This article reports an evaluation study of the Bangalore project, a content-based approach to language learning. It examines the question, Do teachers attend to. meaning or to form, and is such a distinction observable in classroom practice? The study investigated the way that error treatment in Bangalore classrooms was realized in practice in a sample of 21 lesson transcripts. It was found that the treatment of linguistic error was largely consonant with the project's statements about the kinds of attention that are appropriate to a focus on meaning and that this could be distinguished from the ways of treating linguistic error that are attributable to a focus on form. However, the most likely explanation for this finding is that as the project developed, task types were selected that would curtail learner production and, thereby, the risk of making linguistic error. This finding has implications for teachers interested in content-based curricula.

A series of problem-solving activities evolved through a sustained period of trial and error; the result was a task-based syllabus that was intended to foster a preoccupation with meaning: "not 'English for communication' but 'English through communication'; not 'learn English so that you will be able to do and say things later' but 'do and say things now so that as a result you will learn English"' (Prabhu, 1980, p. 23).

The project was conducted in eight schools for durations of up to 3 years between 1979 and 1984. Although it is no longer in operation (it was intended as a 5-year project), it is timely to consider the CTP now for three reasons: (a) The evaluation of it has only recently been completed (Beretta, 1987); (b) the person most influential in initiating the project and its Director has only recently made available a substantial account of it (Prabhu, 1987); and (c) most important, it may be seen as one of a series of efforts toward taskor content-based teaching that are currently generating a great deal of interest (e.g., Edwards, Wesche, Krashen, Clement, & Krudenier, 1984, and Wesche & Ready, 1985, in Canada; Nunan, in press, in Australia; Long, 1985, in Hawaii; Yalden, 1987). The CTP has itself attracted much comment in the literature on syllabus design and methodology (Brumfit, 1984; Crookes, 1986; Ellis, 1984; Johnson, 1982; Richards, 1984).

This article reports on one of several strands of the evaluation of the Bangalore project: error treatment. Other strands of the evaluation are reported elsewhere. (For a comparison of experimental and control groups, see Beretta & Davies, 1985; for a qualitative study of implementation, using CTP teachers' accounts of their experience as a data base, see Beretta, in press; and for an investigation of CTP teachers' anxieties in coming to terms with an innovation in the South Indian setting, see Beretta, in press. The archive document for all of these studies, including the present article, is Beretta, 1987.)

As noted, the CTP abjures attention to linguistic form; lessons are planned only with problem- solving activities in mind but not with an accompanying plan for language input.

With a structural syllabus, the intention is that certain language forms should be practiced. In the CTP, however, there is no such intention; the plan is to present learners with an activity that will engage their interest and impel them to use whatever language they have at their disposal. The most obvious question is, Are these different intentions realized in tokens of the teacher's classroom talk?

Unfortunately, there is no way of determining what might have been in a teacher's head during any given lesson. The following anecdote illustrates the difficulty of distinguishing between planning to teach language form and planning to teach meaning (or content) when actual classroom practice is examined.

The problem is that for the seminar participants the CTP lesson did indeed reflect a plan to teach language form, whereas for Prabbu it did not. The debate ended with all parties apparently entrenched in their views. (The anecdote and the lesson transcript may be read in full in Regional Institute of English, 1980, pp. 44-53).

Clearly, then, the question of intentionality or planning cannot be pursued with any expectation of success. However, in relation to the CTP, Prabbu has made some fairly clear statements about what is acceptable and what is not with regard to actual teacher treatment of pupil error (both linguistic and content). This makes it possible to investigate teachers' attention to form and attention to meaning in terms defined by the project.

In other words, because we can ask whether or not CTP treatment of error corresponds to Prabbu's prescriptions, we can address the fundamental question: Did Bangalore teachers focus on meaning or on form, and is such a distinction observable in classroom practice? That is what this article aims to examine.

(a) Incorrect language from learners is corrected (i.e. rephrased, restated, or drawn attention to) in roughly the same way that interested adults do with children-or the subject-teacher in an English- medium class does in teaching his subject.

(b) This is done more in the context of writing (either on the blackboard, as part of the pre-task, or on paper in performing the task) than in oral work, as being more natural in that context.

(c) All such attention to language is limited to facts (as against generalisations) and treated as contributory to the successful performance of the task on hand.

The notion of transitoriness would be difficult to examine (though see van Lier, 1987, for suggestions). just how brief and how fleeting exchanges should be allows considerable latitude for interpretation, so no related hypothesis is proposed. Similarly, the claim that CTP error treatment is intended simply to facilitate the task at hand is unfalsifiable. If it were shown that a teacher's utterance contravened Prabhu's strictures regarding the use of exemplification, it might be argued that on this occasion exemplification contributed to the task at hand; that is, in the teacher's perception, it cleared up some doubt in the student's mind and allowed the task to proceed. However, an unfalsifiable claim cannot be used as an argument, or we are reduced to the kind of debate witnessed in the anecdote related above. Therefore, it seemed justifiable to leave the "facilitative" claim out of the present inquiry.

However, some elements did seem reasonably tangible. It was clear that explanation, generalization, and exemplification are unacceptable as treatments of linguistic error and that error treatment should focus on content and not on linguistic accuracy. Thus, from Prabhu's descriptions cited above and on the basis of the dichotomies derived from those statements, the following simple hypotheses seem permissible:

In addition, it appeared to be worth documenting in as much detail as possible just what kinds of treatment are given. This required the selection of a checklist of treatment categories. Chaudron's (1977) framework was modified for this purpose.

It will be clear from the descriptions of the categories that they overlap and that most of the error treatments require multiple coding. Thus, one treatment might be associated with more than one category. Wherever multiple coding was applicable, every possible categorization was counted; this procedure explains why there are many more treatments than errors.

In addition to the four hypotheses and the description of treatments, it seemed helpful from the perspective of external validity to try to explain the incidence of particular kinds of error and error treatment and the relationship between them.

The most obvious candidate for investigation would have been the differences between the regular teachers in the schools and the nonregular teachers (highly qualified personnel drafted for the project), since other strands of the Bangalore evaluation (Beretta, 1987, in press) had shown that to be an important variable. However, our data did not include lessons given by regular teachers, so such an inquiry, though particularly relevant, was denied to us. At this point, it should be made clear that the 21 lessons analyzed in this investigation are simply those that were recorded and made available to the author. Of the 21 lessons, 18 were taught by the Director and a close associate so it is unlikely that there would be as much focus on form as visitors had noted in the lessons of regular teachers on the project.

However, four other potentially revealing lines of inquiry seemed pertinent: First, a reasonable variable for investigation was whether errors were treated differently in the early stages of the program than later on. The CTP was a developmental program; that is, it did not start life with a ready-made methodology but evolved one over time through a process of trial and error. In view of this fact, it seemed quite feasible that error was treated differently in the early and late phases of the project. To make this distinction, it was important to find a watershed, a period when the project methodology had become relatively stable, when it would have been unlikely to undergo any further major modification.

Barnes (1982), who visited the project in March and April of 1982 , reported that the lessons were "impressively consistent" (p. 2) and, in a letter to Prabhu, sought his views on the question. Prabhu (personal communication, May 1986) responded as follows:

The pattern of classroom activity was least settled in the first year of the project (June 1979-April 1980), most settled in the last two years (June 1982 to April 1984) and was settling steadily in the second and third years (June 1980 to April 1981 and June 1981 to April 1982).

Another potentially important variable was the length of different lessons. If some of the recordings were shorter than others or incomplete, this might have a bearing on our interpretation.

Yet another possible explanation of the incidence of different kinds of error treatment was the matter of the teacher's personal style. As noted above, 18 of the 21 lessons at our disposal were taught by two teachers. Although it was possible to compare error treatment by these teachers, with only 2 out of 16 teachers represented, inference to a wider population is extremely tenuous.

The fourth variable that demanded attention was that of task type. Was the incidence of error and type of error treatment dependent on the nature of the task? Again, this would influence any interpretation of treatment of errors in CTP classes.

Thus, explanation of error treatment comprised the following elements: (a) location in time (pre- 1982 versus post-1982), (b) length of recording, (c) personal style, and (d) task type.

With reference to task type (see Table 1), five of the lessons were lecturettes (A, B, F, G, and 1); four involved letter writing (C, D, E, and M); four revolved around timetables (H, K, 0, and R); three were based on drawing figures (P, Q, and S); two focused on town maps (L and T); one on distances (N); one on telling the time (U); and one on short narratives plus questions (J).

Of the 21 lessons, 10 were taught by Teacher X, 8 by the Project Director, 2 by Teacher Y, and 1 (Lesson H) by an unidentified male teacher (see Table 1).

As for the length of the lessons, some were clearly incomplete. Also, recordings of the task phase of each lesson were inaudible as the teacher moved around the class checking the work of individuals. Therefore, it was only possible to investigate the pretask phase of each lesson. Even here, the range varied from 6 minutes to 39 minutes, with a mean of 22.67 minutes and a standard deviation of 8.67 minutes. These computations include only the recorded lessons; for the six lessons for which only transcripts were available, the duration is not known.

Lessons A to N all dated from February and March 1981 (14 lessons). Lessons 0 to U were from 1983 and 1984 (7 lessons).

The pupils being taught were aged between 8 and 13, and their first language was either Kannada or Tamil. Schools in the project were either state schools or those founded by religious orders and were selected by the project team because they were not Englishmedium schools but were still well enough managed to accommodate an innovation.

The project teachers ranged from British Council and teachertraining college staff to regular teachers in the school system. As noted, the sample of 21 lessons includes 8 lessons by the Project Director, 10 by a close associate, but none by regular teachers.

The first step was to identify errors. The following three strategies were adopted: (a) A judgment was made concerning the linguistic accuracy of a student's utterance; (b) a judgment was made as to the accuracy of a student's response to a teacher's question in terms of content (i.e., was the solution to a problem correct or incorrect?); and (c) if a teacher clearly disapproved of a student's utterance, the utterance was considered to contain error (in practice, this meant that attention was alerted to the likelihood of error, and attempts were made to identify it).

The second step was to categorize the error as either linguistic or content. A linguistic error was defined as a morphosyntactic or phonological error. In the event, only four phonological errors were noted; in these few cases, the teacher failed to understand the students' utterances. This failure to understand was used as a criterion.

A content error was defined as any response by a student to a teacher's question that was unsatisfactory in terms of its propositional content. Thus, if a student answered a question that was not asked or simply answered the right question incorrectly, a content error was coded. For example, if the teacher's question required a calculation to which the answer was "6" and a student answered "7," a content error was made, irrespective of the linguistic accuracy of the utterance. (Four lexical errors were noted and were included as content errors.) Any error that was both linguistic and content was classified as both.

The third step was to calculate the percentages of linguistic and content errors that were treated (to permit a judgment with reference to Hypothesis 1, that more content errors are treated than linguistic).

Next, treatments were identified in terms of the 20 treatment categories adapted from Chaudron (1977). This required a great deal of working back and forth between codings given at different sittings and for different lessons to try to ensure a reasonable level of consistency. Since this analysis was carried out only by the author, some form of self-checking protocol seemed appropriate. Thus, after all of the lessons had been coded, three lessons were recoded months later. The agreement was only 71%. The difficulties of such analysis are well attested to in Chaudron (1977), Allwright (1975), and Fanselow (1977), but the lack of intracoder consistency needs to be stressed, as intercoder reliability would probably be somewhat lower.

Once the error correction strategies had been coded, the transcripts were scanned for treatment of linguistic error that involved explanation, exemplification, or generalization (in order to test Hypotheses 2, 3, and 4). The categories in the framework that were intended to track these strategies are explanation and clue. Explanation includes both giving information about the cause of the error and making a generalization about the type the error represents. Clue includes the use of further examples of the same error type.

Finally, the incidence of various strategies was calculated for descriptive purposes, and the analyses were arranged for examination of potentially explanatory variables.

Once all errors had been identified and classified as either linguistic or content, it was possible to calculate how many linguistic and content errors were treated. The total number of error treatments for the 21 lessons was 926 (M = 44 per lesson); the total number of content error treatments was 599 (M = 29 per lesson); and the total number of linguistic error treatments was 327 (M = 16 per lesson).

Treatment includes the categories of ignore, acceptance, and acceptance* (see Appendix A for definitions). If, however, these categories (which are really nontreatrnents) are deducted from the number of treatments, a more accurate picture emerges of the comparative attention given to linguistic and content errors. We find that only 211 out of 327 linguistic errors were treated (65%), whereas 529 out of 599 content errors were treated (88%). The difference is significant, X (1, N = 926) = 73-10, p <.001, which indicates support for the first hypothesis: In CTP lessons, more content errors were treated than linguistic.

There were very few examples (only 4) of explanation in the treatment of linguistic error when compared with the far greater frequency (50 instances) of this strategy to deal with content errors:

T: Requested? When?

S: Request

T: [Request. He's requesting now [1]. If he requested yesterday, then we can say'requested' [2].

The teacher's response quite clearly focuses on the cause of the error and generalizes to the extent of invoking the concept of the past.

In regard to the question of exemplification, it was found that clue was coded six times for linguistic error, but closer inspection revealed that only one of these was a matter of exemplification:

Exemplification of the type of error is given when the teacher says, a? ... a apple? Do we say a apple?"

The above three exchanges demonstrate that the nature of the focus on form contravenes Prabhu's statements about how error should be handled to be consistent with CTP principles. Explanation, generalization, and exemplification are all usedalthough rarely-in ways that do not conform with Hypotheses 2, 3, and 4. If the strict wording of those hypotheses is adhered to ("no explanation," "no generalization," "no exemplification"), then the hypotheses are rejected.

On the other hand, it is quite clear that the incidence of such violations" is very low. Moreover, all except one of the exchanges occurred in one lesson. However, this lesson was taught by the Project Director, and if the chief proponent of the CTP could slip into a focus on form, then it is hardly surprising that other CTP teachers did. (This is confirmed by CTP teachers' own accounts of their experience; see Beretta, in press.) From the point of view of external validity, it is important to note that it was not only regular teachers who reverted to more familiar classroom practice.

The incidence of different treatment types for linguistic and content errors is displayed in Table 2. With regard to linguistic error treatment, by far the most common strategy was repetition with change (n = 116), which is to say, when a student made a linguistic error, almost 36% of the teachers' correction strategies consisted of repeating the student's utterance in an accurate form and moving on. This is entirely consistent with the CTP perceptions of appropriate error treatment.

Apart from those mentioned, the rest of the strategies used to correct linguistic error are few in number and more or less neutral with regard to CTP policy.

The overall picture provided by Table 2 is fairly clear. A few major strategies dominate in correction of linguistic error, and these reflect a general willingness to allow errors to pass with a simple rephrasing, without comment or even with apparent acceptance. This is in keeping with expressed CTP attitudes to error correction. Content error, on the other hand, receives far more sustained and varied treatment, possibly indicating an emphasis on solving the problem at hand. However, this overall description hides the contribution of potentially relevant factors, such as location in time, length of recording, teacher style, and task type.

Location in time (pre- 1982 versus post- 1982). As elaborated above one possible explanation of the spread and frequency of types oi corrective treatment is that the CTP methodology had not achieved the relative stability of later years until about 1982. Of the lessons available to this study, 14 took place in 1981 (A to N) and 7 in 1983 or 1984 (0 to U). Table 3 details the frequency of linguistic and content errors and the percentages of correction (i.e., deducting the ignore, acceptance, and acceptance* strategies) for the two sets of lessons.

It can be seen from Table 3 that there is a slightly higher incidence of content error in the post- 1982 lessons (an average of 19.43) than in the pre-1982 lessons (an average of 17.64). There is also a higher rate of correction for the later lessons. With regard to

Analysis revealed that length of pretask could not explain the differences between early and late lessons: Mean duration for preand post-1982 lessons was practically equivalent. Nor could teacher style shed any light: The number of errors and the incidence of correction were very similar for the two teachers whose lessons comprise the majority of the 21 transcripts. Task type, however, did have a tale to tell.

Task type. Although the mean number of linguistic errors for the later lessons is 5.14, closer inspection of the data reveals that 25 of the 36 errors occurred in one lesson. This lesson was different in kind from the others; instead of just receiving instructions, students gave them too. Teacher X, who taught the lesson, explains in her account (see Beretta, 1987) that the intention was to increase learners' productive ability in English; she points out that this was atypical of CTP teaching. If this atypical lesson is discounted in the pre- and post-1982 analysis, the later lessons average only 1.83 errors, which, compared with the 17.36 of the earlier lessons, is a very pronounced difference, t (18) = 2.982, p < .01. Thus, on a first examination of task type, it would appear that there are barely any linguistic errors in the later lessons.

Barely any errors require barely any treatment. So the question now becomes, What is the association between task type and the incidence of linguistic error? And once this is answered, it would be germane to ask whether task types that have little association with linguistic error occur more often or solely in the later period. In other words, did the CTP phase out tasks that appeared to draw student responses that might contain morphosyntactic error?

Table 4 indicates the mean incidence of linguistic error associated with each task type (not including, because of its atypicality, Teacher X's figure-drawing lesson that contained student-student instructions). Table 5 arranges the task types according to pre-1982, post-1982, or both periods and includes in parentheses the frequency of their occurrence in each period.

A closer examination of the data reveals that 198 of the 200 linguistic and content errors in the letter-writing tasks are verbal, whereas in the figure-drawing tasks, only 12 out of 46 are verbal (and 10 of them are simply wrong answers to yes/no questions or wrong letters or numbers called out). The difference is significant, X' (1, N = 246) = 153.41, p < .001.

In the lecturettes, the teacher talks at length on a theme and then asks a series of comprehension questions that prompt verbal responses, usually of phrase length but sometimes longer. In letter writing, students are presented with scripts containing error that they are required to improve, so that language itself becomes the focus of the task and greater demands are placed on students' verbal productive capacities. The figure-drawing lessons, by contrast, call for mainly physical responses (i.e., writing or drawing on the blackboard); thus, whether the response is right or wrong, there is very little possibility of the student making phonological or morphosyntactic errors.

Because the sample of 21 lessons could well be unrepresentative, further clues were sought in RIE Newsletters and Bulletins, lesson reports produced by project members, and Prabhu's writings. There was no mention at all of the lecturettes and no post-1982 mention of letter writing. All of the tasks demanding little verbal production, on the other band, received frequent mention. Figure drawing, telling the time, and timetables were referred to as "beginners' tasks" (Prabbu, 1987, pp. 37-39), presumably because they focus on receptive skills. However, it is possible that as the CTP matured, a tendency grew to protect learners from the risk of linguistic error attendant upon production beyond the beginner level (1) and that tasks such as lecturettes and letter writing were dispensed with. Certainly, they are not included in the long list of task types presented in Prabhu (1987, pp. 138-143).

The story that seems to be emerging is that the CTP started out with a range of tasks that permitted or encouraged not only receptive but also productive skills. However, as the project methodology stabilized, it moved toward delayed production and physical rather than verbal responses, probably not just at the beginner level, as indicated by Teacher X's perception that "generally our tasks were cognitively very challenging, but linguistically did not make adequate demands on the learners' productive abilities" (see Beretta, 1987, Vol. 2, p. 246). But it is stressed that this cannot be considered the full story; a sample of 21 lessons, although endorsed by all available evidence, is still only capable of supporting tentative conclusions.

It was found that there was a significantly greater incidence of treatment of content than linguistic error, consonant with the CTP focus on meaning rather than form. The descriptive analysis revealed that the majority of treatments for linguistic error involved minimal intervention or none at all. Content errors, by contrast, were treated in a wide variety of ways, indicating that more sustained attempts were made to secure the correct answers to problems. The descriptive analysis thus broadly supported the stated CTP approaches to error correction. (It should be recalled, however, that only Prabhu and his closest associates were represented in the 21 transcripts, and visitors to the project had noted that regular teachers tended to revert to a focus on form; see Brumfit, 1984; Davies, 1983.)

The stipulation that explanation, generalization, and exemplification be avoided was not strictly observed; the fact that not only regular teachers but also nonregular teachers could slip into a focus on form was noted.

The most salient finding to emerge from an examination of variables contributing to treatment type was that there was very little error treatment in post-1982 lessons. This was due to the simple fact that there was very little error. The best explanation seemed to be that task types calling for production were phased out and that only those that stressed reception were retained. When this interpretation is considered along with Teacher X's testimony that the CTP did not attend sufficiently to production, confidence in the interpretation is increased.

I would like to thank the Free University of Berlin and Sultan Qaboos University, Oman, for providing facilities for most of this research to be carried out.

Alan Beretta is Assistant Professor in the Department of English at Michigan State University.

Corder, S. P. (1981). Error analysis and interlanguage. Oxford: Oxford University Press.

Crookes, G. (1986). Task classification: A cross-disciplinary review (Tech. Rep. No. 4). Honolulu: University of Hawaii.

Davies, A. (1983). Evaluation and the Bangalore project. Unpublished manuscript, University of Edinburgh.

Edwards, H., Wesche, M., Krashen, S., Clement, R., & Krudenier, B. (1984). Second language acquisition through subject-matter learning: A study of sheltered psychology classes at the University of Ottawa. Canadian Modern Language Review, 41, 268-282.

Ellis, R. (1984). Classroom second language development. Oxford: Pergamon.

Fanselow, J. F. (1977). The treatment of error in oral work. Foreign Language Annals, 10, 583-593.

Johnson, K. (1982). Communicative syllabus design and methodology. Oxford: Pergamon.

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

Long, M. H. (1985). A role for instruction in second language acquisition: Task-based language training. In K. Hyltenstam & M. Pienemann (Eds.), Modelling and assessing second language acquisition (pp. 77-99). San Diego: College-Hill Press.

Nunan, D. (in press). The learner-centred language curriculum in theory and practice. Cambridge: Cambridge University Press.

Prabhu, N. S. (1980). Theoretical background to the Bangalore project. In Regional Institute of English, South India, Bulletin, 4(l), 17-26. Bangalore, India: Author.

Prabhu, N. S. (1982). The communicational teaching project, South India. Madras, India: British Council.

Prabhu, N. S. (1987). Second language pedagogy. Oxford: Oxford University Press. Regional Institute of English, South India, staff. (1980). Bulletin, 4(l). Bangalore, India: Author.

Richards, J. C. (1984). The secret life of methods. TESOL Quarterly, 18, 723.

van Lier, L. (1987). The classroom and the language learner: Ethnography and second-language classroom research. London: Longman.

Wells, G. (1985). Language development in the pre-school years. Cambridge: Cambridge University Press.

Wesche, M. B., & Ready, D. (1985). Foreigner talk in the university classroom. In S. M. Gass & C. G. Madden (Eds.), Input in second language acquisition (pp. 89-114). Rowley, MA: Newbury House.

Yalden, J. (1987). Principles of course design for language teaching. Cambridge: Cambridge University Press.

APPENDIX A

Treatment Categories Based on Chaudron's (1977) Framework

Acceptance: Teacher says a simple approving or accepting word (often as a sign of reception of the utterance) but may proceed to correct an error.

Provide: Teacher provides the correct answer when learner has been unable to or when no response is offered.

Repetition with no change: Teacher repeats learner utterance, with neither change of error nor omission of error.

Repetition with no change and emphasis: Teacher repeats learner utterance with no change of error, but the emphasis locates or indicates fact of error. Repetition with change: Usually, teacher simply adds correction and continues to other topics.

Repetition with change and emphasis: Teacher adds emphasis to stress location of error and its correct formulation.

Explanation: Teacher provides information as to cause of error, possibly including a generalization of the type of error.

Repeat: Teacher requests learner to repeat utterance with the intention of having the learner self-correct.

Loop: Teacher honestly needs a replay of learner utterance due to lack of clarity or certainty about its form.

Prompt: Teacher uses a lead-in cue to get learner to repeat utterance, possibly at point of error and possibly with a slight rising intonation.

Clue: Teacher reaction provides learner with isolation of type of error or the nature of its immediate correction, without actually providing the correction (e.g., further examples of the same error type may be given).

Original question: Teacher repeats the original question that led to the incorrect response.

Altered question: Teacher alters original question syntactically but not semantically.

Questions: Teacher uses numerous ways of asking for a new response, but not just original or altered questions; that is, when error occurs, a new line of questioning is taken up.

Transfer: Teacher asks another student or group of students to provide correction.

Acceptance*: Teacher shows approval of learner utterance and then repeats the error, apparently confirming that it is correct.

Verification: Teacher attempts to make sure that the class has understood the correction.