of the Psycholinguistic Rationale of the Output Hypothesis Shinichi Izumi Sophia University The output hypothesis claims that production makes the learner move from `semantic processing' prevalent in comprehension to more `syntactic processing' that is necessary for second language development. The purpose of this article is to illuminate the psycholinguistic mechanisms that underlie this claim by reviewing previous literature in language acquisition and cognitive psychology on the comprehension and production processes in language use and language learning. In speech comprehension, the interactive and compensatory nature of the human comprehension system can both promote comprehension and hinder language development for second language learners, unless the learners are somehow pushed to attend to formmeaning connections during input processing. In elucidating the mechanisms by which output promotes SLA, it is argued, by drawing on Levelt's (1989, 1992, 1993) speech production model, that the processes of grammatical encoding during production and monitoring to check the matching of the communicative intention and the output enable language learners to assess the possibilities and limitations of their inter- language capability. This may, under certain conditions, serve as an internal priming device for consciousness raising for the learners, which in turn creates an optimal condition for language learning to take place. It is argued that understanding of the constraints and potentials for learning created by input and output processing is crucial for devising pedagogical tasks that eectively promote interlanguage development. In the teaching of second/foreign languages (L2) all over the world, producing the target language (TL), or output, has long been considered as forming an important part of language learning. Such a favourable view of output may be reected in advice which may be commonly heard in conversations between a language teacher and his/her students, such as, `you have to use the language if you want to become good at it', or `speak more actively in class and outside if you want to improve your English'. However, precisely what these `words of wisdom' may mean and how benecial it is to produce output are often left quite vague. This paper presents an attempt to grapple with these challenging yet important questions for both second language acquisition (SLA) theory and L2 pedagogy. Specically, it reviews relevant literature in language acquisition and cognitive psychology on the comprehension and production Applied Linguistics 24/2: 168196 # Oxford University Press 2003 processes in language use and learning with the ultimate aim of advancing our understanding of the output hypothesis in SLA. THE ROLE OF OUTPUT IN SECOND LANGUAGE ACQUISITIONSWAIN'S OUTPUT HYPOTHESIS First, we start with a general background of the output hypothesis. In the SLA literature, it has often been assumed that output is only a sign of SLA that has already taken place and that it does not serve any signicant function in language acquisition processes (e.g. Krashen 1985, 1989). However, such a limited view of output has been questioned since the publication of Swain's seminal article in which the `output hypothesis' was rst proposed (Swain 1985). The output hypothesis postulates active roles played by output in the overall SLA processes. It was formulated essentially in reaction to Krashen's claim about the major role of `comprehensible input' in SLA and is based on many years of research on Canadian immersion programmes. The immersion programmes, which aim at the achievement of both academic and L2 learning through an integration of language teaching and content teaching, generally have great success in many areas of the students' language development (e.g. listening comprehension, uency, functional abilities, condence in using the L2); however, these learners have also been found to have problems in some aspects of the TL grammar, especially in morpho-syntactic areas, even after many years in these programmes (Harley and Swain 1984; Harley 1986, 1992; Swain 1985). Swain (1985) argued that one of the important reasons for this is that these learners engage in too little language production, which prevents them from going beyond a functional level of L2 prociency. Immersion students, Swain (1985) argues, lack output opportunities in two ways: First, the students are simply not givenespecially in later grades adequate opportunities to use the target language in the classroom context. Second, they are not being `pushed' in their output. That is to say, the immersion students have developed, in the early grades, strategies for getting their meaning across which are adequate for the situation they nd themselves in: they are understood by their teachers and peers. There appears to be little social or cognitive pressure to produce language that reects more appropriately or precisely their intended meaning: there is no push to be more comprehensible than they already are (Swain 1985: 249). Observational studies of interaction in French immersion classrooms have indicated that immersion classes are largely teacher-centred and that students are not required to give extended answers (Allen et al. 1990). This permits students to operate successfully with their incomplete knowledge of the language; communication between students and between the teacher and students is quite satisfactory in spite of numerous errors in the students' speech. Observations such as these have led Swain to conclude that SHINICHI IZUMI 169 comprehensible input, while invaluable to the acquisition process, is not sucient for these students to fully develop their L2 prociency. What these students need, Swain argued, is not only comprehensible input, but `comprehensible output' if they are to improve both uency and accuracy in their interlanguage (IL). The construct of comprehensible output posits that when learners experience communication diculties, they will be pushed into making their output more precise, coherent, and appropriate, and this process is said to contribute to language learning. In general terms, the importance of output in learning may be construed in terms of the learners' active deployment of their cognitive resources. That is, the output requirement presents learners with unique opportunities for processing language that may not be decisively necessary for comprehension. As Swain states, [i]n speaking or writing, learners can `stretch' their interlanguage to meet communicative goals. They might work towards solving their linguistic limitations by using their own internalized knowledge, or by cueing themselves to listen for a solution in future input. Learners (as well as native speakers, of course) can fake it, so to speak, in comprehension, but they cannot do so in the same way in production. . . . [T]o produce, learners need to do something; they need to create linguistic form and meaning and in so doing, discover what they can and cannot do (Swain 1995: 127). Thus, it is claimed that producing the TL may serve as `the trigger that forces the learner to pay attention to the means of expression needed in order to successfully convey his or her own intended meaning' (Swain 1985: 249). Since the output hypothesis was rst proposed, Swain has rened her hypothesis and specied the following four functions of output (Swain 1993, 1995, 1998). First, output provides opportunities for developing automaticity in language use. This is the uency function. In order to develop speedy access to extant L2 knowledge for uent productive performance, learners need opportunities to use their knowledge in meaningful contexts, and this naturally requires output. The second function of output is a hypothesis-testing function. Producing output is one way of testing one's hypotheses about the TL. Learners can judge the comprehensibility and linguistic well-formedness of their IL utterances against feedback obtained from their interlocutors. Third, output has a metalinguistic function. It is claimed that `as learners reect upon their own TL use, their output serves a metalinguistic function, enabling them to control and internalize linguistic knowledge' (Swain 1995: 126). In other words, output processes enable learners not only to reveal their hypotheses, but also to reect on them using language. Reection on language may deepen the learners' awareness of forms, rules, and form function relationships if the context of production is communicative in nature. Finally, output serves a noticing/triggering (or consciousness-raising) function. Namely, in producing the TL `learners may notice a gap between 170 COMPREHENSION AND PRODUCTION PROCESSES what they want to say and what they can say, leading them to recognize what they do not know, or know only partially' (Swain 1995: 1256). The recognition of problems may then prompt the learners to attend to the relevant information in the input, which will trigger their IL development. In sum, Swain's output hypothesis claims that output can, under certain conditions, promote language acquisition by allowing learners to try out and stretch their IL capabilities. In so doing, learners may recognize problems in their IL through internal feedbackoutput promotes syntactic processing and self-monitoringor external feedbackoutput invites feedback from inter- locutors, teachers, etc. This recognition may prompt the learners to generate alternatives by searching existing knowledge or to seek out relevant input with more focused attention and with more clearly identied communicative needs (cf. Swain and Lapkin 1995). IN SEARCH OF THE PSYCHOLINGUISTIC RATIONALE OF THE OUTPUT HYPOTHESIS Swain's output hypothesis is now widely recognized as an important extension of approaches that consider input as the only crucial aspect of SLA. As such, it has generated some empirical research into the roles of output in SLA. These studies have reported positive and promising, though not unconditional (see below for discussion), ndings for the specic functions of output: for the uency function (e.g. Bygate 2001; DeKeyser 1997), the hypothesis-testing function (e.g. Ellis and He 1999; Nobuyoshi and Ellis 1993; Pica 1988; Pica et al. 1989; Shehadeh 1999, 2001), the metalinguistic function (e.g. Kowal and Swain 1994; LaPierre 1994; Swain 1995, 1998; Swain and Lapkin 2001), and the noticing function (Izumi 2000, 2002; Izumi and Bigelow 2000, 2001; Izumi et al. 1999; Swain and Lapkin 1995). Despite the recent increase in empirical investigation of output, however, what has been scarce is a discussion of the psycholinguistic basis of the output hypothesis (for notable exceptions, see Bygate 2001; de Bot 1996). What, for example, is the psycholinguistic mechanism underlying the output hypothesis? What, in cognitive terms, is unique in output production that may be lacking in input comprehension and that is relevant for SLA? How are dierent functions of output related to each other? Greater explication of these questions will be necessary for both further advancement of the theoretical construct of output and input processing in SLA and for psycholinguistically guided applications of the output hypothesis in L2 pedagogy. To address these issues, this paper will focus on the general processes and mechanisms of comprehension and production and their relevance to language learning (for a review of empirical studies on the output hypothesis, readers are referred to Izumi 2000 and Shehadeh 2002). In what follows, an integrated model of SLA will be presented rst in order to gain an overview of SLA processes in which the contribution of output to language learning may be properly situated. Then, general characteristics of SHINICHI IZUMI 171 comprehension processes will be described, followed by a discussion of L2 input processing. The question tackled here is why it is said that input comprehension is not sucient to develop one's IL competence. In the ensuing section, a speech production model will be examined in order to gain insight into output processes. The focus here will be on how output may be related to language learning. Finally, some factors that are likely to constrain the acquisitional eect of output will be discussed. GASS'S INTEGRATED MODEL OF SLA In examining the psycholinguistic rationale of the output hypothesis, it is useful, rst of all, to have a general learning model that captures the overall process of how learners derive their L2 grammatical knowledge in SLA. One such model is proposed by Gass (1988, 1997; Gass and Selinker 1993), which is schematically represented in Figure 1. Among other similar models of SLA (e.g. Chaudron 1985; Ellis 1990, 1993; Frch and Kasper 1986; Sharwood Smith 1986; VanPatten 1995, 1996), Gass's model is selected here because it provides a detailed (though not necessarily denitive) description of each component stage and depicts the interrelated and dynamic processes of language acquisition. The model proposes ve stages whereby the learner converts input to output: apperceived input, comprehended input, intake, integration, and output. Gass claims that what learners must do rst with ambient input is to perceive it in light of their past experiences and currently held knowledge. This so-called apperception serves as a priming device, so that later analysis of the input can be conducted. Apperception, Gass claims, relates to the `potentiality of comprehension of the input' (1997: 4). As such, it may be seen as the rst hurdle where the ambient input is ltered for an initial selective processing, capturing the fact that not all of input is automatically used for comprehension, let alone for intake or integration. The input that is apperceived is processed to derive some form of meaning representation, or what is referred to as comprehended input in Gass's model. Gass argues that comprehension represents a continuum of possibilities ranging from semantic analyses to detailed structural analyses. One important factor that determines whether input converts to intake is the level of analysis of the input that the learner achieves. It is claimed that analysis at the level of meaning is not as useful for intake as an analysis made at the level of syntax. What is comprehended, then, can feed into the intake component. Alternatively, it may not be used for any further grammatical analysis if the learner discards the information after using it for the purpose of immediate communication (cf. Sharwood Smith 1986, and Frch and Kasper 1986). If input becomes intake, the intake data may be used for the formation of new IL hypotheses. The hypotheses thus formed are subject to testing upon further exposure to input. If the input data conrm an existing hypothesis, it will facilitate the 172 COMPREHENSION AND PRODUCTION PROCESSES integration of new knowledge into the developing system. If the information contained in the input is already part of one's knowledge base, the intake data may be used for hypothesis re-conrmation or rule strengthening. If the hypothesis is disconrmed by the input data, it will be rejected and will no longer be relevant for grammar formation, and learners will have to seek more input to derive further intake. The intake that is thus integrated causes restructuring in the IL grammar, which is a reorganization of the learner's internal knowledge system. Alternatively, the intake data may be stored as unanalysed or partially analysed items which may be re-analysed when more relevant input becomes available. Finally, there is an output component. Gass sees output as playing an active role in the dynamic, interrelated acquisition processes. Following Swain (1985), Gass stresses the importance of comprehensible output in testing hypotheses. This creates a feedback loop from output into the intake component, where hypothesis formation and testing is considered to take place. The output component is also related to the levels of analysis made at the stage of comprehended input. It is claimed that learners cannot rely on external cues and general world knowledge in production in the same way they do in comprehension and that they would need greater syntactic processing in production. Language production is thus seen as one important means of moving the learner from comprehended input to intake. SHINICHI IZUMI 173 Figure 1: A model of SLA (from Gass 1988: 200) In sum, SLA involves overlapping, yet distinguishable sets of processes. First, exposed to the ambient input, learners perceive selected aspects of the input, from which they derive some form of meaning representations of the input messages. Comprehension and intake are considered to represent dierent processes, of which only the latter is used for further processing for learning. Through the processes of hypothesis formation, testing, modica- tion, conrmation, and rejection, the intake may subsequently be integrated into the developing system. Finally, learners selectively use their developing system in their output. The output process is seen here not only as a product of acquisition, but represents an active component in the overall acquisition processes. SPEECH COMPREHENSION PROCESSES General characteristics Gass's SLA model distinguishes, among other things, comprehension and intake in SLA processes. Why is it that input that is used for comprehension is not directly related to intake? In order to answer this question, we need to address rst the kind of information that is utilized in human speech comprehension. While debate still continues as to the autonomy of syntactic modules from semantic and pragmatic modules, 1 psycholinguistic research over the past decades has accumulated enough evidence to suggest some general characteristics of human speech comprehension processes (see Frch and Kasper 1986; Fender 2001; Garrett 1991; Harrington 2001; Rost 1990; Scovel 1998; Tyler and Tyler 1990, ch. 5; Wingeld 1993, for detailed discussion). These characteristics include the following: . Comprehension is not the passive recording of whatever is heard or seen. . Comprehension processes rely on three types of information: linguistic input, contextual information, and the recipient's linguistic and other general knowledge of the world, including semantic and pragmatic knowledge. . Comprehension is dierentially aected by the linguistic devices used in the sentence (e.g. passive vs. active sentences). The use of linguistic cues (be they syntactic, semantic, morphological, or phonological/orthographi- cal) in comprehension processes is often referred to as bottom-up processing. . Comprehension is dierentially aected by the existence, type, and the amount of contextual clues provided. People tend to seek contextual consistency in comprehending speech. . Comprehension is dierentially aected by the general world knowledge possessed by the recipients and can dier among individuals depending on the amount of such knowledge available for each individual. The use of contextual clues and world knowledge in comprehension processes is referred to as top-down processing. . While there is a possibility that syntactic parsing operations are conducted 174 COMPREHENSION AND PRODUCTION PROCESSES initially before semantic-pragmatic factors come into play (see e.g. Forster 1979), people have a strong tendency to seek semantic plausibility in comprehending speech. Such a semantic bias holds particularly strongly when comprehension of units larger than a sentence (i.e. short or long discourse and texts) is considered. . Comprehension is selective because humans possess limited processing capacities. This selection process is guided by a number of factors, which include, for bottom-up processing, salience of input elements (which may be dened in acoustic/visual or semantic terms), and, for top-down processing, the recipient's expectations (which are inuenced by the contextual clues present in the input and/or the recipient's general knowledge of the world). These characteristics of the human speech comprehension system suggest that highly complex processes underlie how people make sense of the language data presented to them. People do not rely on only one general knowledge source or strategy to understand speech, such as syntactic parsing or applications of semantic and pragmatic knowledge. Instead, they utilize various resources available to them, using both top-down and bottom-up approaches, to arrive at a comprehension of the input messages. Speech comprehension processes for language learners While the above observations are insightful for understanding how human language comprehension takes place, it is important to note that these observations are based on extensive studies of adult language users with already highly established sets of knowledge of the language being used. Therefore, they may not be directly applicable to children acquiring their rst language (L1) or to child or adult L2 learners. It is possible for these populations that the two approaches of bottom-up and top-down processing are not equally and/or as eectively utilized for comprehension (Fender 2001; Pienemann 1998). In fact, some researchers argue that even adult L1 listeners or readers do not utilize the two general approaches of syntactic and semantic processing equally in comprehending speech. Clark and Clark (1977), for example, argue that syntactic information may be circumvented in comprehension processes in listening and reading. Listeners know a lot about what a speaker is going to say. They can make shrewd guesses from what has been said and from the situation being described. They can also be condent that the speaker will make sense, be relevant, provide given and new information appropriately, and in general be cooperative. Listeners almost certainly use this sort of information to select among alternative parses of a sentence, to anticipate words and phrases, and sometimes even to circumvent syntactic analyses altogether (Clark and Clark 1977: 72). In accounting for the use of semantic knowledge in comprehension processes, SHINICHI IZUMI 175 Clark and Clark posit the `reality principle,' according to which listeners interpret sentences in the belief that what the speaker is saying makes sense to them. A primary strategy under this principle is: `Using content words alone, build propositions that make sense and parse the sentence into constituents accordingly' (1977: 73). An example of this strategy in use is illustrated by the ways people paraphrase complex sentences such as the following: (1) The vase that the maid that the agency hired dropped broke on the oor. (2) The dog that the cat that the girl fought scolded approached the colt. Sentence (1) is highly constrained semantically. By using the content words alone, one can reach an accurate interpretation of the sentence. Vase, maid, agency, hired, dropped, and broke on the oor can easily be sorted into three reasonable propositions: the vase broke on the oor; the maid dropped the vase; and the agency hired the maid. This is not true for sentence (2). Dog, cat, girl can all do any of the actions, ght, scold, and approach colts. It is reported that people correctly paraphrased sentence (1) more often than sentence (2), suggesting that people rely on semantic knowledge in interpreting dicult sentences (Stolz 1967, as cited in Clark and Clark 1977: 73). In the realm of reading research, Stanovich (1980) claims that interactive models of reading can provide a more accurate account of reading performance than do strictly bottom-up or top-down models. The reader is seen not merely as a passive recipient of the printed information, but as an active subject in the whole process who utilizes all the knowledge resources available to him/her at a given point in time. What is particularly interesting about Stanovich's model of reading is not just the interactive nature of the reading processes, but its proposal of compensatory mechanisms. If there is a deciency in any particular process (e.g. weak syntactic knowledge), other processes (e.g. higher-order knowledge structures, such as contextual or general world information that the reader has access to) can compensate for the weak knowledge source. Thus, with information provided simultaneously from several knowledge sources, `a decit in any knowledge results in a heavier reliance on other knowledge sources, regardless of their level in the processing hierarchy' (Stanovich 1980: 63). This interactive-compensatory model is seen as an `integrative' model of reading, as it can provide a successful theoretical account of seemingly conicting ndings of many research studies in this area (e.g. studies showing dierential contextual eects of good and poor readers). In L1 acquisition literature, it has been claimed that children typically rely on general world knowledge to comprehend what is uttered to them (Clark and Hecht 1983). They rely on, for example, their general knowledge about the instigators of actions which are typically animate, probable relations between nouns in a sentence (e.g. The baby was fed by the girl is interpreted correctly with the simple knowledge about the relationship between adults and babies without necessarily having the knowledge of the passive 176 COMPREHENSION AND PRODUCTION PROCESSES construction), and the knowledge of the usual routine in particular circumstances to decide how to act. As a result of such strategies, children often appear to understand more language than they actually do. In the case of SLA as well, the restricted L2 knowledge of the learners makes them rely on certain strategies (e.g. use of semantic and contextual cues) more than others (e.g. syntactic cues) in order to overcome their linguistic limitations. Skehan (1996, 1998), for example, argues that L2 learners use a variety of strategies of comprehension that may obviate careful attention to form. There is natural and unavoidable use of strategies of comprehension . . . , in that non-deterministic and non-exhaustive methods are used to recover intended meaning, with the success of this operation often being dependent on only partial use of form as a clue to meaning. . . . In other words, processing language to extract meaning does not guarantee automatic sensitivity to form and the consequent pressures for interlanguage development (Skehan, 1996: 401). Furthermore, Skehan draws attention to the fact that L2 learners are those who have `schematic knowledge' (i.e. factual and sociocultural background knowledge and discoursal procedural knowledge), but have limited `systemic knowledge' (i.e. syntactic, semantic, and morphological knowledgecf. Anderson and Lynch 1988). Such learners may be likely to exploit their schematic knowledge to overcome limitations in their systemic knowledge. This can lead to a reduced chance for the engagement of the IL system. In general, Skehan's claim seems to be supported by the results of previous research which indicate that comprehensible input does not always guarantee learners' grammatical development (see Ellis 1994; Larsen-Freeman and Long 1991; Long 1996, for reviews). In a recent study, Tyler (2001) investigated whether non-native listeners rely more on topic knowledge to aid their speech comprehension than do native listeners. Using a dual-task technique in which listeners had to comprehend an auditory passage while concurrently verifying the totals of single-digit calculations, Tyler found that while access to the topic of the passage had a small eect on the adult native listeners' calculations, it had a large eect on the performance of the adult non-native listeners. That is, non- native listeners had much greater diculty with calculations in the non-topic condition than did native listeners, even though they performed equally as well on the task as the native listeners in the topic condition. It seems that eective use of topic knowledge helps the learners to function eectively in everyday situations in the L2, while it may at the same time inhibit further development of their linguistic knowledge. In some SLA studies, researchers divided linguistic knowledge into dierent sub-components and investigated how each of these sub-components aects comprehension. Mecartty (2000), for example, examined the relationship between lexical and grammatical knowledge to reading and listening SHINICHI IZUMI 177 comprehension by adult L2 learners and found that while both types of knowledge are signicantly related to comprehension, only lexical knowledge explained the variance in both reading and listening comprehension. This suggests that there is an imbalance in the use of dierent sub-components of knowledge sources even within the linguistic knowledge itself, which in turn implies that development of dierent sub-components may be stimulated dierently during comprehension processes. To summarize, although the resourceful nature of the comprehension system is highly useful in making comprehension of sentences containing yet- to-be acquired items possible, this also implies that L2 learners can attain an adequate level of comprehension without necessarily focusing on many formal features in the input. This can lead to a reduction in the amount of intake that can be used for nal integration in the developing system. Characteristics of L2 input processing The preceding discussion reveals a complex interplay among various factors in comprehension processes and suggests that language learners may not use all these factors eectively and equally. What elements in the input, then, do L2 learners focus on as they process input? Is there any bias as to what they process and what they do not process in the input? Answers to these questions are proposed by VanPatten (1995, 1996), who has formulated a model of L2 input processing. Crucial to VanPatten's model of input processing is the assumption that humans possess limited processing capacities. That is, it is held that learners are not capable of attending to all the information in the input; only some of it becomes the object of focused or selective attention, while other information is processed only peripherally (cf. apperception in Gass's SLA model; see McLaughlin 1987; McLaughlin et al. 1983, for similar information-processing views; see also Robinson 1995, for a discussion of alternative views of attention). VanPatten assumes, as do many other researchers (e.g. Gass 1988; Robinson 1995; Schmidt 1990, 1995, 2001; Slobin 1985; Tomlin and Villa 1994), that attention is a prerequisite for learning to take place. He argues, however, that learners' attention tends to be drawn to certain parts of the input, particularly those that are immediately relevant to the message content. Operating with limited processing capacities, L2 learners rst search the input for content words. If resources are not depleted at this point, they may try to make formmeaning mappings by attending to grammatical forms with `high communicative value'. If resources are still not depleted, then further processing of `less communicative value' can occur. Communicative value is dened here as `the relative contribution a form makes to the referential meaning of an utterance . . . based on the presence or absence of two features: inherent semantic value and redundancy within the sentence-utterance' (VanPatten 1996: 24). A form that has inherent semantic value and is not redundant will tend to have high communicative value (e.g. progressive 178 COMPREHENSION AND PRODUCTION PROCESSES morphology, -ing, in English), whereas a form that lacks (or is light in) inherent semantic value and is redundant tends to have low communicative value (e.g. third person present singular morphology, -s, in English). VanPatten claims that forms with low communicative value are made processable by the learners only when their L2 capacities develop to such an extent that their attentional resources are not consumed by the processing of forms with high communicative value. It is not clear, however, whether the learners really attend to less meaningful items when they can spare their attentional resources. It is possible that they may never attend to purely formal, functionally redundant forms unless some form of instructional intervention forces them to do so (see arguments for focus on form by Long 1991; Long and Robinson 1998; Doughty and Williams 1998). Apart from the processing of content and grammatical items in the input, learners also need to assign semantic or grammatical roles to the words they hear or read. Based on research ndings in both L1 and L2 acquisition, VanPatten suggests that the rst noun strategy is a prevalent, possibly a universal, strategy utilized by language learners. This strategy dictates that the rst NP encountered is generally labelled as the agent, while the second NP is assigned the role of patient. This strategy, however, may be overridden if other factors such as lexical semantics and event probabilities are in strong opposition to it. As their L2 competence develops, learners may learn to rely more on grammatically related cues such as morphological markings and syntactic structures. However, this process is generally gradual and slow, particularly if the learners are exposed to sentences containing cues that are in harmony with each other, as opposed to those in conict (cf. Bates and MacWhinney 1989; Gass 1987; Harrington 1987; MacWhinney 1987; Sasaki 1994). In sum, in VanPatten's model of input processing, certain principles are believed to guide the ways in which learners process grammatical form in their attempt to comprehend input strings. These processing principles, in turn, shape the intake data available for accommodation by the learners' developing system. Driven to get the meaning out of the input, learners rst attend to meaningful elements in the input, follow the rst noun strategy as a general strategy for parsing input sentences, and rely on their semantic and pragmatic knowledge to compensate for the lack of sophisticated syntactic parsing mechanisms in the L2. Although VanPatten's model is still in need of more empirical substantiation and accommodation of other factors that are also likely to aect the acquisition of dierent language forms (e.g. semantic complexity, rule complexity, and frequency: cf. Goldschneider and DeKeyser 2001), it does seem to capture some important insights that need to be incorporated into any theories of L2 input processing. In terms of pedagogical applications, VanPatten and his colleagues have developed, based on the understanding of input processing, a pedagogical technique known as `processing instruction'. Processing instruction aims to facilitate better intake from the input by manipulating task demands in such a SHINICHI IZUMI 179 way that the use of the default strategies would not be the best way to go about completing the given task. It generally involves three stages: (1) explaining to the learners the relationship between the given form and the meaning it conveys; (2) providing them with information about good and poor processing strategies; and most importantly, (3) providing learners with `structured input' activities which encourage them, in controlled situations, to pay attention to the relevant grammatical cues so that they can form better formmeaning connections. The positive eects of processing instruction over a more traditional instruction that focused on grammar explanation and output practice are reported in a series of studies conducted by VanPatten and his colleagues (see VanPatten 1996, for a review of relevant research). VanPatten and Oikkenon's (1996) study, in particular, showed that the provision of structured input activities alone was as eective as the regular processing instruction that included explanation components. This suggests that the structured input activities and the formmeaning connections made during the activities are responsible for the positive eects of processing instruction. While we need to be cautious about extrapolating the advantage of processing instruction over other types of instruction and about generalizing the results to all language structures, 2 it is important to ask what makes processing instruction eective at least for the acquisition of the morpho- syntactic structures that have been investigated so far. It seems that the eectiveness of processing instruction lies essentially in the fact that it pushes the learners to attend to crucial formmeaning relationships in comprehend- ing input. In other words, processing instruction is eective because it creates a `pushed input' condition. On further thinking, one may wonder what `pushed output' can do to enhance learning (Swain 1985). If, for example, learners are pushed to produce output and immediately provided with relevant usable input, it is possible that the sensitivity towards the form may be heightened through the production process, which may, in turn, prompt them to attend to formmeaning relationships. This may bring about a shift in processing strategies from meaning-oriented towards a more syntactically sensitive one. In light of the predictions made by the output hypothesis as discussed earlier, it can be posited that output has the potential for altering the manner in which learners process input. How does this occur in psycho- linguistic terms? What are the cognitive mechanisms involved? It is to these topics that we now turn. SPEECH PRODUCTION PROCESSES Levelt's speech production modela general sketch Of several psycholinguistic models of speech production proposed in the literature, the most inuential is the one developed by Levelt (1989, 1992, 1993; Levelt et al. 1999). Levelt's production model, originally developed to 180 COMPREHENSION AND PRODUCTION PROCESSES account for the speech production by L1 adults, is based on decades of psycholinguistic research and is supported by considerable empirical research, both experimental and observational. The model has also been adapted to account for L2 data (Bygate 2001; de Bot 1992; de Bot et al. 1997; Do rnyei and Kormos 1998; Kormos 1999). A brief sketch of the model is provided below. The relevance of this model to SLA will be discussed subsequently. In Levelt's production model, there are ve distinct components: the conceptualizer, the formulator, the articulator, the audition, and the speech comprehension system; and three sources of knowledge: lemmas and forms contained in the lexicon and discourse model, situation and encyclopedic knowledge that is connected to the conceptualizer (see Figure 2). A message to be conveyed is rst generated in the conceptualizer, which produces a preverbal message as its output. The formulator takes the preverbal message as its input and converts it into a phonetic plan. The lexicon, which SHINICHI IZUMI 181 Figure 2: Levelt's speech production model (from Levelt 1989, MIT Press) feeds into the formulator, provides necessary information in this conversion process and consists of two parts: the lemma, which contains semantic and syntactic information of lexical items, and the form (or lexeme), which represents morphological and phonological specications. Using these two types of information in the lexicon, the formulator generates a phonetic/ articulatory plan in two steps. First, grammatical encoding of the message takes place by rst accessing lemmas through a process of matching the meaning of the preverbal message with the semantic specications provided in the lemma. The activation of a specic lemma makes available the syntactic information relevant to it, which activates syntactic building procedures. With the use of the syntactic specications provided by the selected lemma, the grammatical encoder produces the surface structurean ordered string of lemmas grouped in phrases and sub-phrases. As a second step, the phonological encoding takes place by accessing morpho-phonological infor- mation stored in the lexeme, which produces a specic phonetic plan (or internal speech). The phonetic plan is internally scanned by the speaker via the speech-comprehension system. Then the articulator takes the phonetic plan as its input and converts it into actual speech. At this point, the speech-comprehension system connected to the auditory system plays a feedback role: the overt speech is guided through the audition into the speech-comprehension system to check for any anomalous output. The speech-comprehension system, having access to both the form and lemma information in the lexicon, recognizes words, retrieves their meanings and parses the incoming speech. The output of the speech-comprehension system is parsed speech, which is a representation of the input speech in terms of its phonological, morphological, syntactic, and semantic composition. The main work of monitoring is done by the conceptualizer, which attends to the output of the speech-comprehension system. The monitoring is done both covertly prior to articulation or overtly subsequent to articulation. As this brief description of Levelt's model indicates, his model of production is heavily lexically driven. Lexical selection is considered to drive grammatical encoding. As lemmas are retrieved when their semantic conditions are met in the message, they activate syntactic procedures that correspond to their syntactic specications. Thus, a verb will instigate the construction of a verb phrase, a noun the construction of a noun phrase, etc. Lemmas may be conceptually driven or grammatically driven. The latter type of lexical items belongs to the closed class vocabulary. For instance, in the phrase the woman that arrived, the retrieval of the relative pronoun that is not semantically driven such as is the retrieval of woman. Rather, that is called by the syntactic procedure that constructs relative clauses. Here, in other words, `grammatical encoding drives lexical selection' (Levelt 1992: 6). The preverbal message, in this case, dictates that woman here cannot be any woman, but a particular woman that arrived. The NP categorical procedure with woman as head looks for modifying information attached to the concept `woman' in the preverbal 182 COMPREHENSION AND PRODUCTION PROCESSES message. As the preverbal message contains this information, it calls for further specication of which woman this woman refers to in the grammatical encoder. Such phrasal specication is made possible by the close coordination between the conceptualizer and the formulator. Relevance of Levelt's model to the output hypothesis Production processing is not language learning. It is a process in which a concept is encoded in a speech form that is to be communicated. This, as we saw, involves conceptualizing, formulating, articulating, and monitoring, the process of which may recycle depending on the success of the outcome of the processing. Although Levelt's production model is a `steady-state' model and is not intended to account for language learning per se, it can nevertheless provide some important insights into how learning may be brought about through production processes (de Bot 1992, 1996; Kormos 1999). Levelt's model illustrates the putative process in which the grammatical encoder syntacticizes the preverbal message using the syntactic specications provided in the retrieved lemma in order to derive a surface structure of the message. The surface structure is then processed in the phonological encoder for exact form specications, which is then sent to the articulator to derive overt speech. The grammatical encoding in this process, in particular, requires a focus on syntactic form on the part of the language producer. Although essentially the reverse process is believed to take place in the speech comprehension system for any incoming language input (i.e. grammatical decoding), the additional knowledge source stored in the discourse models and situational and encyclopedic knowledge can often compensate the lack of L2 knowledge in decoding the input data. The grammatical decoding, therefore, may eectively be bypassed in the course of input comprehension, as we have seen earlier. In production, on the other hand, the speaker is responsible for message generation and formulation that requires grammatical encoding. There is much less chance (though by no means no chance, as we will see below) for the speaker to escape syntactic operations in the course of production. It is in this sense that output is said to force the learner to move from `the semantic processing prevalent in comprehension to the syntactic processing needed for production' (Swain and Lapkin 1995: 375). In Levelt's model, it is assumed that grammatical encoding in production by adult native speakers occurs subconsciously and automatically. 3 However, this may not be the case for language learners who are still in the process of learning a language and whose language use requires a great deal of controlled processing and attention (Do rnyei and Kormos 1998; Kormos 1999, 2000). It is possible that the very process of grammatical encoding in production sensitizes the learners to the possibilities and limitations of what they can or cannot express in the TL. Such sensitization is bolstered by the feedback system available for monitoring speech. In Levelt's model, both SHINICHI IZUMI 183 internal and overt speech are fed into the speech-comprehension system and back to the conceptualizer to be monitored for matching between the semantic specications in the preverbal message and the outcome of the formulation and articulation. This monitoring mechanism allows for attention to be given to the well-formedness and appropriateness of the production outcome (Do rnyei and Kormos 1998; Kormos 1999, 2000). These processes particularly, grammatical encoding and monitoringcan, under certain circumstances, serve as an `internal priming device' for grammatical consciousness raising for the language learners. In L1 acquisition, some researchers contend that part of the task of language acquisition is to coordinate comprehension and production (Clark and Clark 1977; Clark and Hecht 1983). For example, it is observed that children at the telegraphic stage are more likely to respond to adult commands, such as throw me the ball, than child-like throw ball. This suggests that in comprehension children initially rely on more adult-like representations of words and phrases not yet reected in their own production. These representations, it is claimed, `provide a standard to which they will eventually match their own productions of those same linguistic units' (Clark and Hecht 1983: 338). This matching or coordination mechanism requires that children be able to monitor what they produce and check it against the standard which is their representation for comprehension. Indeed, children have been observed to monitor their own speech actively and to try to repair their utterances (Clark 1982). Furthermore, their repairs are generally made toward the adult norm, not away from it. Observations such as these suggest that children's active monitoring and detecting mismatches between what they understand and what they themselves produce may provide part of the impetus for language development (Clark 1982; Clark and Hecht 1983). This account underscores the importance of the inputoutput interactions in language acquisition processes. The mechanism of speech monitoring advocated here is consistent with Levelt's model of speech production where speech generated in the formulator is fed into the speech-comprehension system and then back to the conceptualizer for monitoring of output. In SLA, drawing on Levelt's production model and Anderson's (1982) information-processing approach to skill acquisition (also see Johnson 1996), de Bot (1992, 1996) proposes that, while output by itself does not create completely new declarative knowledge, it can facilitate the process of the transition of declarative knowledge to procedural knowledge: Specic information in the lemma activates certain procedures, and the system does not get error messages about the result of this connection; hence the strength of this connection increases. When this connection is made repeatedly, the activity becomes automated, and therefore more rapid and more precise. Probably, focused attention to specic production processes stimulates the development of connections in memory (de Bot 1992: 54950). 184 COMPREHENSION AND PRODUCTION PROCESSES Focusing on the feedback system described in Levelt's model, de Bot (1996) further notes that internal speech is matched against internal standards that are formed by the speaker's receptive knowledge about the use of specic rules. Operating under this matching mechanism, `if what is produced and what is correct do not match according to the internal norm, [internal] negative feedback will hamper the development of the connection' (1996: 549). Again, such a comparison is made possible via a feedback loop from internal speech to the speech comprehension system, as indicated in Levelt's model. The speech generated by the formulator is examined internally for both content and form. The overt speech is also fed back into the speech comprehension system for further checking of anomalous output. Such monitoring mechanisms are supported by various ndings in psycholinguistic research. Scovel (1998) neatly summarizes ndings of psycholinguistic research in this area: [c]ommunication is not a one-way broadcast of a signal, but it is an interactive process, involving not just the interaction between the interlocutors but also the interaction within each individual speaker. . . . Speech production (or written composition) is not a linear `one- way' process; it is a parallel, `two-way' system involving both output and the concurrent editing and modulation of that output (Scovel 1998: 49). Although de Bot (1992, 1996) allows for the possibility that new knowledge may be generated in the production process when the learner forms new words through the application of existing rules or the combination of previously acquired morphemes (cf. Swain and Lapkin 1995), he sees the main role of output in strengthening already-stored knowledge representa- tions, which would fall under the scope of the uency function of output. While this is an important role of output in SLA, the argument advanced here is that output has a wider role to play in the overall acquisition processes, the uency function being but one of its roles. In particular, given that the learners' existing L2 linguistic system is not likely to provide sucient information to enable the monitor to decide with certainty whether their output was anomalous or not, decision problems may be experienced in the monitoring process (Kormos 1999, 2000). Thus, even if the external speech passes through the comprehension system without any apparent warning given from the internal norm, the learners may still be left with uncertainty with the correctness of their speech. As argued above, the mechanisms of monitoring both internal speech and overt speech enable the speakers to assess the degree of success in the outcome of the formulation, or more specically, the matching of the message specications and the nal output. This monitoring process permits `the interaction within each individual speaker,' as Scovel puts it, and through this process learners may be prompted to recognize the hole or gap in their IL knowledge, which is an important step for language development (Swain 1998). SHINICHI IZUMI 185 When facing problems in their production process, learners have several alternative routes to take depending on the given situation at the time (and perhaps depending on the individual learners' idiosyncratic preferences as well). For example, despite the uncertainty, learners may try out the outcome because they do not have any other means available to express their communicative intention and/or they want to try it out and see whether it works (de Bot 1992). In interactive situations where communication is taking place with an immediate interlocutor, the learners may receive negative feedback from him/her and conrm, reject, or modify their hypothesis (i.e. the hypothesis-testing function of output: Ellis and He 1999; Nobuyoshi and Ellis 1993; Pica 1988; Pica et al. 1989; Shehadeh 1999, 2001). If an `authoritative' gure, such as a teacher or a native speaker (or even a dictionary or grammar book), is available, learners may ask him/her questions or consult with the available information sources in an eort to understand better how the TL works. Alternatively, in situations where external feedback is not immediately available, as in monologues or communication in writing, learners can resort to other means. If they are communicating amongst themselves, as in the collaborative task situations reported in Kowal and Swain (1994) (see also LaPierre 1994; Swain 1995, 1998; Swain and Lapkin 2001), specic problems encountered in the process of production may be brought to the forefront of the learners' attention and various solutions to the problems may be discussed. The elicitation of relevant input in the collabor- ative work may then trigger language learning (i.e. the metalinguistic function of output). If, on the other hand, the learner is left on his/her own to solve the immediate production diculties, as was tested in Swain and Lapkin (1995), he/she may engage in various thought processes that can consolidate existing knowledge or possibly generate some new knowledge on the basis of their current knowledge (see also Kormos 2000, for a discussion of learners' self- repair behaviours). If relevant input is immediately available, however, the heightened sense of problematicity during production may cause the learners to process the subsequent input with more focused attention; they may try to examine closely how the TL expresses the intention which they just had diculty expressing on their own (i.e. the noticing function of output: e.g. Izumi 2000, 2002; Izumi and Bigelow 2000, 2001; Izumi et al. 1999). For teachers who wish to take an active interventionist approach to help their students develop their L2 knowledge, a good intervention point is obviously when the learners' IL system is most open to change, and this is most likely to be found when the learners are grappling with the specic means of expression to convey their meaning. Output produced in meaningful contexts may create this potential `learning space', which can be lled in a timely manner by the teacher (Samuda 2001). In all cases, learning may be enhanced through the act of producing language, which, by its mechanisms, increases the likelihood that learners become sensitive to what they can and cannot say in the TL, leading to their reappraisal of their IL capabilities. 186 COMPREHENSION AND PRODUCTION PROCESSES These several functions of output are summarized in Figure 3. The processes that intervene between the rst output and the second output, which are depicted in the squares, are believed to constitute an important part of SLA (Swain and Lapkin 1995). To summarize, output, by itself, can contribute to learning by strengthening the IL knowledge base that may still be only weakly established, that is, solidifying the knowledge connections or increasing the automaticity of language use. Equally or perhaps more importantly (depend- ing on one's view of what `acquisition' entails; cf. Ellis 1999: ch. 10), output triggers chains of psycholinguistic processes that are conducive to language learning. In other words, output processing engages important internal procedures such as grammatical encoding and monitoring, which prompts the learners to interact actively with the external environment to nd a solution (e.g. attend selectively to certain aspects of the input) or to explore their internal resources for possible solutions. Output, thus, serves as a useful means to promote the interaction between learner internal factors (including selective attention and their developing L2 competence) and environmental factors (input, interaction, and pedagogical intervention), or the interaction within the learners themselves for internal metalinguistic reection. The outcome of all cases is language acquisition in a broad sense of the term, that SHINICHI IZUMI 187 Figure 3: Output and second language development (adapted from Swain and Lapkin 1995: 388) is, the development of the knowledge base, its restructuring, and the strengthening and increase in the access to the stored knowledge. 4 Relating the roles of output discussed here to the overall SLA processes that were discussed earlier, Figure 4 illustrates such relationships in terms of three arrows connecting output to other SLA components. Specically, the arrow intersecting at the point between the comprehended input and the intake is meant to imply that output generated through the production processes can help to mediate between comprehension and acquisition processes by facilitating noticing of the mismatches between the learners' IL output and the TL input. This function may be variously called, depending on the focus of the emphasis: intake facilitation (Terrell 1991), noticing or noticing of the gap (Schmidt 1990, 1995, 2001), or consciousness-raising (Rutherford and Sharwood Smith 1985; Sharwood Smith 1991). Opportunities for output can also serve as grounds for hypothesis testing and/or metalinguistic reection for the learners, which could lead to intake or integration upon receiving conrmation (here the arrow is connected with the point between the intake and integration components, assuming that the hypothesis being tested has already been taken in but still waits further conrmation for nal integration into the system). And nally, if the same structure is used in the output repeatedly, the output can also serve to promote more eective and faster access to the integrated knowledge by the learners, leading to 188 COMPREHENSION AND PRODUCTION PROCESSES Figure 4: Output as an active component in the overall SLA process automatization of the IL knowledge (as indicated by the arrow intersecting the integration and the output components). In this way, output can serve to promote the learners' intake of the form, its integration, and its speedy access for eective language use. Some factors affecting output effects on learning With these psycholinguistic mechanisms available to language learners, an important caveat needs to be mentioned. That is, not all circumstances of production may provide language learners with ideal grounds in which to encourage syntacticization and sensitization to language forms. In many ways, this is similar to the case of comprehension, which does not always guarantee automatic sensitivity to form, but instead requires some conditions for a focus on form to occur. Just as the availability of rich semantic, contextual, or situational information allows the learner to bypass careful syntactic analysis in comprehension, some production circumstances are not particularly conducive to inducing learners' sensitivity to form; hence, the need for `pushed' output to drive language development (Swain 1985, 1993, 1995, 1998). For instance, it is said that in `loose' conversational contexts, learners can avoid problematic lexical and grammatical structures, yet nevertheless achieve their immediate communicative goals (Bygate 1999; Gary and Gary 1981; Skehan 1998). This would be the case of learners using `reduction strategies,' in Frch and Kasper's (1983) terms. In general, the need for syntacticization would be diminished in situations where one can readily rely on external support such as interlocutors' scaolding or contextual cues available in the environment or through gestures (e.g. having a face-to-face conversation with a familiar interlocutor as opposed to writing to an unfamiliar recipient). In her longitudinal study of the development of the past-time marking in English by two Vietnamese speakers, Sato (1986) concludes that the `compensatory nature of discourse-pragmatics' facilitates learners' communicative performance, but it simultaneously makes the past- time morphological marking `expendable' (1986: 42). A form like the past tense marking, which has inherent semantic content yet is often redundant in many communication contexts may be particularly susceptible to dierences in contextual factors. Thus, both situational and linguistic variables can aect the degree to which production forces learners to allocate their attention to form features (Izumi and Bigelow 2000). Task demands also inuence what aspects of L2 performance (e.g. accuracy, uency, and complexity) may be most attended to by the learners (Bygate 1999; Foster and Skehan 1996; Skehan and Foster 1997, 1999; Skehan 1998). Some of the features of the task that are known to aect L2 performance are: availability of the planning time prior to task performance, specic goals and requirements set for the task, the task directions given to the learner, and the type, amount, and details of the linguistic, as well as non-linguistic (e.g. SHINICHI IZUMI 189 visual), information to be dealt with (cf. see Long in press; Crookes and Gass 1993a, b; Bygate et al. 2001, for many issues in task-based language teaching). All of these factors determine the overall cognitive demand of the task, which inuences the degree to which the learners allocate their attentional resources to form. From an information-processing perspective, the limited capacity of the human attentional system is likely to aect the eciency of the encoding and monitoring processes in production. Such eects, moreover, are likely to be more pronounced for language learners than for mature L1 users due to the greater need by the former to exercise controlled processing that requires attentional control (Kormos 1999). Careful consideration of the task demands, therefore, is essential if output (or input, for that matter) is used in the task to promote IL development. The level of learners' L2 prociency is another related factor that may play a role in how much output stimulates learning mechanisms leading to IL development. It is possible, for instance, that lower prociency learners who are struggling with the production of one-word utterances may not be able to engage much of grammatical encoding during production because their cognitive eort may be spent primarily on the retrieval of lexical items (Bygate 1999). As a result, such learners may not be able to attend to grammatical forms in either the output they produce or the input they receive, though they may pay attention to individual lexical items. For output to exert facilitative eect on grammatical acquisition, then, it is necessary to take into account the timing issue vis-a -vis learners' L2 prociency level in general and with regard to the specic form that may be targeted for instruction in particular (see Izumi 2000, 2002, for a study that found positive eects of output on the acquisition of relative clauses by ESL learners whose developmental readiness (cf. Pienemann 1998) was matched to the target form). In general, if the facilitative impact of output for learning requires the engagement of psycholinguistic processes such as grammatical encoding and monitoring as argued above, it should follow that the eectiveness of output- based activities can be assessed in large part by how successfully these processes are engaged in these activities. A mechanical production task, for instance, does not likely involve genuine production mechanisms as described above; accordingly, its impact on SLA cannot be expected to be large. A fundamental consideration for pedagogy is that, for output to have any signicant impact on learning, a meaningful context for language use needs to be created so that learners can acquire proper formmeaning connections in the L2a focus-on-form consideration (cf. see Doughty and Williams 1998; Long 1991; Long and Robinson 1998). In Levelt's model, this means that the coordination between the conceptualizer and the formulator needs to be involved. Dissociating the two, or reversing the order of the involvement of the two, from the conceptualizer (concept generation)-to-the formulator (its grammatical formulation) to the formulator (grammatical formulation)-to-the conceptualizer (concept generation), as is done in some predominantly form- 190 COMPREHENSION AND PRODUCTION PROCESSES focused instruction, would go against natural production processing. 5 In light of the foregoing discussion, it is hardly a surprise if such approaches to L2 instruction do not lead to the acquisition of any usable knowledge. Elucidating the psycholinguistic mechanisms of the output hypothesis, as has been attempted in this paper, gives us many insights into the specications of optimal conditions for language learning. CONCLUSION This paper reviewed previous literature relevant to the comprehension and production processes in order to illuminate, ultimately, the psycholinguistic rationale for the output hypothesis in SLA. It was argued that comprehension of language input is a complex process involving multiple resources of information, linguistic information being but one source. The resourceful nature of the comprehension system is highly useful in L2 comprehension, but at the cost of reducing the amount of intake that can be used for integration in the developing system. In elucidating the mechanisms by which output promotes SLA, it was argueddrawing on Levelt's production modelthat the processes of grammatical encoding during production and monitoring to check the matching of the communicative intention and the output enable the language learners to assess the possibilities and limitations of what they can or cannot express in the TL. These processes are hypothesized to serve as an internal priming device for consciousness raising for language learning. The resultant state of alertness may then prompt the learners to take several alternative routes depending on the given production circumstance, which leads to dierent functions of output as specied by Swain. It was also argued that situational, linguistic, task, and learner variables can all aect the extent to which these psycholinguistic mechanisms are engaged. Guided by the knowledge of relevant psycholinguistic mechan- isms underlying the output and input processing, future research should aim to identify the optimal conditions under which successful L2 learning is induced through output and input. ACKNOWLEDGEMENTS An earlier version of this paper was presented at JPacSLRF held in Kitakyushu, Japan in November of 2001. This paper is based on part of my doctoral dissertation, completed at Georgetown University. I am grateful to Catherine Doughty for her guidance throughout the entire process of my doctoral work. My thanks also go to Cristina Sanz and Je Connor-Linton, the readers on my dissertation committee, and anonymous Applied Linguistics reviewers for insightful comments. (Final version received October 2002) SHINICHI IZUMI 191 192 COMPREHENSION AND PRODUCTION PROCESSES NOTES 1 In one approach to human sentence process- ing, known as the autonomous modular model (e.g. Forster 1979), the comprehen- sion system is believed to consist of three separate and sequential processes: lexical, structural, and interpretive processes. These processes are believed to occur in the given order. On the other hand, the interactive model claims that these processes are not strictly modularly insulated, but interact in every stage of sentence comprehension (e.g. Taraban and McCelland 1988). While research evidence indicates that both seman- tic and pragmatic factors do aect speech comprehension, the major disagreement lies in when such eects come into playduring on-line parsing operations or after parsing operations are completed. Since evidence in support of either position is various and to review it here is beyond the scope of this paper, readers are referred to surveys of these studies reported in Fender (2001), Garrett (1991), Harrington (2001), Tyler and Tyler (1990), and Wingeld (1993). Without getting bogged down with the psycholinguistic debate over the modularity issue, the present discussion focuses on the general characteristics of speech comprehen- sion processes and their relevance to lan- guage learning. 2 See DeKeyser and Sokalski (1996) for criti- cisms against VanPatten and his colleagues' studies on processing instruction and Allen (2000) for some results contrary to those of VanPatten and others'. 3 Not only is it assumed that speech produc- tion occurs subconsciously and automati- cally, it is also assumed by Levelt that various stages of lexical retrieval and pho- nological encoding are modularly encapsu- lated. In SLA, Doughty (2001) reviews relevant literature in cognitive psychology and concludes that the speech plan is largely modular yet amenable to modication; that is, there are `small cognitive windows of opportunity for ``intrusions''' (2001: 249), that is, focus on form. 4 Obviously, output is also related to the role of negotiated interaction in SLA, which is a much discussed topic in SLA research (see e.g. Ellis 1999; Gass 1997; Long 1996). Research suggests that negotiation helps to draw the learners' attention to the ILTL discrepancies or to the area of language which they know little about yet, thereby contributing to L2 development. In this context, output is considered to play a crucial part in the negotiation because it serves both as the initial trigger of the learning sequence and the ultimate forum of uptake and incorporation. 5 In many EFL classrooms in Japan, for instance, the focus of the English class is often on the teaching of grammatical struc- tures and the associated vocabulary. In this context, a frequently used pedagogical approach is to explain a grammar point, do exercises to consolidate the learned know- ledge, and engage in (semi-)communicative activities to use the learned structure. In other words, learners are asked to think of their message content by rst specifying which grammatical structure to use for their message generationa formulator-to- conceptualizer approach. However, in normal speech processing, one rarely, if ever, decides on which grammatical struc- ture to use rst and then think of what can be said with it. Rather, one thinks of what to say rst, and that naturally leads one to work on how to say it, as described in Levelt's model. REFERENCES Allen, L. Q. 2000. `Formmeaning connections and the French causative: An experiment in processing instruction.' Studies in Second Language Acquisition 22: 6984. Allen, P., M. Swain, B. Harley, and J. Cummins. 1990. `Aspects of classroom treatment: toward a more comprehensive view of second language education' in B. Harley, P. Allen, J. Cummins, and M. Swain (eds): The Development of Second Language Proficiency. Cambridge: Cambridge University Press, pp. 5781. SHINICHI IZUMI 193 Anderson, A. and T. Lynch. 1988. Listening. Oxford: Oxford University Press. Anderson, J. 1982. `Acquisition of cognitive skill.' Psychological Review 89: 369406. Bates, E. and B. MacWhinney. 1989. `Func- tionalism and the competition model' in B. MacWhinney and E. Bates (eds): The Cross-linguistic Study of Sentence Processing. Cambridge: Cambridge University Press, pp. 77117. Bygate, M. 1999. `Task as context for the framing, reframing and unframing of lan- guage.' System 27: 3348. Bygate, M. 2001. `Effects of task repetition on the structure and control of oral language' in M. Bygate, P. Skehan, and M. Swain (eds): Researching Pedagogic Tasks: Second language learning, teaching, and testing. Harlow: Pearson Education, pp. 2348. Bygate, M., P. Skehan, and M. Swain. (eds). 2001. Researching Pedagogic Tasks: Second language learning, teaching, and testing. Harlow: Pearson Education. Chaudron, C. 1985. `Intake: On models and methods for discovering learners' processing of input.' Studies in Second Language Acquisi- tion 7: 114. Clark, E. 1982. `Language change during language acquisition' in M. Lamb and A. Brown (eds): Advances in Child Develop- ment, 2. Hillsdale, NJ: Erlbaum, pp. 17397. Clark, E. and B. Hecht. 1983. `Comprehen- sion, production, and language acquisition.' Annual Review of Psychology 34: 32549. Clark, H. and E. Clark. 1977. Psychology and Language. New York: Harcourt Brace Jova- novich. Crookes, G. and S. Gass. (eds) 1993a. Tasks and Language Learning: Integrating theory and practice. Clevedon, Avon: Multilingual Mat- ters. Crookes, G., and S. Gass. (eds) 1993b. Tasks in a Pedagogical Context: Integrating theory and practice. Clevedon, Avon: Multilingual Mat- ters. de Bot, K. 1992. `A bilingual production model: Levelt's `speaking' model adapted.' Applied Linguistics 13: 124. de Bot, K. 1996. `The psycholinguistics of the output hypothesis.' Language Learning 46: 52955. de Bot, K., T. Paribakht, and M. Wesche. 1997. `Toward a lexical processing model for the study of second language vocabulary acquisition: Evidence from ESL reading.' Studies in Second Language Acquisition 19: 30929. DeKeyser, R. 1997. `Beyond explicit rule learning: Automatizing second language morphosyntax.' Studies in Second Language Acquisition 19: 195221. DeKeyser, R. and K. Sokalski. 1996. `The differential role of comprehension and pro- duction practice.' Language Learning 46: 613 41. Doughty, C. 2001. `Cognitive underpinnings of focus on form' in P. Robinson (ed.): Cognition and Second Language Instruction. Cambridge: Cambridge University Press, pp. 20657. Doughty, C. and J. Williams. 1998. `Pedago- gical choices in focus on form' in C. Doughty and J. Williams (eds): Focus on Form in Classroom Second Language Acquisition. New York: Cambridge University Press, pp. 197 261. Do rnyei, Z. and J. Kormos. 1998. `Problem- solving mechanisms in L2 communication: A psycholinguistic perspective.' Studies in Second Language Acquisition 20: 34985. Ellis, R. 1990. Instructed Second Language Acquisition. Oxford: Blackwell. Ellis, R. 1993. `The structural syllabus and second language acquisition.' TESOL Quar- terly 27: 91113. Ellis, R. 1994. The Study of Second Language Acquisition. Oxford: Oxford University Press. Ellis, R. 1999. Learning a Second Language through Interaction. Philadelphia: John Benja- mins. Ellis, R. and X. He. 1999. `The roles of modified input and output in the incidental acquisition of word meanings.' Studies in Second Language Acquisition 21: 285301. Frch, C. and G. Kasper. 1983. `Plans and strategies in foreign language commun- ication' in C. Faerch and G. Kasper (eds): Strategies in Interlanguage Communication, London: Longman. Frch, C. and G. Kasper. 1986. `The role of comprehension in second-language learn- ing.' Applied Linguistics 7: 25774. Fender, M. J. 2001. `A review of L1 and L2/ ESL word integration skills and the nature of L2/ESL word integration development involved in lower-level text processing.' Language Learning 51: 31996. Forster, K. 1979. `Levels of processing and the 194 COMPREHENSION AND PRODUCTION PROCESSES structure of the language processor' in W. Cooper and E. Walker (eds): Sentence Processing: Psycholinguistic studies. Hillsdale, NJ: Erlbaum. Foster, P. and P. Skehan. 1996. `The influence of planning on performance in task-based learning.' Studies in Second Language Acquisi- tion 18: 299324. Garret, M. 1991. Sentence processing. In D. Osherson and H. Lasnik (eds), Language: An invitation to cognitive science, Cambridge, MA: The MIT Press. Gary, J. O. and N. Gary. 1981. `Caution: Talking may be dangerous to your linguistic health: The case for a much greater emphasis on listening comprehension in foreign lan- guage instruction.' International Review of Applied Linguistics 19: 114. Gass, S. 1987. `The resolution of conflicts among competing systems: A bidirectional perspective.' Applied Psycholinguistics 8: 329 50. Gass, S. 1988. `Integrating research areas: A framework for second language studies.' Applied Linguistics 9: 198217. Gass, S. 1997. Input, Interaction, and the Second Language Learner. Mahwah, NJ: Lawrence Erlbaum Associates. Gass, S. and L. Selinker. 1993. Second Language Acquisition: An introductory course. Mahwah, NJ: Lawrence Erlbaum Associates. Goldschneider, J. M. and R. M. DeKeyser. 2001. `Explaining the ``natural order of L2 morpheme acquisition'' in English: A meta- analysis of multiple determinants.' Language Learning 51: 150. Harley, B. 1986. Age in Second Language Acquisition. Clevedon, UK: Multilingual Matters. Harley, B. 1992. `Patterns of second language development in French immersion.' Journal of French Language Studies 2: 15983. Harley, B. and M. Swain. 1984. `The inter- language of immersion students and its implications for second language teaching' in A. Davies, C. Criper, and A. Howatt (eds): Interlanguage. Edinburgh: Edinburgh Univer- sity Press, pp. 291311. Harrington, M. 1987. `Processing transfer: Language-specific processing strategies as a source of interlanguage variation.' Applied Psycholinguistics 8: 35177. Harrington, M. 2001. `Sentence processing' in P. Robinson (ed.): Cognition and Second Language Instruction. Cambridge: Cambridge University Press, pp. 91124. Izumi, S. 2000. `Promoting noticing and SLA: An empirical study of the effects of output and input enhancement on ESL relativiza- tion.' Unpublished Ph.D. dissertation, Geor- getown University. Izumi, S. 2002. `Output, input enhancement, and the Noticing Hypothesis: An experimen- tal study on ESL relativization.' Studies in Second Language Acquisition 24: 54177. Izumi, S. and M. Bigelow. 2000. `Does output promote noticing and second language acquisition?' TESOL Quarterly 34: 23978. Izumi, S. and M. Bigelow. 2001. `Methodolo- gical and theoretical issues in testing the effects of focus on form.' TESOL Quarterly 35: 1819. Izumi, S, M. Bigelow, M. Fujiwara, and S. Fearnow. 1999. `Testing the output hypothesis: Effects of output on noticing and second language acquisition.' Studies in Second Language Acquisition 21: 42152. Johnson, K. 1996. Language Teaching and Skill Learning. Oxford: Blackwell. Kormos, J. 1999. `Monitoring and self-repair in L2.' Language Learning 49: 30342. Kormos, J. 2000. `The timing of self-repairs in second language speech production.' Studies in Second Language Acquisition 22: 14567. Kowal, M. and M. Swain. 1994. `Using col- laborative language production tasks to promote students' language awareness.' Lan- guage Awareness 3: 7393. Krashen, S. 1985. The Input Hypothesis: Issues and Implications. New York: Longman. Krashen, S. 1989. `We acquire vocabulary and spelling by reading: additional evidence for the input hypothesis.' Modern Language Journal 73: 44064. LaPierre, D. 1994. `Language output in a cooperative learning setting: Determining its effects on second language learning.' Unpublished MA thesis, University of Tor- onto. Larsen-Freeman, D. and M. Long. 1991. An Introduction to Second Language Acquisition Research. London: Longman. Levelt, W. 1989. Speaking: From intention to articulation. Cambridge, MA: The MIT Press. Levelt, W. 1992. `Accessing words in speech production: Stages, processes and represen- tations.' Cognition 42: 122. Levelt, W. 1993. `Language use in normal SHINICHI IZUMI 195 speakers and its disorders' in G. Blanken, J. Dittman, H. Grimm, J. Marshall, and C. Wallesch (eds): Linguistic Disorders and Pathologies: An International Handbook. Berlin: de Gruyter. Levelt, W., A. Roelofs, and A. Meyer. 1999. `A theory of lexical access in speech produc- tion.' Behavioral and Brain Sciences 22: 175. Long, M. 1991. `Focus on form: A design feature in language teaching methodology' in K. deBot, C. Kramsch, and R. Ginsberg (eds): Foreign Language Research in Cross- cultural Perspective. Amsterdam: John Benja- mins, pp. 3952. Long, M. 1996. `The role of the linguistic environment in second language acquisition' in W. Ritchie and T. Bhatia (eds): Handbook of Research on Second Language Acquisition. New York: Academic Press, pp. 413468. Long, M. in press. Task-based Language Teach- ing. Oxford: Blackwell. Long, M. and P. Robinson. 1998. `Focus on form: Theory, research, and practice' in C. Doughty and J. Williams (eds): Focus on Form in Classroom Second Language Acquisition. New York: Cambridge University Press. pp. 1541. MacWhinney, B. 1987. `The competition model' in B. MacWhinney (ed.): Mechanisms of Language Acquisition. Hillsdale, NJ: Lawr- ence Erlbaum, pp. 249308. McLaughlin, B. 1987. Theories of Second Lan- guage Learning. London: Edward Arnold. McLaughlin, B., T. Rossman, and B. McLeod. 1983. `Second-language learning: An infor- mation-processing perspective.' Language Learning 33: 13558. Mecartty, F. 2000. `Lexical and grammatical knowledge in reading and listening compre- hension by foreign language learners of Spanish.' Applied Language Learning 11: 32348. Nobuyoshi, J. and R. Ellis. 1993. `Focused communication tasks and second language acquisition.' ELT Journal 47: 20310. Pica, T. 1988. `Interlanguage adjustments as an outcome of NSNNS negotiated interaction.' Language Learning 38: 4573. Pica, T., L. Holliday, N. Lewis, and L. Morgenthaler. 1989. `Comprehensible output as an outcome of linguistic demands on the learner.' Studies in Second Language Acquisition 11: 6390. Pienemann, M. 1998. Language Processing and Second Language Development: Processability theory. Amsterdam: Benjamins. Robinson, P. 1995. `Attention, memory and the ``noticing'' hypothesis.' Language Learn- ing 45: 283331. Rost, M. 1990. Listening in Language Learning. New York: Longman. Rutherford, W. and M. Sharwood Smith. 1985. `Consciousness-raising and Universal Grammar.' Applied Linguistics 6: 27482. Samuda, V. 2001. `Guiding relationships between form and meaning during task performance: The role of the teacher' in M. Bygate, P. Skehan, and M. Swain (eds): Researching Pedagogic Tasks: Second language learning, teaching, and testing. Harlow: Pearson Education, pp. 11940. Sasaki, Y. 1994. `Paths of processing strategy transfers in learning Japanese and English as foreign languages: A competition model approach.' Studies in Second Language Acquisi- tion 16: 4372. Sato, C. 1986. `Conversation and interlan- guage development: Rethinking the connec- tion' in R. Day (ed.): Talking to Learn: Conversation in second language acquisition. Rowley, MA: Newbury House, pp. 2345. Schmidt, R. 1990. `The role of consciousness in second language learning.' Applied Linguistics 11: 20626. Schmidt, R. 1995. `Consciousness and foreign language learning: A tutorial on the role of attention and awareness in learning' in R. Schmidt (ed.): Attention and Awareness in Foreign Language Learning. Hawaii: Second Language Teaching and Curriculum Center, pp. 163. Schmidt, R. 2001. `Attention' in P. Robinson (ed.): Cognition and Second Language Instruc- tion. Cambridge: Cambridge University Press, pp. 332. Scovel, T. 1998. Psycholinguistics. Oxford: Oxford University Press. Sharwood Smith, M. 1986. `Comprehension vs. acquisition: Two ways of processing input.' Applied Linguistics 7: 23956. Sharwood Smith, M. 1991. `Speaking to many minds: On the relevance of different types of language information for the L2 learner.' Second Language Research 7: 11832. Shehadeh, A. 1999. `Non-native speakers' production of modified comprehensible output and second language learning.' Lan- guage Learning 49: 62775. 196 COMPREHENSION AND PRODUCTION PROCESSES Shehadeh, A. 2001. `Self- and other-initiated modified output during task-based inter- action.' TESOL Quarterly 35: 43357. Shehadeh, A. 2002. `Comprehensible output, from occurrence to acquisition: An agenda for acquisitional research.' Language Learning 52: 597647. Skehan, P. 1996. `A framework for the implementation of task-based instruction.' Applied Linguistics 17: 3862. Skehan, P. 1998. A Cognitive Approach to Language Learning. Oxford: Oxford Univer- sity Press. Skehan, P. and P. Foster. 1997. `The influence of planning and post-task activities on accuracy and complexity in task-based learn- ing.' Language Teaching Research 1: 185211. Skehan, P. and P. Foster. 1999. `The influence of task structure and processing conditions on narrative retellings.' Language Learning 49: 93120. Slobin, D. 1985. `Crosslinguistic evidence for the language-making capacity' in D. Slobin (ed.): The Crosslinguistic Study of Language Acquisition. Hillsdale, NJ: Lawrence Erlbaum. Stanovich, K. E. 1980. `Toward an interactive- compensatory model of individual differ- ences in the development of reading flu- ency.' Reading Research Quarterly 16: 3271. Stolz, W. 1967. `A study of the ability to decode grammatically novel sentences.' Jour- nal of Verbal Learning and Verbal Behavior 6: 86773. Swain, M. 1985. `Communicative competence: Some roles of comprehensible input and comprehensible output in its development' in S. Gass and C. Madden (eds): Input in Second Language Acquisition. Rowley, MA: Newbury House, pp. 23553. Swain, M. 1993. `The output hypothesis: Just speaking and writing aren't enough.' The Canadian Modern Language Review 50: 158 64. Swain, M. 1995. `Three functions of output in second language learning' in G. Cook and B. Seildlhofer (eds): Principles and Practice in Applied Linguistics: Studies in Honour of H. G. Widdowson. Oxford: Oxford University Press, pp. 12544. Swain, M. 1998. `Focus on form through conscious reflection' in C. Doughty and J. Williams (eds): Focus on Form in Classroom Second Language Acquisition. New York: Cam- bridge University Press. 6481. Swain, M. and S. Lapkin. 1995. `Problems in output and the cognitive processes they generate: A step towards second language learning.' Applied Linguistics 16: 37191. Swain, M. and S. Lapkin. 2001. `Focus on form through collaborative dialogue: Exploring task effects' in M. Bygate, P. Skehan, and M. Swain (eds): Researching Pedagogic Tasks: Second Language Learning, Teaching, and Testing. Harlow: Pearson Edu- cation, pp. 99118. Taraban, R. and J. McCelland. 1988. `Con- stituent attachment and thematic role assignment in sentence processing: Influ- ences of content-based expectations.' Journal of Memory and Language 27: 597632. Terrell, T. 1991. `The role of grammar instruc- tion in a communicative approach.' The Modern Language Journal 75: 5263. Tomlin, R. and V. Villa. 1994. `Attention in cognitive science and second language acquisition.' Studies in Second Language Acqui- sition 16: 183203. Tyler, I. and M. Tyler. 1990. Psycholinguistics: Learning and using language. Englewood Cliffs, NJ: Prentice-Hall. Tyler, M. 2001. `Resource consumption as a function of topic knowledge in nonnative and native comprehension.' Language Learn- ing 51: 25780. VanPatten, B. 1995. `Cognitive aspects of input processing in second language acquisition' in P. Hashemipour, R. Maldonaldo, and M. van Naerssen (eds): Studies in Language Learning and Spanish Linguistics: In Honor of Tracy D. Terrell. New York: McGraw-Hill, pp. 170 83. VanPatten, B. 1996. Input Processing and Gram- mar Instruction in Second Language Acquisition. Norwood, NJ: Ablex. VanPatten, B. and S. Oikkenon. 1996. `Ex- planation versus structured input in process- ing instruction.' Studies in Second Language Acquisition 18: 495510. Wingfield, A. 1993. `Sentence processing' in J. Berko Gleason and N. B. Ratner (eds): Psycholinguistics. Fort Worth: Harcorurt Brace Jovanovich College, pp. 199235.