A cognitive view of the bilingual lexicon: reading and speaking words in two languages.

Abstract

In this paper we review recent research on experimental psycholinguistic approaches to the bilingual lexicon. The focus in this work is to understand how it is that lexical access in both comprehension and production is fundamentally nonselective with respective to language, yet bilinguals are able to control the use of their two languages with relatively high accuracy. We first illustrate the nature of the data that support the claims of nonselectivity and then consider some of the factors that may modulate the resulting cross-language competition. These include differences in lexical parsing strategies across languages, in lexical cues that signal one language rather than another, in the ability to allocate cognitive resources, and in the nature of the tasks that initiate spoken production. We argue that the competitive nature of processing across the two languages of the bilingual provides an exquisite model to examine cognitive activity and its control.

Key words

bilingual comprehension bilingual production lexical access

1 Introduction

Although proficient bilinguals rarely make the error of speaking words in the wrong language or thinking that they are reading text in a language other than the one intended, recent cognitive research on lexical access in word recognition and in spoken production suggests that information about both languages is active, at least briefly, in even highly skilled tasks such as reading and speaking (e.g., Brysbaert, van Dyck, & van de Poel, 1999; Colome, 2001; Hermans, Bongaerts, de Bot, & Schreuder, 1998; van Heuven, Dijkstra, & Grainger, 1998). The absence of a simple mechanism to switch off one of the two languages when using the other makes the problem more complicated from the perspective of bilingual performance, but also more interesting from the perspective of elucidating cognitive mechanisms. For this reason, bilingualism has become an important tool for psychologists who wish to model developing systems, the competition between them, and the consequences for executive control (see Bialystok, this issue). Identifying the codes across the two languages that are activated by input from each provides important information about the degree to which language representations are constrained by the course of acquisition of open to later modification. Likewise, understanding how bilinguals negotiate the inherent nonselectivity of lexical access provides critical evidence for models of selective attention and control.

In this paper we first review the evidence on lexical access in bilingual word recognition and production. Because a number of recent papers and chapters provide extensive reviews of this material, we summarize these findings only briefly (see Brysbaert, 1998; Dijkstra & van Heuven, 2002; Kroll & Dijkstra, 2002; Kroll & Dussias, 2004). We then consider briefly the implications of the findings that we review for models of comprehension and production and examine some of the factors that we believe may be informative with respect to identifying the locus of language selection. These include linguistic information that provides cues to language membership, such as language-specific codes, differences in lexical parsing preferences, the influence of language context and nonlinguistic information that provides cues in the perceived environment of the bilingual. The latter includes the form of the tasks that initiate reading and speaking and the ability of the language user to allocate attention and memory resources to the intended language.

1.1 Lexical access in bilingual word recognition

Psycholinguistic studies of reading and word recognition within the native language have shown that during this process information becomes active not only for the target word itself but also for other words that share aspects of lexical form with the target word. In this sense, word recognition appears to be characterized as a parallel process in which information at different levels interacts until a single lexical candidate emerges (e.g., McClelland & Rumelhart, 1981). What happens then when the reader is a bilingual? Recent studies of bilingual word recognition have extended the demonstration of parallel activation of lexical form representations to information associated with each of the bilingual's two languages. Thus, a Dutch-English bilingual reading words in English, the second language (L2), will briefly activate orthographic and phonological codes associated with lexical candidates in Dutch, although English is the target language (e.g., Dijkstra, Grainger, & van Heuven, 1999; Dijkstra, van Jaarsveld, & ten Brinke, 1998; van Heuven et al., 1998). Even more surprising is that similar effects are observed when the more dominant first language (L1) is the target language (e.g., Jared & Kroll, 2001; Jared & Szucs, 2002; van Hell & Dijkstra, 2002) and when the bilingual's two languages do not share the same alphabetic or orthographic form (e.g., Gollan, Forster, & Frost, 1997).

Word recognition can be characterized by a set of signature empirical results. In general, words are recognized more rapidly when they are frequent (Forster & Chambers, 1973) and when the mapping of spelling to sound is unambiguous (e.g., Jared, McRae, & Seidenberg, 1990). The presence of alternative pronunciations for the same orthographic input has been hypothesized to lead to increased processing time because lexical and/or sublexical codes compete and the process of resolving that competition is assumed to take time. Likewise, word recognition is sensitive to the number of orthographic neighbors that resemble the target word. In languages like English that have deep orthography and many exception words, larger orthographic neighborhoods typically facilitate response times. In languages like Spanish or Dutch that have shallow orthography, larger orthographic neighborhoods typically inhibit response times (see Andrews, 1997, for a review).

The logic of research on bilingual word recognition has been to exploit the cross-language similarity of words as a means to determine whether these signature effects in word recognition are observed across languages as well. To illustrate, in languages that share the same alphabet and aspects of the orthographic representation, there are often words that resemble one another. Sometimes these words also share the same meaning, in which case they are considered cognates (e.g., bed in English and Dutch). In other cases they correspond to different meanings and are considered homographs or false friends (e.g., room in English means cream in Dutch). To determine whether word recognition for the bilingual is selective with respect to language, past studies have manipulated these cross-language properties to determine whether the unintended version of the word is also activated even when the task requires recognition in one language only. For example, Dijkstra et al. (1998) reported a series of experiments in which Dutch-English bilinguals performed variants of the lexical decision task (i.e., is the input letter string a real word?). When the task was purely in English, the L2 for these bilinguals, there was little evidence of activation of the alternative sense of homographs but there was significant facilitation for cognates relative to controls. When the lexical decision task was modified by including real word distractors from the non-target native language, there was a clear inhibitory effect for the homographs relative to unambiguous controls. Moreover, the word frequency effects that are typically observed in these tasks appeared to modulate the magnitude of homograph interference so that it was particularly difficult for a Dutch-English bilingual to accept an English word that was low frequency in English but high frequency in Dutch. When the task was then modified into generalized lexical decision, with the instruction to accept any letter string that is a real word in either language, the homographs produced facilitation relative to controls. In all, this pattern of results suggests that the bilingual lexicon is fundamentally nonselective with respect to language.

The results of other studies using different aspects of cross-language similarity converge on the same conclusion. When bilinguals read words in one of their two languages, information about the orthography, phonology, and meaning of words in the other language becomes active. The extent of this language nonselectivity appears to depend on a range of factors that influence the relative activation of the two languages and the manner in which the output of the lexical system itself interacts with the goals that are instantiated for a particular task (see Dijkstra & van Heuven, 2002; von Studnitz & Green, 2002).

1.2 Lexical access in bilingual word production

Finding that word recognition is language nonselective is not surprising given the bottom-up nature of early perceptual processes. What is more counter-intuitive is that similar evidence for nonselectivity has been demonstrated in language production, where spoken words are the result of a process that is initiated by a conceptually-driven event. Logically, it would seem that the intention to speak an idea or to name a pictured object in a specified language should be under the control of the speaker. However, even under these circumstances, there is evidence that the intention to speak a word in one language only does not eliminate the activation of related words in the unintended language. Like word recognition, the evidence for nonselectivity in production is based on demonstrations of cross-language activity. However, unlike word recognition, the top-down conceptually driven nature of production makes it more likely that the competing alternatives that are active in the nontarget language consist of words related to the meaning of the intended utterances, rather than words that share lexical form.

Much of the research on spoken word production in bilinguals has used a cross-language version of the Stroop (1935) task in which a picture has to be named in one language and a distracting word is presented at some point before, during, or after the picture's presentation. Many past studies on lexical access in language production within a single language have used the time course of these distractor effects to map out the stages of production and their associated time course (e.g., Levelt et al., 1991; Schriefers, Meyer, & Levelt, 1990). The typical finding is that semantically related distractors produce interference in picture naming, whereas phonologically related distractors produce facilitation. The issue for bilingual production then is whether the same effects observed within language are also observed across languages. If lexical access is selective, then distractors in the nonselected language should have little effect. …