An Introduction to Applied Linguistics

participate, or they may occur unintentionally, for example when an L1 
word is accidentally accessed instead of an intended L2 word. Poulisse and 
Bongaerts (1994) argue that such accidental switches to the L1 are very similar 
to substitutions and slips in monolingual speech. In addition to such code 
switches, L2 speech also contains traces of the L1 which are due to transfer or 
cross-linguistic influence.
*Aphasia is the condition where language centres of the brain have been physically damaged through 
illness or accident. In a ‘tip-of-the-tongue’ state, a person is trying to remember a word, but cannot quite 
recall the complete word form. The individual is likely to remember some elements of the form, however, 
such as the number of syllables in the word.

---

128 An Introduction to Applied Linguistics
Poulisse (1997) argues that the incomplete L2 knowledge base and the lack of 
automaticity of L2 speakers can be handled adequately by existing monolingual 
production models, but that the occurrence of L1 traces in L2 speech poses 
problems for such models. Paradis (1981), on the other hand, claims that neither 
switches to the L1 nor cross-linguistic influence phenomena call for adaptations 
of existing models. Paradis (1981) claims that a phenomenon which is very similar 
to cross-linguistic influence is operating in monolingual speech production, for 
example, when monolinguals use words from another style in an incorrect way 
(informal words in formal speech). In terms of processing, Paradis (1981) argues, 
cross-linguistic influence 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 cross-linguistic influence may sometimes be 
undesirable in terms of an external model of the target language is not relevant here.
Keeping Languages Apart
Psycholinguistically, code-switching and keeping languages apart are different 
aspects of the same phenomenon. In the literature, a number of proposals have 
been made on how bilingual speakers keep their languages apart. Earlier proposals 
suggested that there were ‘switches’ controlling the input and output of different 
languages, but these have been abandoned for models based on activation spreading. 
On the basis of research on bilingual aphasia, Paradis (1981) has proposed the ‘sub-set 
hypothesis’, which, it is claimed, can account for most of the data found. According 
to Paradis (1981), words (or syntactic rules or phonemes) from a given language 
form a sub-set of the total inventory. Each sub-set can be activated independently. 
Some sub-sets (for example, from typologically related languages) may show 
considerable overlap in the form of cognate words. The sub-sets are formed and 
maintained by the use of words in specific settings: words from a given language 
will be used together in most settings, but in settings in which code-switching is the 
norm, speakers may develop a sub-set in which words from more than one language 
can be used together. A major advantage of the sub-set hypothesis is that the set of 
lexical elements from which a selection has to be made is reduced dramatically as a 
result of the fact that a particular language or sub-set has been chosen. Our claim is 
that the sub-set hypothesis may explain how languages in bilinguals may be kept 
apart, but not how the choice for a given language is made.
According to the sub-set hypothesis, bilingual speakers have stores for lemmas, 
lexemes, syntactic rules, morpho-phonological rules and elements, and articulatory 
elements that are not fundamentally different from those of monolingual speakers. 
Within each of these stores there will be sub-sets for different languages, but also 
for different varieties, styles and registers. There are probably relations between 
sub-sets in different stores, that is, lemmas forming a sub-set in a given language 
will be related to both lexemes and syntactical rules from that same language, 
and phonological rules from that language will be connected with articulatory 
elements accordingly.
Language Choice
Returning to the model, we will now discuss how language choice is implemented. 
In speaking, the step which is probably most crucial is the matching of chunks from 
the pre-verbal message with the meaning part of lemmas, because here the transition 

---

129
Psycholinguistics
from (language-independent) conceptualization to language-specific coding takes 
place. In Levelt’s description, the lemma consists basically of three parts: a semantic 
specification, syntactic information and a pointer to a particular lexeme.
The semantic specification is ‘the set of conceptual conditions under which the 
lemma can be appropriately used’ (Levelt, 1993: 4), which is matched with a chunk 
from a pre-verbal message. It is likely that in lexical retrieval a single concept can 
temporarily activate more than one semantically related lemma, which suggests 
that the lemma store is organized according to semantic principles.
The syntactic information refers to the syntactic category of a lemma and 
its grammatical functions. When a lemma is activated, its particular syntactic 
environment is defined as well: for example, the verb sell will involve a subject, 
an object and a prepositional phrase. Other lemmas will be labelled as ‘recipient’ 
or ‘agent’. The lemmas that have been activated will ‘search’ for other lemmas 
that fit, that is, the verb will ‘search’ for a subject (and sometimes a direct object/
indirect object). ‘Grammatical encoding is like solving a set of simultaneous 
equations: the surface structure must be such that for all lemmas the required 
syntactic environments are realized’ (Levelt, 1993: 4).
The third type of information in the lemma is a pointer to a lexeme. Lexemes 
contain the phonological specifications of a lemma and the morphological 
makeup, although the exact relation between the lemma and the lexeme is not 
entirely clear.
Thus there are a number of steps in the process of lexical access where choices 
have to be made. When choosing lemmas, Poulisse and Bongaerts (1994) argue that 
‘language’ is one of the features used in the selection process. So, for the selection of 
the lemma ‘boy’, not only do the semantic features ‘male’ and ‘young’ have to match 
relevant conceptual information in the pre-verbal message, but, for a bilingual speaker 
who has English as one of his languages, the lemma ‘boy’, will also need to contain 
information about which language it belongs to (English) and this information will 
have to match the language cue in the pre-verbal message. Translation equivalents 
such as ‘boy’ and ‘jongen’ (Dutch) show considerable overlap in their semantic 
specifications, but differ mainly with respect to the ‘language’ feature.
In the preceding sections we gave a short description of the production model that 
represents the state of the art at the moment. However, many aspects of bilingual 
processing are still unclear. One has to do with ‘timing’: to what extent is the precise 
timing of the sub-processes in our production system (as measured in milliseconds) 
based on characteristics of our L1? Do languages that are structurally different require 
different internal timing between the sub-processes? If so, might there be a mis-
match between the timing in place due to the L1 and the timing required to use an 
L2 effectively? A great deal of current research on bilingual language production also 
shows that it is virtually impossible for bilinguals to selectively activate candidates 
in the target language alone (for example, Costa, 2005). This observation suggests 
that a mechanism other than the intention to speak one language only must be in 
place to permit control over production. In the section that follows we consider 
the implications of cross-language activation during speech planning for models of 
speech production and for cognitive control more generally.
Experimental Studies of Language Production in L1 and L2
Compared to research on language comprehension, there has been less 
experimental research on language production. One important reason is, no 

---

130 An Introduction to Applied Linguistics
doubt, that the kind of careful manipulations of the stimuli that may be made 
in comprehension studies, as described later on, cannot be done as directly in 
language production. In studies of comprehension, a word, sentence or text can 
be presented and we can examine the way in which processing reflects its structure 
and meaning. However, it is much more difficult to elicit speech with particular 
characteristics, even in response to a simple picture or scene.
Recent studies have used a set of experimental tasks to constrain the words that 
speakers produce in order to investigate the planning of utterances in real time. 
For example, in a simple picture-naming task, participants are shown a picture 
of a drawing and asked to speak the name of the picture aloud as quickly and 
as accurately as possible. By measuring the time to begin to speak the picture’s 
name in L1 or L2 it is possible to infer the bilingual’s relative proficiency in the 
two languages. Typically, even proficient bilinguals are faster to name pictures 
in L1 than L2. However, the time difference alone does not reveal the source 
of the language difference. One possibility is simply that bilingual speakers are 
slower to access the phonology of L2 than L1 and therefore they are slower on 
any production task in L2. Evidence from single-word translation and word- 
naming (Kroll and Stewart, 1994) is consistent with this view, although the delay 
in L2 naming relative to L1 appears to be greatest in production tasks which 
require lexicalization, that is, the selection of a single word on the basis of initial 
conceptual activation. Thus, an alternative account is that L2 is not only slower 
to speak but also harder to select for output. L2 lemmas may be more weakly 
activated than the corresponding L1 lemmas or they may be more vulnerable 
to competition from the more active L1 alternatives. Green (1998) proposed an 
‘inhibitory control model’ in which L1 lemmas are suppressed to allow bilinguals 
to speak words in L2. A focal issue in this area of research is to understand the 
source of this control. Does it arise from within the processing dynamics of the 
lexicon itself? Or is it externally imposed by general cognitive mechanisms that 
modulate the allocation of attentional resources as a function of the task and 
context in which it is placed?
The main empirical approach to language production in monolinguals has been 
to examine the pattern and timecourse of interference effects in a variant of the 
picture-naming task known as picture–word interference. A picture is presented 
to be named, just as in the simple picture-naming task, but now a word distractor 
is also presented, either visually or auditorily, and the participant is instructed to 
ignore the word and name the picture. By varying the time at which the word 
is presented relative to the picture (before, during or after the picture) and the 
relation of the word to the picture’s name (whether the word is identical to the 
picture’s name, phonologically or semantically related to the name, or completely 
unrelated), it is possible to infer the nature of the processes that must have been 
operating at different moments in time prior to speaking. The results of studies 
taking this approach have shown that semantically related distractors appear 
to produce the greatest effects early in the process of planning the picture’s 
name, whereas phonological effects are largest later in planning, although they 
are sometimes observed at earlier points as well (Schriefers, Meyer and Levelt, 
1990; Levelt et al., 1991; Starreveld and La Heij, 1995; Levelt, Roelofs and Meyer, 
1999; Starreveld, 2000). At a general level, the empirical results of these time 
course studies support the claims of production models such as the one outlined 
above in suggesting that first the meaning of the intended utterance needs to be 
established, and only later can the form of the utterance be planned. However, 

---

131
Psycholinguistics
there has been a great deal of debate about the fine tuning of this process. Some 
studies (Jescheniak and Schriefers, 1998; Peterson and Savoy, 1998) have shown 
that concepts that can be named in two alternative ways (for example, close 
synonyms such as couch versus sofa in English) compete with each other for quite 
a long time during speech planning, to the point where the phonology of both 
alternatives appears to be active.
Although close synonyms may be the exception rather than the rule for 
monolingual speakers, for individuals who speak more than one language, the 
situation may be more complicated because translation equivalents may actively 
compete for selection. A number of studies have examined this issue in bilingual 
speakers using the picture–word interference task described above (Hermans, 
Bongaerts, de Bot and Schreuder, 1998; Costa, Miozzo and Caramazza, 1999; 
Hermans, 2000). Although they come from different theoretical positions, the 
empirical results that they report converge closely. Perhaps most significant is 
that they find evidence for cross-language semantic interference. That is, picture-
naming in either of the bilingual’s two languages is slowed when a semantically 
related word is presented, regardless of whether or not the word is in the language 
they are about to speak. This observation suggests that lemmas in both languages 
are active during speaking.
The question in subsequent research has been whether activation reaches the 
level of the phonology and, if so, how the bilingual avoids making the error or 
speaking the word in the wrong language. A number of past studies have provided 
evidence that the phonology of a picture’s name in the language not in use is on 
the tip of the bilingual’s tongue. For example, naming a picture whose name is 
a cognate in the bilingual’s two languages is faster than naming a picture with a 
non-cognate translation, suggesting that activation of shared phonology facilitates 
naming (for example, Costa, Caramazza and Sebastián-Gallés, 2000). Furthermore, 
phonological facilitation for naming cognate pictures is not restricted to bilinguals 
whose two languages share the same written form. Japanese–English bilinguals 
produce cognate effects that are identical to those observed for Spanish–English 
bilinguals (Hoshino and Kroll, 2008).
If under some circumstances there is activation of the phonology of the 
language not being spoken, then how does a bilingual eventually select the 
intended language alternative? Two types of explanation have been considered. 
The language specific model, following the work of Costa et al. (1999), suggests 
that the bilinguals establish a mental firewall of sorts in which lexical alternatives 
in both languages become active but attention is directed only to the selection 
of candidates in the target language cued for selection. In this view, the language 
cue becomes critical in establishing the basis on which the separation between 
the two languages functions. The alternative, following Green’s (1998) Inhibitory 
Control model, is a competition-for-selection model, in which all activated 
candidates compete for selection, requiring eventual inhibition of alternatives in 
the unintended language (see Kroll, Bobb, Misra and Guo, 2008 for a detailed 
review of these alternatives). Although the debate on this issue is ongoing, an 
exciting new development is the use of neuroscience methods, including both 
Event Related Potentials (ERPs) and functional Magnetic Resonance Imaging 
(fMRI) to investigate the time course and localization of brain function during 
bilingual speech planning (for example, Abutalebi et al., 2007; Christoffels, Firk 
and Schiller, 2007; Verhoefs, Roelof and Chwilla, 2009). Like the behavioural 
evidence, the neuroscience evidence provides mixed support for the two models.

---

132 An Introduction to Applied Linguistics
Illustrative Research on Second Language Acquisition 
and Bilingualism
The Non-selective Nature of Lexical Access
The topic of selectivity of lexical access mentioned above in the discussion of 
language production research is also a key issue in understanding how knowledge 
of the bilingual’s two languages is organized and accessed, in particular for 
understanding the role of the L1 during L2 acquisition. Early research on this 
issue suggested that lexical access was indeed selective by language. One approach 
to this problem was to ask bilinguals to make lexical decisions about letter strings 
that might be words in one or both of their languages. In the lexical decision 
task, letter strings are presented and the participant must decide whether they 
are real words or not. On some trials the letter strings form real words but on 
others they are non-words that are possible but not actual words. The participant 
must make the decision as quickly as possible and indicate his or her response 
by pressing a ‘Yes’ or ‘No’ button. Gerard and Scarborough (1989) used the 
lexical decision to test the selectivity of lexical access by having English–Spanish 
bilinguals judge whether letter strings were real words in their L2. The condition 
of interest consisted of interlingual homographs or false friends – words that exist 
in both of the bilingual’s languages, but that have different meanings in the two 
languages. For example, in Spanish the word red means ‘net’, whereas in English 
the same letter string refers to a colour. If lexical access is selective then it should 
be possible for a bilingual to retrieve only the language-appropriate reading of 
the homograph. Gerard and Scarborough (1989) found support for the selective 
hypothesis because bilinguals were able to accept an interlingual homograph as a 
real word as quickly as a control word that was exclusively a word in one language 
only. That is, it appeared that the non-intended reading of the word did not affect 
processing, suggesting that it was unavailable.
Subsequent research has challenged the conclusion that lexical access is selective 
by language. In a series of studies that initially used a slightly modified version of 
the above procedure, Dijkstra, Van Jaarsveld and Ten Brinke (1998) manipulated 
the presence of words in the L1 in the task. When the L1 was required to be active, 
there was significant interference for the interlingual homographs relative to 
their control ‘words’. The result suggests that when the non-target language was 
sufficiently active, the alternative reading of the word was also available. In the 
same series of studies, when the task was changed from English lexical decision to 
generalized lexical decision, with a ‘Yes’ decision indicating that the letter string 
is a real word in either English or Dutch, there was facilitation for the interlingual 
homographs relative to control ‘words’, suggesting again that both readings of the 
word were active.
Subsequent research has supported the claim that lexical access is language 
non-selective in comprehension (Dijkstra and Van Heuven, 1998; Dijkstra, Van 
Heuven and Grainger, 1998; Dijkstra, Grainger and Van Heuven, 1999; Dijkstra, 
de Bruijn, Schriefers and Ten Brinke, 2000; de Groot, Delmaar and Lupker, 2000; 
Jared and Kroll, 2001; Marian and Spivey, 2003; see Dijkstra, 2005 for a review). 
One aspect of these results that may seem a bit surprising from the perspective of 
L2 acquisition, is that these studies have almost all examined the performance of 
highly proficient bilinguals. Even skilled bilinguals appear to be unable to control 
the consequences of activating information in the unintended language, at least 

---

133
Psycholinguistics
in these out-of-context word recognition tasks. One implication is that learners 
may be even more vulnerable to the consequences of the effects of L1 lexical 
form on processing in L2. Furthermore, we might ask whether these findings are 
confined to bilinguals for whom the two language share orthographic properties. 
Will Hebrew–English or Chinese–English bilinguals also show evidence for non-
selective access? The few studies that have examined these effects across languages 
that do not share the same alphabet or script suggest that there are still persistent 
interactions attributable to shared phonology (Gollan, Forster and Frost, 1997; 
Jiang, 1999).
Developing Lexical Proficiency in a Second Language
If competent bilinguals activate lexical forms in both languages when presented 
with information in one language alone, then what about learners? Relatively few 
studies have taken a developmental approach to this issue to ask how the nature 
of activated lexical information changes with increasing proficiency in L2. The 
few that have compared performance across proficiency groups have observed 
differences consistent with the view that initially the high degree of activation of 
L1 influences processing in L2, but that effects of L2 on L1 that can be obtained 
with competent bilinguals are less likely to be seen (Bijeljac-Babic, Biardeau and 
Grainger, 1997; Jared and Kroll, 2001).
The main focus in psycholinguistic research on the development of L2 expertise 
has instead been on the availability of the L1 translation equivalent during L2 
processing. An important paper by Potter, So, van Eckardt and Feldman (1984) 
used the comparison between picture naming and single word translation as a 
means of determining whether bilinguals were able to access concepts directly 
for L2 or whether access proceeded through the L1 first. Potter et al. (1984) 
observed similar picture naming and translation performance and concluded that 
bilinguals conceptually mediate L2 without L1 influence. However, a subsequent 
series of studies (Kroll and Curley, 1988; Chen and Leung, 1989) showed that the 
pattern of results depended on the level of L2 proficiency. The results for skilled 
bilinguals replicated the findings of Potter et al. (1984), suggesting that at this 
level of proficiency concepts can be accessed directly for L2. However, the results 
for L2 learners suggested that at earlier stages of L2 development there was indeed 
lexical mediation whereby L1 translation equivalents were activated to facilitate 
access to concepts.
Subsequent research has considered the implications of this developmental 
course, for example, is the early reliance on L1 something that one outgrows 
when one gains sufficient knowledge and automaticity in L2? Kroll and Stewart 
(1994) reported a set of results which suggest not (and see Thierry and Wu, 
2007, for related ERP evidence on the activation of the translation equivalent 
in proficient bilinguals). They showed that the performance of even a group 
of highly proficient Dutch–English bilinguals revealed the use of direct lexical-
to-lexical connections to perform translation from L2 to L1. When bilinguals 
translated words from L1 to L2, there were strong effects of a semantic variable, 
whether the words appeared in lists that were organized by semantic category 
or randomly mixed. However, when they translated from L2 to L1, there were 
no apparent effects of the semantics of the list, suggesting that they were able to 
bypass conceptual processing in this direction of translation. These findings have 
been a focus of debate because other studies suggest that conceptual processing 

---

134 An Introduction to Applied Linguistics
is directly available for L2 for both proficient bilinguals and learners (La Heij, 
Kerling and Van der Velden, 1996; Altarriba and Mathis, 1997).
A recent study by Sunderman and Kroll (2006) provided evidence for early access 
to semantics for L2 learners but, at the same time, reliance on the translation 
equivalent during the initial stages of L2 learning. The results of that study suggest 
that the role of the L1 translation during L2 learning may be more complex than 
initially understood. It will remain for future research to map out a complete 
account of lexical development that traces the role of the L1 translation equivalent 
(see also the Hands-on Activity at the end of this chapter for an opportunity to 
consider these issues in more detail).
The costs of code switching
One of the research areas that has provided us with rich data on bilingual language 
production is code switching. Code-switching (CS) is defined here as the use of 
more than one language in an utterance. There is a wealth of data and literature on 
CS and the focus has gradually moved from a primarily linguistic approach (how 
do languages as formal systems interact) to a more psycholinguistic approach 
that focuses on the mechanisms of language processing involved. Here we limit 
ourselves to two aspects that have drawn researchers’ attention. One is the issue 
of ‘switching costs’: the costs in terms of time and effort involved in switching 
between languages, the other is the study of CS in interaction.
Switching costs
In 1999, Meuter and Allport published an article that triggered an extensive 
discussion on the mental costs involved in CS. In an experimental task, they 
had bilingual speakers produce words that had to be produced in one of two 
languages, depending on visual cues. Their most important findings were that 
switching does cost time, and that is takes more time to switch from the weaker 
language into a stronger language than the other way around. Their explanation 
was that more effort is needed to inhibit the stronger language and that it 
takes accordingly more time and effort to reactivate again. Jackson, Swainson, 
Cunnington and Jackson (2001), Christoffels et al. (2007) and Verhoef et al. 
(2009) provide neuro-imaging data that seem to support Meuter and Allports’s 
experimental findings.
More recently, Abutalebi et al. (2007) studied CS and switching costs in an Event-
Related Functional Magnetic Resonance (er-fMRI) study. This technique provides 
data on both the timing of brain activity and the location. Participants listened 
to sentences in their L1 and L2 with switches between languages. There were 
four types of switches: semantically acceptable or unacceptable, and syntactically 
acceptable or unacceptable.
An example of an acceptable switch between Italian and French they present is: