The Routledge Handbook of Second Language Acquisition and Speaking

Tracey M. Derwing, Murray J. Munro, Ron I. Thomson, Tracey M. Derwing, Murray J. Munro, Ron I. Thomson

• Verlag: Routledge
• Jahr: 2022
• ISBN:
  9781000532739
• Auflage: 1
• Thema: Filología
• Thema: Lingüística

This Handbook is a comprehensive volume outlining the foremost issues regarding research and teaching of second language speaking, examining such diverse topics as cognitive processing, articulation, knowledge of pragmatics, instruction in sub-components of speaking (e.g., grammar, pronunciation, and vocabulary) and the attrition of the first language. Outstanding academics have contributed chapters to provide an integrated and inclusive perspective on oral language skills. Specialized contexts for speaking are also explored (e.g., English as a Lingua Franca, workplace, and interpreting). The Routledge Handbook of Second Language Acquisition and Speaking will be an indispensable resource for students and scholars in applied linguistics, cognitive psychology, linguistics, and education.

PART I Theoretical Foundations and Processes Underlying Speaking

1 Bilingual Models of Speaking

Kees de Bot and Szilvia Bátyi

DOI: 10.4324/9781003022497-3

1 Introduction

In this contribution, we discuss bilingual language production and some of the theoretical concepts related to this. Our starting point will be the 1989 Speaking Model by Willem Levelt. Despite the fact that it is now more than 30 years old, it still stands as the most elaborate and empirically founded model for language production. The original model did not aim to elucidate how bilingual production differs from monolingual production, however, in the past few decades a number of bilingual variants have been developed, although these variants do not alter the fundamental characteristics of the model: it is lexically based, modular in type, incremental, and skills oriented. It is not a model of change, but parts of the model can change due to use and learning. One of the major issues in applying the model is that it is unclear how individual differences such as motivation, attitudes, and anxiety can be built into the blue print.

Bilingual models have been developed on the basis of the Levelt model. They have the same characteristics, however, should help in understanding phenomena such as code-switching and cross-linguistic influence (CLI).

2 Historical Perspectives

The question is: is the default model a monolingual or a bilingual one? On various occasions, Levelt has indicated that in his view bilingualism is a fascinating topic, but not one he wants to work on claiming that “monolingualism is already complex enough.” Several authors have argued that taking a monolingual model as a starting point does not do justice to the fact that an overwhelming majority of the world's population is bilingual and that therefore the default model should be a bilingual one (Grosjean, 2008, 2010). First of all, it has to be assessed to what extent the current model can deal with various aspects of multilingual processing. Bilingual and multilingual speech production models are usually derivations of Levelt's speaking model (Figure 1.1) or at least they borrow some elements from it. Consequently, bilingual speaking models cannot be discussed without mentioning the Speaking model. In the following parts, we describe the main components and processing mechanism of the Levelt model followed by a discussion of bilingual versions of the model and a brief outline of how code-switching is accounted for by the different models.

Figure 1.1 Levelt's SPEAKING model (1995). Reprinted with permission

The Levelt Model

Speaking is an integral part of our everyday activity and it is often considered a “gift of evolution to mankind” (Levelt, 1995, p. 13). Despite this all-pervading nature of speech, it was not until the end of the 20th century when the first comprehensive model of the components and processes of speech was developed by Willem J.M. Levelt (1989, 1993, 1995, 1999). It is still the most widely used theoretical framework sketching the mental mechanisms that generate speech from communicative intentions and is based on the theoretical and empirical knowledge that had accumulated until the creation of the model (Figure 1.1). The model is composed of processing components (boxes) and knowledge components (circles and ellipses). The processing components are “modular,” which means they are quite autonomous and the intelligence of the system comes from the cooperation between them. The three main components of speech production are as follows: the conceptualizer, the formulator, and the articulator. Processing in the formulator and the articulator is highly automatic, while planning processes in the conceptualizer require attention from the speaker.

A full description of the Speaking Model would take more space than allowed for the present contribution. Here, we offer a condensed version of the main characteristics. Production starts in the conceptualizer where a communicative intention is turned into lexical concepts. In the generation of the message, information about the conversational setting and the discourse model are taken into account which includes the selection of a linguistic register. Hesitation markers (e.g., silent or filled pauses) are often taken as indicators of the amount of mental activity going on: the expression of new or complex ideas are often preceded by greater hesitancy manifested in pauses, that is, more attention is directed to the planning stage, therefore, resources to be used to execute the act of speaking are limited. Conceptualization involves macroplanning and microplanning, including the rendering of ideas in the right order (linearization) and the plan of achieving communication goals (instrumentality). Within the conceptualizer, the message is generated and monitored internally whether the (preverbal) plan coincides with the intended message. Finally, the series of lexical concepts is turned into the preverbal message (see Figure 1.2 and Table 1.1 as an example), which is fed forward to the formulator.

Figure 1.2 The dog caught the ball

Table 1.1 The speech production process

Level	Task	Example
Conceptualizer	Concepts are chosen.	Catching (of something by someone); dog (the entity carrying out this action); ball (the entity on which the action is carried out.
Formulator	Lemmas are accessed and retrieved from the mental lexicon.	{catch}{dog}{ball}
	Grammatical roles are given to the lemmas.	VERB = {catch}; SUBJECT = {dog}: singular, definite; OBJECT = {ball}: singular, definite; TIME = past
	The selected set of lemmas is organized into an ordered string.	(DETERMINER) {dog} [singular; definite] {catch} [past] (DETERMINER) {ball} [singular; definite]
	The lexemes or word-forms are made available via links with the lemmas.	e.g., {dog} is linked in the mental lexicon both to the written form <dog> and to the spoken form /dɒɡ/
Articulator	The utterance is pronounced.	The dog caught the ball.

Here, the essential process of turning lexical concepts into a surface structure takes place which is done by matching lexical concepts with lemmas in the lexicon. Lexical items consist of two parts: the lemma in which the entry's meaning and syntax are represented, and the lexeme that contains the morphosyntactic and phonological information. The matching of a lemma with a lexical concept also leads to the activation of the syntactic procedures that are part of the lemma. For instance, if a transitive verb (e.g., caught) is activated, it will start the syntactic procedures for the generation of a direct and an indirect object (e.g., ball). The activation of the lexical item also leads to the lexeme becoming available. This process is not always successful, as the well-known tip-of-the-tongue phenomenon shows; sometimes the lexical item is activated through the lemma, but the lexeme part does not come up in time. Interestingly, some properties of the intended word form do become available; speakers often know how many syllables the word has and what the rhythmic pattern is. The selection of the lemmas and lexemes also leads to the formation of a surface structure. While the surface structure is being formed, the morpho-phonological information belonging to the lemma is activated and encoded. The output of the formulator is the input of the articulator which converts the speech plan into actual speech. There are two feedback loops, one internal that checks the inner speech, and one external that checks the overt speech. Syllables are the building blocks of speech. The outputs of the articulator are motor-plans to execute the assembling of syllables into running speech. The two feedback loops monitor the speech and articulation.

As this short description may show, speaking is not primarily syntactically based but lexically. It is modular in nature though Levelt has always carefully avoided calling the processing components modules, since that would imply that these components are modular and therefore innate. In later publications, Levelt followed the view that the modular character of these components is emergent, it is the result of use, not the origin. Several commentators (e.g., de Bot et al., 2007) have argued that the modularity of the model is one of its main characteristics and also one of its weaker points, because a strict modular view does not allow for a view on language that is more dynamic in nature.

The system works “from left to right,” that is, the information enters the system and is processed from intention to articulation without feedback or feedforward. It is only at the level of the phonetic plan that internal speech is monitored and corrections can be made. This means that errors in speaking can only be detected fairly late in the process. If, for instance, the wrong lexical item has been activated, this can only be detected through the internal feedback loop that monitors the internal speech. This also means that there are no in between mechanisms that can correct the error. Basically, error correction is redoing the same procedures and hoping that this time the intended meaning is actually expressed correctly.

Bilingual Models of Speaking

Bilingual and monolingual speech production shares many features, however, as outlined by Kormos (2006), there are some distinctive features as well. They share one main characteristic: all of them are directly or indirectly derived from the Levelt model. It was used as a starting point because the model is based on decades of psycholinguistic research and empirical data. While it could be argued that the Speaking model is not really monolingual, because it can deal with different registers and styles, which are not essentially different from languages, the model is not aimed at providing a model for bilingual production. The main question is, what a bilingual model should have that is not needed for a monolingual version. Such a model should be able to explain issues such as code-switching and CLI. Many models deal with part of the speaking process but only a few address speaking from conceptualization to articulation. In the following, we will concentrate on three models (de Bot, 1992; Poulisse & Bongaerts, 1994; Kormos, 2006), which offer suggestions on how the bilingual version of the Levelt model should be composed. In addition, the phenomena of code-switching will be discussed in relation to the speech production process.

The first adaptation of the Levelt model for bilingual speakers was carried out by de Bot (1992). The empirical basis was very limited to evaluate such a model, but it triggered a host of new findings and several new model versions. With as little modification to the original model as possible, de Bot suggested that the conceptualizer is language specific, that is, while the macroplanning phase is not language specific, in the microplanning phase, a language is activated in which the intended message is to be produced and it is part of the preverbal message (a similar process to choosing between registers in unilingual speech production). de Bot proposes that two alternativ...