Learners with large lexical needs are not difficult to find in second-language settings, especially where the goal is to perform academically in the second language with minimal delay. Many studies have determined that academic performance in a second language mainly involves building up a large vocabulary (Saville-Troike, 1984; Laufer, 1992). The subjects of the present study are first-year students from the College of Commerce at Sultan Qaboos University (henceforth SQU) in Oman. The following analysis shows that these students do indeed have large lexical needs, and further that these needs can in principle be met with a computer tutorial based on contextual inferencing from a corpus.
All language students need to learn words, and most of them know it. As Krashen (1989) observed, learners walk around with dictionaries in their pockets, not grammar books. Arabic-speaking learners have an especially pressing need to learn words because Arabic has so few cognates with European languages. This fact was noted by Praninskas (1972) when she began teaching English at the University of Beirut, was experimentally confirmed by Ard and Homburg (1983), and has been the ever-present background to English instruction in the Arab world ever since. Even so, few learners have as pressing a need to learn words as students in the College of Commerce, for reasons relating to a particular language test they are required to take.
Since September 1993, first-year students in the College of Commerce have been required to reach Band 4 on a standardized English proficiency test, the Preliminary English Test (Cambridge University, 1990) before proceeding to their English-medium academic courses. Students test in at Band 1, 2, or 3, and have three terms (one academic year) to reach Band 4. The Preliminary English Test (PET) had been used as a placement measure at SQU since 1991, but its use as an exit measure was an experiment, to be tried first in the College of Commerce and then used in other colleges if shown to be a useful measure. When the PET was adapted as an exit measure, there were few guidelines as to what would be involved in getting Band 1 students to Band 4 within a year.
Although not particularly difficult by international standards, at SQU the PET has been an enormous burden on students, instructors, and even the institution as a whole. Many students have been simply unable to reach Band 4 in any reasonable amount of time using the instructional resources available. Here is a letter from a commerce student caught in the PET process, giving some of the flavour of the experience (copied with permission):
Dear Nawal,
I heard that you are going to join the College of Commerce and Economics after you finish your high school. I have a lot to tell you about this college. The first and important thing is the PET test. You must pass this test so you can continue your studies in the College. The PET test is not easy as it seems. It is so difficult and we have to do a lot to pass it.... The English that we learned at school is too easy and it's nothing compared with the English in the University. Let me tell you about myself as an example.
I thought that I knew English and really in the school I was from the three best students in the class in English. But here my English is nothing, then I thought I learned nine years English in the school but I don't have any knowledge and I don't know anything about real English. I really don't know the fault from who.
My advice to you is you must think carefully before you make your decision and think about the PET that tired all the students and I don't know how to get rid of it. Again before you join this college take a course in English. I hope you will understand what I meant. I hope for you good life.
Your friend, F.
This student is unwilling to point any fingers, but "the fault from who" is not a difficult question.
First, the PET is a proficiency test (measuring a learner against a native speaker) not an achievement test (measuring a learner against the content of a course), and used as an exit measure it is likely to test students on things they were never taught. The PET was designed for placement, and its use as an exit measure is controversial.
Second, Omani students arrive at university vastly underprepared for academic work in English. Their high school English is taught almost exclusively by non-native speakers, in a rigidly structured memory-based approach involving almost no communicative use of English in any medium. This use of non-native teachers and dated methods is hardly a matter of economic necessity. As noted in a recent World Bank report (1994), primary and secondary education in Oman are underfunded while money is lavished on the high-profile University, with predictable results. The only change since 1994 is that the public school system has now gone onto half-day shifts to accommodate the burgeoning youth population (median age in Oman is 16).
Third, once arrived at university students are placed in a time squeeze. Cambridge recommends an average 250 hours of language instruction to move a student from Band 3 to 4 of the PET, while SQU grants only 170 hours (Scott, Gerber, Salem, Marzouqa, and Sherazee, 1995). This squeeze is no anomaly, but consistent with a long-standing approach. Even before the University opened in 1986, Adams Smith (1984) reported that the administration wished to grant 150 hours of language instruction to students entering English-medium academic courses, while 500 hours was the area average (as established by numerous British Council experiences). Commerce students are given one year to reach Band 4, or leave the College.
So the task of the CALL developer is to rescue the students from some part of this instructional non-design.
The PET tests separately the skill areas of reading, listening, and writing , and the main problem has consistently been reading, even when modest success has been achieved in listening and writing. Table 6.1 shows a typical score profile from a PET testing session. Even when Band 4 is reached, the students enter their academic areas with weak reading ability.
| Listening | Reading | Writing | |
|---|---|---|---|
| Student 1 | 4 | 3 | 4 |
| Student 2 | 4 | 3 | 4 |
| Student 1 | 4 | 3 | 4 |
| Student 1 | 4 | 3 | 4 |
| Student 1 | 4 | 3 | 4 |
| Student 1 | 4 | 3 | 4 |
| ... | |||
| Student 100 | 4 | 3 | 4 |
Beyond the short term need to deal with the PET, reading is generally viewed as the key skill for academic performance in a second language.
In fact, the PET merely documents a reality about the students' reading ability that has been true since the beginning of the University in 1986. The Language Centre started out with a very ambitious plan to teach the students English via their academic course materials in a scheme known as "content-based English" (Holes, 1985). But as actual students were injected into the plan, it quickly became clear that they were much weaker than course planners had expected, particularly in reading. When tested with a British test called the IELTS (roughly equivalent to the TOEFL) "most students failed to achieve higher than band one and even the best only reached band two, where at least band six would be required for entry to a British degree course" (Flowerdew, 1993a, p. 122). As a result, between 1987 and 1990, the scientific texts the students used were continually simplified and shortened (Flowerdew, 1993a), and content-course lecturers were forced to spend an inordinate amount of class time explaining scientific words (Flowerdew, 1992), doing for the students what students normally do for themselves, at home or in the library. Oddly, during this period it was never thought useful to inquire into the students' knowledge of general or sub-technical English terms.
The arrival of the PET in 1991 marked the final dissolution of the content-based approach and a recognition of prior needs at the level of general English. Of course recognizing and testing those needs in itself said nothing about how to meet them. (Some of these background issues are discussed further in the context of computer-assisted learning in Chapter 5.)
But is the weak reading caused by weak vocabulary? There are general and specific reasons for treating vocabulary needs as primary, both of them admittedly correlational.
Generally, it has long been established that vocabulary size correlates more highly than either syntax or culture knowledge with reading comprehension. A long series of factor analyses in the 1940s and 1950s found vocabulary size to be the highest loading factor in reading comprehension (ranging from .41 to .93, discussed in Anderson and Freebody, 1979). Further, vocabulary size correlates higher with reading than it does with listening (which relies on a smaller word stock) or writing (which relies on the imagination and syntax a student has for recombining whatever words happen to be available). Of course this research does not prove that a large vocabulary guarantees skilled reading, but it does suggest that a small vocabulary makes it quite difficult.
Specifically, there is a huge gap between the number of words these students know and the number they need. On one side of the gap, the PET is a lexically explicit measure; all its texts, tapes, and questions are constrained to a particular list of words. This list consists of the most frequent 2387 words of English according to Hindmarsh's (1980) Cambridge English Lexicon. On the other side of the gap, students entering Band 1 often know as few as 350 English words-a shortfall of about 2000 words. This figure was determined through repeated administration of Nation's Vocabulary Levels Test with entering commerce students between 1993 and 1996, and confirmed using the Eurocentres Vocabulary Size Test (Meara and Jones, 1990).
Nation's test is a simple but well-researched instrument (Nation, 1983; 1990) and arguably the most reliable of the various vocabulary size measures available (Schmitt, 1995), particularly with learners whose first languages are not derived from Latin. The test samples from several vocabulary frequency zones, asking learners to match the words to simple definitions phrased in terms from the next lower frequency zone. The test is quick to administer so that it can be repeated with the same subjects without building up covert resistance. The format is a variant of multiple-choice, with six choices for each word reducing the role of chance.
Figure 6.1 is Nation's Levels Test, slightly compressed, showing most of the 2000 and 3000 levels (but not the 5000, University, and 10,000 levels). Even if this test measures a fairly crude type of word knowledge, definitional rather than transferable, it has been of enormous use at SQU in delineating some approximate task dimensions for helping learners deal with the PET.
A problem raised by the use of the Levels Test is that 2000 does not equal 2387. The Levels Test, at the zone of relevance to the learners in question, is based on West's (1953) General Service List (GSL) of the 2000 most common English words, while the Cambridge Lexicon is a list of 2387 words. These lists are only about 65% overlapping (see Appendix A). However, all the sample items of the Levels Test are also Cambridge Lexicon words, so the provenance of the test does not affect its usefulness. In the present study, the 2000-word Levels Test will be the yardstick for measuring the task facing the students and their progress in getting control of it.
Scores on the Levels Test correlate well with PET success at SQU over the course of the band process. Students entering Band 1 have a remarkably consistent profile of about 15-20% at the 2000-word level. Admittedly, there are some questions about how to interpret a figure this low. It is probably not as simple as extrapolating a vocabulary size of (2000 x .2 =) 400 words, since the test does not sample from the 100 or so prepositions and other function words that the students probably know to some extent. But even supposing the average student knew 500 words, that would still leave 1500 to learn in one academic year .
Admittedly, some number of words less than the complete 2000 will probably see a learner through Band 4. The Word Levels test appears to show students reaching Band 4 (but with Band 3 reading ability) with scores as low as 70-75% at the 2000 level, or 1400-1500 words. In that case the to-be-learned number drops to about 1000 words. But this is still a formidable number; to put it in perspective, the average lexical growth for English learners in western Europe is about 275 new words per six-month term, or 550 words a year, by a recent calculation (Milton & Meara, 1995). And this is achieved in an environment permeated with English pop-culture, by learners whose first languages share many cognates with English.
Over the course of repeated testing in the College of Commerce, it became apparent that vocabulary size correlated consistently with PET success and failure. The groups selected for presentation below were intact groups of students who had either just passed or just failed a PET band. In spite of some predictable noise in the data, there appears to be a remarkably steady rise in vocabulary size for successful students, and a stagnation for failing students.
| PASS Band 1 | FAIL Band 2 | PASS Band 2 | PASS Band 3 | ||||
|---|---|---|---|---|---|---|---|
| FEB94 | FEB95 | FEB94 | FEB95 | FEB95 | FEB95 | FEB95 | |
| Group | "1A" | "1B" | "2D" | "2B" | "2C" | "3A" | "3B" |
| 33% | 39% | 27% | 33% | 44% | 88% | 50% | |
| 33 | 22 | 39 | 33 | 50 | 61 | 94 | |
| 22 | 5 | 33 | 50 | 50 | 66 | 83 | |
| 22 | 33 | 33 | 33 | 66 | 77 | 77 | |
| 22 | 39 | 27 | 39 | 61 | 66 | 83 | |
| 39 | 17 | 27 | 44 | 61 | 66 | 72 | |
| 39 | 39 | 50 | 72 | 33 | 61 | 66 | |
| 16 | 33 | 27 | 44 | 33 | 88 | 83 | |
| 50 | 39 | 33 | 55 | 66 | 72 | 72 | |
| 33 | 28 | 39 | 83 | 44 | 61 | ||
| 39 | 27 | 33 | 61 | 72 | 72 | ||
| 22 | 28 | 33.5 | 77 | 61 | |||
| 61 | 61 | 7.1 | 55 | 44 | |||
| 44 | 17 | ||||||
| 11 | |||||||
| MEAN % | 33.65 | 31 | 33.5 | 49.7 | 52.7 | 70.7 | 75.5 |
| S. Dev. | 12.4 | 13.9 | 7.1 | 18.8 | 12.7 | 9.9 | 12.6 |
| # Words | 672 | 620 | 670 | 994 | 1054 | 1414 | 1510 |
There is a consistent pattern of 30-50-70% in the 2000-test results over the course of moving through the PET bands, replicated with hundreds of students over several terms. Multiplying percentages against 2000, it appears that successful students know about 600 words at the end of Band 1, 1000 words at the end of Band 2, and 1400 words at the end of Band 3. Repeating groups consistently weigh in one size-range below par, so that the Band 2 failing group in Table 6.2 has the same mean vocabulary size as the Band 1 groups. These vocabulary size differences by PET band are real, in the sense that within-band means are not statistically different while between-band means are. By this measure, then, successful students seem to be learning about 400 words per four-month term, well above the European average. The problem, of course, is that many students are not successful.
But what resources are available for learning these words? In the era of the lexical syllabus, one might assume that the authors of major English course books would know which words were the lexical core of English and then make a point of exposing learners to all of them. In fact, most course books do nothing of the kind.
Given that the PET is a creation of Cambridge University, which is based on the Cambridge Lexicon, and that the students prepare for the PET with the New Cambridge English Course (Swan and Walter, 1990), one might think they would have every opportunity to learn all the words they need. In fact, the Cambridge English Course presents learners with about half the PET list by the end of its third volume. (This makes some sense when one remembers that the PET is a proficiency test, not an achievement test.)
This information was obtained by the following method (discussed in greater detail in Cobb, 1994a): 20 consecutive words from the PET list were selected at 10 random sampling points, and each of these was checked against the word lists of the Cambridge course books. For the first Cambridge book, 50 of the 200 sampled PET words appeared in the course word list, or 25%, by extrapolation about 600 of the 2387. For the second book, 80 of the 200 sampled PET words appeared in the course word list, or 40%, about 950 of the PET words. Even the third (advanced) book would take learners only a little over half way. Similarly, the non-Cambridge courses also occasionally used at SQU to prepare students for the PET, such as Oxford's Headway (Soars and Soars, 1991), also come up short of PET words even in their final course books. Only the COBUILD course presents the full 2500 words by its final book, but it is not used at SQU.
| CA1 | CA2 | CA3 | |
|---|---|---|---|
| Sample 1 | 6 out of 20 | 9 out of 20 | 11 out of 20 |
| Sample 2 | 3 | 6 | 10 |
| Sample 3 | 6 | 9 | 11 |
| Sample 4 | 2 | 6 | 8 |
| Sample 5 | 8 | 8 | 10 |
| Sample 6 | 4 | 12 | 14 |
| Sample 7 | 6 | 9 | 11 |
| Sample 8 | 2 | 5 | 10 |
| Sample 9 | 5 | 5 | 10 |
| Sample 10 | 8 | 11 | 12 |
| TOTAL | 50 / 200 | 80 / 200 | 107 / 200 |
| MEAN | 5 / 20 | 8 / 20 | 10.7 / 20 |
| S. Dev. | 2.2 | 2.4 | 1.6 |
| % of PET Words | 25 | 40 | 53.5 |
(CA1 = Cambridge English Course Book 1, etc.)
The lack of 2000-level coverage is not confined to the particular commercial courses that happen to be used at SQU. A decade and a half ago, Nation (1982) complained that the lexical coverage of coursebooks was excessively light. Even graded readers are surprisingly weak in the lexical ranges they claim to cover; Wodinsky and Nation (1988) examined a popular set of readers finding that stories graded for 2000 words in fact contained a poor sampling of those words, so that a learner would have to read several stories to run into most of them even once. In other words, most commercial materials have poor provision for learning the words that comprise 80% of English.
To summarize, these students are forced to take an external English test based on 2000 words or more; this test can and does determine academic fate; for some this is a learning task roughly double that expected of European learners; and the instructional resources available to these students expose them to about half these words-effectively far fewer, since the final course books in these series are almost never reached.
Nation (1990) argues that no academic English course can afford to neglect instructing its learners in the highest-frequency words of the language, and the point has never been contested. However, it is often ignored, and in fact very few English courses ensure that their students will meet a significant portion of these words by their end.
Of course, coverage could be guaranteed by giving students supplementary word lists, but this solution is anathema to most instructors, because it presents words out of context and invites students to learn translation equivalents. Nonetheless, the idea of supplementation may be on the right track. What seems to be needed is some sort of supplementary vocabulary course, running concurrently with the normal syllabus and guaranteeing introduction to core lexis if the syllabus does not provide it, further practice if it does, and plenty of recycling in either case. Arguments for a supplementary approach to vocabulary coverage are made by Beck, McKeown, and Omanson (1984), and Carrell (1988), although not necessarily involving computers. But a computer tutorial would be the obvious medium to deliver such a supplement, especially if one could be found that guaranteed full coverage of particular word lists and yet allowed words to be met in context-a paradox concordance technology should be able to crack.
Another reason for considering the computer as the medium of choice for this instruction is the usual one that it makes individual learning possible. Students exiting Band 1 seem to have about 800 words to learn for entry into the passing zone of the PET, but that does not mean they all have the same 800 words to learn. Vocabulary instruction, at least at the sub-technical level, is a classic candidate for some sort of individualized instruction such as a computer tutorial where students can determine for themselves which words to work on.
But how suitable for Arab learners is a computer tutorial based on inferential learning from text? These students belong to a Muslim culture that endorses memorization as the main learning strategy, often to the detriment of comprehension. Non-Arabic-speaking Muslims in countries like Pakistan and Indonesia regularly memorize the Koran in Arabic without knowing the language. Even native Arabic speakers are not necessarily able to understand the classical Arabic of the Koran very well, and even if they do, interpretation and comprehension questions are not on the instructional agenda (see Gardner, 1993, for a relevant discussion of traditional education). Surely by university age these learners have developed a learning style that suggests a more deductive approach to their vocabulary expansion. For example, they could be given short definitions for the words they need to learn. This might be a low-transfer kind of learning, but nevertheless the kind they are capable of.
There is actually no reason to think that Arabs' learning-style repertoires have been significantly restricted by their upbringing. In their famous study of the supposed cognitive effects of different literacies, Scribner and Cole (1981) empirically tested the notion that Koranic scholars would remember the surface of a text more than its semantic depth. The test turned on whether a reader can distinguish between sentences actually read and propositions merely inferred (good readers cannot). For example, a good reader cannot remember whether he read "three turtles rested on a log, and a fish swam under them," or "and a fish swam under it," since both decompose to the same situation model. Scribner and Cole expected Koranic memorizers to fixate on the surface of a text, failing to comprehend or remember the gist in significant ways. However, they found that these scholars, while indeed excellent text memorizers, nonetheless constructed deep semantic representations of text under normal circumstances just like anyone else:
Contrary to our speculations, Arabic literates, along with all other populations, made false recognition errors on true inference sentences while correctly accepting or rejecting other recognition items. While this result is impressive evidence for the universality of a tendency to make sense of what one is hearing by constructing an integrated representation of incoming information, it was a disappointing failure to anticipate a specific effect of practice in the kind of Arabic literacy encountered [i.e. practice in Koranic memorisation] ... (p. 222).
So there is no reason to think that Arabic speakers would be worse at inferential tasks than anybody else.
There is even some reason to think they might be better, because Arabic readers have had a lifetime of practice in contextual inference by virtue of how their writing system works. Arabic script does not encode most of its vowels, with the effect that most written Arabic words in isolation could actually point toward several spoken words, just as the de-vowelled English word "ct" could be shorthand for any of "cat," "cot," "cut," "cute," or "coot" although the reference would be obvious in even a short sentence ("The ct sat on the mat.") Of course, any script, vowelled or not, contains ambiguous words that can be resolved only by context, like "bank" (of a river) and "bank" (your money), but an unvowelled script has far more; Sampson (1985, p. 93) estimates that such ambiguity, "rare" in a vowelled script like English, characterises 3 out of 10 words in an unvowelled script.
In English, it is established that it is not ability to use context that characterises skilled reading, but rather ability in word recognition out of context (Stanovich, 1986). But this finding has no meaning when applied to a script like Arabic. Out of context, "ct," "tlt," "brk" are only skeletons of words to which a reader must add flesh from his understanding of the ongoing context. A study by Abu Rabia and Siegel (1995) establishes empirically that skilled reading in Arabic is largely a matter of contextual interpretation from minimal graphic input, the converse of English. Of course, one could argue that the contextual sensitivity required for reading Arabic would not necessarily transfer to reading a vowelled script like English, but it appears that at least some reading habits and strategies indeed do travel from first language to second, usually with disadvantages but maybe here with advantages. Koda (1988) provides empirical evidence for "cognitive process transfer" between Arabic and English reading. So, in principle, a lifetime of reading Arabic may well have made SQU learners sensitive to words' contexts and so good contextual learners.
Some "mystery" information about Arabic readers cropped up in one of the studies of inferencing mentioned in Chapter 3. Haynes (1983) had students from four language groups infer word meanings from text, where the meanings were implicit in either the local or the global context. Local context was moderately useful to most second-language readers, but global context, where crucial information often lies, was much less. But there was a surprise exception: for monolingual Arabic speakers, global context was as useful as local for inferring new word meanings.
All this is only to say that Arabic-speaking students are as capable as anyone of learning words inferentially. In fact, however, any instructor in the Middle East knows that his students are actually quite poor at exercising this skill (demonstrated empirically at SQU by Arden-Close, 1993a, and Horst, 1995). There are two reasons for this that could have little to do with ability. The first is that to learn words you need words; to make an inference about a word, you need to know most of the other words in the context, which of course these students almost never do. For students who know 1000 words, one word in four in any text is unknown (Nation, 1990), so any inferential skill their first language might give them can hardly enter into play.
Second, to say that memory-based education does not definitively destroy the cognitive ability to construct deep representations is not to say that cultural influences can not produce some unuseful learning habits. SQU students' learning habits have been the subject of many studies since the University opened in 1986. A recent example is a survey conducted by the College of Education and Islamic Sciences (Barwani, Yahya, and Ibrahim, 1994) which asked faculty members to rate 34 learning-related skills and attitudes in terms of two dimensions, perceived importance in academic success and actual availability in SQU students. Their responses were entered into a factor analysis, and only one of the 34 skills came out as both desirable and available: "ability to memorize and recall information." Highly desirable, but totally unavailable, was "knowing how to learn."
Many of the constituent colleges of SQU have expressed similar views about their students and proposed measures for dealing with the study skills problem. The College of Commerce and Economics was set up in 1993 as a high-flyers' college under the advice of INSEAD in Paris (the European Institute of Business Administration), including an information centre with four networked computer labs and the PET as a stiff guarantee of students who could cope with English. But at the end of a disappointing first term (only four students arrived in Band 4) a task force was set up to look into the students' study skills (Scott, Gerber, Salem, Marzouqa, and Sherazee, 1995). Faculty members perceived these to be effectively non-existent and made several recommendations to members of the language unit, who of course meet the students first and set their initial academic course: Language instructors should encourage students to rely less on rote memorization, depend less on instructors, take responsibility for their own learning, and so on, in preparation for their upcoming academic courses.
Three of the recommendations are specifically relevant to the present study because they feed directly into the design specifications for the computer tutorials about to be described:
- Students should adopt a problem-solving approach with emphasis on the transfer of information and the application of rules or principles in new situations.
- Students should learn to synthesize information from several sources.
- Students should become responsible for their own work by using the Information Centre's resources to find things out for themselves (Scott and colleagues, 1995).
One way to help students use the Information Centre is to make sure there is some information there that they need in a form that they can use. And one way to make students more independent is to download some significant part of the curriculum into an self-access activity that nonetheless "counts" back in the classroom.
Many students in the College of Commerce are in desperate need of rapid
lexical growth, and their courses are unlikely to help them sufficiently.
Some sort of individualized course supplement seems indicated, such as a
self-access computer tutorial, for which the concepts and machinery are
available. An inference-based lexical tutor would probably be challenging
for these students, but there is nothing in either cognitive research or
local task analysis to rule against it. Students have been working with
texts on computer screens for several years at SQU, and there is extensive
experience in CALL development to aid in the design of a 2000-size lexical
tutor. This experience will be discussed in the next chapter.
| contents |