This book presents the most comprehensive study to date of the starting point of second language acquisition. With its focus on the language input that learners receive and what they do with this input, the study sheds light on questions still unanswered in second language acquisition literature, such as what knowledge is brought to the acquisition process and how learners use this knowledge to process new linguistic information.
"synopsis" may belong to another edition of this title.
Rebekah Rast teaches English and Linguistics at The American University of Paris. She has published in the field of second and third language acquisition and conducts research in collaboration with the interdisciplinary CNRS research team Language, Cognition et Développement (UMR 7023), based at the Université Paris VIII. Her research interests include second and additional language acquisition, as well as the interface between language acquisition and teaching.
Acknowledgements, ix,
Abbreviations, xi,
Introduction, xiii,
Part 1: Theoretical Preliminaries,
1. Input and Intake Revisited, 3,
2. First Exposure Studies, 29,
Part 2: The Study,
3. Polish-French Contrastive Analysis, 47,
4. Research Methodology, 66,
5. The Adult's Available Knowledge at First Exposure to an Unknown Language, 83,
6. Case Studies: Two Learners with Similar Linguistic Profiles. ..., 108,
7. Speech Perception, 143,
8. Speech Comprehension, 166,
9. Grammatical Analysis, 192,
10. Concluding Remarks, 226,
Conclusion, 236,
Appendix 1, 238,
Appendix 2, 240,
Appendix 3, 241,
Appendix 4, 242,
Appendix 5, 245,
References, 246,
Index, 258,
Input and Intake Revisited
Reflections on 'Input' and 'Intake'
In the field of applied linguistics, and more specifically in the field of second language acquisition, the terms 'input' and 'intake' can be traced back several decades. Corder (1967: 165) was the first to use the term 'intake' as distinct from 'input' in his renowned quote:
The simple fact of presenting a certain linguistic form to a learner in the classroom does not necessarily qualify it for the status of input, for the reason that input is 'what goes in' not what is available for going in, and we may reasonably suppose that it is the learner who controls this input, or more properly his intake.
In the 1950s, Skinner (1957) and Chomsky (1959) had already made reference to these same phenomena without proposing the specific terms 'input' and 'intake'. At that time, the debate focused on the degree to which external factors influenced language acquisition, or rather, whether an internal innate structure, known as Universal Grammar guided language acquisition (cf. Chomsky, 1959, 1965). In his criticism of Skinner's work in which Skinner reflects on the notion of 'stimulus' (the environment) and 'response' (individual behaviour), Chomsky remarks that, 'We cannot predict verbal behaviour in terms of stimuli in the speaker's environment, since we do not know what the current stimuli are until he responds' (Chomsky, 1959: 32). In other words, we cannot know what the individual has taken in until the moment of response, leaving an important gap in our ability to observe language processing.
Decades later, Hatch (1983: 81) reflects on this distinction between input and intake:
If we wish to keep both terms, we may say that input is what the learner hears and attempts to process. That part that learners process only partially is still input, though traces of it may remain and help in building the internal representation of the language. The part the learner actually successfully and completely processed is a subset called intake. That part, then, is the language that is already part of the internal representation.
Hatch extends Corder's definition of intake from that which is controlled by the learner and actually 'goes in' to that which the learner 'successfully and completely' processes. That which is only 'partially' processed remains input. The distinction, in fact, is found in the role that this processed information will play in building internal language representations. This is all fine and well, but we are still faced with the problem of identifying what information is 'taken in' or 'processed'. Hatch's explanation is quite representative of our understanding of input and intake in the 1980s and 90s. A distinction between the two having already been made, researchers then attempted to articulate this difference more clearly. The essence of the problem, however, remains in our inability to measure the phenomena involved in these processes. How can we know what elements of the linguistic environment are processed completely, partially or not at all? How can we know if learners even 'hear' a signal in the input and if they process what they heard? How can we know what part of the signal is processed and what part is not? These are questions that continue to challenge us in the field of language acquisition today.
Numerous researchers concerned with the study of input and intake in SLA express a shared concern. VanPatten (2000: 294) articulates this in his article entitled 'Thirty years of input': '... in spite of the significant advances made by SLA research and the diversification of theoretical and research frameworks in which to conduct this research, our knowledge of the role of input has remained relatively unchanged during the last 30 years'. Carroll (2001: 1) confirms VanPatten's concerns when describing input as '... one of the most under-researched and under-theorized aspects of second language acquisition'. This lack of research has become apparent in part because everyone, regardless of the theoretical framework, seems to agree on the importance of input in language acquisition. As VanPatten (2000: 295) points out, 'We seem to concur that input is somehow central to SLA, that without it successful SLA is not possible'. Carroll (2001: 2) makes a similar remark: '... one point on which there is consensus is that SLA requires exposure to the second language'.
The study of 'intake' evokes similar concerns. According to Carroll (2001: 2), 'There is no agreement on what kind or how much exposure a learner needs. Indeed, we know very little still about the kinds of linguistic exposure learners actually get'. Consequently, we know very little about what learners do with what they get. In order to move forward in our understanding of input and intake, we need to analyse the divergence that exists between the different research perspectives, revealing at the same time the complexity of these two terms. According to the current dominant definition, 'input' refers to the linguistic environment of the learner, that is, to that which is available to be taken in, or rather, to everything in the TL that the learner is exposed to and has the opportunity to either hear or read. The dominant definition of 'intake' is less clear. Researchers disagree on how input is processed and on when and how intake enters the picture. Thus, their definitions reflect their research agendas. Wong and Simard (2000) provide a comprehensive overview of research conducted on intake, showing how researchers differ in their views as to whether intake is a process, a product or a combination of both. In fact, we appear to be in a deadlock in that the true debate focuses on the distinction between product and process, a distinction for which we have not yet found appropriate terminology.
It is unlikely that one approach alone will be able to resolve all the aspects of this debate. It is rather through collaboration and exchange of methodology, results and knowledge that we will come to discover what is actually going on in the mind of an L2 learner when exposed to TL input. Diverse theoretical frameworks have already contributed and will continue to contribute in their own way to our understanding of input and intake. This study in no way intentionally eliminates research findings due to difference of approach. On the contrary, it attempts to incorporate what we already know about learners' processing, regardless of the source (theoretical framework) of this knowledge. We will now review four of these sources: connectionism (including the Competition Model), interactionist approaches, generativist models and cognitivist models.
Connectionism
Researchers working within a connectionist framework must consider the distinction between input and intake for the simple reason that their approach is all about providing stimulus (input) to a network in an attempt to observe what the network can or cannot do with this stimulus under certain conditions. The following section provides a general description of connectionist architecture and its usage in language acquisition research.
Connectionist models
Connectionism involves the use of computer processing to simulate the functions of the mind and to predict what humans will do under specific conditions. The first specificity designated to a model depends on research objectives and the cognitive processes under investigation. In our case, for example, we will examine models designed to simulate language processes as opposed to other cognitive processes such as face recognition or serial recall. A connectionist architecture comprises a network of a large number of interconnected elements called 'nodes'. The 'knowledge' of the network resides in the information given to these nodes (e.g. phonetic or orthographic) and the strength of the connections between the nodes.
Generally we can distinguish between two types of connectionist models: localist symbolic models and distributed subsymbolic models. In the localist tradition, representations are coded for distinct pieces of information, such as visual information for all letters in the alphabet or for entire words. These models are designed to explain the functional mechanics of skilled human performance, such as word recognition processes (cf. Grainger & Jacobs, 1998). In distributed subsymbolic models, such as the well-known pattern associator of 'Parallel Distributed Processing' (Rumelhart & McClelland, 1986), information or knowledge is coded as a pattern activation across many processing units, and these units contribute to many different representations. The focus of these models is on the emergence of skilled human performance through learning. Distributed models generally comprise one or more internal layers of nodes in addition to an input and an output layer. The internal or 'hidden' layer is where the input is processed before becoming output. This layer is therefore adjusted over time following the learning process. Localist models, on the other hand, are concerned more with performance and therefore refer to 'bottom' or 'lower' layers, where a pattern of resources is stored (e.g. orthographic information), and 'upper' layers, where this pattern is categorised (e.g. entire words).
We will focus here on distributed connectionist models and their contribution to the study of language acquisition and, in particular, to the study of input and intake. At the outset, a network has no set parameters. No connections between input and output nodes have been set, nor has the strength of the connections been set. The network represents a 'tabula rasa' (Pinker & Prince, 1988: 90). Before initiating a simulation, researchers must set the nodes in the network, set the parameters or variables, and set the connections between the nodes. The strength of the connections will be developed during training. This generally occurs over a series of training sets. In order to train the network, the researchers provide it with input (linguistic, in this case) by activating the input layer, which then produces activation throughout the entire network, first on the internal layer(s) followed by the output layer. When the network is completely trained, each connection receives a scalar value that serves as input to the next node. According to Taraban et al. (1989: 172), 'This value is the product of the current activation level of the node on the input side of the connection and the strength or "weight" associated with the connection'. The 'output' of the system refers to the state of the output layer once the network has stabilised. A comparison between the 'real' output of the system and the 'desired' output (i.e. the correct form in the language of the study) allows for training of the system. In other words, the network 'learns' by example and by recognition of similarities. 'Similarities' here refer to elements that the network considers as similar relative to the training it received. The strength of connections can be changed according to the process involved, modifying the input/output function as it satisfies the network. For certain tasks, it is possible to train the network to calculate the function of a given input-output (cf. Cummins & Schwarz, 1992; Dreyfus, 1992; Taraban et al., 1989).
To better understand the problem of input in these models, we turn to a study of the acquisition of gender in L1 German (Taraban et al., 1989) in which the authors developed a connectionist architecture as a model for the acquisition of the declensional paradigm of the German definite article. Two simulations of child acquisition of L1 German were performed with a view to predicting whether German-speaking children can assign word classes through inferences based on the declensional paradigm. In both cases, a certain number of German nouns were presented one by one to the network. 'The input to the network consists of the kind of input considered to be available to a learner and the desired output is the correct form of the German article' (Taraban et al., 1989: 173). The input used for the study consisted of a selection of words taken from a German corpus of 80,000 words. As the nouns were presented to the network, the input nodes encoded the presence or absence of cues associated with a given noun. Each input node described a cue which contained, at minimum, phonological, morphological and semantic information. When a specific cue was present in a given noun, the input node was completely activated. On the contrary, if the cue was absent, the node remained inactivated. In short, this system provides a concrete example of what the authors use as input that 'enters'. If the network recognises a cue as a particular word, it can then use this 'representation' to process the input and produce the correct output; if it does not recognise the given cue, the 'representation' will not be activated, and consequently the output will not be a correct form of the TL. Words become a sort of representation of a series of cues. The activation of the input layer triggers the activation of internal layers, which in turn triggers the activation of the output layer. 'We think of these internal layers as forming a useful internal representation of the input. In our simulations we would expect these to correspond to the grammatical categories that describe German nouns presented to the network' (Taraban et al., 1989: 172). In essence, this analysis proposes an internal representation of the input by which the learner accesses corresponding grammatical categories as soon as a lexical item is presented. In other words, the authors suggest that input processing leads to an internal representation that corresponds to grammatical categories in the series presented to the network. The results of their study show that their network made use of a series of cues in the input about noun stems and noun usage in sentences to correctly select the appropriate form from the six possible forms of the definite article. The authors conclude that, 'We show not only that such a network can learn to correctly assign definite articles to a set of training items, but also that it forms consistent internal representations of its knowledge and is able to generalize this knowledge to new instances' (p. 173).
It follows that connectionism may be the framework that most directly treats the question of first contact with the input of any given TL in that researchers working within this perspective must formulate hypotheses about the 'learner's' knowledge before setting the parameters of the network. The problem of characterising the input, however, continues to trouble us. In the study by Taraban et al. (1989: 173), for example, the kind of input provided to the network is described by the authors themselves as, '... considered to be available to a learner...' (our emphasis). A related problem lies as well in the fact that in order to test their hypotheses about input processing, studies must be designed to show that learners themselves process input in the manner proposed by connectionist models. To put it simply, when one asks a computer to 'take in', it will 'take in', but when the same is asked of human learners, there is no guarantee that they will 'take in', or are even capable of 'taking in'.
In essence, those working within the connectionist framework are faced with similar questions to those posed in this book: what is the nature of the input available to a learner, and is it possible that learners fail to perceive information that we consider to be generally available or perceivable? How can we know what information should be presented to the network for a valid ecological simulation of what German children receive from their linguistic environment? Taraban et al. (1989: 173) are well aware of the methodological problems posed by using a corpus based on the average frequency of items in the environment: '... the input to the simulation can only be viewed as an approximation to the actual input received by the German-speaking child'. They propose the CHILDES database as a solution for future studies. With the help of CHILDES, they foresee designing simulations that more accurately represent real acquisitional situations. Establishing tools that provide more precision proves a positive step toward resolving the problem of what is 'available' for further processing.
Excerpted from Foreign Language Input by Rebekah Rast. Copyright © 2008 Rebekah Rast. Excerpted by permission of Multilingual Matters.
All rights reserved. No part of this excerpt may be reproduced or reprinted without permission in writing from the publisher.
Excerpts are provided by Dial-A-Book Inc. solely for the personal use of visitors to this web site.
"About this title" may belong to another edition of this title.
Seller: Majestic Books, Hounslow, United Kingdom
Condition: New. Seller Inventory # 355801192
Quantity: 1 available
Seller: Books Puddle, New York, NY, U.S.A.
Condition: New. Seller Inventory # 26356689847
Quantity: 1 available
Seller: Romtrade Corp., STERLING HEIGHTS, MI, U.S.A.
Condition: New. This is a Brand-new US Edition. This Item may be shipped from US or any other country as we have multiple locations worldwide. Seller Inventory # ABNR-174545
Quantity: 1 available
Seller: Basi6 International, Irving, TX, U.S.A.
Condition: Brand New. New. US edition. Expediting shipping for all USA and Europe orders excluding PO Box. Excellent Customer Service. Seller Inventory # ABEJUNE24-244059
Quantity: 3 available
Seller: Biblios, Frankfurt am main, HESSE, Germany
Condition: New. Seller Inventory # 18356689853
Quantity: 1 available
Seller: ALLBOOKS1, Direk, SA, Australia
Seller Inventory # SHUB244059
Quantity: 1 available
Seller: Asano Bookshop, Nagoya, AICHI, Japan
Condition: Brand New. Presents the comprehensive study of the starting point of second language acquisition. This book focuses on questions in second language acquisition literature, such as what knowledge is brought to the acquisition process and how learners use this knowledge to process linguistic information. Seller Inventory # 89816
Quantity: 1 available
Seller: moluna, Greven, Germany
Condition: New. Über den AutorRebekah Rast is Professor of English and Linguistics at the American University of Paris, where she also holds the position of Director of Teaching, Learning and Faculty Development. A member of the French research . Seller Inventory # 905667415
Quantity: 1 available