Kees de Bot* Bilingual speech: from concepts to articulation Abstract. In this contribution the process of language production will be discussed from a psycholinguistic perspective. For the description of the process we will make use of existing ideas and models which have been developed with the monolingual speaker as the normal case. lt is argued that in bilinguals and multilinguals the multiple languages play a role at all levels of the production system, even at the conceptual system. The role of the language node as a monitoring and control device is discussed in the framework of the Multilingual Processing Model. Finally, some new developments in theories on lexical access in language production are presented which cast some doubt on existing models of the bilingual lexicon. 1. Introduction As has been argued over and over again in the last decade, a monolingual focus on language production retlects a biased perspective on the world, since the vast majority of speakers are .in fact bilingual and often multilingual, even in countries which view themselves as fundamentally monolingual. This also holds for traditionally monolingual countries like Germany and the USA, in which speaking the national language is the official norm, while in fact in both countries there are large sections of the population that either have a migrant background, or have learned one or more foreign languages in education, or are speakers of both the standard language and a dialect. Few people in small villages in Bavaria would call themselves bilingual and might even be offended by this label, but from a sociolinguistic and psycholinguistic perspective their command and use of their standard language and local dialect are prime examples of bilingualism: there are two codes th_at each have their function and pattem of usage, that are different in many respects, and that may sometimes be used in a mixed form but sometimes need to be kept strictly apart. How this processing of multiple languages in the brain takes place, is the focus of the present article and indeed of the project on "Mündliche Produktion in der Fremdsprache" reported on by various authors in this issue of FLuL. Language production rarely takes place in a situation without interaction. Speech is typically interactive in nature. Any model that claims to present a ·valid picture of the language production process should take that into account. As DAUSENDSCHÖN-GAY (in this volume [178-195]) argues, the interactive nature of the production process is of crucial importance and should be part of the models. In the model presented here, LE- Korrespondenzadresse: Prof. Kees DE BOT, chair of Applied Linguistics, _Univetsity of Groningen, Department of Language and Communication, Oude Kijk in 't Jatstraat 26, 9712 EK GRONINGEN. E-mail: c.l.j.de.bot@let.rug.nl Areas ofwork: Multilinguism, Psycholinguistics. lFLllL 32 (2003) Bilingual speech: from concepts to articulation 93 VELT's 'Speaking' model, this interactive basis is fully represented. LEVELT makes this point very clearly in the opening section ofhis monumental 1989 book: "lt is [... ] not enough to study the functions of speaking the kinds of intentional acts a language user can perform through speech, such as referring, requesting, and explaining. Nor is it enough to study the pattems of spoken interaction between interlocutors the way they engage in conversation, take tums, signal misunderstanding, and so forth [...] (... ). But they do not suffice. Developing a theory of any complex cognitive skills requires a reasoned dissection of the system into subsystems, or processing components. lt also requires a characterization of the representations that are computed by these processors and of the manner in which they are computed, as well as specification ofhow these components cooperate in generating their joint end product" (LEVELT 1989: 1). In other words, researchers who focus on lexical access do not deny the importance of interaction as the basis for production, but they limit their attention to one of the subprocesses. How interaction has an impact on word choice need not be their prime concern, since the selection criteria for lexical access are defined in a part of the system these researchers are not working on. Of course there is a risk of such a narrow view. An overwhelming part of the research on bilingual processing has focused on rather minute parts of access, without considering lexical access as a part of the larger system. This has led to a stream of publications in which e.g. the processing of pseudo-homophones (like 'coin' in French and English) has emerged as a complete subfield with its own research questions and paradigms, but increasingly little connection to larger models or theories of multilingual processing. Here the aim is to present the larget picture without losing sight of some of the more fine-grained processes that are crucial in multilingual processing. The structure of the article is as follows: first the Levelt model will be presented briefly, then some issues that are specific to multilingual processing in such a model will be discussed. Three issues will be focused on: the impact of multiple languages on utterance planning, the role of the language node, and the issue of language specific/ non-language specific access. 2. The Levelt Model Although the model has developed over time, the basis structure as described in LEVELT (1989, 1993) is still intact. The following processing components can be distinguished: A conceptualizer, in which communicative intentions are translated into messages that can function as input for the speech production system. Levelt distinguishes 'macroplanning', which involves the planning of a speech act, the selection of information to be expressed and the linearization of that information, from 'microplanning' which involves the propositionalization of the event to be expressed, the perspective taken and certain language-specific decisions that have an effect on the form of the message tobe conveyed. When a language (such as English) has a tense system, it is obligatory to specify the relevant temporal relations, even if they don't contribute to conveying the speaker's intention. lFLlllL 32 (2003) 94 Kees de Bot The output of the conceptualizer is a preverbal message, which consists of all the information needed by the next component, the formulator, to actually turn the communicative intention into speech. Crucial aspects of the model are that there is no external unit controlling the components, that there is no feedback from the formulator to the conceptualizer, and that there is no feedforward from the conceptualizer to other components of the model. This means that all information that is relevant for the 'lower' components has tobe included in the preverbal message. - Aformulator where the preverbal message is converted into a speech plan (phonetic plan) by selecting lexical items and applying grammatical and phonological rules. Lexical items consist of two parts, the lemma and the phonological form or lexeme. In the lemma the lexical entry's meaning and syntax are represented, while morphological and phonological properties are represented in the lexeme. In production, lexical items are activated by matching the meaning part of the lemma with the semantic information in the preverbal message. Accordingly, the information from the lexicon becomes available in two phases: semantic activation precedes form activation. Tue lemma information of a lexical item concerns both conceptual specifications of its use, such as pragmatic and stylistic conditions, and (morpho-)syntactic information, including the lemma's syntactic category and its grammatical functions, as well as information that is needed for its syntactic encoding (in particular: number, tense, aspect, mood, case and pitch accent). Activation of the lemma immediately provides the relevant syntactic information which in turn activates syntactic procedures. The selection of the lemmas and the relevant syntactic information leads to the formation of the surface structure. While the surface structure is being formed, the morpho-phonological information in the lexeme is activated and encoded. The phonological encoding provides the input for the articulator in the form of a phonetic plan. This phonetic plan can be scanned internally by the speaker via the speech-comprehension system, which provides the first possibility for feedback. An articulator which converts the speech plan into actual speech. The output from the formulator is processed and temporarily stored in such a way that the phonetic plan can be fed back to the speech-comprehension system and the speech can be produced at normal speed. - A speech-comprehension system connected with an auditory system which plays a role in the two ways in which feedback takes place within the model: the phonetic plan as well as the overt speech are guided to the speech-comprehension system to find any mistakes that may have crept in. In comparison to the original (1989) version of the model, the 1993 version gives more information about speech understanding, but that will not be discussed here. Suffices to say that the speech understanding is modeled as an mirror image of language production, and that the lexicon is shared. - A monitoring system that compares the intended meanings with the meanings as conveyed in speech. The solution chosen for the monitoring system is straightforward: the speaker' s own speech is perceived by the perceptual system and translated back into meaning parts that can be interpreted by the conceptual system. A crucial aspect IF[,uJL 32 (2003) Bilingual speech: from concepts to articulation 95 of this solution is that there is no in-betw.ren monitoring of subprocesses: only the resulting speech is evaluated. Language production as it is modeled here, is typically viewed as a form of skilled behaviour in terms of e.g. SHIFFRIN/ SCHNEIDER (1977) and ANDERSON (1983). This means that there is a hierarchy of sk: ills, with higher level sk: ills being subserved by lower level skills, and that lower level skills/ components will be fully automatic. Word retrieval, syntactic proceduralization, the formation of an articulatory plan and the articulatory mechanisms will take place without conscious control. In language production almost all the attentional resources go to the conceptual level, with some attention left for errorcontrol through the feedback mechanisms. Over the last decade the Levelt model has undergone a few changes. Following proposals by ROELOFS (1992) the feature based structure of concepts, i.e. concepts viewed as bundles of conceptual features, was replaced by what was called 'lexical concepts'. These are non-decomposable concepts that match with lemmas in the lexicon. As will be discussed later on, this position has recently been challenged by LAHEIJ (2003). A second change has been the introduction of the syllabary, a storage and retrieval system for syllable-sized units that contain articulatory programs for parts of words (LEVELT/ ROELOFS/ MEYER 1999). This replaces or rather complements a system in which for each lexical item individual sounds had to be assembled time and time again. The model is based in various types of data: experimental data mainly, but also spontaneous speech data and analyses of speech errors and self-corrections. More recently, data using neuro-imaging techniques have been used to validate the model. INDEFREY/ LEVELT (2000) show that there are neural correlates for the various subcomponents proposed, lending further support for the validity of the model. 3. Speech production in bilingual speakers The Levelt model was developed as a monolingual model, and if we want to apply it for the processing of multiple languages, we need to clarify to what extent the present model is capable of handling bilingual speech. PARADIS (1977, 1999) has claimed that there is nothing specific to the bilingual speaker that would call for additional components or mechanisms. With respect to code-switching, one of the aspects supposed to be unique for bilingual speech, he claims that "There is no need to postulate an anatomical localization or even a specific functional organization, other than that which every speaker already possesses and which allows him, among other things, to switch registers within the same language" (1977: 91). In their discussion of learners of a foreign language as bilingual speakers, POULISSE/ BONGAERTS (1994) mention the following factors that have tobe taken into account in a bilingual model: 1. The L2-knowledge of learners is not as complete as that of the native speaker: they know fewer words and rules, which leads to the use of compensatory strategies, and to avoidance of 'uncertain' words and structures. lFJLwL 32 (2003) 96 Kees de Bot 2. The speech is more hesitant, and contains more errors and slips, depending on level of proficiency. 3. The performance shows traces ofthe Ll, sometimes deliberately, sometimes unintentionally. According to POULISSE/ BONGAERTS (1994) the L 1-traces are very similar to substitutions and slips in monolingual speech. They also show that the incomplete knowledge base and the lack of automaticity of L2 speakers can be handled adequately by existing monolingual models. This is not the case for cross-linguistic influence (CLI) phenomena in their view. On the other hand, PARADIS (1977) claims that there is also interference in monolinguals e.g. when using words from another style in an incorrect way (informal words in formal speech). In terms of processing, CLI-phenomena cannot be distinguished clearly from code-switching-phenomena: both result from the working of the production system in an individual speaker, and the fact that CLI may sometimes be undesirable in terms of an external model of the target language is not relevant here. In the remainder of this article we will focus on three specific aspects of multilingual processing that have consequences for the model in its present form. The first is the impact of different languages on the formation of the preverbal message. The second concerns the role of the language node as a monitoring device, and the third is the way in which lexical items are selected and accessed. 4. The role of multiple languages in the conceptualizer As mentioned earlier, a distinction is made at the level of the conceptualizer between macroplanning ("the selection of information whose expression may realize the communicative goals") (LEVELT 1989: 5), and microplanning ("[...] the selection and molding of information in such a way that its expression will be an appropriate means for conveying the intention" (ibid.). In an earlier version ofthe bilingual version ofthe Levelt model (DE BOT 1992) it was proposed that the conceptualizer functions largely language-independent. Recent work on utterance planning makes it clear that that is probably not correct. VON STUTTERHEIM (1999, 2002) reports on a number of empirical studies that aim to elucidate the processes of utterance construction form both a cross-linguistic and a multilingual perspective. V ON STUTTERHEIM (2002) mentions the following steps in utterance planning: selection of information about the external world in which the utterance will be used, the segmentation of the complex knowledge structure to be conveyed, the selection of information components for verbalization, the taking of a perspective relevant for the construction of relevant discourse (selection ofreferential frames and topic domains), the linearization and interconnection of the informational units. The data that have been gathered suggest that these steps may be different for speakers of different languages. V ON STUTTERHEIM (2002) presents various types of data (film retellings, verbalization of single events, voice onset data and eye tracking data) from both native speakers of English, German and Spanish, and learners of those languages with lFLllliL 32 (2003) Bilingual speech: from concepts to articulation 97 different Ll backgrounds. Here we will discuss only one aspect from those complex and intriguing experiments, the effect of the perspective taken on reported events. For this, there appear to be clear differences between native speakers of German and English: 'While German speakers present events holistically, which means including some kind of end point of an action or motion event, English speakers tend to segment or decompose events into different phases, preferring a temporal perspective point which views an event as ongoing' (VON STUTTERHEIM 2002: 3). In other words, speakers of German tend to include the endpoint in their description while speakers of English describe the event as ongoing, without such an endpoint. This reflects differences between the grammars from these two languages: the ongoingness is expressed by verbal aspect, in particular the progressive from. So this aspectual distinction is gralillilatised in English while is it expressed with lexical expressions in German. For the description of events this means that German speakers have to wait for the endpoint of the event, while English speakers can start describing it right away. These differences come out nicely in the descriptions of the events and the Voice Onset Times (VOTs): native speakers of German systematically report more endpoints and show langer VOTs because they have to wait for the endpoint in order to describe the event. Building on this, von Stutterheim and her colleagues tested groups of very advanced learners with German Ll/ English L2 and English Ll/ German L2. The outcomes show that even those advanced learners tend to apply their Ll-based principles of information organization when speaking their L2, though there is some influence of the L2. What these researchers did not look into, is to what extend speakers who use the L2 very frequently, show signs of L2 interference in their Ll on this level. Whether the strong impact of the Ll-based principles is an effect of a deeply ingrained link between utterance structure and information structure in Ll development, or results from the more frequent use of Ll needs to be studied by looking at individuals who switch from their original Ll to another language as their dominant language. The research presented shows that also at the conceptual level multilingualism has an impact. This research has so far been restricted to bilinguals, and it will be interesting to see how speakers who have an colilliland of three or more languages deal with such potentially conflicting principles or routines. 1 5. Monitoring and control: the role of the language node As mentioned earlier, the monitor in the Levelt model had as its most important function to check whether the intended meaning was actually expressed. In multilinguals, checking for the use of the right language should also be done by the monitor. lt could be argued however, that the role of the monitor is not just to give feedback, but also to provide feedforward. In other words, the monitor could play a role in preparing the language production system for the use of one or more specific languages. In DE BOT (2003) we Additional support for the impact of L1 on conceptualization in L2 is provided by several of the papers in STARRENIDIMROTH (forthcoming). lFLIJllL 32 (2003) 98 Kees de Bot present a multilingual processing model in which a language node is included that serves both to control the output with respect to the language or languages used and to inform the production system about the language to be used. A schematic overview of the model is given in Fig. 1: Communicative Intentions + language Lexical concepts Lemmas Lexemes Syllables / sounds / gestures Syntactic procedures Fig. 1: The multilingual processing model JFLIIIL 32 (2003) Bilingual speech: from concepts to articulation 99 The basic assumption behind the model is that there are three stores with information: conceptual features, syntactic procedures and form elements (sounds, syllables or gestures).2 Within each of these stores, there are language-specific subsets. These subsets show overlap reflecting the cognateness of the languages involved. The language node controls the various processing components with respect to the language to be used. The intention to use a specific language originates from the conceptual/ communicative intention level and is relayed to both the system generating lexical concepts and the language node. For the subsequent components information on the language to be used now comes from two sources: through the lexical concepts and directly from the language node. An extemal language node system can more locally control language choice. When a particular language is called for, the language node will inform all relevant components, that is those components in which syntactic or form information needs to be selected, about the subset to be activated. This higher activation will lead to the selection of elements from the right language. Crucially, information about the activation of subsets (either through the lexical concepts or through the language mode) will be exchanged between subsets of the various languages. In other words, when the subset of syntactic procedures of language A is activated, this information will be relayed to both the language node (which will then forward that information again to the rest of the system), and t,o the subset of form-related elements from language A. So activation of a part of a language will activate elements from that same language at other levels. At the same time, because subsets overlap, elements that are shared by more than one language, may also activate the other subset (or subsets) they are part of. The language node conveys information about language selection both from the conceptual level to lower level components and between components at these lower levels. lt accumulates information of the state of activation of various languages and acts in that sense as a monitoring device which compares the intended language with the language actually used. To what extent the language node as described above and the monitor as it functions in the Levelt model overlap is unclear. DE BOT/ JESSNER (2003) argue for a more expanded role for the monitor in language use and language acquisition. They argue that the monitor is also the device that notices gaps in knowledge, e.g. the inability to express a given meaning in L2 that can be expressed in L 1, suggesting a lack of linguistic means in the L2. In that way the monitor can become part of the language learning system. A further elaboration of this role of the monitor is beyond the scope of the present article (but see SCHADE's contribution in this issue, page 104-115). 2 Following the line of argumentation presented with respect to the structure of utterances and their planning at the conceptual level, it could be argued that there are also language-specific mechanisms at the conceptual level that are controlled by the language node. lFLl.lL 32 (2003) 100 Kees de Bot 6. Lexical access in language production The above description of the language production system is more or less in line with the literature on monolingual processing, but there are trends in the literature that seem to call for a review of current positions. LAHEIJ (2003) presents an overview of more recent developments in language production models and its relevance for bilingual processing. He argues for what is named a 'complex access, simple selection' model in which all relevant information for the selection oflexical items is included in the preverbal message in such a way that only the right word will be selected, so circumventing the old 'convergence' problem in theories on language production. The preverbal message needs to contain all sorts of information about the communicative setting, addressee, topic of conversation, style and so on. In line with earlier proposals (DE BOT/ SCHREDDER 1993, PODLISSE/ BONGAERTS 1994) 'Language' is seen as a cue that has tobe included as well. In this model, concepts are featurebased, which contrasts with proposals by ROELOFS (1992) and LEVELT [et al.] (1999), who defend an non-decompositional view of concepts. LAHEIJ mak: es a distinction between message selection ('what conceptual information has tobe selected for production') and response selection ('which lexical items must be accessed'). He concludes that "lexical selection is based on a complex preverbal message that contains all relevant information to arrive at the correct word. Response selection can then be a simple process that selects one word from the set of activated words on the basis of the activation levels only" (24). This conclusion leads him to reject models of bilingual production like the ones proposed by de BOT/ SCHREDDER (1993) and GREEN (1993) which are based on mechanisms for selective attention and inhibition of language sets. In his view, there is no need for such mechanisms, because the detailed information in the preverbal message, which includes language information as a cue, will automatically select the right word in the right language. While LAHEIJ' s proposals are attractive because of their simplicity and because they meet the constraint that such models should have as few control mechanisms as possible, there are quite some problems with it. Interestingly, one is of particular relevance for the study of language attrition. This concems the problem that the conceptualizer has no information about what words are stored in the lexicon. In other words, the specifications in the preverbal message may not match with any lexical item. This is clearly the strength of postulating the existence of lexical concepts as Levelt et al. do, because these concepts are based on existing lexical items. In language attrition the decline in availability of lexical items of the attriting language is one of the most prominent features. There are many examples in both foreign language leamers (PODLISSE 1997) and rnigrants (e.g. BACKDS 1992) who appear to select a word from the other language because they cannot find the word they were looking for in the attriting language. Another problem is that research in code-switching has shown that languages are activated and inhibited as sets. MEDTER/ ALLPORT (1999) present convincing experimental evidence for this effect, and JACKSON [et al.] (2001) provide supporting neuroimaging data for this. A final problem is that LAHEIJ' s model would predict sirnilar effects for within and between language prirning, while there is sufficient evidence to show that lFlLlllllL 32 (2003) Bilingual speech: from concepts to articulation 101 in most bilinguals within language priming effects are larger than between priming effects (HERMANS [et al.] 1998). The likelihood of a mechanism that impacts on levels of activation of languages is also supported by data from spontaneous speech in migrants. CLYNE (1987) analysed speech data from Dutch and German migrants in Australia and mentions many cases of the use of English during the interviews: despite their conscious efforts to keep the other language (English) away, this slipped through frequently. This cannot be reduced to a individual item effect of frequency, because quite often the intended word in Dutch was used in the same conversation. This means that the interference and switching is not the result of the frequency effects of individual items, but from the inability to suppress elements from that other language as a set. Finally, recent analyses of data on triggering, the tendency of code-switchers to switch to the other language when they encounter a word that is highly similar in the two languages can be interpreted as evidence against Laheij 's views. BR0ERSMAIDE BOT (in prep.) looked at two corpora of code-switching data to test CLYNE's original trigger hypothesis and concluded that the word following a triggerword is likely to be switched more often as compared to non-triggerwords. The mechanisms they propose are based on feedback at the phonological level and the role of a language node that informs the language production system of the language subsets to be activated at various levels. The central question most of the research on bilingual processing is concerned with is language-specific versus non-specific access. In other words, when we are confronted with a word, e.g. in a lexical decision task, do we first access the lexicon from one language and then the next, or is there a parallel search through all languages, words not being organized primarily through language, but e.g. through frequency. In the past the language selective position was favored (GERARD/ SCARB0R0UGH 1989, MACNAMARAI KUSHNIR 1971), but in the last two decades research has accumulated to support the nonselective view. A number of experimental paradigms have been used to study this question, including the use of neighborhood effects (V AN HEUVEN/ DIJKSTRAIGRAINGER 1998), the processing of interlingual homographs (DDKSTRAIGRAINGER/ V AN HEUVEN 1999) , withinand between-language semantic priming and repetition tasks (WOUTER- SEN 1997) and eye-tracking studies (MARIAN/ SPIVEYIHIRSCH 2002). With respect to language production, much less is done. In a series of experiments by Hermans (HERMANS [et al.] 1998) using a complicated picture word interference task in which the task is to name a picture while at different SAOs (Stimulus Onset Asynchrony, the time lapse between the presentation of the prime word and the the target word or picture), words with similar meaning of similar form characteristics are presented on the screen, he showed that also for production non-selective access is the better solution, although CosTAICARAMAZZA (1999) present data which they interpret as support for a more selective access procedure. Overall, the evidence in support of the non-selective access hypothesis is substantial, and much stronger than for the selective access hypothesis. The LAHED (2003) proposal even takes this a step further by assuming that there is no language-based organization of the multilingual lexicon and therefore language does not have a special pre-selecting function as a cue in access. JFILIJIL 32 (2003) 102 Kees de Bot 7. Conclusion In this article a model for multilingual processing is presented based on the monolingual model developed by LEVELT and his colleagues. lt is argued that there is no need to make substantial changes in the monolingual model to make it work for multiple languages too. The use of more than one language appears to have an impact on processing on all levels, including the conceptual level, which in earlier proposals was taken to function languageindependently. A special role is played by the language node, a monitoring device that not only seems to check errors in output, but is also likely to play a role in activating language-specific information on different levels of the process. Finally, discussions on lexical access now seem to favor language non-specific access, and there are even proposals that state that there is no need to postulate a prestructuring of the lexicon according to language, because the information in the preverbal message is detailed enough to select the one candidate aimed at. The model described here is clearly a steady state model and not an acquisition model. For the future, we need to develop such models further and include an acquisitional device in it. References ANDERSON, John R. (1983): The architecture of cognition. Cambridge, MA: Harvard University Press. BACKDS, Ad (1992): Patterns of language mixing: a study of Turkish-Dutch bilingualism. Wiesbaden: Harrassowitz. BROERSMA, Mirjam/ DE BOT, Kees (in prep.): An Alternative for the triggering hypothesis. CLYNE, Michael (1987): "Constraints on code switching: how universal are they? " In: Linguistics 25, 739-764. COSTA, Albert/ CARAMAZZA, Alphonso (1999): "Is lexical selection in bilingual speech production language-specific? Further evidence from Spanish-English and English-Spanish bilinguals". In: Bilingualism: Language and Cognition 2.3, 231-244. DADSENDSCHÖN-GAY, Ulrich (2002): "Producing and learning to produce utterances in social interaction". In: Eurosla Yearbook 2003 [in print]. DE BOT, Kees (1992): "A bilingual production model: Levelt's Speaking model adapted". In: Applied Linguistics 13.1, 1-24. DE BOT, Kees (2003): "The multilingual lexicon: modeling selection and control". In: CENOZ, Jasone / HUFEISEN, Britta/ JESSNER, Ulrike (eds.): The Multilingual Lexicon [forthcoming]. DE BOT, Kees/ JESSNER, Ulrike (2003): Monitoring and control in multilinguals. Ms. University of Nijmegen/ University of Innsbruck. DE BOT, Kees / SCHREDDER, Rob (1993 ): "Word production and the bilingual lexicon". In: SCHREDDER, Rob / WELTENS, Bert (eds.): The Bilingual Lexicon. Amsterdam: Benjamins, 191-214. DIJKSTRA, Ton/ GRAINGER, Jonathan/ V AN HEDVEN, Walter (1999): "Recognition of cognates and interlingual homographs: The neglected role of phonology". In: Journal of Memory and Language 41, 496-518. DIMROTH, Christine/ STARREN, Marianne (eds.) (2003): Information structure and the dynamincs of language acquisition. Amsterdam/ Phildelphia: John Benjamins. GERARD, Linda/ SCARBORODGH, Don (1989): "Language-Specific Lexical Access of Homographs by lFLllL 32 (2003) Bilingual speech: from concepts to articulation 103 Bilinguals". In: Journal of Experimental Psychology: Learning, Memory and Cognition 15.2, 305- 315. GREEN, David (1993): "Towards a model of L2 comprehension and production". In: In: SCHREDDER, Rob / WELTENS, Bert (eds.): The Bilingual Lexicon. Amsterdam: Benjamins, 249-278. HERMANS, Daan / BONGAERTS, Theo/ DE BOT, Kees/ SCHREDDER, Rob (1998): "Producing words in a foreign language: can speakers prevent interference from their first language? " In: Bilingualism: Language and Cognition 1, 213-229. INDEFREY, Peter/ LEVELT, Willem (2000): "The Neural correlates of language production". In: GAZZANI- GA, Michael (ed.): The New Cognitive Neurosciences. Cambridge, MA: MIT press, 845-865. JACKSON, Georgina/ SWAINSON, Rachel / CUNNINGTON, Ross/ JACKSON, Stephen (2001): "ERP correlates of executive control during repeated language switching". In: Bilingualism: Language and Cognition 4.2, 169-178. LAHEIJ, Wido (2003): "Monolingual and bilingual lexical access in speech production: issues and models". In: KROLL, Judy / DE GROOT, Annette (eds.) [forthcoming]. LEVELT, Willem (1993): "Language use in normal speakers and its disorders". In BLANKEN, Gerhard/ DITTMAN, Jürgen/ GRIMM, Hannelore/ MARSHALL, John/ WALLESCH, Claus (eds.): Linguistic disorders and pathologies. An International Handbook. Berlin: Walther de Gruyter, 1-15. LEVELT, Willem / ROELOFS, Ardi / MEYER, Antje (1999): "A theory of lexical access in speech production". In: Behavioral-and-Brain-Sciences 22.1, 1-38. LEVELT, Willem (1989): Speaking. From Intention to Articulation. Cambridge, Mass.: The MIT Press. MACNAMARA, John/ KDSHNIR, Seymour (1971): "Linguistic independence of bilinguals: The input switch". In: Journal ofVerbal Learning and Verbal Behavior 10, 480-487. MARIAN, Viorica / SPIVEY, Michael/ HIRSCH, Joy (2002): "Shared and separate systems in bilingual language processing: converging evidence from eyetrack: ing and brain imaging". In: Brain and Language [in press]. MEUTER, Renata / ALLPORT, Allan (1999): "Bilingual language switching in naming: Asymmetrical costs oflanguage selection". In: Journal of Memory and Language 40.1, 25-40. PARADIS, Michel (1977): "Bilingualism and aphasia". In: WHITAKER, Haigonoosh/ WHITAKER, Harry (eds.): Studies in Neurolinguistics. Vol. 3. New York: Academic Press, 65-121. PARADIS, Michel (1999): Neuroimaging studies of the bilingual brain: Some words of caution. Paper presented at the 25th Lacus Forum, University of Alberta, Edmonton, August 1999. POULISSE, Nanda (1997): "Language production in Bilinguals". In: DE GROOT, Annette/ KROLL, Judy (eds.): Tutorials in bilingualism. Mahwah, N.J.: Lawrence Erlbaum, 201-224. POULISSE, Nanda/ BONGAERTS, Theo (1994): "First language use in second language production". In: Applied Linguistics 15, 36-57. ROELOFS, Ardi (1992): "A spreading-activation theory of lemma retrieval in speak: ing". In: Cognition 41.1/ 2, 107-142. SHIFFRIN, Richard/ SCHNEIDER, Walther ( 1977): "Controlled and automatic human information processing: II Perceptual learning, automatic attending and a general theory". In: Psychological Review 84, 127- 190. STUTTERHEIM, Christiane v. (1999): "How language specific are processes in the conceptualiser"? In: KLABUNDE, Ralph/ STUTTERHEIM, Christiane v. (eds.): Representations and Processes in Language Production. Wiesbaden: Deutscher Universitäts-Verlag, 153-179. STUTTERHEIM, Christiane v. (2002): "Linguistic structures and information organisation: the case of very advanced leamers". In: Plenary Address, Eurosla conference. Basel. v AN HEDVEN, Walter/ DIJKSTRA, Ton/ GRAINGER, Jonathan (1998): "Orthographie neigborhood effects in bilingual word recognition". In: Journal of Memory and Language 39, 468-483. WOUTERSEN, Mirjam (1997): The organisation of the bilectal lexicon. University of Nijmegen. lFLlllL 32 (2003)