node in the non-target language receives activation from two sources –
the picture and the distractor word – and becomes highly activated as a
competitor. On the other hand, Hermans et al. (1998) argue for language-
nonspecific selection based on the result that when the distractor word was
phonologically related to the picture’s name in the non-target language (e.g.
the Dutch distractor berm ‘verge’ which sounds like the Dutch word berg
‘mountain’ for the English picture name ‘mountain’), there was interference
rather than facilitation. The results from these studies suggest that different
manipulations can support either the language-specific selection mecha-
nism or the language-nonspecific selection account despite using a very
similar experimental paradigm.
Another paradigm that has been often used in past behavioral studies on
bilingual production is language switching, in which bilinguals are cued to
speak in one of their two languages in a mixed language sequence. Past
research on language mixing/switching has shown that language mixing
affects L1 and L2 word production differently if participants are not equally
proficient in their languages (e.g. Kroll et al., in preparation; Meuter &
Allport, 1999; but see Finkbeiner et al., 2006). Meuter and Allport (1999)
found that there was a greater switch cost from L2 to L1 than from L1 to L2,
which suggests that the suppression of L1 during L2 production requires
more effort than the suppression of L2 during L1 production. The greater
suppression of L1 when speaking L2 would make it harder to reactivate the
L1 on the subsequent trial and produces an asymmetrical pattern of switch
costs. This finding supports language-nonspecific selection models in which
bilinguals need to inhibit the activation of the non-target language. In con-
trast, balanced bilinguals produce a symmetrical pattern of switch costs
(Costa & Santesteban, 2004). This finding itself can be understood as evi-
dence for an inhibitory control mechanism for language-nonspecific selec-
tion in which the degree of inhibition is a function of the relative activation
of the two languages. Because balanced bilinguals are equally proficient in
their L1 and L2, the amount of suppression that is necessary to speak in the
intended language is similar for L1 and L2, which results in symmetrical
switch costs. However, a critical result in their study was that balanced bilin-
guals also produced symmetrical switch costs when switching between L1
and L3, although L3 is not nearly as active as L1 or L2. If the amount of sup-
pression were the account for the symmetrical switch cost from L3 to L1 in
balanced bilinguals, then the balanced bilinguals should have shown an
asymmetrical switch cost when they switched between L1 and L3. Based on
these results, Costa and Santesteban claimed that balanced bilinguals who
have attained high proficiency in two languages can apply a language-spe-
cific selection mechanism in which no inhibition is involved, not only to
their highly proficient L1 and L2 but also to their less proficient language,
L3.^2 Such findings have clear implications not only for theories of bilingual
language production and transfer (see Chapters 1 to 4), but also for studies
Language Selection During Speech Production in Bilingual Speakers 205