REPRESENTATION AND PROCESSING

Various researchers have invoked information-processing models to describe L2 learning as the acquisition of complex cognitive skills (e.g., DeKeyser, 1998, 2001; Hulstijn, 1990; Johnson, 1996; McLaughlin, 1987, 1990; O'Malley & Chamot, 1990; Towell & Hawkins, 1994). The notion of information processing was developed in cognitive psychology to describe skill acquisition as a gradual change in knowledge from declarative to procedural mental representations (Anderson, 1983, 1985) and in executive control from controlled to automatic processing of those mental representations (Shiffrin & Schneider, 1977). Regarding language, declarative knowledge refers to knowledge of the language system, such as word definitions and rule-based representations, whereas procedural knowledge refers to knowledge about how to perform cognitive activities, including language comprehension and production. Skill development depends on transforming declarative representations, through practice, into production rules that represent procedural knowledge. This development is described as a transition from controlled processing, which requires a great deal of attention and use of short-term memory, to automatic processing, which operates on routinized procedures available in long-term memory. The proceduralization of rule-based representations occurs through practice and feedback (Anderson, Corbett, Koedinger, & Pelletier, 1995), which are considered crucial elements in information-processing models of L2 development, because they engage learners in processes of restructuring interlanguage representations and, thus, have particular relevance in immersion contexts (Lyster, 1990; Ranta & Lyster, 2003).

An alternative view of skill acquisition is Logan's (1988) instance theory, whereby automatization involves, rather than proceduralization of rule-based representations with increasingly less attention, a transition from rule- to memory-based performance (see DeKeyser, 2001; Robinson & Ha, 1993; Schmidt, 1992, 2001). In this view, procedures initially deriving from rule-based representations become available as memory-based chunks, which then operate autonomously. With minimal computational demands, retrieval from the memory-based system involves more efficient processing, enables fluent performance, and is thus considered synonymous with automaticity. However, to account more specifically for L2 learning and variability in L2 performance, still from an information-processing perspective, Skehan (1998) described a dual-coding system, which instead reconciles both rule- and memory-based systems as equally important representational systems for L2 learners.

According to Skehan (1998), the dual-mode system comprises two interrelated representational systems: an analytic rule-based system and a memory-driven exemplar-based system. The rule-based system is generative and parsimoniously organized, with the advantage of its propensity for creative and precise computations but the disadvantage of a heavy processing burden during ongoing language use. In contrast, the exemplar-based system is composed of lexicalized chunks, stored in a redundant unstructured memory system, which have the advantage of being retrieved quickly during ongoing communication because they do not require excessive internal computation but the disadvantage of having only limited potential for expressing new and precise meanings. Access to the dual-mode system thus affects fluency, as learners engage differentially in retrieval, leading to either computed rule-based performance or memory-driven exemplar-based performance. Skehan argued that during online communication, “where communicative pressure and accessibility are paramount,” the exemplar-based system will be the system of choice, thus “reducing the likelihood that the more open and generative rule-based system will be used” (p. 62). These two representational systems, however, are not entirely separate; instead, they are “in constant dialect,” enabling L2 learners to engage in complementary processes of analysis and synthesis: “The analysis is necessary to enable the learner to gain generativity and flexibility, and the synthesis is necessary to enable fluency and control to be achieved” (Skehan, p. 92).

With respect to L2 development, Skehan (1998) argued that interlanguage changes are more effectively activated through the rule-based system than through the exemplar-based system. Moreover, he claimed that the rule-based system “is also more likely to be more sensitive to feedback since the precision and system which accounts for rule-organization will make the feedback more informative” (p. 88), whereas feedback may not be so effective in effecting change in the exemplar-based system, because exemplars are not part of a structured system with connections to rules that can produce general change. In further consideration of the potential effectiveness of feedback, or what he called “feedback appreciation,” Skehan suggested that conscious awareness of feedback (“consciousness enhanced processing”) predisposes the learner toward a rule-based perspective, which in turn is more likely to lead to longer-term change.

RECASTS AND PROMPTS

This section introduces a distinction between two very different types of feedback, which have been observed in immersion classrooms (Lyster & Ranta, 1997) and presented as complementary teacher moves with different purposes for different learners in different contexts (Lyster, 2002a, 2002b). Notwithstanding the pedagogical justification for teachers to use both these moves “in accordance with their students' language abilities and content knowledge … without abandoning one at the expense of the other” (Lyster, 2002b, p. 251), this section presents a theoretical justification, by drawing on the information-processing framework described in the previous section, for comparing recasts with prompts in terms of their relative effectiveness in a classroom experiment. In the next section, the rationale is developed further by drawing on processing and representational difficulties specific to grammatical gender attribution.

Long (1996) and others (e.g., Gass, 1997; Pica, 1994) have proposed that the negotiation of meaning, defined as various input modifications and interactional moves including semantically contingent feedback such as repetition and reformulation, provides learners with implicit negative evidence and thus serves to benefit L2 development. One type of implicit reformulation move that has received increasing attention in both L1 and L2 contexts is the recast—a well-formed reformulation of a learner's nontarget utterance with the original meaning intact, an example of which follows (this and subsequent examples of feedback types were observed in either the present study or Lyster & Ranta's [1997] study and were selected because they target grammatical gender; M and F in parentheses indicate masculine and feminine determiners, respectively):

There is nothing disconfirming in the teacher's recast (La guimauve) that actually incites the learner to notice that le guimauve is ungrammatical; the learner could even infer from this and other classroom input that le and la are interchangeable variants. Some researchers, however, have argued that the juxtaposition of the learner's ungrammatical utterance with the teacher's reformulation provides the learner with an ideal opportunity to make a cognitive comparison and to notice the gap between the targetlike and nontargetlike forms, especially because meaning is held constant and so the learner's processing resources are freed up to focus on form (e.g., Doughty, 2001; Long; Long & Robinson, 1998).

A sampling of studies investigating the use of recasts demonstrates their effectiveness in some contexts, but not in comparison with other clearly identified feedback. For example, during experimentally controlled interaction with adults, Long, Inagaki, and Ortega (1998) showed that recasts can be more effective for some target features than models, which are provided to learners before they speak and, thus, are not a comparable type of semantically contingent feedback. Mackey and Philp (1998) showed that experimentally controlled interaction during which a researcher provides L2 learners with intensive recasts, which focus consistently on one type of structure, is more effective, in the case of adults developmentally ready to learn the target form, than interaction without intensive recasts (see also Han, 2002). With young learners in content-based classrooms, Doughty and Varela (1998) showed that corrective recasting, which includes a recast preceded by a repetition of the learner's error and intonational stress added to both repetition and recast to emphasize the incorrect and correct forms, is also more effective than no feedback. Ohta (2000) showed that four adult learners who produced private speech with moderate to high frequency while wearing lapel microphones in a form-oriented foreign language classroom tended to repeat recasts, regardless of whether they were the intended recipients of the recasts (cf. Nabei & Swain, 2002). For a comprehensive review of recast studies, see Nicholas, Lightbown, and Spada (2001).

Based on my own observation studies of immersion classrooms (e.g., Lyster, 1998b, 2002b), I have argued that recasting—as defined in the L1 literature (e.g., Marcus, 1993; Saxton, 1997) and as observed in immersion classrooms (i.e., an implicit targetlike reformulation of a learner's utterance)—is not the most effective way of providing young L2 learners with negative evidence in classrooms where the primary focus is on subject matter, especially in comparison with other feedback options. My analyses of immersion classroom discourse suggested that recasts are, for the most part, embedded in meaning-focused negotiation and are thus ideal for facilitating the delivery of complex subject matter, because they provide supportive, scaffolded help, which serves to move the lesson ahead when the target forms in question are beyond the students' current production abilities. Recasts used in this way compete with many other demands on attention during content-based instruction and appear to be ambiguous, because they share discourse functions with a similar proportion of teacher repetitions of well-formed utterances (Lyster, 1998b); that is, recasts of ill-formed utterances and repetitions of well-formed utterances together appear to confirm or disconfirm the meaning of a learner's message, not its form. Moreover, recasts often co-occur with signs of approval (i.e., positive feedback moves including affirmations such as Oui, C'est ça, and OK and praise markers such as Très bien, Bravo, and Excellent; see Lyster, 1998b), as teachers respond to the veracity of learners' messages. Similarly, Musumeci (1996) and Pica (2002) documented the difficulty of effectively drawing attention to form specifically during content-based instruction, even in the case of adult L2 learners. Indeed, researchers discussing a wide range of classroom contexts have noted the ambiguity of implicit feedback (e.g., Allwright, 1975; Chaudron, 1977; Fanselow, 1977; Long, 1977; Netten, 1991).

In addition to recasts, I observed that teachers have at their disposal a set of interactional moves that provide less ambiguous and more cognitively engaging feedback, referred to interchangeably as the “negotiation of form” (Lyster, 1998a, 2002a; Lyster & Ranta, 1997), “form-focused negotiation” (Lyster, 2002b), and “prompts” (Lyster, 2002b; Ranta & Lyster, 2003). Prompts provide a solution to Swain's (1985) call for immersion teachers to “push” their students to be more accurate in their output. They include the following four types of teacher response:

Although these four prompting moves, used separately or in combination, represent a wide range of feedback types, they have one crucial feature in common: They withhold correct forms (and other signs of approval) and instead offer learners an opportunity to self-repair by generating their own modified response. This approach resembles the “clueing” procedure or “withholding phenomenon” identified by McHoul (1990) in his study of feedback in subject-matter classrooms. In contrast, recasts provide learners with correct target forms, which frequently co-occur with signs of approval.

In their study of cognitive processes generated by output, Swain and Lapkin (1995) posited that feedback enables learners to notice problems in their output and pushes them to conduct an analysis leading to modified output. Swain and Lapkin contended that what occurs between the first and second outputs is part of the process of L2 learning. I would add that the extent to which restructuring processes are activated between the learner's first and second outputs depends on the type of feedback. That is, not all types of feedback lead to similar degrees of pushing. Recasts, for example, tend not to push learners to modify their nontarget output, at least not immediately following feedback (e.g., Chaudron, 1977; Lyster & Ranta, 1997; Panova & Lyster, 2002; Pica, 1988, 2002). From a discourse perspective, a student's repetition of a teacher's recast can be seen as a redundant move in an exchange in which the teacher, by virtue of recasting, both initiates and completes the repair within a single move. On the small number of occasions when learners do modify their ill-formed utterances following recasts, the modification may simply be a repetition of the alternative form, involving retrieval from short-term memory rather than from long-term memory.

Ellis (1997) and others (e.g., Bialystok & Sharwood Smith, 1985; de Bot, 1996) have distinguished between two types of acquisition: acquisition as the internalization of new forms, and acquisition as an increase in control over forms that have already been internalized. In this view, as exemplars of positive evidence (see Braidi, 2002; Leeman, 2003), recasts occurring in appropriate discourse contexts can facilitate the encoding of new declarative knowledge. Prompts, on the other hand, given their aim to elicit modified output, can enhance control over already-internalized forms—that is, prompts serve to assist learners in the transition of declarative to procedural knowledge (regarding the psycholinguistic effect of output, see de Bot, p. 549).

In the immersion context, because learners have had years of exposure to L2 input, including the target forms that they consistently have problems acquiring, they need to be pushed, when their focus is on academic content, to use target forms that are in competition with highly accessible interlanguage forms (Ranta & Lyster, 2003; Swain, 1985). Prompts, therefore, may be particularly beneficial in immersion classrooms and other meaning-focused instructional contexts where continued recasting of what students already know may prove to be less effective for promoting the restructuring of interlanguage representations and the proceduralization of competing targetlike representations. This was the case in Ammar's (2003) recent classroom study, which revealed superior effects for prompts over recasts in the acquisition of possessive determiners by Grade 6 Francophone learners of English as a second language (ESL). She also found that prompts were particularly effective for lower proficiency learners, whereas higher proficiency learners appeared to benefit similarly from both recasts and prompts. Other studies as well have suggested that low-proficiency learners might be at a disadvantage in their limited ability to notice recasts as corrective feedback (Lin & Hedgcock, 1996; Mackey & Philp, 1998; Netten, 1991).

Whereas Doughty (2001) argued that recasting is an ideal type of feedback, because L2 learners are able to store the target reformulation in working memory and to make a direct comparison between input and output, de Bot (2000) suggested that “there is never a direct comparison between input and output because the input information is immediately processed and not stored in memory in that form” (p. 228). Conversely, in response to prompts, learners must attend to the retrieval from long-term memory of previously encoded representations, retrieving either an alternate exemplar or a rule for computing a more targetlike form. De Bot (1996, 2000) argued that this increased level of activation enhances the likelihood of the retrieved item being selected again, because the attention required for retrieval from long-term memory and subsequent production stimulates the development of strong connections in memory. By retrieving target forms stored in long-term memory, therefore, L2 learners are more likely to restructure existing interlanguage representations than by merely hearing the forms in the input (de Bot, 1996). The restructuring of knowledge representations and concomitant processing constraints specific to grammatical gender are addressed in the next section.

GRAMMATICAL GENDER

Grammatical gender, which is best defined in contrast to biological (or natural) gender, encompasses the gender of generic nouns—that is, inanimate nouns (e.g., un chapeau “a hat,” une casquette “a cap”) as well as a class of animate nouns, that refer generically, without identifying sex, to many living entities such as insects (e.g., une mouche, “a fly”), marine life (e.g., une crevette “a shrimp,” une truite “a trout”), birds (e.g., un moineau “a sparrow,” une alouette “a lark”), and various other animals (e.g., une gazelle “a gazelle,” un serpent “a snake”). Biological gender, on the other hand, encompasses gender determined by sex, in reference to humans (e.g., une fille vs. un garçon “a girl” vs. “a boy”; une actrice vs. un acteur “an actress” vs. “an actor”) and many animals (e.g., un chat vs. une chatte “a cat”; un loup vs. une louve “a wolf”).

Carroll (1989) described a useful distinction between gender agreement and gender attribution. On the one hand, gender agreement, a variable characteristic of modifiers, is systematically derivative and rule-based and depends on the context of occurrence of modifiers with nouns. Gender attribution, on the other hand, is an inherent feature of nouns that remains invariable and independent of the context of occurrence. Through exposure to language arts materials often designed for and by native speakers (NSs) of French (Lyster, 1990), immersion students develop some knowledge of gender agreement, as they typically engage in grammatical exercises requiring the transformation of adjectival forms from masculine to feminine (nouveau → nouvelle “new”). Similarly, with respect to gender attribution, immersion students become familiar with morphological rules governing animate nouns whose inherent gender is determined by sex (e.g., un chanteur → une chanteuse “a singer”; un chat → une chatte “a cat”), again because these types of transformational exercises appear in their language arts materials. However, their materials do not focus on the inherent gender of generic nouns, because this information is acquired early by NSs and does not involve rules of grammatical spelling, which permeate grammar-based materials designed for NSs of French and their language arts classes (e.g., Fazio & Lyster, 1998). Consequently, the resulting gap in immersion students' knowledge of gender attribution seriously constrains their ability to compute appropriate matches between determiners and generic nouns.

Carroll (1989) shed further light on the processing difficulties experienced by immersion students by arguing that Anglophone learners of French, unlike NSs of French, do not learn determiners as parts of nouns but learn them instead as distinct syntactic words and independent phonological units. The morphosyntactic representations of nouns are thus different for Francophone and Anglophone children, and so it follows that the processing of these different representations entails different types of computation. In the case of learning French L1, a phonological representation initially co-occurs with a nominal constituent to form an unanalyzed chunk composed of determiner + noun (e.g., laporte

= the door). “When the learner figures out that the determiners are distinct lexical items, the phonological representations will be reduced and the morphosyntactic representation will be augmented to include [gender specification]” (Carroll, p. 578). Thus, NSs of French process determiners and nouns, at least initially, as coindexed chunks; when these chunks are later analyzed as separate constituents, the noun still retains its inherent gender specification. Anglophone learners of French L2 appear instead to acquire nouns and their determiners as separate entities, applying, on the one hand, their knowledge of definiteness, indefiniteness, and possession to acquire determiners such as le/la, un/une, and mon/ma, and, on the other, their conceptual and morphosyntactic knowledge of nouns in English to acquire corresponding lexical entries in French. Thus, Carroll suggests, in the case of Anglophone learners of French L2, the functions deriving underlying phonological representations do not chunk determiners with nouns, nor do the functions deriving morphosyntactic representations of nouns include gender specification. In her view, this explains the difficulty that Anglophone learners, both children and adults, have in retrieving the gender specification of nouns during online production, “although they might be able to resort to some heuristic, a testable prediction” (Carroll, p. 578).

The problem with resorting to some heuristic or rule-driven mnemonic device is that many French grammarians (e.g., Bosquart, 1998; Grevisse & Goosse, 1995) and, thus, L2 teachers and their instructional materials claim that gender attribution is arbitrary and unsystematic in the case of generic nouns (i.e., to be learned on an item-by-item basis). Yet, considerable counterevidence to the claim that gender attribution is not rule governed suggests instead that gender attribution, even in the case of generic nouns, is largely rule driven, based on word-internal structural properties. For example, based on lexicographical analyses of nouns appearing in Le Petit Larousse, Tucker (1967, p. 13) concluded that “gender does co-occur, in a systematic and predictable manner, with noun ending” (see also Batchelor & Offord, 1993; Huot, 1991; Surridge, 1986, 1995; Tucker, Rigault, Lambert, & Segalowitz, 1968). Some predictors are primarily orthographic, showing up in the spelling of suffixes. For example, the suffix -elle is reliably feminine, whereas the homophonous ending -el reliably predicts masculine gender. Other predictors are phonological, such as the reliably masculine ending /o/ and the reliably feminine ending

, each with variable orthography: /o/ = -eau, -ot, -op;

= -aite, -ète, -ette.

Research undertaken in preparation for the present study also confirmed an unexpectedly high number of orthographic and phonological patterns at play in predicting grammatical gender (Lyster, 2004). Using a CD-ROM version of Le Robert Junior Illustré, I compiled a corpus of more than 10,000 common nouns in French organized according to some 100 endings that are either reliably feminine, reliably masculine, or ambiguous. Close to 80% of the nouns in this corpus were found to have endings that reliably predict their grammatical gender. This finding, which is at odds with traditional grammars, provided the impetus for the present study to investigate the extent to which FFI and corrective feedback effect change in immersion students' rule-based representations of grammatical gender.

RESEARCH QUESTIONS AND PREDICTIONS

The purpose of the present study is to test the prediction that immersion students will improve in their use of grammatical gender, if provided with helpful opportunities to notice the co-occurrence of particular noun endings with appropriate gender attribution. The study aims to effect change in learners' rule-based representational system by providing learners with opportunities for consciousness-enhanced processing via FFI, including opportunities for feedback appreciation induced to varying degrees through the provision of different types of feedback (prompts and recasts), which are hypothesized to trigger different degrees of cognitive processing and awareness. The study's research questions are thus formulated as follows:

1. Will FFI improve French immersion students' ability to accurately assign grammatical gender?
2. Is FFI more effective with feedback than without feedback?
3. Which type of feedback is more effective in FFI—recasts or prompts?

With respect to question 1, it is predicted that immersion students participating in form-focused activities will outperform students whose attention is not drawn to grammatical gender through such activities. This is because, without FFI, immersion students are left to their own devices to process grammatical-gender patterns, which may occur frequently in classroom input but without salience (in the absence of FFI) and are thus not easily noticeable—especially not in a sufficiently robust way that might make the rules accessible in subsequent production.

With respect to question 2, it is predicted that immersion students receiving feedback in addition to FFI will outperform immersion students receiving FFI without feedback. This prediction is based on the premise that feedback can be effective at drawing learners' attention to mismatches (Ellis, 1994; Schmidt & Frota, 1986) and on Skehan's (1998) claim that feedback appreciation effectively contributes to changes in the learner's rule-based representations.

It remains difficult to predict outcomes pertaining to question 3 because, if the FFI is effective, then it seems reasonable to expect that students will be sufficiently primed to notice and to process either type of feedback. However, a slight advantage for learners receiving prompts is predicted, for three reasons. First, prompts will provide more opportunity for output practice and thus for proceduralization of newly acquired rule-based representations. Second, prompts are more likely than recasts to draw learners' attention to feedback, thus creating conditions for conscious awareness, which adds efficiency to working memory operations (Schmidt, 1990) and predisposes the learner toward a rule-based perspective that is more likely to effect change in long-term memory (Skehan, 1998). Third, according to Carroll (1989), the processing demands necessitated by accurate gender attribution during a learner's online production involve the coordinated retrieval of previously disconnected elements: a noun plus its gender attribution, manifest in a separately stored determiner but also apparent in the noun's internal structural properties (as highlighted by the FFI). One could thus expect learners with more opportunities to process these elements in a productive mode in response to prompts to benefit more than students provided with opportunities to process (or not) the same information in a receptive mode via recasting. This prediction is in keeping with Skehan's argument that interlanguage change is unlikely to arise in a receptive mode and then later become available to production (see also Swain, 1985).

Given the binary nature of grammatical gender in French, it could be argued that the learner's retrieval of more targetlike representations of grammatical gender does not require much conscious awareness, because it involves a simple on-the-spot computation (i.e., if un is wrong, then it must be une).¹

MEASURES

There were four tests used to assess the students' ability to assign grammatical gender: two written tasks and two oral tasks.