sense of the meaning of sentences as they go along, creating and adjusting their mental picture of what the sentence means along the way. Imagine, for example reading a sentence that starts out, “The blue sailboat bobbed up and down in the rippling water...” An experienced reader would not hold off on creating a mental image, but would rather start to form a mental picture that adjusted itself as the sentence swung through even such abrupt changes as, “beneath the faucet dripping into the bathtub.” Meaning is not something that waits to form until the words have been properly decoded, it develops along the way, adapting to the changes that take place in the text as it unfolds.
The insistence upon decoding prior to meaning actually helps suppress the distinction between the reliability of vowel and consonant generalizations long acknowledged in the literature. If decoding is to take place absent meaning altogether – as a mechanical process whereby letters yield sounds that add up to the accurate pronunciation of a not-yet-recognized word which can only then be identified – then every letter in the word is crucial to the process. The vowels must somehow be made to provide meaningful and reliable cues as to how they are to be pronounced. Thus, the decision to teach phonics has been construed as an all-or-nothing one: if you teach the consonant generalizations, you must also teach vowel generalizations, and if that is to work, then a bottom-up decoding process that distinguishes decoding from comprehension becomes necessary. It is this all-or-nothing approach that has pitted phonics instruction against meaning and posited decoding as prior to comprehension. The power of context is sacrificed to the principle that both vowel and consonant rules must be taught.

The Trouble With Decoding
It has perhaps become obvious that the thrust of this article is to challenge the notion of decoding, at least as it is currently understood as a necessary intermediate step between exposure to written text and comprehension. Before exploring some of the empirical evidence in further detail we seek to raise some issues of logic. Taught as a skill antecedent to meaning – that is, as a skill that simply links written letters with spoken sounds – and hence as a skill that can lead readers from letters to spoken words in the absence of content or context, decoding faces impossibly high hurdles. 
The notion that decoding is prior to comprehension is a necessary presupposition of approaches that teach decoding as a meaning-free process. The underlying assumption of such approaches is that letters can, on their own and with the proper developmental sequence of training, yield correct word pronunciation.  It is worth making this assumption concrete. It presupposes, for example, that there is something in a written word that provides guidance, on its own, as to how it is to be pronounced. This seems, on its face, to be the essence of common sense, but further consideration renders it suspect. The letter combination that comprises the word “plow,” for example, cannot provide, on its own, guidance that it should be pronounced to rhyme with the word “brow” and not the word “slow.”  There is nothing in the word “bow” that indicates when it is to be pronounced so to as to rhyme with “how” and when it should be pronounced to rhyme with “tow” (and thus to mean something different).  What might it mean to decode the word “bow” in a sentence prior to grappling with the question of what it means? Is there a correct pronunciation for the three-letter combination “b-o-w” that is antecedent to the meaning of the word it spells?
It might be possible to argue that these words are merely exceptions that do not undermine the general utility of decoding as a process that precedes comprehension. However, the list of so-called exceptions goes on and on. How do we know whether the word “come” is to be pronounced as in “some” or “home”? What about the word “tear” – should it be pronounced as in “near” or “bear” or even as in “heart” or “learn”; does “move” indicate that it should be pronounced differently from “love” or “stove,” and so on. We could make the point somewhat more simply, and without having to calculate percentages of “regularly” pronounced words by simply highlighting what experienced readers already know: that, given a written word they have never seen before and do not know, the spelling cannot serve as a dependable guide to its correct pronunciation. Such a word cannot reliably be “decoded” (in the sense described above) even by the most experienced of readers. That is why dictionaries have pronunciation guides (these would be redundant if spelling, on its own and out of context, indicated the correct pronunciation). When we ask children to arrive at the correct pronunciation of an unknown word solely from its letters, we are asking them to do what experienced readers cannot. It is important to clarify this point – when we say “unknown” in this context, we do not mean a word that is not in a beginning readers’ spoken vocabulary – but rather a word that has not yet been identified (and is thus unknown to them at the moment they are trying to read it). To put this differently, knowing the meaning of a word is presumably irrelevant to the initial step of decoding, which, according to its own account, precedes meaning. Decoding is supposed to yield the correct pronunciation (independent of context), which in turn is supposed to lead to meaning, assuming the spoken word is in the reader’s vocabulary.
If, on the other hand, one were to make the case that meaning is an integral part of decoding, the notion of decoding as antecedent to and distinct from comprehension would disappear. Decoding as currently understood is founded on the assumption of the existence of an internalized knowledge of spelling patterns that, absorbed over time, allow readers to predictably and accurately pronounce words prior to knowing what they are.  Anyone who wants to make the case for the existence of such patterns, needs to be able to explain two things: 1) why is it that these patterns cannot be clearly articulated as rules (as they are in other languages such as, for example, German, Russian, Spanish, etc.); 2) why is it that experienced readers cannot be expected to accurately pronounce English words that they have not previously encountered based solely on their spelling.

Actual Teaching Practice
For lack of a better system, a reliance on variants of decoding remains a staple of contemporary reading practice. Our survey of 110 teachers involved in beginning reading revealed that several very unreliable pronunciation rules are routinely taught during the first six months of instruction.  This section takes a closer look at the rules that are actually taught in the classroom and follows up on the fact that consonant-based generalizations are significantly more reliable than vowel-based generalizations. The remainder of the article seeks to offer an explanation for why rules that are often unreliable continue to serve as staples of classroom instruction, and to suggest a possible alternative.
As part of the background research for this article, we surveyed 110 teachers across the U.S. on the letter-sound correspondences they used in teaching beginning reading. The survey was conducted in the summer of 2009 and posted on several e-mail lists for reading teachers, reading specialists, and elementary school teachers in the U.S.[1] The rules listed in the survey were taken not just from Clymer’s original list, but also from interviews with teachers about contemporary teaching practice, and from contemporary reading programs. A complete version of the rules included in the survey is included in the appendix. We supplemented the responses to the listserv with appeals to randomly selected school districts by region and demographics (urban, rural, and suburban). Respondents came from 35 states and included first grade teachers (43 percent), kindergarten teachers (27 percent) and reading coaches and specialists (including resource and special education teachers) (30 percent). Almost all of the respondents (98 percent) indicated that they believed the ability to name letters was either useful (58 percent) or a necessary prerequisite (40 percent) for learning to read. In keeping with these findings, the most commonly taught pronunciation rules, according to our survey, were associated with individual letter names. The majority of respondents (90 percent) indicated that they taught the consonant sounds within the first six months of reading instruction, whereas 97 percent indicated that they taught the short vowel sounds during the same period (almost two-thirds -- 63 percent -- also taught the long vowel sounds during the first six months).
In both cases these letter-sound correspondences were the most commonly taught by the teachers surveyed. The findings are suggestive: they strongly indicate what those in the teaching profession already know well: that letter sounds – of both the consonants and the vowels – serve as the backbone for current reading teaching approaches. Indeed, the short vowel sounds are actually taught more often than the consonant sounds. While we are strong believers in the importance of knowing the alphabet for beginning readers, we are also concerned that the reliance on vowel-sound associations in current forms of decoding instruction necessitates teaching unreliable generalizations.
Theodore Clymer’s (1963) simple and straightforward assessment of the effectiveness of vowel generalizations remains a widely acknowledged study whose implications have yet to be fully absorbed by the field of reading instruction. Johnston (2001) describes Clymer’s study of the utility of phonic generalizations as a “classic” work that “continues to be frequently cited almost 40 years after it was originally published” (p. 132).  The import of Clymer’s study is that it calls into question a significant number of the phonics generalizations he found in contemporary teacher manuals. After eliminating those rules that he deemed too general or unclear to either “aid or hinder in the pronunciation of a particular word” (186), Clymer found that only 18 of 45 generalizations (many having to do with syllable division rather than pronunciation of particular vowels) were useful at least 75 percent of the time when applied to a list of words children were likely to encounter. Applying the rules to a list of some 2,600 words compiled from the same readers from which he selected the phonics generalizations (along with the Gates Reading Vocabulary for the Primary Grades), he discovered that the majority of phonics generalizations mislead students more than a quarter of the time. Clymer’s response to this somewhat stunning fact was tentative. He observed that his standard of 75 percent utility for accepting a generalization, “may be too high. Classroom research might reveal that generalizations with much lower percentages of utility should be taught because they encourage children to examine words for sound and letter relationships” (1963, p. 187). 
            We think that Clymer’s findings provide a greater challenge to the teaching of vowel generalizations than he realized, for reasons that will be developed later in this section. First, however, it is worth considering some of the academic responses to Clymer’s findings in the intervening years.  Caldwell, Roth and Turner (1978) noted that studies of phonics generalizations, including Clymer’s, tended to overlook the importance of considering how the rules applied  to running text, rather than relying on word lists like those compiled by Clymer. As they put it, “The real world of English may be considerably more or less consistent than these studies suggest…A more appropriate measure would be the frequency of occurrence of the rule in running text” (p. 91). The main difference between their findings and Clymer’s was that the utility level for the two of the broadest vowel generalizations (the “silent e” generalization and the vowel-pair generalization) dropped significantly: from 63% to 42% and from 45% to 26% respectively (pp. 94-5).  Nevertheless, they conclude with the observation that, “The most apparent result is the fairly high agreement” among all of the various studies they consider (p. 93). Johnston’s (2001) overview of attempts to replicate Clymer’s findings reaches a similar conclusion: “Overall, however, no matter the number of words or the source, the results still show that less than half of the generalizations, as stated, meet Clymer’s criteria of 75% utility applied to at least 20 words across all types of text” (p. 133).
            We take a different approach to updating Clymer’s research, drawing not on primers but on a survey to see what generalizations are most commonly taught to beginning readers. We then test these generalizations on samples of running text selected from popular children’s books – not course readers, but actual trade books.  For the sake of comparison we divided the generalizations into vowel-based and consonant-based categories. Unsurprisingly, and in keeping with past research, we found the consonant-based generalizations were significantly more reliable than the vowel-based ones, but the latter are still taught by a majority of the teachers and specialists we surveyed. We take up the reasons for this, along with possible alternatives to current teaching practice, in the concluding section.  
Our findings for the teaching of vowel-based generalizations are listed in Chart 1, which also includes the results of applying these generalizations to randomly selected samples of running text from children’s books.  We tested the generalizations on a total of 1,870 words in samples of running text taken from eight books ranging from pre-school to fourth grade reading levels.[2]  The eight samples of running text ranged in length from 164 words to 287 words. The chart indicates the percentage of teachers who use a particular rule as well as the percentage of words for which the generalization was relevant and then the smaller percentage to which it applied accurately. Thus, for example “when two vowels go walking, the first does the talking,” is relevant (that is, it can be applied) to the word “great” (and not to the word, “bird,” for example), but it does not provide an accurate cue. A child applying the rule to “great” would mispronounce it as “greet,” (a word to which the rule does apply accurately) and would likely have a hard time making sense of it in the context of a meaningful sentence. Thus the chart is meant to give a sense not just of how commonly taught a rule is, but of how useful it is for a child attempting to read meaningful sentences in a story.

Table 1: List of Phonics Based Generalizations Taught in US Schools during the first six months of reading instruction (the most frequently taught rules are listed first)

Percentage who teach it
Percentage of words to which it is relevant
Reliability (percentage of the time it is correct when applied to relevant words)
Short vowel sound (as the most common way of pronouncing a vowel)
100 (every word has a vowel)
Typically taught in the context of consonant-vowel-consonant words (“hat,” “pot,” etc.).  Even in this context the rule will lead to many mispronunciations (“put” etc.) and thus hinder work identification.
The final “e” in a word is not pronounced (silent “e” rule)



The 222 final e’s that do not follow the rule are all accounted for by the following six words: be, he, me, we, she, the.
The silent “e” rule: “e at the end of the word makes the preceding vowel in the word long.
Even though this rule was taught by more than two-thirds of the teachers surveyed, it has a low level of reliability. Students who use it will be provided with a misleading cue four-fifths of the time.
When a word has only one vowel and this vowel is in the middle, it is usually short.


Again, not a particularly useful rule, despite the fact that it is commonly taught.  It applies to a quarter of the words encountered in the samples and is correct only a bit more than half of the time.
The long vowel sounds
See rules specifying when this sound is to be used
See rules specifying when this sound is to be used
See rules specifying when this sound is to be used
The combination “AR” is pronounced as in “car,” “far,” etc.

Applies to a small sub-section of the words, and is reliable in more than a third of the cases considered.
The combinations “IR,” “UR,” and “ER,” are pronounced as in “Her”

“IR” and “UR” are more reliable; “ER” brings down the percentage reliability. But since these are taught together as one rule, it will still provide an incorrect cue about one-third of the time.
The combination “OR” is pronounced as in “For”

This is a reasonably reliable rule, but helpful for only a very small percentage of words encountered in the running text samples.
When two vowels go walking, the first does the talking (in a two-vowel pair, the first is long, the second silent).

It is striking that almost half of the teachers surveyed taught a rule that works less than one-third of the time.
The combination “AY” is pronounced as in “Day”

This is quite a reliable rule – it will, however, mislead readers in the case of two quite common words: “says” and “always.”
If “Y” is the only vowel in a word, and comes at the end, it is pronounced as in “Fly”

A very reliable rule for a small percentage of words.
The combination “OU” is pronounced as in “Found”

A misleading generalization for more than half of the words in the sample, but still taught by more than one-third of the teachers surveyed. 
A single vowel at the end of a word makes its “long” sound.
All but three of the words that end in long vowels are accounted for by “be,” “he,” “she,” “we,” “me,” as well as “go.”  The only other words that followed the rule in the sample were “hero” (2x) and “Eskimo.”
The combination “EW” is pronounced as in “Flew”
Again, reliable, but only for a miniscule percentage of words encountered.

The pattern revealed by the chart is clear: the most commonly taught generalizations are not useful for reading words in running text. It would be incorrect to claim that a rule which works half the time is better than no rule at all, because this is also a rule that misdirects the beginning reader half of the time, obstructing the attempt to correctly pronounce and thus identify a word. Some of the more accurate generalizations are among those less frequently taught, most likely because these tend to only apply to a small fraction of words likely to be encountered by beginning readers. A generalization that only applies to one word in a book, for example, is only a tiny bit better than no generalization at all. If each word becomes its own “generalization” the result would be dispensing with generalizations as such.
By contrast, consider the reliability of the most commonly taught consonant generalizations, as illustrated in Table 2. Perhaps the simplest and most striking finding is that just teaching the sounds of the consonants (setting aside the other rules for the moment) yields 91 percent reliability. Interestingly, the reliability is lowered if the “g” rule (“g” is soft when followed by “e,” “i,” or “y”) is included – because this rule was accurate less than half of the time in our sample. Removing one rule improves accuracy (although knowing that there are only two common sounds for “g” is surely useful). Weighting the reliability of these rules by the percentage of words to which they apply suggests a reliability level well over 90 percent for the commonly taught consonant-based generalizations.

Table 2: List of Consonant Based Generalizations Taught in US Schools during the first six months of reading instruction (the most frequently taught generalizations are listed first)

Percentage who teach it
Percentage of words to which it is relevant
Reliability (percentage of the time it is correct when applied to relevant words)
Pronunciation of the consonant sounds
98 (“I” and “A” do not have consonants)
This assumes that “g” and “c” are taught as in “go” and “cat,” h as in “how” and w as in “way”
The consonant digraph  “TH” is pronounced as in “There”

The digraph  “SH” is pronounced as in “Show”

The digraph “CH” is pronounced as in “Chair”

The digraph “WH” is pronounced as in “Where”
Three instances of  “who” lowered the reliability of this rule
The consonant digraph  “PH” is pronounced as in “Phone”
No instances in the chosen samples
When “C” is followed by “E,” “I,” or “Y” it is soft (as in “Cell”)

When “G” is followed by “E,” “I,” or “Y” it is soft (as in “giant”)
The words that lowered the reliability of this rule included: "get," "give,"   "girl," "begin," and "tiger" (a key character in one of the books)

Discussion and Implications
There is nothing particularly new about the finding of a big discrepancy between vowel- and consonant-based generalizations. This is a well-known fact of phonics-based and decoding-based forms of instruction – although our findings illustrate the systematic difference between generalizations currently used in the classroom. What is striking about our findings is the fact that this knowledge seems to have little impact on classroom practice. Why are teachers continuing to teach generalizations that do not work and mislead beginning readers? Most likely, this choice was made in the name of consistency in decoding: if we are going to teach consonant rules and the letter names, the fear is that readers will be confounded if they don’t have rules to go along with all of the letters they encounter as they read.  If decoding is presumed to be antecedent to meaning and context is thereby removed as a cue for beginning readers, all that is left are the letters. That is, the reliance on a particular understanding of decoding goes a long way to explain why vowel-based generalizations continue to be taught despite their alarmingly low reliability.
The real question we should be asking is why the insistence on a notion of decoding as antecedent to meaning? This accords neither with the experience of skilled readers nor with the evidence adduced on its behalf.  Nevertheless, it is an insistence which has led recent reading instruction to choose vowel generalizations over meaning, in the full knowledge that these generalizations are at best unreliable. In practice, our experience in the classroom indicates that beginning readers can identify words far more effectively by using extra-linguistic information such as the subject of the story, real-world experience and other context clues such as knowledge of syntax, real-world information that would suggest what the sentence he or she is reading is likely to be saying, and his ability to judge whether or not a sentence makes sense. These contextual cues dramatically narrow the word choices available to them and make it possible for them to identify to identify words in meaningful text using the consonants alone.
The efficacy of context is demonstrated by psychologist George Miller’s observation (as summarized in Pinker, 1999) that, “[I]f speakers keep their sentence perfectly grammatical and sensible as they choose their words, their menu at each point offers an average of about ten choices [for the next word in the sentence]” (p. 7).  In other words, meaning allows the beginning reader to engage in a dramatic reduction of uncertainty: narrowing the possible options for the next word in the sentence down from the 10,000 or more words in the readers’ vocabulary to about 10. A reliance on consonants helps choose from among these 10 or so possibilities.
Systematic research needs to be done on such an approach, but first the underlying causes for the profound problems that plague the existing system need to be recognized. While this article does not undertake such research, it does attempt to outline the reasons for pursuing it and to demonstrate the impasse to which the current insistence on a decoding –based system that necessitates the teaching of unreliable generalizations has led.
To get a sense of how it might be possible to read a sentence without bringing any vowel generalizations to bear, consider the following sentence taken from a story about a dog named Ribsy:
R-bsy,  dr-pp-d  th-  b-ll  -nd  l-t  -t  r-ll.
With only a little effort, once the reader knows that this is the story of a dog named Ribsy, it is possible to read every word in this sentence, even without the vowels. It is no accident that we find it possible to read meaningful sentences even when the vowels are missing. The original writing system from which all alphabetic systems are descended had no symbols for the vowels and was able to function for centuries without them (Healy, 1990).  Alphabetic systems such as Hebrew, Farsi and Arabic continue to function effectively without vowel characters. This is not to say that vowels have no use – they do make it possible for those who already know how to read to identify words out of context.  Rather it is to claim that in English (as opposed to several other languages – including the European ones from which some reading methods are borrowed and adapted) they are not useful guides for helping beginning readers sound out words.  
Those who argue that any rule for the vowels is better than none might suggest the sentence about Ribsy would be even easier to read with the help of vowel generalizations. However, commonly taught vowel rules provide the wrong cues for three of the words: “ball,” “roll,” and “dropped,” introducing confusion and doubt where there needn’t be any. In such situations, the context suggests one word, which is contradicted by the preferred pronunciation indicated by the vowel generalizations. This contradiction might be readily resolvable in favour of context, were it common practice to teach phonics generalizations in meaningful sentences and stories.
When words are read in context, the beginning reader needn’t have a rule for every letter in a word, since the combination of the cues provided by the consonants and the context is sufficient and more effective than the often misleading information provided by vowel generalizations. The fact that phonics instruction has found it necessary to abstract away from meaning in order to teach vowel generalizations suggests just how incompatible these rules are with meaningful written language. Were it possible for vowel generalizations to serve as useful supplements to context, there would be no need to teach them in words taken out of context or in nonsense words or sentences.         
In many Continental European languages children are typically introduced to reading through little stories written in normal sentences. It is only in English that children are introduced to single words written out of context (in lists for example). Of all the alphabetic languages, it is only in English that beginning reading instruction may, for several years, consist of trying to teach children to read words by their letters alone.
Children taught this way approach print with nothing in mind but the sound they have been taught for each letter. Beginning readers taught this way have no anticipation of meaning for these words, but only seek to pronounce them through the specific rules they have been taught. As researchers including Hebert (Hebert, 1990, pp. 170, 172, 180) and McCandliss et al. (2003) have noted, the rules they have been taught will frequently produce the wrong pronunciation. Even one wrong word can make a sentence incomprehensible, as in the example (taken from an actual classroom example) of a student who correctly read, “The bunny hid in the bushes” and then stopped short, saying that the last word couldn’t be “bushes” because “bu” is pronounced “buh.”
The mental attitude of a child taught to approach print in complete sentences is wholly different then the mental attitude of children who believe that reading is a matter of sounding out single words. The first group approaches a written text expecting a sensible sentence.  These children already know a great deal about that sentence.  They know that it will advance the story in a meaningful way.  They also know that the words of a sentence will be in the normal order that they have known and used since they were toddlers (O’Grady, 2005). Using this information and any words in the sentence that they already happen to know, along with the sounds of the consonants, they achieve their goal of following the story they are reading. By contrast, children taught to abstract reading from meaning because they have been taught to read words in lists or in incomprehensible sentences like, “Will a whale wipe its wet hands?,” find themselves facing a world of text very different from the language world to which they have grown accustomed: a world in which any word might follow any other – and in which successful reading does not result in meaningful comprehension (there is no meaningful answer to the whale question – it is a nonsense question).
The goal of this article has been to argue that the distinction between vowel and consonant generalizations long acknowledged by the literature has not yet been accorded its full significance. It is a distinction that allows us to clarify the choices we make in deciding how best to use the letters in reading instruction. To date, this choice has largely been interpreted in ways that overlook the vowel-consonant discrepancy. That is to say, the decision to teach phonics has been construed as an all-or-nothing one: if you teach the consonant generalizations, you must also teach vowel generalizations. It is this all-or-nothing approach that has pitted phonics instruction against meaning in a fashion detrimental to reading instruction and resulted in an insistence upon teaching misleading letter-sound generalizations. The power of context is sacrificed to the principle that both vowel and consonant rules must be taught.
Recognizing the vowel-consonant discrepancy opens up the possibility of deciding differently – and teaching differently. The fundamental truth that all sides in the reading discussion need to grasp is that consonant-based phonics can be distinguished from vowel-based phonics and that the former can be used in conjunction with context and meaning in ways that the latter cannot.
Past attempts to develop a system based on context and consonants (see, for example, McKee, 1966) have not been able to move beyond an idea that plagues phonics instruction generally: the notion that beginning readers can use letters alone to arrive at the word they are trying to read. In practice, beginning readers need several different cues, since vowel sounds, even if they follow phonics generalizations (which they do not, in most cases), do not reproduce the word as it is heard in speech. The approach we are advocating recognizes that successful beginning readers rely on contextual cues to generate possible words while relying on consonant cues to narrow down their choices. In other words, beginning readers aren’t engaged in the process of building a word out of nothing from its constituent sounds, but in using the available cues to find a whole word that would make a meaningful sentence or phrase advance sensibly. 
The reason phonics has become, for all practical purposes, a complex and ultimately ineffectual science of the vowels is because using the consonants is so simple it can be taught easily and effectively – or even figured out by a child who merely knows the names of the letters. Dispensing with the vowel generalizations does not significantly hamper the ability to read a word in context.  It allows teachers to avoid the endemic practice of teaching highly misleading generalizations and it removes a barrier to using one of the powerful tools available to the beginning reader – a tool that is so effective not just because it narrows down the choice of words, but because it restores the point of reading to the child: the joy of making sense out of the world in which he or she is immersed.

Works Cited:

Adams, M.J. (1994). Beginning to read: Thinking and learning about print. The MIT Press.
Adams, M.J. et alia (1995). Step-by-Step Practice Stories: Grade 1. Open Court Publishing Company.
Bailey, M. (1967). The utility of phonic generalizations in grades one through six. The Reading Teacher, 20, 413-418.
Betts, E, (1957) Foundations of Reading Instruction: With Emphasis on Differentiated Guidance. American Book Company.
Burmeister, L. (1968). Selected word analysis generalizations for a group approach to corrective reading in the secondary school. Reading Research Quarterly, 4(1) 71-95.
Caldwell, E., S. Roth and R. Turner (1978). A Reconsideration of phonic generalizations. Journal of Reading Behavior, 10, 91-96.
Chall, J. (1967) Learning to read: The great debate. McGraw-Hill.
Clowes, G. (1999). Reading is Anything But Natural. School Reform News, July. Retrieved online 4 May, 2008 at: 
Clymer, T. (1963). The utility of phonic generalizations in the primary grades. The Reading Teacher, 16, 252-258.
Clymer, T. (1996). The utility of phonic generalizations in the primary grades (RT Classics). The Reading Teacher, 50(3), 182-187.
Cole, R, (1979) Current Issues in Linguistic Theory.  Indiana University Press.
Cunningham, P.M. (1978). Decoding polysyllabic words: An alternative strategy. Journal of Reading, 21, 608-614.
Cunningham, P.M. (1979). A compare/contrast theory of mediated word identification. The Reading Teacher, 32, 774-778.
Emans, R. (1967) The usefulness of phonic generalizations above the primary grades. The Reading Teacher, 419-425.
Flesch, R. (1955). Why Johnny Can’t Read: And What You Can Do About It. Harper & Brothers.
Gaskins, R.W., Gaskins, J.C. & Gaskins, I. (1992). Using what you know to figure out what you don’t know: An analogy approach to decoding. Reading and Wring Quarterly: Overcoming Learning Disabilities, 8, 197-221.
Gates, L. (1983). A face lift for the silent e. The Reading Teacher, 37, 102-103.
Goodman, K. (1982). Language and Literacy: The Selected Writings of Kenneth Goodman, Vol. II (F. Golasch, ed.). Routledge and Kegan Paul.
Hall, N. and R. Price (1985). Explode the Code, 3 ½. Educators Publishing Service.
Healy, J. (1990). Reading the past: The Early Alphabet. University of California Press/British Museum.
Horn, E. (1929). The child’s early experience with the letter A. The Journal of Educational Psychology, XX(3), 161-168.
Huey, B Edmond, (1908). The Psychology and Pedagogy of Reading. The M.I.T. Press.
Jakobson, R, (1985). Verbal Art, Verbal Sign, Verbal Time. The University of Minnesota Press.
Johnston, F. (2001). The utility of phonic generalizations: Let’s take another look at Clymer’s conclusions. The Reading Teacher, 55(2), 132-143.
Mann, H, (1957) The Republic and the School. Teacher’s College Press.
Manzo, K. (2008) “Federal Path for Reading Questioned,” Education Week, Dec. 3, pp. 1,16.
McCandliss, B., I. Beck, R. Sandak and C. Perfetti (2003). “Focusing Attention on Decoding for Children with Poor Reading Skills.” Scientific Studies of Reading, 7(1), 75-104.
McKee, P. (1996). Reading: A Program of Instruction for the Elementary School. Houghton-Mifflin.
Owen P. and Pumfrey, P. (1995) Emergent and Developing Reading: Messages for Teachers, London: Routledge.
Pinker, S. (1999). Words and Rules. New York: Basic Books.
Samuels, S.J. (2007) The DIBELS tests: Is speed at barking at print what we mean by fluency?
Reading Research Quarterly, 42(4), 546-567.
Saussure, Ferdinand De [C, Bally & A, Sechehaye & A Riedlinger] (1972). Cours de Linguistique Generale.  Payout Paris
Smith, F. (1979). Reading without nonsense. New York: Teachers College Press.
Stahl, S, A. Duffy-Hester, K. Stahl (1998). Everything You Wanted to Know About Phonics (But Were Afraid to Ask). Reading Research Quarterly, 33(3), 338-355.
Stanovich, K. (1984). “The Interactive Compensatory Model of Reading.” Remedial and Special Education, 5(3), pp. 11-19.
Vygotsky, L, (1962) Thought and Language. The M.I.T. Press.
Yatvin, J. (2000). “Minority View,” National Reading Panel Report, a publication of the National Institute of Child Health and Development, retrieved online 4 April, 2010 at:
Yatvin, J. (2002). “Babes in the Woods: The Wanderings of the National Reading Panel.” The Phi Delta Kappan, Vol. 83, No. 5, Jan. 2002, pp. 364-369

[1] The e-mail lists to which we sent an online version of the survey were the following: (, Kindergarten Teachers Mailring (, First Grade Teachers Mailring (, Second Grade Teachers Mailring (, Reading Teachers Mailring (, Remedial Reading Teachers Mailring (, The Reading Teacher listserv (, the Read Across America Listserv ( Additionally, we emailed links to the survey to reading instructors (where email addresses were posted online) and principals (with a request to forward the link to teachers involved in reading instruction) at fifty randomly selected school districts across the U.S.     
[2] The books from which the samples of running text were taken are:
Bulla, R.C. (1954). Squanto: Friend of the Pilgrims. Scholastic Book Services;
Hoff, S. (1958). Danny and the Dinosaur. HarperCollins Publishers;
Lindgren, A. (1997). Pippi Longstocking. Puffin Books;
Little, E. (1987). David and the Giant. Random House;
Minarik, E.H., M. (1957). Little Bear. Harper & Row.
Perkins, A. (1969). King Midas and the Golden Touch. Beginner Books;
Phillips, J. (1986). Tiger is a Scaredy Cat. Random House;
Ziefert, H. and Nicklaus, C. (1985). So Sick. Random House.

No comments:

Post a Comment