Prediction of difficulty by means of contrastive procedures was not without glaring shortcomings. For one thing, the process was oversimplified. Subtle phonetic, phonological, and grammatical distinctions were not carefully accounted for. Second, it was very difficult, even with six categories, to determine exactly which category a particular contrast fit into. For example, when a Japanese speaker learns the English /r/, is it a case of a level 0, 1, or 3 difficulty? A case can be made for all three. The third and most problematic issue centered on the larger question of whether or not predictions of difficulty levels were actually verifiable.
The attempt to predict difficulty by means of contrastive analysis is what Ronald Wardhaugh (1970) called the strong version of the САН, аversion that he believed was quite unrealistic and impracticable. Wardhaugh noted (p. 125) that "at the very least, this version demands of linguists that they have available a set of linguistic universals formulated within a comprehensive linguistic theory which deals adequately with syntax, semantics, and phonology." He went on to point out the difficulty (p. 126), already noted, of an adequate procedure, built on sound theory, for actually contrasting the forms of languages: "Do linguists have available to them an overall contrastive system within which they can relate the two languages in terms of mergers, splits, zeroes, over-differentiations, under-differentiations, reinterpretations?" And so, while many linguists claimed to be using a scientific, empirical, and theoretically justified tool in contrastive analysis, in actuality they were operating more out of mentalistic subjectivity.
Wardhaugh noted, however (p. 126), that contrastive analysis had intuitive appeal, and that teachers and linguists had successfully used "the best linguistic knowledge available ... in order to account for observed difficulties in second language learning." He termed such observational use of contrastive analysis the weak version of the САН. The weak version does not imply the a priori prediction of certain degrees of difficulty. It recognizes the significance of interference across languages, the fact that such interference does exist and can explain difficulties, but it also recognizes that linguistic difficulties can be more profitably explained a posteriori— after the fact. As learners are learning the language and errors appear, teachers can utilize their knowledge of the target and native languages to understand sources of error.
The so-called weak version of the САН is what remains today under the label cross-linguistic influence (CLI), suggesting that we all recognize the significant role that prior experience plays in any learning act, and that the influence of the native language as prior experience must not be overlooked. The difference between today's emphasis on influence nth ' than prediction, is an important one. Aside from phonology, which remains the most reliable linguistic category for predicting learner performance, as illustrated at the beginning of the chapter, other aspects of language present more of a gamble. Syntactic, lexical, and semantic interference show far more variation among learners than psychomotor-based pronunciation interference. Even presumably simple grammatical categories like word order, tense, or aspect have been shown to contain a good deal of variation. For example, one might expect a French speaker who is beginning to learn English to say "I am in New York since January"; however, to predict such an utterance from every French learner of English is to go too far.
The most convincing early criticism of the strong version of the CAD was offered by Whitman and Jackson (1972), who undertook to test empirically the effectiveness of contrastive analysis as a tool for predicting areas of difficulty for Japanese learners of English. The predictions of four separate contrastive analysis rubrics (including that of Stockwell, Bowen, S Martin 1965) were applied to a forty-item test of English grammar to determine, a priori, the relative difficulty of the test items for speakers of Japanese. The test was administered to 2500 Japanese learners of English who did not know the relative predicted difficulty of each item. The results of the test were compared with the predictions. The result: Whitman and Jackson found no support for the predictions of the contrastive analyses so carefully worked out by linguists! They concluded (p. 40) that "contrastive analysis, as represented by the four analyses tested in this project, is inadequate, theoretically and practically, to predict the interference problems of a language learner."
Another blow to the strong version of the САН was delivered by Oller and Ziahosseiny (1970), who proposed what one might call a "subtle differences" version of the САН on the basis of a rather intriguing study of spelling errors. They found that for learners of English as a second language, English spelling proved to be more difficult for people whose native language used a Roman script (for example, French, Spanish) than for those whose native language used a non-Roman script (Arabic, Japanese). The strong form of the САН would have predicted that the learning of an entirely new writing system (Level 4in the hierarchy of difficulty) would be more difficult than reinterpreting (Level 3) spelling rules. Oller and Ziahosseiny (p. 186) found the opposite to be true, concluding that "wherever patterns are minimally distinct in form or meaning in one or more systems, confusion may result."
The learning of sounds, sequences, and meanings will, according to Oller and Ziahosseiny's study, be potentially very difficult where subtle distinctions are required either between the target language and native language or within the target language itself. In the case of their research on spelling English, there were more differences between non-Roman writing and Roman writing, but learners from a non-Roman writing system had to make fewer subtle distinctions than did those from the Roman writing system. Examples of subtle distinctions at the lexical level may be seen in false cognates like the French word parent, which in the singular means "relative" or "kin," while only the plural (parents) means "parents." Consider the Spanish verb embarazar, which commonly denotes "to make pregnant," and has therefore been the source of true "embarrassment" on the part of beginners attempting to speak Spanish! In recent years, research on CLI has uncovered a number of instances of subtle differences causing great difficulty (Sjoholm 1995).
The conclusion that great difference does not necessarily cause great difficulty underscores the significance of intralingual (within one language) errors (see subsequent sections in this chapter), which are as much a factor in second language learning as interlingual (across two or more languages) errors. The forms within one language are often perceived to be minimally distinct in comparison to the vast differences between the native and target language, yet those intralingual factors can lead to some of the greatest difficulties.
Today we recognize that teachers must certainly guard against a priori pigeon-holing of learners before we have even given learners a chance to perform. At the same time, we must also understand that CLI is an important linguistic factor at play in the acquisition of a second language (Jaszczolt 1995). CLI implies much more than simply the effect of one's first language on a second: the second language also influences the first; moreover, subsequent languages in multilinguals all affect each other in various ways. Specialized research on CLI in the form of contrastive lexicology, syntax, semantics, and pragmatics continues to provide insights into SLA that must not be discounted (Sharwood-Smith 1996; Sheen 1996). Sheen (1996) found, for example, that in an ESL course for speakers of Arabic, overt attention to targeted syntactic contrasts between Arabic and English reduced error rates. Indeed, the strong form of the САН was too strong, but the weak form was also perhaps too weak. CLI research offers a cautious middle ground.
Fred Eckman (1977, 1981) proposed a useful method for determining directionality of difficulty. His Markedness Differential Hypothesis (otherwise known as markedness theory) accounted for relative degrees of difficulty by means of principles of universal grammar. Celce-Murcia and Hawkins (1985:66) sum up markedness theory:
It distinguishes members of a pair of related forms or structures by assuming that the marked member of a pair contains at least one more feature than the unmarked one. In addition, the unmarked (or neutral) member of the pair is the one with a wider range of distribution than the marked one. For example, in the case of the English indefinite articles (a and an), an is the more complex or marked form (it has an additional sound) and a is the unmarked form with the wider distribution.
Eckman (1981) showed that marked items in a language will be more difficult to acquire than unmarked, and that degrees of markedness will correspond to degrees of difficulty. Rutherford (1982) used markedness theory to explain why there seems to be a certain order of acquisition of morphemes in English: marked structures are acquired later than unmarked structures. Major and Faudree (1996) found that the phonological performance of native speakers of Korean learning English reflected principles of markedness universals.
In recent years, the attention of some second language researchers has expanded beyond markedness hypotheses alone to the broader framework of linguistic universals in general (Major & Faudree 1996; Eckman 1991; Carrolland Meisel 1990; Comrie 1990; Gass 1989). Some of these arguments focus on the applicability of notions of universal grammar (UG) to second language acquisition (White 1990; Schachter 1988; among others). As we saw in Chapter 2, many of the "rules" acquired by children learning their first language are presumed to be universal. By extension, rules that are shared by all languages comprise this UG. Such rules are a set of limitations or parameters (Flynn 1987) of language. Different languages set their parameters differently, thereby creating the characteristic grammar for that language. The hope is that by discovering innate linguistic principles that govern what is possible in human languages, we may be better able to understand and describe contrasts between native and target languages and the difficulties encountered by adult second language learners. Research on UG has begun to identify such universal properties and principles, and therefore represents an avenue of some promise.
Markedness theory and UG perspectives provide a more sophisticated understanding of difficulty in learning a second language than we had previously from the early formulations of the САН, and fit more appropriately into current studies of CLI. But we do well to remember that describing and predicting difficulty amidst all the variables of human learning is still an elusive process. Teachers of foreign languages can benefit from UG and markedness research, but even in this hope-filled avenue of research, an instant map predicting learner difficulties is not right around the corner.
The САН stressed the interfering effects of the first language on second language learning and claimed, in its strong form, that second language learning is primarily, if not exclusively, a process of acquiring whatever items are different from the first language. As already noted above, such a narrow view of interference ignored the intralingual effects of learning, among other factors. In recent years researchers and teachers have come more and more to understand that second language learning is a process of the creative construction of a system in which learners are consciously testing hypotheses about the target language from a number of possible sources of knowledge: knowledge of the native language, limited knowledge of the target language itself, knowledge of the communicative functions of language, knowledge about language in general, and knowledge about life, human beings, and the universe. The learners, in acting upon their environment, construct what to them is a legitimate system of language in its own right—a structured set of rules that for the time being bring some order to the linguistic chaos that confronts them.
By the late 1960s, SLA began to be examined in much the same way that first language acquisition had been studied for some time: learners were looked on not as producers of malformed, imperfect language replete with mistakes but as intelligent and creative beings proceeding through logical, systematic stages of acquisition, creatively acting upon their linguistic environment as they encountered its forms and functions in meaningful contexts. By a gradual process of trial and error and hypothesis testing, learners slowly and tediously succeed in establishing closer and closer approximations to the system used by native speakers of the language. A number of terms have been coined to describe the perspective that stresses the legitimacy of learners' second language systems. The best known of these is interlanguage, a term that Selinker (1972) adapted from Weinreich's (1953) term "interlingual." Interlanguage refers to the separateness of a second language learner's system, a system that has a structurally intermediate status between the native and target languages.
Nemser (1971) referred to the same general phenomenon in second language learning but stressed the successive approximation to the target language in his term approximative system. Corder (1971:151) used the term idiosyncratic dialect to connote the idea that the learner's language is unique to a particular individual, that the rules of the learner's language are peculiar to the language of that individual alone. While each of these designations emphasizes a particular notion, they share the concept that second language learners are forming their own self-contained linguistic systems. This is neither the system of the native language nor the system of the target language, but a system based upon the best attempt of learners to bring order and structure to the linguistic stimuli surrounding them. The interlanguage hypothesis led to a whole new era of second language research and teaching and presented a significant breakthrough from the shackles of the САН.
The most obvious approach to analyzing interlanguage is to study the speech and writing of learners, or what has come to be called learner language (Lightbown & Spada 1993; C.James 1990). Production data is publicly observable and is presumably reflective of a learner's underlying competence—production competence, that is. Comprehension of a second language is more difficult to study since it is not directly observable and must be inferred from overt verbal and nonverbal responses, by artificial instruments, or by the intuition of the teacher or researcher.
It follows that the study of the speech and writing of learners is largely the study of the errors of learners. "Correct" production yields little information about the actual linguistic system of learners, only information about the target language system that learners have already acquired. Therefore, our focus in the rest of this chapter will be on the significance of errors in learners' developing systems, otherwise known as error analysis.
Human learning is fundamentally a process that involves the making of mistakes. Mistakes, misjudgments, miscalculations, and erroneous assumptions form an important aspect of learning virtually any skill or acquiring information. You learn to swim by first jumping into the water and flailing arms and legs until you discover that there is a combination of movements—a structured pattern—that succeeds in keeping you afloat and propelling you through the water. The first mistakes of learning to swim are giant ones, gradually diminishing as you learn from making those mistakes. Learning to swim, to play tennis, to type, or to read all involve a process in which success comes by profiting from mistakes, by using mistakes to obtain feedback from the environment, and with that feedback to make new attempts that successively approximate desired goals.
Language learning, in this sense, is like any other human learning. We have already seen in the second chapter that children learning their first language make countless "mistakes" from the point of view of adult grammatical language. Many of these mistakes are logical in the limited linguistic system within which children operate, but, by carefully processing feedback from others, children slowly but surely learn to produce what is acceptable speech in their native language. Second language learning is a process that is clearly not unlike first language learning in its trial-and-error nature. Inevitably learners will make mistakes in the process of acquisition, and that process will be impeded if they do not commit errors and then benefit from various forms of feedback on those errors.