A Principled Consideration of Computers & Reading in a Second Language.

In M. Pennington (Ed.), The power of CALL (pp. 115-136). Houston: Athelstan.

By Tom Cobb & Vance Stevens, Language Centre, Sultan Qaboos University, Sultanate of Oman.



It has often been noted that CALL (computer assisted language learning) lacks a solid research base (Dunkel, 1991; Wyatt, 1988; Dunkel 1987; Roblyer 1988). The problem is mainly in two areas: lack of sustained reference to theories of language acquisition (Hubbard, 1992), or to sustained description of what if anything students actually do with specific CALL programs (Chapelle, 1990). In developing an argument promoting the use of text manipulation in the development of reading skills, this chapter addresses both these areas, with the support of evidence from research into student use of such programs.

On reproche souvent à ceux qui oeuvrent dans le domaine de l'enseignement assisté par ordinateur leur manque d'une bonne base de recherche. Le problème semble se localiser dans deux domaines: l'absence ou la quasi-absence de références systématiques à la recherche en acquisition du langage et l'absence de descriptions de la performance réelle des apprenants lorsqu'ils travaillent sur des logiciels. En se référant à la recherche en acquisition du langage et à l'observation actuelle des apprenants qui travaillent sur des logiciels, ce chapitre explique comment la manipulation des textes par les apprenants lors des séances de lecture peut faciliter l'apprentissage de la lecture.



It is not at all clear what language teachers expect reading skills development courseware to do. Although making insightful predictions concerning the impact of CD-ROM and laser printers, two devices neither widely used nor understood when his article was written, Wyatt (1989) placed use of computers in reading on a continuum constrained by introduction and development of alphabet recognition skills at the low end and mechanical and meaningful reading tasks such as comprehension exercises at the other, with "revolutionary" applications extended only to "annotation" (i.e. hypertext), modeling of productive reading strategies, and interactions with branching-plot and adventure/simulation programs. While extolling the "raw potential" of the medium, Wyatt noted that "almost none of the existing courseware for second and foreign language reading skills has moved beyond the stage of directly paralleling the activities found in printed textbooks." (p.64)

Teachers often assume that reading courseware might do something similar to what they do to teach reading as a classroom activity. Indeed, much reading courseware does attempt to emulate what might be done in a classroom; hence the "reading comprehension" approach, where a passage is presented, and questions are provided at the end. In such courseware, computers can make existing techniques more efficient for the learner, in that feedback is immediate and interactive, possibly highlighting areas of the text where attention could be most productively directed. The main drawback with the "efficiency" of this approach is the inordinate amount of time needed by developers to prepare each CALL lesson. Wyatt, for example, points out that software highlighting context clues assumes that help has been set up for "every relevant word in every reading passage" (italicized in original, p.73). Adding to the frustration is the work wasted if the content (e.g. text) the software is tied to is later deemed inappropriate and replaced in the curriculum. For these reasons, tools for producing this type of courseware are prone to lie fallow on developers' shelves after only one harvest.

But more recently uses have been made of computers in reading which do not emulate traditional methods of teaching and learning reading. Much of this "activity" has been directed not so much at the development of new "courseware," but at devising ways of making connections between an emerging battery of software tools and a proliferation of machine-readable text. One focus of this chapter is thus to examine such connections in light of recent thinking on how reading skills are developed in a second or foreign language.

Hypertext is one means of making such connections. In its simplest form, hypertext allows annotations to on-screen text to be displayed on request. However, hyperlinks can in more sophisticated implementations be almost anything imaginable: e.g. video or sound segments, pathways into reference databases, annotations made by other readers, etc. These links might give students access to background and reference information; e.g. on-line access to tools such as dictionaries and encyclopedias.

An example of the evolution of such courseware can be seen in the development of WHERE IN THE WORLD IS CARMEN SAN DIEGO (Broderbund) and its offshoots (WHERE IN SPACE ..., WHERE IN THE USA ...). In this program, users try to solve a crime by discerning clues that enable them to track down a criminal moving freely throughout the world (or space, or the USA). Solution of the mystery depends on "world knowledge" which, if lacking, may be augmented from a database of information supplied on disk, containing appealing sound and animated graphics. More recently, CD-ROM versions of the program have come out, greatly increasing the amount of information and imagery that can be made available to crime-stoppers, as well as enhancing the sophistication with which this information can be accessed. In the CD-ROM version, the screen becomes a mouse-driven console providing a video telephone, a computer sub-screen for database access, a notepad, and a video window where pictures are displayed. The program produces a plethora of spoken discourse via the sound card, and whatever is spoken is generally printed out on the computer sub-screen (giving students who read it the benefit of vocalization). Many other CD-ROM-based multimedia packages offer similar rich mixes of reading and sound. THE ANIMALS!, for example, offers hyper-linked video and still-image tours and explorations of The San Diego Zoo. Complex and authentic instructional text is spoken to users, but also printed on the screen for those who prefer to read it, or in the case of language learners, who may have difficulty in following the spoken discourse.

Along these lines, we might envision students solving similar language-enriched learning tasks by accessing authentic real-world databases over local or global networks, exploring the databases via hyper-links, and annotating the materials or reading the annotations of others before achieving some result or resolution. The potential of these media in providing both a text-rich substrate for language learning and the means and motivation for these materials to be used is becoming more apparent to those engaged in teaching and learning languages, as these powerful tools become more readily available and commonplace on networked microcomputers and stand-alone PC's.

When readers have widespread access to such tools, the concept of reading itself may change. Tuman (1992) argues that an "on-line literacy" is emerging which, while empowering readers by allowing them to interact in compelling ways with text and with each other, will also lead to the demise of the author as qualified and ever-present guide to a reader's private, sustained, and critical reading experience. Of course, this would advantage some readers more than others. Reading for some could soon be characterized by zapping their way aimlessly around the 'docuverse' of available materials. Thus, as with any application of technology to pedagogy, research will be needed to characterize the nature of the reading that takes place when learners are granted access to corpora and databases (ease of access would be an important variable), and what affect this might have on L2 reading in particular. Initial observation suggests there is no guarantee that making available large and varied amounts of on-line text, even focused on a motivating and pleasurable game-task and supplemented by other types of information, automatically promotes particularly deep or effortful processing. Before we turn our students loose to go play on the information highway, we need to decide what they can use from it and roughly to what effect.

Having moved tentatively into speculation about on-line reading in the not-so-distant future, we would like to step back to a point where we are more certain of our position. The remainder of this chapter will suggest how students can be presented with copious amounts of text, along with exercises which we believe train strategies in comprehension of that text for L2 learners. In developing a theory supporting such implementation, we expand somewhat the notion of "courseware" for reading in the sense that Wyatt meant it, taking the concept beyond what is typically done in classes where reading is "taught." We support in particular the text manipulation concept as an L2 reading activity, as it is readily implementable on most present-day computer-based learning configurations, and as it is of particular value to students learning to read in a second or foreign language. Moreover, it can make use of the large amounts of text now becoming available without departing totally from a pedagogy that we know at least a little about.



One of the most significant recent developments with impact on computer-based reading is the proliferation of and improved ease of access to machine-readable text. Text comes in over e-mail, or is scanned from printed documents, or is downloaded from CD-ROM databases in university libraries, or is purchased as huge corpora from commercial suppliers, or is captured in endless streams from close-captioned television broadcasts. Thus experienced as well as less-skilled readers can anticipate increasingly wider access to text in a format exploitable in computer-based programs of reading instruction.

One of the most interesting aspects of such text is that almost all of it is authentic discourse. In light of Higgins's (1991) definition of authentic text as any text not prepared by a teacher for the purpose of teaching a target language, the question then arises whether second and foreign language learners can cope with it. Indications are that they can.

Bacon and Finnemann (1990) examined whether, for first-year Spanish students at two U.S. universities, perceptions of general language learning (attitudes, motivation, choice of strategy), gender, and willingness to deal with authentic input could be associated with comprehension, satisfaction, and strategy-use in situations of authentic input. Results suggest that students are willing to accept and deal with authentic text; e.g. that they perceive its value to their learning and are not unduly constrained (e.g. by a desire to analyze it) in processing what they can. Kienbaum, Russel, and Welty (1986) found from an attitudes survey that elementary level foreign language learners express a high degree of interest in authentic current events materials. Allen, Bernhardt, Berry, and Demel (1988) found in a study of 1500 high school foreign language students that subjects were able to cope with all authentic texts they were presented with at three levels of difficulty. In an offshoot of that study, Bernhardt and Berkemeyer (1988) found that high school level learners of German could cope with authentic texts of all types, and "that target language and level of instruction was a more important correlate of comprehension than was text difficulty." (p. 460 in Bacon and Finnemann). These results all suggest that use of authentic text in L2 reading can be motivating and not unduly daunting to second language learners.

The foregoing is of particular interest in light of Kleinmann's (1987) findings of no significant difference when a basket of 20 CAI reading programs was used to teach reading as opposed to conventional reading materials. Kleinmann suggested that the drill-and-practice nature of the CALL material prevented greater strides in learning by failing to address higher order reading skills. The solution, he suggests, is programs providing more challenging levels of comprehensible input.

In his words:

If we accept the notion that comprehensible input in the form of text material that is interesting, relevant, and at an appropriate level of complexity is crucial to second language development (Krashen & Terrell) then the nonsignificant findings with respect to the effect of CAI compared to non-CAI in the present study are easily understood. Very little of the available reading skills software meets these criteria of comprehensible input, especially for more advanced learners. ... Moreover, it will be necessary to develop software that stimulates general learning strategies that have been correlated with successful language learning, e.g. guessing, attending to meaning, self-monitoring (Rubin, Stern), as well as more specific strategies relating to particular skill areas. For reading skills development, strategies such as skimming, scanning, and context utilization will be important. (p.272)

So there is a prima facie case for the feasibility of flowing some part of the textstream through our reading courseware in some way. However, if beginning or intermediate learners are to be exposed to large amounts of authentic text, clearly they will need something to do with this text besides attempting to read or browse it or use it for writing papers in their academic courses as if they were native speakers. Intermediate learners may be able to search through various kinds of on-line textbase, perhaps seeking answers to questions on a worksheet, but as scanning for specific information requires a fairly low level of engagement with either high-level themes or low-level details of a text, this is not the type of reading development most beneficial to learners.

This chapter argues that text manipulation templates can engage students at higher cognitive levels while presenting them with virtually limitless amounts of comprehensible input in the form of authentic texts. Although scanning is not a skill that cloze encourages (Nunan, 1985; Alderson, 1980; Windeatt, 1986; Feldmann and Stemmer, 1987), working text manipulation (TM) exercises such as on-line cloze may exercise context utilization (Jonz, 1990; Bachman, 1982, 1985??) while exposing learners to high levels of comprehensible input, assuming that learners take advantage of the amount of text that can be made available. And it appears from the results of the studies noted above that use of authentic, ungraded text, rather than posing insurmountable problems for second language learners, might instead be an appropriate matrix for exercise of higher order processing skills called for by Wyatt, Kleinmann, and others.



It is not hard to see the attractions of linking text manipulation technology to the stream of on-line text becoming available. Copious amounts of machine readable text, on the one hand, coupled with ease of implementation, on the other, makes appealing a template approach, where the courseware incorporates an algorithm which can be applied to any text supplied, realizing quantum savings in implementation time. Indeed, the distinctive feature of TM program design is that the program be able to deal with any text whatever.

TM systems can be quite varied, although they all have in common the algorithmic deconstruction of on-screen text for a learner to put back together. Some common types are cloze and other systematic deletions (suffixes, auxiliaries, transition markers, all function words, etc); scrambled elements (chunks, sentences, or paragraphs); sentence-boundary identification; crosswords; hangman or concentration-type word matching or supplying; and many others. The developer's task is to find machine-readable features of text that correspond to something readers need to pay attention to, as indicated by either observation or theory. For example, if readers are observed to pay little heed to sentence boundaries, then an algorithm can be written to detect the surface features of sentence boundaries and then eliminate them throughout a chosen text for the reader to pay attention to hands-on, i.e. through re-insertion. Because such features are common to all text, one great advantage of a template approach is that texts of almost any genre can be shared among a set of driver TM programs.

On-line help can also be designed to take advantage of this commonality of generic text. Help can be any kind of information the text can provide that is relevant to the task at hand, from giving a peek at the target reconstruction to granting access only to that part of the context that will enable a learner to make an inference. The only limitation is that the help must come from the text itself (or from the larger textbase the text comes from) and be computable by an algorithm rather than coded ad hoc or "canned." One of the authors' present experiments looks at user responses to on-line concordance as a help system for various word-level TM activities.

Text manipulation ideally uses any text, 'raw' from its authentic source. However, the TM concept extends to cases where text might be altered or annotated slightly to adhere to the particular requirements of the template, but where such alterations don't necessarily render the text unusable by other text manipulation programs. For example, John Higgins's program HOPALONG, an implementation of the "speed read" approach to reading instruction, highlights text in such a way that the eye is guided from chunk to chunk at a measured speed. All that the developer (e.g. teacher, curriculum specialist) must do, after selecting the text, is denote the chunks with carriage returns (and, in the case of HOPALONG, write in the comprehension questions, but this is in a file separate from the text itself). The already-chunked text can be used directly in another of Higgins's programs, SEQUITUR, which displays the first chunk of text and has the student rebuild the entire passage by discerning the follow-on chunks from among several proposed (i.e. the correctly sequenced chunk plus two distractors taken at random from the pool of chunks from further down the text).

The chunked text can in turn be used in a variety of other text manipulation programs which format the text according to sentence and paragraph boundaries (sentence ending strings and blank lines respectively), so that the integrity of sentences and paragraphs is essentially unaffected by the chunking required by SEQUITUR and HOPALONG. By the same token, the text can be part of a larger corpus used in concordancing or other forms of text analysis, from which again still other text-based activities may be drawn. Thus a wide variety of reading activities can be performed on any text considered relevant to the learners, who might be more motivated to re-suffix or de-scramble the sentences from an article on a topic of interest in a recent Times of London than in equivalent operations in their graded reading workbooks (or vice-versa) -- whether on screen or off.

From a developer's point of view, the advantages of this approach to CALL implementation are obvious. However, the history of technology in education should alert us to the potential dangers of too-easy marriages of technology and instruction. The apparent ease often hides the fact that one partner has been made to adapt to the other, usually instruction to technology in this case, rather than a difficult compromise worked out. Dick (1991) has noted, with regard to the development of interactive multi-media instructional systems generally, that as the technology becomes more sophisticated the pedagogy tends to become more simplistic, often becoming detached entirely from any basis in instructional research. The question to be addressed in this chapter is whether the TM approach yields corresponding benefits to L2 learners, and particularly to their skills in reading. In arguing that it does, the authors will show how the activities students perform in text manipulation exercises are commensurate with current theories regarding productive reading strategies and environments favoring the development of L2 reading ability.



For most people, reading is more agreeable and efficient on paper than on screen (Heppner et al, 1985). However, onscreen reading has the potential compensation of interactivity. A reader can send a message via the screen to a text, and then the text, properly coded, can send a message back to the reader. A paper text, by comparison, suggests a reader passively 'responding' to a text whose fixed and independent meaning he or she must discover. Of course, for a skilled reader the process is interactive whether on paper or screen, except that with a paper text the interaction is mainly invisible, in the head.

Some notion of interactivity between reader and text characterizes virtually all post-behaviorist models of the reading process. (Various applications of the term 'interactive' to the study of reading are discussed in Lesgold & Perfetti, 1981). In these models, the skilled reader is far from a passive responder to print stimuli, but rather a questioner, judge, summarizer, comparer, predictor, hypothesizor, and elaborator, depending somewhat on the type of text and a great deal on the reader's prior knowledge and current goals. A text's meaning for a particular reader is gradually constructed through the dynamic flow of information to and fro, or top-down and bottom-up in the more usual metaphor. And of course no two readers are likely to construct identical mental models of a given text, inasmuch as they bring to it different knowledge-bases, purposes, and information processing strategies.

However, interaction with a text, although characteristic of skilled L1 reading, is often problematic for L2 readers even at relatively advanced levels. The L2 reader characteristically resembles Skinner's reader, passive before the text and attempting to extract its secret meaning. This characterization often holds true even for L2 readers whose reading in their L1 is nonetheless highly interactive. The potential reasons for such a reversal are many. L2 readers may not have automatized one or more of the component processes of reading in the L2, such as word decoding and recognition, resulting in working memory overload and diversion of attention away from the construction of a text model. Or at a higher processing level, readers may not be familiar with semantic or discourse schemata specific to the culture of the L2, so that they have no preactivated scaffolding to help them summarize and organize the details of the incoming text, and quickly face a combinatorial explosion. For these and related reasons, many L2 readers experience reading as a one-way flow of information coming from the text to them, and never get close to sending any messages of the types suggested above back to the text. So one objective for L2 reading courseware might be to encourage the automatization of certain controlled processes such as decoding, or it might inform the learner about certain discourse schemata, or in some other way attempt to establish the preconditions for eventual interaction. Perfetti (1983) has advocated such a role for courseware with regard to young L1 readers, and Frederiksen (1986) has implemented and tested related ideas with L2.

Text manipulation courseware attacks the problem in a different, possibly complementary, way. TM simulates the target activity itself, rather than giving practice in any of its precondition or component processes. At any of a number of levels, text manipulation externalizes the otherwise invisible reader-text interaction and gives the reader supported practice

in actually interacting. Readers faced with a text that has been deconstructed in one of the ways described above must operate on it by questioning it, hypothesizing about what it might mean, or how it might fit together. Readers have no choice but to interact if they want to play; passive meaning-extraction is not an option.

Admittedly, the simulations of interaction provided by a TM system may not be perfect ones. Many of the typical TM operations that must be performed to reconstruct a text will not be far below the surface level, whereas the target interaction will be mainly semantic. Nonetheless, we assume that an L2 reader who, for example, drags the boxed sentences of a text into place with the mouse is doing something akin to what skilled L1 readers do unconsciously when they read - for instance, puzzling out the logical connection between two sentences, or supplying a bridging inference from memory or elsewhere in the preceding text. Further, when the box has been placed, we assume that the TM system's mechanical feedback then simulates the far more subtle confirmatory or disconfirmatory feedback supplied for the skilled L1 reader by subsequent text itself. How successfully TM operations simulate the high-level interactions that will eventually characterize skilled reading, and with what eventual degree of transfer, are empirical questions. The best-case scenario is that the habit of interaction is transferable to on-paper reading regardless of the exact level of the interactions provided by a TM system. In any case, the competition is not stiff; many L2 readers get no interaction with text from solo reading, and only second-hand and/or delayed interaction from classroom reading.

So far, then, we are arguing that Text Manipulation is capable of tapping the mounting tide of text in ways that we can currently implement, and that the interactive model of skilled reading can provide the goals to guide, control, and evaluate development. However, the alert reader will have noticed that this interactive-simulation idea of TM is phrased in a particular conceptual framework, that of information processing or cognitive psychology, and may be aware that adapting such a framework raises some controversies. In the battle with behaviorism, cognitivism may have been or seemed unified, but now that "we are all cognitivists," the subdivisions are assuming more importance. For example, even given the interactive view of skilled reading, how do we know that readers who are skilled interactors in L1 need support for a similar target interaction in L2? Why isn't it enough to give them practice in automatization and component sub-skills peculiar to the L2? With those mastered, then, what they can do at a high level in L1 they will naturally be able to do in L2. Practice in high-level interaction may thus be redundant, and, worse, a diversion of time and energy away from where it is needed. This is precisely what many argue in L1 reading research (Perfetti, 1983 and 1985; Stanovich and Cunningham, 1991), and the case has been recently extended to L2 (Segalowitz, 1986; Polson, 1992). If true, this would be a serious argument against any further development of TM, especially against developing a whole new generation of it to exploit the voluminous textbase about to present itself. We believe the argument is false, but must dredge up a little history to frame the issue.



The interactive version of reading, with the reader contributing to the construction of text meaning in conjunction with the text itself, is often seen as an "attractive" account of this ultimate human activity. In fact, however, it rests on the rather dismal realization that human working memory is far too limited for behaviorist theory to have much applicability to reading. The constant theme in cognitive studies from Miller (1956) onward is that the mind uses various tricks, like chunking and prediction, to compensate for processing limitations. Experiments showed even simple acts of perception to be "knowledge driven" to varying degrees, and moreso complex information processing like reading. For example, on the level of word perception Tulving & Gold (1963) found that deformed words were better perceived when primed by more context, in other words by more prior expectation. On the level of discourse Bransford & Johnson's 'laundry story' (1972) showed that not only immediate comprehension but also subsequent memory for a story was determined by prior expectation. The studies are legion; the theme is that expectation, especially well-structured expectation (in the form of models, schemas, scripts, grammars, and others) is needed to cope with the otherwise overwhelming flow of incoming information. Also, such structures are important in view of how much typically gets left out of texts and yet is required for their comprehension, to be supplied from the reader's store of default or schema knowledge (Minsky, 1975; Schank & Abelson, 1977).

The pedagogy of reading implied by this version of human information processing seemed straightforward. The application came mainly from Smith (1971) and Goodman (1967) under the heading "reading as a psycholinguistic guessing game." In their model, reading is barely perception driven at all (at least, not after the first few sentences to set the scene). Having made predictions at various levels, from various contextual sources, activating the relevant schema, the skilled reader "feedforwards" through the text, merely "sampling" from the words themselves and stopping for a closer look when predictions are unconfirmed. The role of text is thus changed from authoritative to merely suggestive. In both Smith and Goodman, the reader constructs the text almost as much as writer, and the beginning reader should be encouraged to be as constructive as possible. Moreover, and here is the nub of the pedagogy, he should be discouraged from any major effort to pay closer attention to the text itself, such as careful word decoding.

"Reading as writing" was very much the original basis of the text manipulation concept. The deformed on-screen text simulates and of course exaggerates the limited usefulness that characterizes any text surface as given. A "storyboard," with every word masked apart from the title, is essentially Goodman's idea of what any text "really" looks like to the brain: a set of suggestive symbols encoding a message to be reconstructed through interaction with any prior and contextual information sources available, as opposed to a set of fixed signs whose single meaning is to be determined linearly from the combined independent meanings of the words. (Admittedly, a TM routine insists in the end on a single exact surface reconstruction, but this can be de-emphasized to some extent by imaginative programming.)

The applicability of psycholinguistic reading theory to second-language reading seemed obvious (Clarke & Silberstein, 1977; Coady, 1979), and the theory quickly assumed the status of dogma in EFL/ESL practice (See Grabe, 1991, for more background). Clearly, if native speakers must bring a lot of their own information to the act of reading, then the second language learner perforce brings more. If reading is a guessing game even when most of the words and discourse conventions are familiar, how much moreso with a large proportion of the words and discourse conventions unknown or semi-known. The case seemed quite strong for providing L2 readers with a practice environment in which to develop guessing and related strategies - especially one that feeds back to the guesses in shorter loops than are provided by nature. In other words, in the late 1970s the case seemed strong for developing Text Manipulation: the theory matched the technology becoming available.



Given the enormous influence of the Smith-Goodman view of reading in both L1 and L2, its instructional prescriptions and effects have been remarkably little researched. Perhaps it is because the theory, as a processing model, is actually quite short on specifics, as Perfetti (1985) believes. Perhaps it seemed as if the enormous weight of psychological evidence for top-down processing made testing of the "obvious" instructional application unnecessary (an idea that is almost never true). Least researched of all, of course, has been the CALL extension of the pedagogical application. And many involved in TM believe that to undertake such research now would be somewhat irrelevant, as the reading theory underpinning the idea has already started to unravel.

It was probably inevitable that the Smith-Goodman theory of reading would come in for some criticism over the late 1970s and 1980's, since the pendulum has been swinging in L1 reading all this century between expectation-driven and perception-driven reading theories, with the latter currently in the ascendant (Adams, 1990, provides a good backgrounder). Fashion aside, however, some novel research paradigms and techniques emerged in these years that seemed to produce genuinely new information about the nature of skilled reading, namely the expert-novice comparison (Lesgold, 1984). Unexpectedly, in several studies seeking to identify the actual characteristics that divide skilled readers from unskilled, guessing and predicting often came in quite low on the list.

Sampling from a very large pool: Mitchell and Greene (1978) argued that Goodman's eye-movement data could represent any number of underlying cognitive processes, and that when a less ambiguous measure was used, no evidence at all of the use of prediction in skilled reading would emerge. Their consistent experimental finding was that reading speed is not a function of predictability levels of texts. Balota, Pollatsek, and Rayner (1985) searched the mechanisms of parafoveal vision for right-of-focus predictive influences on word recognition, and found that while such influences exist they rarely take precedence over direct visual information in normal reading. Perfetti, Goldman, and Hogaboam (1979) found that while contextually predictable words are identified a little more quickly than unpredictable, even skilled readers' predictions are accurate at a rate of only 20-30% and therefore cannot be the basis of their success. Graesser, Hoffman, and Clark (1980) found that for good readers, neither speed nor comprehension is significantly affected by the degree of syntactic predictability of each additional word in a sentence, although weak readers are significantly aided by higher predictability. Possibly strongest in this line is Stanovich & West (1979 & 1981), who found an effect similar to Graesser et al but for semantic predictability: good readers are aided by it, moderately and unconsciously, but weak readers rely on it strategically, to the extent that they are thrown off when their predictions are wrong.

The theme to emerge was that poor readers guess and predict a good deal, either because they do not know enough words, or do not know them well enough, or can not recognize visually those they know phonologically, at least not fast enough to beat the decay-rate of information from working memory. A coherent sequence of studies on this subject can be found in Perfetti (1985). Study after study in the 1980s showed speed of context-free, expectation free, word decoding to claim the majority of the variance in multiple regression analyses with numerous reader attributes pitted against comprehension as the dependent measure. The instructional implication here is that practice in rapid word recognition, not practice in guessing, is what can turn weak readers into strong.

The decoding issue was slow to arrive in L2 reading theory, possibly because reading-as-predicting had become so dominant (as suggested by Grabe, 1991). However, a sign that the tide has finally turned can be found in a number of the contributions to Huckin, Haynes, and Coady (1993), who qualify severely the nature, role, importance, and conditions of guessing in L2 reading. Coady, as noted above, was one of original importers of "psycholinguistic" reading into L2 theory. The emergence of findings counter to guessing theory suggest that CALL reading software, rather than promoting student strategies in predicting and hypothesizing, would be better devoted to various schemes to help learners develop automatic decoding abilities for the highest frequency words. In fact, some large-scale CALL projects now seem headed in this direction (for example, Coady et al, 1993).

If L2 theory and practice were to embrace the latest L1 reading theory as quickly and thoroughly as it once did the so-called psycholinguistic theory, then soon we shall all be teaching word-lists and rapid decoding via our various media. This may be inevitable; Selinker (1992) characterizes EFL/ESL as a field fond of throwing out the little it achieves in periodic swings to discover ever newer and more exciting theoretical underpinnings. L2 reading research is bound to follow L1 in significant ways, given the relative size and gravitational pull of the two enterprises. In any case, it is no doubt true that there is a greater role in L2 reading for more specific vocabulary and word recognition training, particularly at the early stages, as argued by many of the contributors to Huckin et al (1993). However, an argument can be made for encouraging L2 reading researchers to be more discriminating about what they borrow from L1 research and how they interpret and adapt it (also the view of Grabe, 1991).


L1 reading research does not necessarily map onto L2 reading in any simple or obvious way. Even Perfetti (1985), the arch-foe of guessing theory, suggests as much:

Skilled reading is, by definition, a very fluent process. If a skilled reader fixates three or four words per second, around the normal rate, where is there time to guess? Moreover, if he is skilled at reading, why bother? Reading is much easier than guessing. The case may be different in, for example, reading in a foreign language that is incompletely mastered. There is plenty of time to guess in such cases and perhaps enough payoff for doing so."  (p. 26)

Actual studies looking into subtle L1-L2 differences are somewhat sparse; however, a number have attempted to replicate some of the L1 reading experiments mentioned above with L2 readers and obtained rather different results. For example, the key Stanovich & West experiment mentioned above was replicated in Quebec by Favreau and Segalowitz (1983) with skilled and less-skilled bilingual readers, and rather different patterns of context sensitivity were obtained. What Stanovich and West characterized as less-skilled readers' over-reliance on and yet poor use of contextual information was found precisely to characterize slow but otherwise highly skilled L2 readers. In other words, skilled L2 readers were found to be still strategically reliant on context to recognize a large proportion of words, and yet not very successful in using the information context offers, just as weak L1 readers are. Skilled, flexible, automatized use of context apparently does not automatically transfer from L1 to L2, even when the foundations for it appear to be in place.

Second-language readers' apparent context-insensitivity even at otherwise high skill levels is not an extensively documented phenomenon, yet it appears to exist. For example, it appears in a series of mainly unpublished studies discussed by McLaughlin (1987). The theme of these studies is summarized in McLeod and McLaughlin (1986). Their study compared the read-aloud errors of both more and less skilled L2 readers against those produced by L1 readers in terms of meaningfulness, or contextual goodness-of-fit. One sentence in the text the subjects read was, "She shook the piggy bank, and out came some money." Predictably, if young L1 subjects did not know the word 'money,' they might replace it with "dimes," a semantically reasonable alternative. But if L2 students did not know "money," they tended to replace it with something orthographically similar but contextually violating, like "many." The main point of interest in the study concerns the advanced ESL students. Advanced students made far fewer errors than beginners, as one would expect, but of those that remained just as large a proportion were context violating or non-meaningful. In other words, if recognition was not automatic, there was no strategy for producing a reasonable guess. A few other experiments confirm the existence of this phenomenon, including Arden-Close's (1993) work here in Oman. It is even beginning to reappear in L1 studies (e.g. Oakhill, 1993).

So direct instruction or practice in reading-as-interaction or even reading-as-educated-guessing makes some sense in principle in L2, whatever other realities may exist in L1. The recent decoding movement has not made the idea, or its applications such as Text Manipulation, untenable. Psycholinguistic reading theory has not unraveled as much as it has been moderated, supplemented, and specified in useful ways - specified in that it has been shown to have a special relevance to L2. So it remains worthwhile to ask the empirical questions: To what extent does work on Text Manipulation software produce context sensitivity in various kinds and levels of L2 readers, i.e. produce a more interactive reader, who habitually integrates text information with his or her own prior knowledge in such a way that 'out fell some many' becomes impossible. Who needs this training, all or merely some readers, and how do we find out? If training in reading at both high and low levels is required, what are the proportions and what is the sequencing?

This excursion into theoretical background, we would argue, builds a plausible case for text manipulation in line with what is currently known about the reading process, and suggests a number of hypotheses for empirical research and a rationale for doing that research. However, as important as it is to seek guidance from the larger questions and promise of eventual answers, the actual "state of the art" for TM is that it is a plausible idea bearing a nice technological opportunity but with empirical examination hardly begun. Where should one begin an empirical examination? We follow Long's (1983) argument that in second-language acquisition research generally the research cycle ideally moves from descriptive to correlational to experimental studies, and that no phase should be skipped. Chapelle (1990) proposes the applicability of this cycle to the CALL area, and as mentioned before complains that the descriptive phase has hardly begun. What, if anything, do learners actually do with reading courseware of the kind we produce and talk about? What are the variables? What strategies if any seem to emerge? We argue that the first step is to build a database on the mainly descriptive level. Of course, the building of such a database is already under way, so a preliminary task is to find out what others have been doing and where our work can slot in.



Studies of the strategies employed by students when working Text Manipulation programs have not been well worked out. Researchers into student strategies tend to make separate lists of perceived strategies, with little or no attempt to reconcile lists or work within a framework, or sometimes even to categorize the strategies according to cognitive load required. Therefore, the picture that emerges is not clear. The following will serve as a sampling of some of the findings to date.

Because of its interest to researchers as a measure of language proficiency, student strategies when working cloze have been much examined, though not necessarily as a computer-based exercise. Feldmann and Stemmer (1987) for example, found that in solving C-tests, solution of gaps was either "automatic" or "non-automatic" (spontaneous vs. considered). In the latter case, recall strategy was used leading either to delay, giving up, or activation of another recall strategy. Once an item was recovered, evaluation strategy was used to check appropriateness (also used for automatic recovery), leading to acceptance or rejection of the item for that blank. Since production problems (e.g. spelling) could still occur after recall of the item, application strategies might also have to be used.

Lee's (1990) survey of the previous decade of research on reading examines several genres of research instrument including cloze. The section on beginners dwells heavily on Nunan, 1985, who finds that "unlike more advanced learners, beginning language learners are less able to perceive (or perhaps utilize) intratextual relationships when carrying out written cloze tests" (p.5). Advanced readers "unlike native readers ... are more reliant on local redundancy in a text in order to complete a cloze test than they are on longer range redundancy" (Douglas, 1981). One conclusion from Nunan's study is that "Beginning language learners are not able to take in the text as an integrated expression of ideas, when the text is violated by blanks. This finding may be a by-product of the fact that the text itself, as presented to readers, is not an integrated expression of ideas." (p.5) Accordingly, Feldmann and Stemmer (1987) found that in processing text, only 2 of 20 subjects attempted to skim the entire text, but gave up quickly, as it was impossible because of the gaps (Cohen, Segal & Weiss, 1985, instructed students to skim cloze passages first, but reported similar breakdown). Alderson (1980) gives further evidence of students not treating cloze passages as integrated readings when he says "the nature of the cloze test, the filling-in of gaps in connected text, forces subjects to concentrate on the immediate environment of the gap ..." (p.74), and he finds that varying the amount of context has no predictable bearing on the ability of either NS or NNS to solve cloze tests. "Neither native nor nonnative speakers were aided in their ability to restore deleted words, or a semantic equivalent, or a grammatically correct word, by the addition even the doubling, of context around the deletion" (p.72). Alderson's findings have been countered in Bachman and Jonz.

One possible advantage to computer-based cloze then might be in taking advantage of enhancements to the reader-to-text interaction possible when the gaps violating the text respond to student attempts at recovering them. In the first place, as Feldmann and Stemmer note, it is possible that the more of the text that is solved, the more redundancy learners have at their disposal to elucidate unsolved blanks; and students working computer-based cloze have the advantage of knowing whether blanks solved have been filled in correctly or not (incorrect words left intact in paper-based cloze might further skew meaning). Secondly, students receive feedback as they go, to whatever degree granted by the program designer, and do indeed show evidence of hypothesis formation, testing, and reformation as they interact with feedback provided in at least one computer-based cloze program (Stevens, 1990).

In the study referred to above, 36% of 100 paragraph-length L2 cloze interactions were substantatively completed (with 22% completing the entire paragraph). However, there is also evidence in the same data of students giving up on passages started (49% of all interactions are essentially nil sessions, where students log on, check things out, and log off again with little or not interaction; and a further 16% quit within the first sentence). Though it is not clear if this is because they were unable to complete the passages or simply didn't want to, the former seems unlikely, as use of the hints and help features built into the program guarantee solution of any problem by anyone who persists. It is therefore more likely that many of the students who logged on to the computer-based text manipulation programs in the Stevens study were simply "window shopping", just looking for something to do for a few minutes, but not in the mood for cognitive engagement.

There is probably nothing inherent in the medium that would provoke this outcome other than the fact that it was possible in the study to gather data unobtrusively, without students knowing that they were being monitored (in this study no attempt was made to identify individual students, thus there were no violations of privacy, and also no compunction on students to concentrate on the task unless motivated from within to do so). An "intrusive" protocol is one in which the act of gathering data interferes with the process under study; for example, where the presence of video equipment or the need to 'think aloud' causes learners to monitor their behavior more closely than they might if left to their own devices. Non-intrusive studies attempt to gather data from students who have no idea that they are being monitored, on the theory that only in this way can self-access use of TM be honestly elucidated.

It seems reasonable that results from non-intrusive studies should contradict those from intrusive ones. For example, Windeatt (1986) finds in an intrusive study (where screens were videoed and students thought aloud and were later interviewed) that students make little use of program help features. The unobtrusive studies of Stevens however (Stevens, 1991/CAELL Journal; Stevens, 1991/ERIC, Cloze; Stevens, 1991/ERIC, Hangman) suggest that Arab university students working under self-access conditions tend to abuse help features rather than to consistently apply their own cognitive abilities toward solving the problems. If it is true that whether students know they are being monitored is a factor in their resort to help, then whether a study is intrusive or not could itself be an important variable determining outcome.

There is some evidence that students who rely excessively on program-supplied help are not learning as much as those who try to solve problems through trial-and-error feedback accompanied by judicious use of help. Pederson (1986) for example showed differences in cognitive processing when comparing students who had access to help in the form of ability to review the reading passages while answering comprehension questions with those for whom such access was denied. "The results indicate that passage-unavailable treatment always resulted in a comparatively higher comprehension rate than occurred in counterpart passage-available treatments regardless of the level of question or level of verbal ability." (p.39) In other words, "greater benefit was derived from the subjects' being aware that they were required to do all of their processing of the text prior to viewing the question" (p.38). It follows then that in using text manipulation as a means of having students engage in "reading as guessing," help should not be given to such an extent that guessing is suppressed.

The possibility (indeed, likelihood) that students may not of their own free will choose a pathway through the CALL materials leading to optimal learning suggests a re-examination of the magister-pedagogue dichotomy (Higgins, 1983, 1988) which has strongly influenced CALL software development over the past decade. Rather than the computer acting as a pliant slave which unquestioningly obeys all student commands (the role favored in the dichotomy), it may be that an entity which aids the learner on demand while exercising enlightened authority over the learning process is more conducive to learning. But how much authority can a program exert without depriving students of benefits of autonomous learning (thus tending to be a magister, in terms of the dichotomy)?

One problem with allowing learners control over their own learning is getting them to take advantage of available options. How, for example, can students be encouraged to select and learn to interpret unfamiliar forms of feedback? Bland, Noblitt, Armstrong, and Gray (1990) discovered in a Système D implementation that although students had access to both dictionary and lexical help, they avoided lexical help for fear of getting lost in it. "We were initially surprised at the very few queries of this nature in the data" (p.445). Furthermore, in an attempt to reverse the outcome of the Stevens' Hangman study (19??), where it was found that 53% of the students were touring the material with unacceptable levels of cognitive engagement, the program was reconfigured to present context surrounding the target word on demand. The demand feature comes at the cost of points, the idea being for students to request just as much context as they need to solve the problem. On examination of the first set of data after the revised program was implemented, it was found that cognitive engagement remained about the same and that the students weren't using the context feature, probably because the program failed to make them aware of it. These are just two examples of the caveat that simply providing options to students by no means ensures that they will use them.

Cobb has found much the same thing in his ongoing doctoral research into learners' use of on-line concordance (with keyword masked) as the help in a systematic deletion exercise: learners' use of this support in self-access was virtually non-existent, in spite of their knowing about it and having tried it in a practice session, and also in spite of the fact that when they did use it in the practice session they had double the success rate of either no-help or dictionary-help. In order for the experiment to continue, the system had to be reconfigured three times to make the concordance window unavoidable. Admittedly, spontaneous use of the concordance increased with familiarity, but not in proportion to the increasing advantages it produced, both on-line and later on in classroom paper-and-pencil cloze tests for the same vocabulary items. The conclusion can only be that the benefits of concordancing, and no doubt other features of Text Manipulation, may not be simple for learners to perceive even when they are working well.

All the evidence seems to point in the same direction: most learners will not find the most useful ways to manipulate components of texts or use text-generated help by themselves. Numerous published studies have shown this, and we have replicated many of them in our own labs. Two examples are enough to make the point. Firsty, one strategy frequently noted when students use TM programs is a tendency to proceed linearly (rather than holistically, as one might be expected to do if reading a passage and drawing inferences from outside the immediate context). Edmondson, Reck, and Schroder (1988) tracked 9 secondary level students doing a combined jumble sentence/paragraph exercise called SHUFFLE and noted a tendency for students to use "frontal-attack" strategies; that is, take the first available sentence and try to place it; or build from the first sentence to the next and so on. Edmondson et al distinguish between coherence discovery strategies and linearization ones. The former refers to moves made while developing a hypothesis that is validated if shuffling text eventually achieves coherence, while the latter means grasping the correct order of words or paragraphs and then shuffling in order to make the meaning take shape. Though either case can yield an identical sequence of tracked events, the two are distinguished according to verbal reports or from noting the timing of movements of elements during reconstruction. competency based strategies for students working a version of Hangman, and found that fully 53% of student moves studied were in the latter category, whereas for a control group of native speaking language instructors, non-competency based moves accounted for only 3% of all those made. Second, Windeatt (1986) found that his subjects completed computer-based cloze blanks in predominantly linear fashion even though the system didn't require it (perhaps because they didn't like to scroll from screen to screen). and these findings are also largely confirmed in the Stevens cloze data.

As pointed out by Chapelle and Mizuno (1989), the issue of optimal degree of learner control over CALL had "not yet been investigated." With investigation tentatively under way, it is still fair to say that the issue of learner control is far from being resolved. However, inital optimism about learner control in Text Manipulation seems largely unsustainable, as has been more or less confirmed for some time in the wider world of computer assisted instruction generally (e.g. Steinberg, 1989).


Text Manipulation reading courseware is technologically capable of exploiting the growing supply of on-line text, mainly in English, that the world is producing. This in itself would be insignificant if the idea had no likelihood of helping learners become better readers. While the interactive reading theory that TM derives from has been challenged, there remains a plausible case for pursuing the development of TM materials, particularly in second language, no doubt in conjunction with other types of materials working at a lower or less holistic level. However, the plausibility of TM must shortly be supported by more definite evidence that TM actually produces differences in skill acquisition over alternatives, on-line or off. For a number of years now we have heard about what TM could do if learners would just use it properly, always on the assumption that TM and learner control were an inseparable duo. It now seems that whatever processes TM instigates, like those of language learning itself, are beneath the learner's level of perception. Really, this should be no surprise. It appears that developers at this point should take advantage of the descriptive data available and feed it back into the design process, particularly that part of the process concerning learner control. If effectiveness research must wait on learners to discover for themselves the best ways to exploit TM, the idea will likely end up on the scrap heap of plausible but unsuccessful educational technologies without ever having been given a fair test. We are finding that we have to make our TM programs somewhat more magisterial to get any research out of them. For the same reasons, they may have to be somewhat more magisterial for learners to get any use out of them.



Adams, M.J. (1990). Beginning to read : Thinking and learning about print. Cambridge, Mass. : MIT Press.

Adams, M.J. & Collins, A. (1979). A schema-theoretic view of reading. In R. O. Freedle (Ed.). New directions in discourse processing (pp. 1-22). Norwood, N.J.: Ablex.

Alderson, J. C. (1979). The cloze procedure and proficiency in English as a foreign language. TESOL Quarterly 13, 219-227.

Alderson, J. C. (1980). Native and nonnative speaker performance on cloze tests. Language Learning 30 (1) 59-76.

Alderson, J. C., & A.H. Urquhart (Eds.) (1984). Reading in a foreign language. Burnt Mill, Harlow: Longman.

Allen, E., E. Bernhardt, M. Berry, & M. Demel. (1988). Comprehension and text genre: An analysis of secondary school foreign language readers. Modern Language Journal 72, 163-172.

Anderson, N. J. (1991). Individual differences in strategy use in second language reading and testing. The Modern Language Journal 75 (4), 460-472.

Arden-Close, C. (1992). Learners' strategies for inferring the meanings of unknown words. Unpublished manuscript. Sultan Qaboos University, Muscat, Oman.

Bachman, L.F. (1982). The trait structure of cloze text scores. TESOL Quarterly 16, 61-70.

Bachman, L.F. (1985). Performance on cloze tests with fixed-ratio and rational deletions. TESOL Quarterly 19, 535-556.

Bacon, S. M. & M. D. Finnemann. (1990). A study of the attitudes, motives, and strategies of university foreign language students and their disposition to authentic oral and written input. The Modern Language Journal 74 (4), 459-473.

Balota, D.A., Pollatsek, A., & Rayner, K. (1985). The interaction of contextual constraints and parafoveal visual information in reading. Cognitive Psychology 17, 364-390.

Bernhardt, Elizabeth, and Victoria Berkemeyer. (1988). Authentic texts and the high school German learner. Unterrichtspraxis, 21, 6-28.

Bland, S. K., J. S. Noblitt, S. Armstrong, & G. Gray. (1990). The naive lexical hypothesis: Evidence from computer-assisted language learning. The Modern Language Journal 74 (4), 440-450.

Bransford, J.D., & Johnson, M.K. (1972). Contextual prerequisites for understanding: Some investigations of comprehension and recall. Journal of Verbal Learning and Verbal Behavior 11, 717-726.

Butler, J. (1991). Cloze procedures and concordances: The advantages of discourse level authenticity in testing expectancy grammar. System 19, (1/2), 29-38.

Chapelle, C. (1990). The discourse of computer-assisted language learning: Toward a context for descriptive research. TESOL Quarterly, 24 (2), 199-225.

Chapelle, C. & S. Mizuno. (1989). Student strategies with learner controlled CALL. CALICO Journal 7, 2:25-47.

Chapelle, Carol, and Joan Jamieson. (1986). Computer-assisted language learning as a predictor of success in acquiring English as a second language. TESOL Quarterly 20 (1), 27-46.

Chavez-Oller, M.A., Chihara, T. Weaver, K.A., and Oller, J.W., Jr. (1985). When are cloze items sensitive to constraints across sentences? Language Learning 35, 181-206.

Clarke, M. & Silberstein, S. (1977). Toward a realization of psycholinguistic principles for the ESL reading class. Language Learning 27, 135-154.

Coady, J. (1979). A psycholinguistic model of the ESL reader. In R. Mackay, B. Barkman, & R.R. Jordan (Eds.), Reading in a second language (pp. 5-12). Rowley, MA: Newbury House.

Coady, J., Magoto, J., Hubbard, P., Graney, J., & Mokhtari, K. (1993). High frequency vocabulary and reading proficiency in ESL readers. In T. Huckin, M. Haynes, & J. Coady (Eds.) Second language reading and vocabulary learning. Norwood, NJ: Ablex.

Cobb, T.M. (1992). TEXPERT: Textbase and question engine. Montreal: Concordia University, software.

Cohen, A., Segal, M., and Weiss, R. 1985. The C-test in Hebrew. Fremdsprachen und Hochschulen. AKS-Rundbrief 13/14, 121-127.

Cousin, W. (1990). Computer clozentropy measures: Groundwork for an investigation. manuscript (supplied via personal communication to V. Stevens).

Cziko, G. A. (1980). Language Competence and reading strategies: A comparison of first- and second-language oral reading errors. Language Learning 30, 101-115.

Darnell, D.K. (1968). The development of an English language proficiency test of foreign students using a clozentropy procedure. Final report. University of Colorado, Boulder. Dept. of Health, Education, and Welfare Research Bureau No. BP-7-H-010TP.

Dhaif, H. 1990. Computer assisted language learning: A client's view. CALICO 7, (4)67-81.

Dick, W. (1991). An instructional designer's view of constructivism. Educational Technology 31 (5), 41-44.

Douglas, D. (1981). An exploratory study of bilingual reading proficiency. In Hudelson, S. (Ed.). Learning to read in different languages. Washington Center for Applied Linguistics.

Dulany, D.E., R.A. Carlson, & G.I. Dewey. (1984) A case of syntactical learning and judgment: How conscious and how abstract? Journal of Experimental Psychology, General 114, 25-32.

Dunkel, P. 1991. Computer-assisted language learning and testing: Research issues and practice. Rowley, MA: Newbury House.

Dunkel, P. (1991). The effectiveness research on computer-assisted instruction and computer-assisted language learning. In P. Dunkel (Ed.), Computer-assisted language learning and testing: Research issues and practice. New York: Newbury House.

Dunkel, P. (1987). The effectiveness literature on CAI/CALL and computing: Implications of the research for limited English proficiency learners. TESOL Quarterly 21, 367-372.

Edmondson, W., S. Reck, & N. Schroder. (1988). Strategic approaches used in a text-manipulation exercise. In Jung, U. O. H. (Ed.). Computers in applied linguistics and language teaching (pp.193-211). Frankfurt: Verlag Peter Lang.

Ehrman, M., & R. Oxford. (1990). Adult language learning styles and strategies in an intensive training setting. The Modern Language Journal 74 (3), 311-327.

Faerch, C. & G. Kasper (Eds.) (1987). Introspection in second language research. Clevedon, UK: Multilingual Matters, 30.

Favreau, M., & Segalowitz, N.S. (1983). Automatic and controlled processes in the first- and second-language reading of fluent bilinguals. Memory and Cognition 11 (6), 565-574.

Feldmann, U., & B. Stemmer. (1987). Thin- aloud a- retrospective da- in C-Te- taking: Diffe- languages - diff- learners - sa- approaches? In Faerch, C. and G. Kasper (Eds.). Introspection in second language research (pp. 251-267.) Clevedon, UK: Multilingual Matters, 30.

Frederiksen, J.R. (1986). Final report on the development of computer-based instructional systems for training essential components of reading. Report No. 6465. Cambridge, MA: BBN Laboratories.

Gaies, S.J. 1986. Validation of the noise test. In R. Grotjahn, C. Klein-Braley, & D.K. Stevenson (Eds.). Taking their measure: The validity and validation of language tests. Quantitative Linguistics, 30.

Garrett, N.. 1988. Software review of Text Tanglers: A Vance Stevens RDA/Mindbuilders Software. Journal of Educational Techniques and Technologies 21 (2/3), 59-61.

Goodman, K.S. (1967). Reading: A psycholinguistic guessing game. Journal of the Reading Specialist 6, 126-135.

Goodman, K.S. (1974). Miscue analysis: Theory and reality in reading. Paper presented at 5th World IRA Congress, 1974. In Merritt, J. (Ed.). Proceedings, New horizons in reading. International Reading Association 19, 1976.

Grabe, W. (1991). Current developments in second language reading research. TESOL Quarterly 25 (3), 375-406.

Graesser, A.C., Hoffman, N.L., Clark, L.F. (1980). Structural components of reading time. Journal of Verbal Learning and Verbal Behavior 19, 135-51.

Harri-Augstein, Sheila, and Laurie F. Thomas. (1984). Conversational investigations of reading: The self-organized learner and the text. In Alderson, J. Charles, and A.H. Urquhart (Eds.). Reading in a foreign language. Burnt Mill, Harlow: Longman.

Hart, R. (1981). Language study and the PLATO system. Studies in Language Learning 3, 1-24.

Heppner, F., Anderson, J., Farstrup, A., & Weiderman, N. (1985). Reading performance on a standardized test is better from print than from computer display. Journal of Reading, January, 321-325.

Higgins, J. 1983. Can computers teach? CALICO 1, (2) 4-6.

Higgins, J. 1988. Language, learners and computers. London: Longman.

Higgins, John. 1991. Fuel for learning: The neglected element of textbooks and CALL. CAELL Journal 2, (2) 3-7.

Hosenfeld, C. 1976. Learning about learning: Discovering our students' strategies. Foreign Language Annals 9, 117-129.

Hosenfeld, C. (1984). Case studies of ninth grade readers. In Alderson, J. Charles, and A.H. Urquhart (Eds.) Reading in a foreign language ( pp. 231-240). Burnt Mill, Harlow: Longman.

Hubbard, P. (1992). A methodological framework for CALL courseware development. In Pennington, M., and V. Stevens (Eds.) Computers in Applied Linguistics (pp. 39-65). Clevedon: Multilingual Matters.

Huckin, T., Haynes, M., & Coady, J. (Eds.) (1993). Second language reading and vocabulary learning. Norwood, NJ: Ablex.

Jamieson, J. & C. Chapelle. 1987. Working styles on computers as evidence of second language learning strategies. Language Learning 37, (4)523-544.

Johns, T. 1991. From printout to handout: Grammar and vocabulary teaching in the context of data-driven learning. In Johns, T., and P. King (Eds.). Classroom concordancing, ELR Journal 4, 27-37. University of Birmingham.

Johns, T. 1988. Whence and whither classroom concordancing? In Bongaerts, T. (Ed.), Computer Applications in Language Learning. London: Foris.

Johns, T. (1991). Should you be persuaded: Two examples of data driven learning. In Johns, Tim, and Philip King (Eds.). Classroom concordancing, ELR Journal 4, 1-16. University of Birmingham.


Jonz, J. (1990). Another turn in the conversation: What does cloze measure? TESOL Quarterly 24 (1), 61-83.

Kienbaum, B., A.J. Russel, & S. Welty (1986). Communicative competence in foreign language learning with authentic materials. Final project report, ERIC 275200.

Klein-Braley, C. 1983. A cloze is a cloze is a question. In Oller, J., (Ed.). Issues in language testing research (pp. 218-228. ). Rowley, MA: Newbury House.

Klein-Braley, C. (1985). C-tests and construct validity. Fremdsprachen und Hochschulen. AKS-Rundbrief 13/14, 55-65.

Klein-Braley, C. (1985). Reduced redundancy as an approach to language testing. Fremdsprachen und Hochschule. AKS-Rundbrief 13/14, 1-13.

Klein-Braley, C. & Raatz, U. (1984). A survey of research on the C-Test. Language Testing 1, 134-146.

Kleinmann, H. (1987). The effect of computer-assisted instruction on ESL reading achievement. The Modern Language Journal 71, 3:267-276.

Krashen, S., & T. Terrell. (1983). The natural approach: Language acquisition in the classroom. Elmsford, NY: Pergamon.

Laufer, B., & H. Osimo. (1991). Facilitating long-term retention of vocabulary: The second-hand cloze. System 19, 217-224.

Lee, J. 1990. A review of empirical comparisons of non-native reading behaviors across stages of language development. Paper presented at SLRF, Eugene, Oregon, (manuscript).

Lesgold A.M. & Perfetti, C.A. (1981). Interactive processes in reading. Hillsdale, NJ: Lawrence Erlbaum.

Lesgold, A.M. (1984). Acquiring expertise. In J.R. Anderson & S.M. Kosslyn (Eds.), Tutorials in learning and memory. New York: W.H. Freeman and Company.

Littlejohn, A. 1983. Increasing learner involvement in course management. TESOL Quarterly 17, 4:595-608.

Long, M. (1983). Inside the 'black box': Methodological issues in classroom research in language learning. In H.W. Seliger & M.Long (Eds.), Classroom oriented research in second language acquisition (pp. 104-123). Rowley, MA: Newbury House.

McClelland, J., D. Rumelhart, & G. Hinton. 1986. The appeal of parallel distributed processing. In D. Rumelhart, J. McClelland, and the PDP Research Group (Eds.), Parallel distributed processing: Explorations in the microstructures of cognition,Vol. 1, Foundations (pp. 3-44). Cambridge, MA: MIT Press.

McLaughlin, B. (1987). Reading in a second language: Studies with adult and child learners. In S. Goldman & H.T. Trueba, Becoming literate in ESL. Norwood, NJ: Ablex Publishing Corporation.

McLaughlin, B., T. Rossman, & B. McLeod (1983). Second language learning: An information-processing perspective. Language Learning 33, 135-158.

McLaughlin, Barry. (1990). "Conscious" versus "unconscious" learning. TESOL Quarterly 24, 4:617-634.

McLeod, B., & McLaughlin, B. (1986). Restructuring or automaticity? Reading in a second language. Language Learning, 36 (2), 109-123.

Miller, G.A. (1956). The magical number seven, plus-or-minus two: some limitations on our capacity for information processing. Psychological Review 63, 81-97.

Millmore, S., & Stevens, V. (1990). SUPER CLOZE (in various versions). Shareware available through CALL Interest Section MS-DOS Users Group, TESOL.

Minsky, M. (1975). A framework for representing knowledge. In P.H. Winston (Ed.), The psychology of computer vision. New York: McGraw-Hill.

Mitchell, D.C., & Green, D.W. (1978). The effects of context and content on immediate processing in reading. Quarterly Journal of Experimental Psychology 30, 609-636.

Mosco, V. (1989). The pay-per society: Computers and communication in the information age. Essays in critical theory and public policy. Norwood, NJ: Ablex.

Nevo, N. (1989). Test-taking strategies on a multiple-choice test of reading comprehension. Language Testing 6, 199-215.

Nunan, David. (1985). Content familiarity and the perception of textual relationships in second language reading. RELC Journal 16, 43-50.

Oakhill, J. (1993). Children's difficulties in reading comprehension. Educational Psychology Review 5 (3), 223-237.

Oller, J.W. (1976). Evidence for a general language proficiency factor: An expectancy grammar. Die Neueren Sprachen, 165-174.

Pederson, K. M. (1986). An experiment in computer-assisted second language reading. Modern Language Journal 70, (1) 36-41.

Perfetti, C.A. (1983). Reading, vocabulary, and writing: Implications for computer-based instruction. In Wilkinson, A.C., Classroom computers and cognitive science. New York: Academic Press.

Perfetti, C.A. (1985). Reading ability. New York: Oxford University Press.

Perfetti, C.A., Goldman, S.R., & Hogaboam, T.W. (1979). Reading skill and the identification of words in discourse context. Memory and Cognition 7 (4), 273-282.

Piper, A. (1986). Conversation and the computer: A study of the conversational spin-off generated among learners of English as a foreign language working in groups. System 14 (2), 187-198.

Polson, C. (1992). Component processes of second language reading. Concordia University, Montreal. Unpublished Master's thesis.

Pritchard, Robert. 1990. The effects of cultural schemata in reading processing strategies. Reading Research Quarterly 25, 273-295.

Reber, A.S., & R. Allen. (1978). Analogic and abstraction strategies in syntactic grammar learning: A functional interpretation. Cognition 6, 189-221.

Roblyer, M.D., Castine W.H., & King, F.J. (Eds.) (1988). Assessing the impact of computer-based instruction: A review of recent research. New York: Hawthorn Press.

Rubin, J. (1975). What the 'good language learner' can teach us. TESOL Quarterly 9, 41-51.

Rutherford, W. (1987). Second language grammar: learning and teaching. Longman.

Schank, R.C. &. Abelson, R.P. (1977). Scripts, plans, goals and understanding: An inquiry into human knowledge structures. Hillsdale NJ: Erlbaum.

Schmidt, R. (1990). The role of consciousness in second language learning. Applied Linguistics 11: 129-158.

Segalowitz, N. (1986). Skilled reading in a second language. In J. Vaid (Ed.), Language processing in bilinguals: Psycholonguistic and neuropsychological perspectives. Hillsdale, NJ: Erlbaum.

Segel, K. 1986. Does a standardized reading comprehension test predict textbook prose reading proficiency of a linguistically heterogeneous college population? Unpublisged dissertation, University of Texas.

Selinker, L. (1992). Rediscovering interlanguage. London: Longman.

Shiffrin, R.M., & W. Schneider. (1977). Controlled and automatic information processing: I. Detection, search, and attention. Psychological Review 84, 1-66.

Stern, H.H. (1980). What can we learn from the good language learner. In Croft, K. (Ed.) Readings in English as a Second Language, 2nd Edition (54-71). Cambridge, MA: Winthrop.

Smith, F. (1971). Understanding reading: A psycholinguistic analysis of reading and learning to read (3rd ed.) New York: Holt, Rinehart, & Winston.

Stanovich, K.E., & Cunningham, A.E. (1991). Reading as constrained reasoning. In R.J. Sternberg & P.A. Frensch (Eds.), Complex problem solving: Principles and mechanisms. Hillsdale, NJ: Erlbaum.

Stanovich, K.E., & West, R.F. (1979). Mechanisms of sentence context effects in reading: Automatic activation and conscious attention. Memory and Cognition 7 (2), 77-85.

Stanovich, K.E., & West, R.F. (1981). The effect of sentence context on ongoing word recognition: Tests of a two-process theory. Journal of Experimental Psychology: Human Perception and Performance 7 (3), 658-672.

Steinberg, E.R. (1989). Cognition and learner control: A literature review, 1977-1988. Journal of Computer Based Instruction 16 (4).

Stevens, V. (1988). A critical but overlooked factor: Not can but will teachers modify the software? C.A.L.L. Digest 4 , (8)3-5.

Stevens, V. (1990). Text manipulation: What's wrong with it anyway? CAELL Journal 1,, (2) 5-8; reprinted in ON-CALL 5, 1:5-10.

Stevens, V. 1990. Text manipulation: What's wrong with it, anyway? CAELL Journal 1, (2)5-8.

Stevens, V. 1991. Classroom concordancing: Vocabulary materials derived from relevant, authentic text. English for Special Purposes Journal 10, 35-46.

Stevens, V. 1991. Computer HANGMAN: Pedagogically sound or a waste of time? Revised version of a paper presented at the Annual Meeting of the Teachers of English to Speakers of Other Languages (24th, San Francisco, CA, March 6-10, 1990). ERIC Document Reproduction Service No. ED 332 524).

Stevens, V. 1991. Concordance-based vocabulary exercises: A viable alternative to gap-fillers. In Johns, Tim, and Philip King (Eds.). English Language Research Journal, Vol. 4 (pp. 47-61), University of Birmingham.

Stevens, V. 1991. Reading and computers: Hangman and cloze. CAELL Journal 2, (3)12-16.

Stevens, Vance. 1991. Strategies in solving computer-based cloze: Is it reading? Paper presented at the Annual Meeting of the Teachers of English to Speakers of Other Languages (25th, New York, NY, March 24-28, 1991). ERIC Document Reproduction Service No. ED 335 952.

Tulving, E., & Gold, C. (1963). Stimulus information and contextual information as determinants of tachistoscopic recognition of words. Journal of Experimental Psychology 66 (4), 319-327.

Tuman, M. C. (1992). Word Perfect: Literacy in the computer age. London: The Falmer Press; Pittsburgh: University of Pittsburgh Press.

Windeatt, S. (1986). Observing CALL in action. In Leech, G. and C. Candlin (Eds.). Computers in English language teaching and research. London: Longman.

Wittrock, M.C. (1979). The cognitive movement in instruction. Educational Researcher 8,5-11.

Wyatt, D. 1989. Computers and reading skills: The medium and the message. In Pennington, M. (Ed.). Teaching languages with computers: The state of the art. La Jolla: Athelstan.