0.2 The background: the problem of describing and classifying errors.

The problem of describing and classifying errors made by second language learners is one which has been recognized for a long time as "what most teachers will agree is the most disorderly of all their problems ... Errors defy classification, for one kind merges into another as grey shades off into blue" (French 1949:13).

To begin with, terminology is a source of confusion. The term "error description" has been used to refer to both the description of the error itself and the description of the possible causes of the error, whereas, if we follow Corder's recommendations, it should have only the first meaning (1971:158). Dulay and Burt (1974:96) distinguish between the error itself and its possible causes as the "level of product and the level of process". To describe an error means to specify what it is about the deviant utterance that does not conform to the grammar of L2. What Corder calls the "explanation" of the error, however, seems to involve two separate, or at least separable, issues. Taking for a moment an analogy from American baseball, the "learner" who gets a hit and runs directly to second base instead of to first can lead us to conjecture about 1) why his action violates the rules of baseball (error description), 2) why he did it (the cause of the error), or 3) whether he is playing baseball or some other game (like cricket), a combination of different games, or some kind of idiosyncratic "intergame" (or whether he is playing any sort of game at all). This last might be called error systemics, the study of the learner's "interlanguage" (Selinker 1972), which, in turn, can be viewed synchronically (its rules at any one point in time) or diachronically (the development of its rules through time). As Corder says, the first question (error description) is purely linguistic, whereas the second (error causation) is psycholinguistic. The third question (error systemics) may be both.

There is a fourth question we can ask about errors which is sociolinguistic--concerning their influence on interpersonal relations and communication--which might be called error effects. Burt's (Burt 1975:56-57, Burt and Kiparsky 1972:6) distinction between "global" and "local" errors is a brave attempt to provide linguistic criteria for errors which "significantly hinder communication" (global errors) and those which do not (local errors), but the criteria just do not hold up. There are many cases where local errors, linguistically defined, hinder communication more than global errors, and vice versa. For example, *Book bought I yesterday contains global errors of word order that affect the relationship of major clausal constituents, but hinders communication less than *I bought it at the library (meaning 'at the bookstore'), which is local because the error affects only one constituent. Furthermore, any definition of what hinders or furthers communication cannot depend on linguistic criteria alone.

The baseball analogy is rather interesting in another respect. What are officially counted as players' errors seem to correspond with language performance or non-systematic errors ("mistakes"): they do not break any rules of the game, and in fact are expected and even essential to play, because a game without errors would be considerably more predictable--and less interesting-if the only question were whether or not the batter would get a hit and where he would hit the ball. One wonders if the unpredictability factor of performance is equally essential to the "game" of language--perhaps just as essential as the (in theory) predictable grammatical rules (competence). If this is true, a model or theory of language, as opposed to a theory of grammar, would have to take into account a much more intricate relationship between performance and competence than has generally been assumed.

The problem of error description seems to have been the most neglected aspect of error analysis, at least from a theoretical point of view. One reason for this may be that most of the second language acquisition research to date (cf. Hatch 1978) has involved the study of beginning or at most intermediate learners, where the linguistic description of errors is not nearly so problematic as in the case of the more complex speech of advanced learners. It is much simpler, for example, to describe an error like *I have two dog, produced by a beginning Spanish-speaking learner of English (Shapira 1978), than one like *She asked for him to keep away from the stock market, produced by an advanced German speaker (Legenhausen 1975:87-88).

It may also be that error description, as a strictly linguistic question, simply does not appeal to many researchers in the field, or appeals to them only insofar as it may have psycholinguistic implications for theories of first and second language acquisition. Such a bias is unfortunate, and does not do justice to the field of language education in general. Of course the questions of error causation and systemics are important to the more general investigation of learning strategies and have far-reaching pedagogical implications, but what is often forgotten is the immediate needs of the foreign language teacher and learner, who are not likely to benefit much in the short run from the pursuit of these ultimate objectives. What they need, particularly in the case of adult students, is not so much explanations of why errors are made as why errors a errors. When a student follows up the teacher's correction of his erroneous utterance with the question "Why?", he doesn't mean "Why did I do it?" but "Why is it wrong?" Anyone who has tried to answer this question conscientiously over a large number of instances knows how difficult and frustrating it can be--just as much, perhaps even more so, when the language being taught is the teacher' 9 L1 rather than his L2. The point is that in addressing the linguistic problem of error description we are also addressing an immediate pedagogical need, one that teachers face almost daily in the classroom.

Particularly with beginning or younger learners, one learns as a teacher to answer the question "Why?" in a way that is tantamount to no answer at all, that is, by saying things like "Because it doesn't make sense", "Because that's not what English speakers say", etc. There is a perfectly good pedagogical-reason for this, of course, which is that young learners do not have the intellectual equipment, nor beginning adults the linguistic equipment (in the target language), to make serious or detailed grammatical descriptions anything but confusing and therefore counterproductive. With advanced learners-and particularly advanced adult learners--the situation is quite different. Here the student deserves an answer appropriate to the high level of his cognitive and linguistic abilities. The problem is that our metalinguistic competence--our ability to describe the language we teach--is very fragmentary and inadequate. The rules typically offered in language textbooks and reference grammars are all too often useless in terms of either explaining an error satisfactorily or predicting correct usage. On the whole, if we are honest, we must admit that the only pedagogical justification (and it is a legitimate justification) for side-stepping the question "Why is it wrong?" in the case of adult learners is that we just do not know the answer.

Providing adequate explanations of why errors are errors that can be used in the classroom (or in the textbook), then, is one purpose of error description. Another purpose, of more interest to those doing empirical research in language acquisition, is to provide a taxonomy for the collection of data on error types and distribution. The practical difficulties involved in classifying errors, as already mentioned, are well-known, but they need to be considered more carefully from a theoretical point of view. Part of the problem may be a lingering prejudice against the word itself, deriving from the attempts of early transformationalists to dismiss certain types of structural linguistics as "mere taxonomy".

Another problem we face in trying to classify errors is that any classification of linguistic structures presupposes a description of those structures. In the study of L2 learners' errors, this descriptive model is lacking. Here we are confronted with a variety of language that cannot be investigated with the tools that are indispensable to linguistic analysis as it predominately has been practiced in the last two decades. These tools are paraphrase and judgments of grammatical acceptability, both of which can be subsumed under the heading of "native speaker intuition". These tools cannot be applied to the study of L2 learners' speech for two reasons. First, there can by definition be no "native" speakers of a second language, and, if we understand "competence" in Chomsky's sense of what the speaker knows about his native language, L2 speakers can never be said to be competent in the L2, nor even, as Corder (1967:166) would have it, "transitionally competent", since they can hardly be in a state of ''becoming'' native speakers. Second, even if we stretch our definition of native speaker competence to include what the L2 speaker knows, not about the L2, but about his own "interlanguage" (and this is quite a stretch, since an interlanguage speaker is still not a native speaker of his interlanguage), the only paraphrases or judgments of acceptability of interlanguage utterances would be those of the speaker himself. This idea, proposed by Corder (1972), is pursued, for example, by Schachter 1976. But how valid can such paraphrases or judgments be when there is no way to corroborate them--i.e. when there are no other "native speakers"? Furthermore, since the rules of the learner's interlanguage are variable, both synchronically and diachronically (cf. Hyltenstam 1977), how can we expect the learner even to corroborate himself, that is, to be consistent in his own self-paraphrases or judgments? Any two "intuitive" statements he might make, in other words, may or may not reflect the same underlying rule or set of rules--i.e. the same grammar--and there is no way to know which is the case.

Since an L2 learner cannot be considered "competent" in the L2 in Chomsky's sense, we can hardly assume that a model of native speaker competence underlies his performance in the L2. For example, to describe an error like. Everyone delights that you won the lottery as "mismanaged extraposition" (Burt 1972:112) is theoretically implausible in two important respects. First, describing a performance error in terms of a transformational derivation makes what is still a rather considerable shift in focus (not to say credibility) from the phenomenon itself to a theory of the phenomenon, and one mustn't forget that transformational grammar was never intended as a model of performance, as opposed to competence, in the first place, even in the case of native speakers. Second, and more importantly, the derivation of a sentence like It delights everyone that you won the lottery from a putative deep structure which, without extraposition, would have generated That you won the lottery delights everyone, is based precisely on the fact that sentences like *Everyone delights that you won the lottery are not produced by native speakers of English. Therefore, the fact of its occurrence in the speech of an English learner invalidates the evidence on which the derivation is based--at least as far as the L2 learner is concerned. In other words, there is no more justification for describing the error as "mismanaged extraposition" than as "mismanaged passivization" (Everyone is delighted that you won the lottery) or even "wrong verb choice" (Everyone rejoices that you won the lottery).

The only formal model for the linguistic description of L2 learners' errors that has been advanced so far is that of Corder 1971 (158):

The methodology of description is, needless to say, fundamentally that of bilingual comparison. In this, two languages are described in terms of a common set of categories and relations, that is, in terms of the same formal model. The technical problems of this are well known ...

In other words, description of the error consists of a combination of translation and contrastive analysis: the learner's deviant utterance is translated into the language of the native speaker (e.g. *I want going = I want to go), and the result of this translation is described in a contrastive statement (e.g. "-ing participle instead of to-infinitive"). The "technical problems" Corder refers to involve basically the same question that is raised in the contrastive analysis of any two dialects or languages. (For a recent discussion see Raabe 1977 and Kohn 1977.) But there are other, more fundamental, problems with this "bilingual comparison" model (cf. Morrissey 1979). First of all, it does not sufficiently isolate and identify the error, even when alternative "translations" are provided for (cf. Legenhausen 1975), since there is no non-arbitrary way to know just which and how many such translations would serve as equivalents of the deviant utterance. To take a simple example, does the error in *I don't agree with you in this point lie with the choice of the preposition (i.e. on this point), or of the noun (i.e. in this respect, in this regard, etc.)? Secondly, the contrastive statement based on this arbitrary translation only serves to label the forms or structures which have been translated, without necessarily revealing anything about the error itself--not the cause of the error, but what it is about the L2 that the learner has failed to understand. For example, *He gave me a good advice (i.e. some good advice, good advice) might be described as "singular instead of plural indefinite article", whereas *I need some informations (i.e. information) would be described as "plural instead of singular noun". This obscures the otherwise obvious fact that in both cases a non-count noun is used as if it were a count noun, a generalization we would surely want to capture in our description.

In order to avoid these theoretical problems, and simply to have a relatively unambiguous classificatory scheme in order to collect data, most researchers in second language acquisition or learning (see Note 1) omit the bilingual comparison part of the description altogether and characterize each error type as a string of surface forms or category symbols. For example, *two dog might be described as "N + 0" (Shapira 1978:251), or *I enjoy to get Presents as "V (finite) + to + V (infinitive)" (Morrissey 1979). This avoids the ambiguities involved in bilingual comparison, but, again, it does not tell us much about the error itself: *I enjoy to get presents surely reflects a different misperception of English grammar than *Literature helps to come to know others (i.e. us to, one to), but this difference is obscured when the error is described in terms of surface structure only. Furthermore, these surface representations may characterize correct structures as well as incorrect ones (e.g. a dog; I like to get presents).

To sum up these critical remarks, what is needed as a model for error description is one which a) is independent of considerations of the putative cause(s), systematicity, or communicative effect of errors, b)) is powerful enough to classify errors on a more profound basis than that of "bilingual comparison", c) does not make assumptions about underlying grammatical operations that are (presumably) part of native speaker competence, and d) serves the pedagogical need of explaining to both teacher and learner why the utterance is wrong. All of these needs seem best addressed by appeal to the notion of rule--not the rules of the learner's interlanguage or the rules of his L1, but the rules of the L2. Such rules could be called "error description rules" (see Note 2). These would be similar to the rules, written either in L1 or L2, that typically appear in pedagogical grammars, and in fact could be considered a subclass of pedagogical rules. The complex and intriguing problem of formulating pedagogical rules has been left almost exclusively to the attention of textbook writers and the individual instructor, and, in keeping with the trend in language teaching theory of the past twenty years or so, most textbooks keep the number of explicit formulations of rules to an absolute minimum, preferring to multiply the number of sentence examples in the form of pattern drills that illustrate a very small number of implicit rules. The result of the unprestigious status accorded to the problem of writing pedagogical rules is that when they do appear, even in the most modern language textbook (and in fact, sometimes the older grammars are considerably better in this respect), they are often oversimplified or inaccurate to the point of being useless, if not detrimental, to student and teacher alike. A lot of energy has been spent on what the notion "rule of grammar" might mean in relation to a theory of language; perhaps more effort is needed to define what the notion "pedagogical rule" entails and what its relationship might be to grammatical theory in general.

Learning how a language works is, after all, a goal that theoretical linguists, language teachers, textbook writers, and language learners all have in common. One might object that "descriptive adequacy" means quite a different thing to the language teacher or learner than to the theorist. Why or if this should be true is certainly not self-evident, however. What, in any science, testifies to the adequacy of a theory? First and foremost it is some practical invention or discovery that "proves" a theory; i.e. if it works, it is correct. So far there has been no practical demonstration of a linguistic theory dramatic enough to prove its correctness vis-à-vis other theories or, for that matter, vis-à-vis so-called "traditional grammar"--e.g. the kind of linguistic description of English found in Jespersen 1909-49, Zandvoort 1972, or Quirk et al. 1972. Perhaps this is because we have been looking for the wrong kind of proof, or for a particular kind of result to the exclusion of others that are equally desirable and valid. In a technological era it is difficult to think in terms of practical results without thinking immediately of computers and other technological hardware, e.g. translation machines, voice typewriters, computerized dictionaries and "grammar testing devices". Nevertheless, the language classroom should not be dismissed as an empirical proving ground. Whether or not a particular grammatical rule helps a student learn or not is a perfectly legitimate criterion of adequacy. And since we are talking about people and not machines, whether or not something works must be evaluated in human terms. -for example, machines respond more readily to numbers and algebraic formulations than to words, but human beings respond more readily to ordinary language. It follows that grammatical rules are better expressed in ordinary language as much as possible, at least if human learners are to be allowed as "grammar testing devices".

It all seems to boil down to who (or what) we are writing grammars for. The prevailing opinion in recent times has been that grammars written for computers (at least, with the characteristics of computers in mind) are the most sophisticated ones and the ones most nearly approaching scientific truth. Grammars written for learners of languages, on the other hand, or for teachers of languages, are considered necessarily oversimplified and watered-down versions of theoretical grammars. tend to forget that this attitude is only justified if we accept the premise that computers are better empirical tests of the validity of a theory than human beings. If we do not accept this premise, and return to the older and simpler notion of rule as a guideline for human behavior (in this case linguistic behavior), the question of whether or not, or to what extent, a rule "works" in the classroom has a much more significant and central role to play in linguistic theory than has been allowed to date.

There are some fundamental differences between what is being proposed here and the conception of the procedures and purpose of error analysis expounded by Corder in 1971. Corder suggests that the first stage in error analysis is the "recognition of idiosyncracy" and the interpretation of the deviant utterance--ie. determining what the speaker intended to say. ("Idiosyncracy", for the present purpose, can be equated with "deviance", although Corder makes--a distinction.) The second stage is error description, which involves the bilingual comparison discussed above. The third and--for Corder--the "ultimate object of error analysis is explanation." By this he means causology and systemics--the attempt "to account for how and why the . learner's idiosyncratic dialect is of the nature it is." He adds:

We must, I think, all agree that there could be no reason to engage in error-analysis unless it served one or both of two objects. Firstly, to elucidate what and how a learner learns when he studies a second language. This is a theoretical object (Corder 1967); secondly, the applied object of enabling the learner to learn more effectively by exploiting our knowledge of his dialect for pedagogical purposes. The second objective is clearly dependent on the first. We cannot make any principled use of his idiosyncratic sentences to improve teaching unless we understand how and why they occur (1971:158).

It is not true, however, that the pedagogical objective is dependent on explaining the nature of the learner's idiosyncratic dialect. If it were, it would mean that in order to teach a German student to say I was born in 1946 instead of *I am born in 1946, we must know that this error is (probably) due to interference from German (Ich bin 1946 geboren). Now, without denying that this knowledge, if it is not obvious to the learner, may help him understand his error, remember the point of difficulty, and hopefully eliminate it, the argument that this understanding is necessary or prerequisite to the learning or teaching objective is patently false, and ignores the obvious and more important point that the learner must know the rule for past tense formation in English in order to produce the correct sentence.

To take a more complicated example, when a German student says *Let's do that: tomorrow we'll call Betty and ask her (with stress and falling intonation on that, followed by a pause), he is most likely again following the German pattern: Machen wir es so (or, machen wir des): morgen rufen wir Betty an und fragen sie . The correct English version is Let's do this ... or Let's do it this way ... If we insist on scientific accuracy in our causal explanation, we would have to devise some empirical procedure--which, needless to say, is still lacking--to determine whether interference is in fact the cause, or the only case, of this error. But even assuming that we can be certain that interference is the cause, is it more important, from a pedagogical point of view, to make the learner aware of this fact (again, if it is not obvious to him) or to explain to him the rule (or rules) of English grammar that make his sentence unacceptable? This rule might take the following form: "This can be cataphoric (refer to a following noun phrase or proposition), but that cannot." The formulation of this rule, and presentation of it to the learner, may be prerequisite to learning, if anything is, but speculation as to the cause of the error is an interesting but at most only ancillary contribution to the pedagogical goal. From this perspective, a rule-based error description places the pedagogical component primarily--though not exclusively--in what Corder calls the second stage of analysis, rather than the third.

In order to illustrate the advantages of a rule-based error description, let us consider a relatively isolatable set of data which we can call errors in tense usage (cf. 2.1-10 and compare Richards 1971). The classification scheme, in the traditional mode, would consist of categories such as "simple present instead of present perfect", "present progressive instead of simple present", etc. Describing and classifying errors in this way does not reveal the relationship between errors that involve the misunderstanding or ignorance of the same rule, and, on the other hand, errors that are classified together may involve quite different and even unrelated rules. For example, the following sentences would be classified in three categories as simple present, present progressive, and simple past instead of present perfect (+ progressive), respectively:

This law exists since 1976 (i.e. has existed)
I am waiting for my sister since Friday (i.e. have waited or have been waiting)
I didn't have a chance to practice English since 1962 (i.e. haven't had)

On the other hand, the rule that "certain temporal adverbials (since, so far, up to now, etc.) must be accompanied by the perfective" applies to all three sentences, and we well have captured a significant gen eralization about these errors if we classify all three sentences as violations of this one rule (See Note 3.)

Formulating an "adequate" error description rule, as in any kind of language description, involves general as well as language-specific questions. What kind of metalanguage (grammatical terminology and logical framework) should be used? What is the scope of any particular rule; that is, how much of the grammar should be incorporated in it? How can the rule be expressed in a way (in either L1 or L2) that will be maximally useful and comprehensible to the learner? It would be difficult to underestimate the task of error description, considering the range of errors--syntactic, lexical, semantic, phonological) that have to be accounted for. On the other hand, it should not be overestimated. First of all, the range of errors actually produced by learners is nothing like the range of possible ungrammatical or unacceptable utterances--which, by comparison, a complete theoretical grammar would have to account for. *Colorless green ideas sleep furiously and similar violations of universal selection restrictions, for example, do not have to be accounted for in an error grammar because the likelihood of such utterances being produced by a learner of English is minimal. Secondly, the corpus of errors on which any particular set of error description rules is based should be limited to those produced by a given population of learners (most logically, one with a common L1), this corpus being further limited by considering only the statistically most frequent errors (what Kielhofer 1975 calls "Standardfehler"). Thirdly, many of the errors which are produced will be describable, to varying degrees of adequacy, by grammatical and lexical rules which are already accessible in standard grammars and dictionaries. Finally, the task of writing rules is simplified considerably if we disburden ourselves of the "explicitness" criterion, as expressed by Chomsky and Halle (1968:60):

The rules of the grammar operate in a mechanical fashion; one may think of them as instructions that might be given to a mindless robot, incapable of exercising any judgment or imagination in their application.

Such a premise, whatever its validity for general linguistic theory, is obviously inappropriate for error and for pedagogical grammar in general, where we are concerned with grammar for people, not for robots. The ultimate criterion for pedagogical rules is: Does it work? Which formulation more effectively helps the learner to avoid errors and their recurrence? These questions are also open to empirical investigation.

What follows is an attempt to formulate some error description rules for English. Most of these rules will be familiar, though they may be formulated differently than in other grammars. For the sake of simplicity, grammatical terminology and the general analytical scheme follow that of Quirk et al. (1972). The formulation of the rules is of course tentative--particularly with regard to how they might eventually be presented to the learner, since age, cognitive maturity and other individual factors would have to be considered, along with the likely possibility that grammatical discussion is more effective when it takes place in the learner's native language. In some cases the analysis of these errors has posed linguistic problems that are quite unfamiliar, perhaps even considered here for the first time (cf. 1.1, 1.2.3, 1.5.2, 1.6.3, 1.6.4, 1.7.2, 1.9.4, 2.7.2, 3.10, 3.11, 3.14, 6.2, 6.3). In any event the rules vary greatly in their manner of formulation, are pragmatic rather than systematic, and await further discussion.

As already mentioned, there will be disagreement-even among native speakers--about how wrong some of the examples are, or indeed about whether they are erroneous at all. This seems unavoidable, since there are simply too many factors affecting acceptability judgments that cannot be controlled in a study such as this. On the other hand, if all the examples about which native speakers would disagree were eliminated, the corpus would be drastically reduced in size and in interest, since this would completely side-step the important question as to why the problematic examples do in fact seem odd or unacceptable to at least some speakers. "Is the sentence wrong or unacceptable, and to what degree?'' is a different question from "Why does it sound wrong or odd?", and it is the latter which is addressed here. In sum the term "error" or "mistake" as used here means not that a given sentence is wrong or unacceptable in any categorical sense, but rather that there is something about the sentence which, considering the context in which it occurred, sounds odd enough that at least some native speakers would question its acceptability or the likelihood of its being formulated in this way by a native speaker of English. This definition is broad enough to include all the cited examples, though "problematic utterances" might be a more accurate term for them than "errors".

Corrections--perhaps more appropriately "revisions" in view of the preceding remarks--of the examples have not been provided because, first, what is wrong or odd in the sentences in most cases should not be difficult to discern--if so, it is probably a bad example. Secondly, since there are always more than one way to correct a deviant utterance, it would be pointless to try to account for all of the possibilities. Thirdly, supplying (arbitrary) corrections would distract attention from the problem of what is wrong with the examples to the quite different problem of how they can be repaired, since the natural tendency is to fix what is wrong and move on rather than dwell on the question of why it is wrong. Leaving the task of correction to the reader is meant both to discourage this tendency and to encourage consideration of the question of why the sentences are problematic in the first place.

Note 1: The distinction between language "learning" and "acquisition" depends on whether the language is taught or is acquired in a natural environment. However, the whole notion of "natural environment" applied to particularly adult) L2 learners may be spurious. Learning an L2, under any circumstances, does not take place in an environment that is at all comparable with the environment in which one learns one's mother tongue, so what is "natural" for an L2 learner cannot be the same as what is "natural" for an L1 learner. So just what can it mean for an adult to acquire a second language in a natural environment? Even though one can establish arbitrary criteria such as classroom walls, it seems to me that the teaching element in L2 contact situations extends far beyond those walls, and may be just as natural a part of that special environment as anything else--whether the teaching takes the form of formal instruction or not.

Note 2: Burt and Kiparsky (1974:72) hint at the formalization of the role of rule in error description, although this is, in their words, "a systematic equivocation in our terminology. The items that are ranked in the hierarchy (i.e. of errors) ... are seen on the one hand as types of mistakes, and on the other hand as the part of the grammar (rules or other description) which is violated to produce such mistakes."

Note 3: German learners tend to use the simple present or present progressive when they consider the predication still in effect at the moment of speaking, but the simple past when they consider it terminated (cf. 2.4.1). This "interlanguage rule" most probably reflects the fact that in German the distinction is made with the Perfekt (Bis jetzt hat sie in einem Kindergarten gearbeitet) and the Prasens (Dieses Gesetz existiert seit 1976). Again, referring to the point made earlier, it is interesting to know this, but primarily because it may help us formulate the English rule more clearly-in this case by pointing out that the English perfective does not depend on this question of whether the predication is still true at he precise moment of utterance. Trying to corroborate our speculation about the learner's interlanguage rule (as opposed to focusing our attention on English) by studying, as Corder (1971:107) recommends, his correct sentences as well as his errors will only yield, at best, relative statements of the kind "Structure A is used in environment X a certain percent of the time correctly and a certain percent incorrectly." (For an example of a study of this kind involving German learners' use of the English present progressive vs. present perfect, cf. ZydatiB 1976, 1977.) Thus any interlanguage rule (if these can be called rules at all) will allow us to predict errors only in this relative statistical sense, which is quite different from what we mean by a rule of language that predicts some (grammatical) structures but not others (ungrammatical ones). This problem is in strict relation to the one discussed earlier--that of learner "competence".