The Production-Comprehension Interface in Second Language Acquisition

An Integrated Encoding-Decoding Model

Anke Lenzing

• Publisher: Bloomsbury Academic
• Year: 2021
• ISBN:
  9781350148758
• Edition: 1
• Topic: Languages & Linguistics
• Subtopic: Syntax in Linguistics
• Index: Languages & Linguistics

Examining a key issue in second language acquisition (SLA) research, this book explores the relation between second language (L2) production and comprehension at the level of processing. The central question underlying this interface is the relationship between grammatical encoding and decoding, namely: are the two modalities of production and comprehension subserved by different types of processors, or by the same syntactic processing module? Proposing an 'Integrated Encoding-Decoding Model' of SLA, Anke Lenzing presents the results of a comprehensive empirical study to demonstrate the extent to which the two modalities rely on shared representations and/or shared processes. Through this detailed analysis The Production-Comprehension Interface in Second Language Acquisition sheds new light on the cognitive architecture of human language processing and offers a deeper understanding of the mechanisms at work in the L2 acquisition process.

1

The architecture of human language processing

This chapter engages with core aspects of human language processing. In order to be able to investigate potential relationships between language production and language comprehension in L2 acquisition, it is crucial to gain a deeper understanding of the processes underlying human language processing in general. Therefore, the chapter introduces the theoretical approaches to language production and comprehension that are relevant for the Integrated Encoding-Decoding Model of SLA. In a first step, the processes involved in language production are presented. The view on language production that I adopt in the model is mainly based on Levelt’s model of language generation (e.g. Levelt 1989, 1999). In a second step, I introduce major approaches to language comprehension. In this regard, the focus is on the Good-Enough Approach to Language Comprehension (e.g. Ferreira 2003) and its extension, the Online Cognitive Equilibrium Hypothesis (Karimi & Ferreira 2016), as some of their core assumptions are integrated in the Integrated Encoding-Decoding Model of SLA.

1.1 Speech production: Levelt’s model of language generation

Speech production theories focus on how humans transform ideas they want to convey into a linguistic form. This process covers the retrieval of words in the mental lexicon as well as the syntactic, morphological and phonological encoding of the various elements in the intended message to be able to produce intelligible overt speech. A central part of sentence production consists of the processes of grammatical encoding: They cover (1) the retrieval of lexical and syntactic forms that express the non-linguistic message and (2) the determination of the actual morphological forms and the assembly of constituent structure. Thus, Ferreira, Morgan and Slevc (2018: 432) argue that ‘it is only a minor indulgence to claim that to understand why and how grammatical encoding carries out its duties is to understand a significant part of the why and how of language itself’.

Approaches to language production in general and grammatical encoding in particular differ in their underlying theoretical assumptions concerning the architecture of the language processing system. In the psycholinguistic literature, a common distinction is between connectionist approaches to speech production (e.g. Chang 2002; Dell 1986; Dell, Chang & Griffin 1999) and approaches that assume a modular architecture of the human processing system (Garrett 1982; Levelt 1989; Levelt, Roelofs & Meyer 1999). However, the line separating these two perspectives is becoming increasingly blurred, as the two approaches have influenced each other over time: Kormos (2006: 6) points out that models with a modular architecture increasingly incorporate connectionist aspects, in particular in modelling lexical access (see e.g. Levelt 1999, 2000; Levelt, Roelofs & Meyer 1999). Connectionist accounts, on the other hand, have adopted aspects of modularity, in that the system is assumed to be globally modular but locally interactive (see e.g. Dell & O’Seaghda 1991; Goldrick 2007).¹

Although there is considerable debate in the field concerning the exact conceptualization of the different processes involved in language production, there seems to be a consensus on some core aspects of sentence generation. This concerns, for instance, the assumption that there are three main steps in language production, namely conceptualizing, formulating and articulating (see e.g. Griffin & Ferreira 2006: 21; Grosjean 2013: 51; Levelt 1999: 88; Pickering, Branigan & McLean 2002: 586). These stages are independent in that each has its own type of representation and works on its own characteristic input.

Levelt’s model of language generation (see e.g. Levelt 1989, 1999, 2000) provides a detailed account of these three stages in language production. The model constitutes the psycholinguistic basis of Processability Theory (PT), and some of its core assumptions form a central part of the Integrated Encoding-Decoding Model presented in Chapter 4. This concerns in particular the psychological aspects of language generation and the processes involved in grammatical encoding.

1.1.1 A blueprint for the speaker: An introduction

Levelt’s (1989, 1999) model of sentence generation, the Blueprint for the Speaker, is arguably the most influential account of speech production in the field of psycholinguistics. According to Kormos (2006: 7), the model is also ‘the most widely used theoretical framework in L2 speech production research’. Initially articulated in 1989, the blueprint has been continually enhanced and expanded in order to incorporate advances in psycholinguistic research that have provided more information on the language production process. This applies in particular to a revised conceptualization of the process of lexical access in speech production (see Levelt, Roelofs & Meyer 1999) but also includes slightly revised labels for the different components (see Levelt 1999).

This chapter introduces the core components of the blueprint as well as those developments of the theory that are central to an understanding of the perspective on the development of language production in SLA taken in this book. For the sake of compatibility with the theoretical approaches to SLA based on Levelt’s model, that is, PT and its extension, the Multiple Constraints Hypothesis, reference is made to its initial conceptualization and the related terminology (Levelt 1989). However, the chapter also introduces relevant revisions of the model, including modifications in both the area of grammatical encoding and the mental lexicon (e.g. Bock & Levelt 1994; Levelt 1999, 2000; Levelt, Roelofs & Meyer 1999).

The Blueprint for the Speaker is depicted in Figure 1.1. It covers the processes from the initial concept a speaker intends to express to the actual articulation of the speaker’s utterance. The model is modular in nature and consists of a number of interrelated processing components. These are the conceptualizer, the formulator, the articulator, the audition and the speech comprehension system. Speech generation starts in the conceptualizer. It is here that the message a speaker intends to convey is conceived, which involves the selection and ordering of the information as well as perspective-taking. The output of the conceptualizer is the preverbal message. In a next step, the preverbal message is transformed into a linguistic structure. This takes place in the formulator and involves both the retrieval of lexical information and the grammatical encoding of the message. The output of the grammatical encoder, the surface structure, is then given a phonological shape, which results in the phonetic plan or internal speech. In a final step, the phonetic plan is executed by the articulator, resulting in overt speech (see Levelt 1989: 8–13).

Figure 1.1 A blueprint for the speaker (from Levelt 1989: 9, fig. 1.1). © 1989 Massachusetts Institute of Technology, by permission of The MIT Press.

Processing components as autonomous specialists

Levelt (1989: 4) conceptualizes the processing components as ‘relatively autonomous specialists’. He argues that each component is a specialist that works (relatively) autonomously on a specific kind of input and produces a specific kind of output. The output of one component serves as input to the next component. For instance, the grammatical encoder takes the preverbal message as characteristic input and translates the conceptual structure into a surface form. The encoder contains specialized procedures to retrieve lemmas from the mental lexicon as well as syntactic building procedures. The characteristic output – the surface structure – is the input to the phonological encoder. The algorithms of the specific components are assumed to be executed in real time. Another tenet of the model is that there is no exchange of information between the components and no feedback from components down the line (Levelt 1989: 15–16).

Automaticity

A further characteristic of the processing components is their automaticity. In contrast to controlled processing, the execution of automatic processes does not require conscious attention. Apart from the conceptualizer, which involves controlled processing as the generation of a communicative intention relies on awareness and conscious attention, all other processing components are assumed to operate largely automatically (Levelt 1989: 21). This view of processing can account for its high speed as well as its efficiency (see Levelt 1989: 22).

Incremental processing

One core assumption underlying Levelt’s model is that language generation is incremental. This means that processing components are triggered by fragments of characteristic input. Processors can thus work on the incomplete output of the previous processing component. This architecture allows for parallel processing in a stage model: The different components can work in parallel on different parts of the utterance or, as Levelt (1999: 88) puts it: ‘[T]he various processing components are normally simultaneously active, overlapping their processing as the tiles of a roof. When we are uttering a phrase, we are already organizing the content for the next phrase, etc.’ Levelt (1989: 27) points out that incremental processing is only possible when processing takes place automatically.

1.1.2 Conceptualizer

The conceptualizer is the locus of message generation. Levelt emphasizes that message generation does not occur in isolation but has to be seen in the context of social and communicative interaction (Levelt 1999: 83–4), as language use can be viewed as some kind of joint action (see also Clark 1996). Thus, when conceptually preparing a message, the speaker draws on various kinds of knowledge sources, including the exercise of social competence.

The generation of the conceptual message involves the planning of the communicative intention including its sequencing into subgoals and the selection of the type of information to be expressed (e.g. assertions, declarations, questions) (Levelt 1989: 107). In sequencing the communicative intention, the speaker is confronted with what is called the linearization problem, which consists of ‘deciding what to say first, what to say next, and so on’ (Levelt 1989: 138). The sequencing or, in other words, the linearization of propositions is guided by the principle of natural order, stating that the information to be expressed should be sequenced ‘according to the natural order of its content’ (Levelt 1989: 138). When expressing events, the natural order relates to the chronological order in which the events occur. The default interpretation of propositions expressing an event is to assume that the order in which the events are presented corresponds to the chronological order in which they occurred. Deviations of the natural order should ideally be marked explicitly in order to facilitate mutual understanding of the message, as in example (1), in which the preposition after indicates a deviation from the natural order of events.

(1) She took off her boots after she climbed the mountain.

In example (1), the second event (she climbed the mountain) occurs before the first one (she took off her boots), which poses a processing problem for the speaker. The linearization problem and the resulting processing difficulties are central to PT, the theory of SLA adopted in this book, as it is assumed that linearization poses a major challenge for L2 learners. Therefore, the linearization problem will be taken up again in Section 3.1 in the context of L2 acquisition and processing.

Conceptualizing additionally involves the assignment of an accessibility status to the referents in the conceptual message, which indicates whether referents have been previously introduced in the discourse or whether they can be assumed to be known by the interlocutor. The message is then assigned a propositional format and perspective. Furthermore, the mood of the message is determined, that is, whether it is declarative, interrogative or imperative (Levelt 1999: 93). The output of the conceptualizer is the preverbal message, which serves as the characteristic input for the next component, the formulator.

1.1.3 Formulator

In the formulator, a conceptual structure is transformed into a linguistic structure. The translation of the preverbal message into an articulatory plan is achieved by the process of grammatical encoding. As this process forms the heart of the formulator and is central to both PT and the Integrated Encoding-Decoding Model, the focus of this section is on grammatical encoding. The second process taking place in the formulator – phonological encoding – will not be introduced (for details, see e.g. Levelt 1999).

According to Levelt (1999: 94), grammatical encoding exhibits the following three properties: ‘it takes preverbal messages as input, it produces surface structures as output, and it has access to the mental lexicon’. In formalizing the process of grammatical encoding, Levelt (1989) draws on aspects of Kempen and Hoenkamp’s (1987) Incremental Procedural Grammar and an early version of LFG (Bresnan 1982).² The process of grammatical encoding can be divided into two major types of processing – functional processing and positional processing (see also Garrett 1980, 1982, 1988) – and is illustrated in Figure 1.2.

Figure 1.2 Grammatical encoding (from Bock & Levelt 1994: 946, fig. 1). © 1994 Academic Press.

Whereas functional processing encompasses the selection of suitable lexical concepts as well as the assignment of grammatical functions, in positional processing, the constituent structure is generated including morphological inflections (Bock & Levelt 1994: 946). Finally, the output of grammatical encoding serves as input to phonological encoding.

Functional processing

The first step in functional processing is lexical selection. A core tenet of Levelt’s model is that speech generation is lexically driven, and he argues that ‘the lexicon is the essential mediator between conceptualization and grammatical and phonological encoding’ (Levelt 1989: 181).

Since Levelt’s initial conceptualization of his model in 1989, the field has seen important advances in theory development of lemma access (see e.g. Roelofs 1992a, b, 1993; Levelt, Roelofs & Meyer 1999), which were incorporated into Levelt’s (1999) revised model. The process from lexical selection to the actual phonological encoding of a word is framed in the computational model WEAVER ++ (Levelt, Roelofs & Meyer 1999). The core of the model is formed by lexical networks consisting of three different levels that represent different types of information. A fragment of a lexical network is presented in Figure 1.3 with the example item select. The first level, the conceptual stratum, includes conceptual information about the lexical item. The conceptual node SELECT stands for the meaning of the verb select with its two slots X and Y representing the two arguments the verb takes, that is, the agent that selects and the patient/theme that is selected. The conceptual node is linked to other concept nodes that are related to the concept SELECT. They can be both lexical and non-lexical. The labelled links express the relations between the concepts. An important characteristic of this approach is its non-decompositional character: Lexical concepts are not decomposed into sets of semantic features but are represented as undivided wholes (Levelt, Roelofs & Meyer 1999: 4).

Figure 1.3 Fragment of a lexical network (from Levelt 1999: 97, fig. 4.4). © Oxford University Press, 1999. Reproduced with permission of the Licensor through PLSclear.

In the process of lexical selection, the concepts that best match the conceptual message are activated and spread their activation to semantically related concepts. In Figure 1.3, the conceptual node SELECT spreads its activation to the concepts CHOOSE and ELECT. The conceptual node SELECT is also linked to one lemma at the second level, the lemma stratum or lemma level. The activated lexical concept (SELECT in the example above) spreads some of its activation to the lemma it is related to (select). The process of lexical selection ‘is a statistical mechanism, which favors the selection of the highest activated lemma’ (Levelt, Roelofs & Meyer 1999: 4).

Once a lemma is selected, its syntactic properties are activated, such as its syntactic category, for example noun or verb. Additionally, the values of the lemma’s diacritic parameters, including its inflectional features for numbe...

Table of contents

• Cover
• Halftitle Page
• Title Page
• Contents 
• List of figures
• List of tables
• Acknowledgements
• List of abbreviations
• Introduction
• 1. The architecture of human language processing
• 2. The relation between production and comprehension in language processing
• 3 The view on second language acquisition: Processability Theory and the Multiple Constraints Hypothesis
• 4 The Integrated Encoding-Decoding Model of SLA
• 5 The English passive as a test case for the Integrated Encoding-Decoding Model of SLA
• 6 The study: Methodological considerations
• 7 Testing the Integrated Encoding-Decoding Model of SLA
• Concluding remarks
• Appendix
• Notes
• References
• Index of subjects
• Imprint