This paper aims to foster discussion of the means by which bilinguals control their two language systems. It proposes an inhibitory control (IC) model that embodies the principle that there are multiple levels of control. In the model a language task schema (modulated by a higher level of control) “reactively” inhibits potential competitors for production at the lemma level by virtue of their language tags. The IC model is used to expand the explanation of the effect of category blocking in translation proposed by Kroll and Stewart (1994), and predictions of the model are tested against other data. Its relationship to other proposals and models is considered and future directions proposed.

David Green's excellent paper “Mental control of the bilingual lexico-semantic system” raises three fundamental questions. My goal is to suggest elaborations on his answers that might make the arguments stronger and more comprehensive.

The schema concept has had a long and varied history in psychology. It was introduced by Bartlett (1932), who observed that subjects' reproductions of a story showed systematic deviations from the original. Elements in the original story that were uncommon or strange were lost or changed and elaborated such that they became more common and made more sense to the subjects, and infrequent words and concepts were replaced by more common ones. Bartlett explained this normalization behavior by assuming that humans adapt incoming information to existing knowledge structures in long-term memory and that they understand new information in terms of these structures. These knowledge structures were called schemas. Schemas may represent our knowledge of stereotypical events such as doing the laundry or cooking a meal (these schema structures are usually called scripts, Schank & Abelson, 1977), of objects and natural categories (see Anderson, 1985, for a discussion), and the knowledge underlying routine behavior (Norman, 1981). They provide a basis for explaining many different phenomena of information processing and memory functioning, such as inferencing, elaborating, stereotyping, reconstruction, false memorization, and the occurrence of slips in task performance. When, for instance, a reader encounters the word forest in a text, he is not surprised to see the noun phrase the trees, with the definite article, in the next sentence even though no trees were explicitly introduced before.

The recent trend in psycholinguistics from autonomous, rule-oriented models with symbolic representations towards more interactive, regularity oriented, and subsymbolic models reflects an increasing awareness that the finer aspects of cognitive processing may be sensitive to “contextual factors.” Apparently, there is systematic variability in the data that depends on task demands and stimulus list composition, and their effect on the strategies and decision criteria maintained by the experimental subject. However, most current (computational) models at best embody a task-independent stimulus identification process with a simple decision process operating in the same rigid manner across different tasks (Dijkstra & de Smedt, 1996).

Green makes the critical observation that models of lexical and semantic representation in bilingual memory fail to specify a mechanism that would enable the bilingual speaker to act. Under the conditions typical of most experimental research, bilinguals are asked to perform a well-defined task, such as word or picture naming, lexical decision, or translation. The question Green raises is how is it that the bilingual effectively performs one of these tasks rather than another? And within a given task such as word translation, how does the individual manage to produce words in one language and suppress the other? The goal of Green's proposal is to provide a preliminary account of the control apparatus that a bilingual would need to possess in order to perform effectively in this environment. An adequate model of these control mechanisms will presumably allow us to understand not only how bilinguals perform simple laboratory tasks, but also how they manage to engage the appropriate language during normal discourse, including code-switching with other bilingual speakers.

In this paper Green proposes an inhibitory control (IC) model of bilingual lexical processing. At the core of Green's arguments is the notion of “mental control,” formulated in terms of inhibition, control schemas, and a supervisory attentional system. The very notion of control, it seems, suggests some sort of intentional, exogenous force at work (e.g., the supervisory attentional system). Presumably mental control differs from automatic processes (Schneider & Shiffrin, 1977), yet in the IC model there is no precise computational specification of how the various parameters of the control system actually interact to determine automatic bilingual processes. In a computational view, the IC model has quite some symbolic AI flavor (e.g., with goal-oriented decision boxes and control schemas), but it also attempts to integrate activation-based accounts (e.g., interactive activation mechanisms). Again, because the model remains at a rather conceptual level as presented, it is difficult to determine how successful it will be in combining symbolic and connectionist approaches in understanding bilingual processing.

The planning of speech involves making successive choices in a hierarchy of options. In the case of bilingualism, these options are provided by two lexicons and grammars rather than one internalized by a speaker. Conceptualization processes map a communicative intention onto a message indicating the conceptual information to be verbalized to reach a speaker's communicative goal. In bilinguals, the illocutionary intention may be to express oneself in one language rather than the other, or to mix languages. Formulation processes activate and select lemmas and forms for the message concepts, and plan a syntactic and a morphophonological structure. Lemmas specify the syntactic properties of words, crucial for their use in sentences. The result of formulation is an articulatory program, which, when executed by articulation processes, yields overt speech. A central theoretical problem is how bilingual speakers manage to keep the options provided by the two languages apart in monolingual conversation, and how speakers are able to integrate the options in bilingual conversation where language mixing (i.e., code-switching or borrowing) may take place.

A surprisingly large part of the population of the world is at least bilingual. The question as to how people are able to control their language system is an important one and has been a topic for highly active research in recent years. Until now, models have been quite simplistic, but Green places his model in a wider framework of attention and control, drawing our attention to the fact that the mechanisms involved in language control share basic properties with the systems in other cognitive domains. Green introduces the (not language-specific) supervisory attentional system in combination with language task schemas which make it possible to adapt to the situation in which the bilingual system is functioning. Models of bilingual processing should be able to explain how lexical processing is affected by, among others, task demands (see also Dijkstra, van Jaarsveld, and ten Brinke, 1998). By adding a general mechanism of attentional control that is independently motivated by general cognitive mechanisms, Green has enriched his previous model considerably. In what follows we will first make some general remarks about the model Green proposes. In the second part we will discuss some recent results from our group that nicely tie in with some of the basic properties of the model outlined by Green.

In his contribution Green proposes a very interesting model of bilingual speech processing, the inhibitory control (IC) model. The model's aim is to account for the way in which bilinguals control their two language systems. Although the model was not developed to account for code-switching, the author explicitly goes into implications of his model for code-switching and this makes it very relevant for linguists working in that field. Until now, psycholinguistic aspects of code-switching have received far less attention in the literature than the syntactic aspects of code-switching. The model therefore offers an excellent starting point for incorporating insights from psycholinguistics into code-switching research and vice versa.

The commentators raise many important issues and suggest fruitful lines of development. I thank all of them for their efforts. My paper focused on how bilingual speakers control the lexico-semantic system rather than on the nature of that system. It glossed over the processes of word recognition (Carr, Dijkstra) and considered only cursorily issues such as the mapping of thought into language (Schreuder & Hermans), the role of conceptual activation in lemma selection (Kroll & Michael) and the range and nature of the semantic relationships between words (Schreuder & Hermans, Treffers-Daller, Li). I will comment on these matters as I address three concerns about the proposed control processes: (1) the nature of schemas and their role in control (Carr, de Groot, Dijkstra); (2) the notion of a tag: its psychological reality (Li, Treffers-Daller), its relationship to a language node (Dijkstra) and its role in selection (Carr); and (3) the question of inhibition as the general means of lemma selection (Carr, de Groot, Roelofs).

While it is now commonly accepted that simultaneous bilingual children can differentiate their two languages from very early in development, it is still not very well understood how they come to understand that there are two languages in their input. The purpose of this study was to examine how a bilingual child might come to an understanding of the existence of two languages in terms of pragmatic differentiation (use of the appropriate language for the context) and lexical differentiation (use of translation equivalents). Using data from a Portuguese-English bilingual child from 1;0 to 1;6, this study showed that the child showed evidence of pragmatic differentiation before lexical differentiation. If these results are replicated, they suggest that bilingual children's understanding of the appropriate social use of their two languages may lead to an understanding that the translation equivalents in their vocabulary belong to two distinct input languages.

Two experiments examine the view that the variation in bilingual short-term memory capacity is determined by differential rates of subvocal rehearsal between the languages. Auditory memory span and articulation time were measured for three bilingual groups who spoke Finnish at home and Swedish at school (FS), and either Finnish (FF) or Swedish (SS) in both the home and the school. The results of Experiment 1 indicate that memory span for words varied in a lawful manner as a function of both articulation time and language dominance for SS and FF. For FS, however, an equivalent memory span between the languages was noted despite a shorter articulation time in Finnish than Swedish. Experiment 2 found that for items with no pre-existing lexical representations (nonwords), articulation time was a more reliable predictor of memory span than language dominance for all three groups. The finding that within-language memory span was greater for short items than long items shows that, ceteris paribus, bilingual short-term memory capacity is sensitive to the effects of word length in both the dominant and non-dominant language. Taken together, these findings moderate the view that bilingual short-term memory capacity is mediated exclusively by subvocal rehearsal and indicate an influential contribution from factors related to language fluency and the strength of lexico-semantic representations.

Because the field of bilingualism is still relatively new, studies in the linguistics, psycholinguistics, language development and neurolinguistics of bilingualism have often produced conflicting results. It will be argued in this paper that some of the difficulties encountered by researchers could have been lessened, if not avoided, had close attention been paid to methodological and conceptual issues. Among the issues covered are bilingual participants, language mode, stimuli, tasks as well as models of bilingual representation and processing. Each issue is dealt with in the following way: first it is explained, then the problems it causes are discussed, and, finally, tentative solutions are proposed. Examples are taken from descriptive and experimental studies of normal bilingual adults and children as well as bilinguals suffering from aphasia and dementia.