the two corpora. We found that there was considerable uniformity in the
Welsh-English data (i.e. the ML was always Welsh) and this ruled out the
possibility of using speaker-based variables to account for differences between
speakers. In the Spanish-English data there was more variation in the choice
of ML, allowing for the possibility that this could be related to speaker-based
variables. However, none of these variables turned out to be significant
statistically.
We then went on to consider the role of particular community character-
istics in determining (a) the uniformity of the ML in the Welsh-English
community and (b) the relative flexibility in the choice of the ML in the
Spanish-English community. The main differences between the two com-
munities as shown by the questionnaire results were that (a) English tended
to be acquired later in Miami than in Wales; (b) a higher level of proficiency
in English was reported by our speakers in Miami than by our speakers in
Wales (but we suggested that the real situation was the reverse); (c) of the
speakers we recorded, more had received English-medium education in
Miami than in Wales; (d) self-perceptions of identity were more homoge-
neous in the Welsh-English community, and (e) social networks were ori-
ented more to one language (Welsh) in the Welsh-English community
whereas close contacts in the Spanish-English community were more bal-
anced between the two languages. Considering first the role of English in the
two communities, it looks as though possibly superior English proficiency in
Wales may be counterbalanced by a bigger role for English in education in
Miami. This may provide a partial explanation for why English is chosen as
an ML at least some of the time in Miami. So why is English not chosen as
an ML in Wales, despite early English acquisition and proficiency in English?
Despite the fact that some factors such as age or gender have been shown to
affect language choice (Chambers, 2002), not all of them have affected the
choice of ML. Here identity, medium of education and social networks may
play a role. The fact that the Welsh-English speakers involved in the study
consider themselves to have a mainly Welsh identity, received education
mostly through the medium of Welsh, and also address their closest contacts
mostly in Welsh could account for their virtually exclusive choice of Welsh
as a matrix language. In the Miami sample, identity is less uniform, and
hence a uniform ML is not predicted. Further research is needed to determine
whether American identity favours the choice of English ML by individuals,
whereas Latin American identity favours the choice of Spanish ML.
It is also possible that language-internal factors may play a role. There
may be a universal tendency to select one ML as suggested by Chan (2009).
It could be argued that it would be more parsimonious for bilinguals to
choose one language as the ML unless the two languages concerned have
similar word order. While Spanish and English have similar order (SVO), the
word order of Welsh (VSO) and English (SVO) contrasts. Hence, inter-clausal
switching between a VSO and an SVO language may be dispreferred. If this
136 Part 3: Bilingual Language Use