1.1 The linguistic phenomenon, L2 & issue
focus of this report is the phenomenon of final obstruent devoicing of (L2) English
words for speakers of L1 German and Swedish. The issue of focus is transfer.
1.2 The learners and the L1s
Data was gathered from four learners; two
German and two Swedish speakers. They all learnt English as an L2 before the
end of the ‘critical period’ (Lenneberg, 1967) in instructed/academic circumstances
and were subsequently immersed in natural language. Final obstruent devoicing
behaves differently in German to Swedish. German speakers devoice all obstruents
word finally. In Swedish there is no such tendency. There are consonants which
appears voiceless due to the Swedish voicing rules, (such as intervocalic
devoicing) but this report will focus just on word final positions. The phoneme /z/ is not realised in Swedish.
Rather, the sound /s/ is used not due to final devoicing rules but the
lack of /z/ in the L1.
1.3 The research question
‘To what extent will L1 German and Swedish final (de)voicing
characteristics transfer when acquiring L2 English?’
2.1 The linguistic phenomenon in the L1 and the
In German, final obstruent devoicing occurs at the end of all words from
all word groups. For example ‘Hund’ becomes Hunt. Grijzenhout writes that
‘all syllable-final obstruents are voiceless, irrespective of the voicing
specification of the neighbouring segment’ (2000: 4). In Swedish, there is no
such phenomenon. Final obstruents contrast from voiced to voiceless. ‘Fyllt’
and ‘fylld’ are minimal pair, for example.
There is also no rule in the L2, English, leading to final obstruent devoicing.
Findings from key publications
on L1 transfer
If two languages have more differences (syntactically, morphologically or
phonologically), it will be more difficult for learners (Hayati, 1998). This is
also supported by Decherts (1983) who notes that L2 learners tend to rely on
their first language structures. If they are different, then errors occur thus
indicating an interference of L1 on L2.
Luz (1990: 73) notes Fries’ words in the foreword of
Lado’s work (1957) that “learning a second language is a very ‘different task’
from learning the first language.” The initial problems arise from the special
“set” created by the L1 habits, not the L2 itself.
Similarities are helpful and make use
of already existing knowledge… facilitating learning Ringbom (1987).
from key publication on final obstruent devoicing
Grijzenhout & Joppen (1999) propose that in phonology, final
devoicing is a ‘relevant constraint in early child speech and English children
have to learn that it is ranked low in their language’ (2000: 11). The
constraint rankings in English and German can illustrate why German speakers
learning English may have difficulty adhering to English phonological rules. With
regards to Correspondence Theory (McCarthy and Prince, 1995), where inputs
correspond outputs, Grijzenhout (2000: 13) outlines the relevant constraints:
English: IDENTSTEM voice >> AGREE
>> IDENT voice >> FINALDEV,
German: IDENTONSvoice >> FINALDEV >> IDENTvoice >> AGREE,
James and Leather (1997) mention how Swedish has also got voicing
contrasts in its obstruents, but lacks a similar rule on final devoicing. Swedish
has a constraint more similar to
English. De Lacy (2006) summarises that the constraint ‘IDENT±voice
is indispensable’ (259)
with regards to obstruent voicing assimilation. It must be active so all
voicing contrasts are retained.
2.4 Research question & hypothesis
what extent will L1 German and Swedish final (de)voicing characteristics
transfer when acquiring L2 English?’
My hypothesis is that because there are more differences in (de)voicing
conventions between English and German, the German speakers will have more
erroneous productions than Swedish speakers. Additionally, the Swedish speakers
will make most errors voicing the particular obstruent /z/ due to it appearing
as /s/ in their L1.
3. The data
of exposure in English speaking environment/ country (years)
of L2 exposure
of sample (minutes/words)
8:41 / 397
Data used in this
report was collected from interviews on YouTube. Web links to the specific
clips are in the references page. The initial steps to the cross sectional data
collection were finding German and Swedish celebrities, then reviewing their
ages/exposure to L2 English and background to learning. It was important to
control variables not being tested as much as possible to keep the data
reliable and conclusions justified. The age, type and length of exposure are
similar and consistent. However, it was
difficult to control the age at testing
variable when using YouTube sources. There was no data from the speakers at all
ages, so the most appropriate ones were selected.
Note: though Becker is younger than the other learners, his exposure by
this age is similar to the other speakers from his English speaking tutoring
and extensive time in England training. Likewise, Ekland is considerably older
but the amount of exposure (in years) is only marginally higher as she returned
to Sweden for a long period.
Once each learner file was deemed reliable, roughly
four minutes of learner speaking time was listened to and transcribed. Instances
where the speaker used word final obstruents were transcribed phonetically
before numbering those cases where it was devoiced, giving the number of
tokens. The number of erroneous productions was converted into a percentage
which allowed the results to be graphed. I also recorded the numbers of times
each particular obstruent was devoiced in order to analyse specific frequency.
Figure 1 shows the most notable feature of the results; the
German speakers used the devoicing phenomenon substantially more than the
Swedish speakers. The German mean is 42.7% erroneous production, compared to
the Swedish mean of only 8.3%. Such a considerable difference, consistent with
all learners, gives support to the hypothesis that German speakers would
produce the final devoicing variant more. There are examples of /d/, /v/, /z/,
/g/ and /b/ final devoicing in the German data. In the Swedish learner files
there are instances only of /d/ and /z/ devoicing:
Figure 2 and 3 show that /d/ and /z/ are the most commonly
devoiced obstruents for the German speakers. Becker devoices /d/ a total of 10
times; Zimmer makes the /z/ à
/s/ change 24 times. The obstruent /v/ is also devoiced in Becker and Zimmer’s
speech (respectively 26.1% and 19.2% of the devoicing tokens). With Zimmer’s we
also see instances of /g/ and /b/ devoiced to become /k/ and /p/, again
following German final obstruent devoicing patterns.
Figure 4 and 5 show the devoicing of /z/ to be most common. The
more marginal devoicing of /d/ is still present, as in German, but much less so
than /z/. This supports the second part of the hypothesis that that the Swedish
speakers will have an issue with /z/ due to it appearing as /s/ in their L1.
This could be why it is 80-90% more common than the other realised devoicing. Interestingly,
there is no devoicing of obstruents /v/, /g/ or /b/ in contrast to the German
learner files. This could be because those phonemes are realised word finally
in the L1 just as they would be in the target L2: voiced.
is a distinct difference between the frequency and amount of devoicing that
occurs between German and Swedish learners (note figure 1). Clearly, this is no
coincidence and supports the suggestions
of Hayati (1998), Decherts (1983), Luz
(1990) and Ringbom (1987), in suggesting that many problems a second language learner has with L2s result from the
interference of habits from L1. This explains why habits of final devoicing in
German have transferred to these speakers’ L2 English. Likewise, Swedish similarities
to English mean speakers do not have a new phenomenon to overcome, causing fewer
errors. As shown, the Swedish make substantially less.
As Grijzenhout (2000) anticipates, the devoicing
constraint is high in German and the transfer could mean that final devoicing
occurs frequently, despite English having this constraint ranked lower. De Lacy’s
(2006) Swedish devoicing expectation is supported as we see both voiced and voiceless final obstruents (take ‘unsophisticated’
and ‘just’ in Ekland’s first sentence) in the data. The evidence we have of
the obstruents devoiced support the literature of both the issue and
Ideas for the future
Using the same
set of L2 learners, an interesting factor to investigate is the influence of
the era. The interviews of Becker and Ulvaeus are from 1985 and 1979; the
others being more recent (2006 and 2016). The input between the two periods would
have changed drastically. The speakers in the 21st century have been
exposed to the internet and global/other English speaking platforms such as
streaming services and social media, intensifying their exposure. A study could
be conducted on the affects of language
Crystal (2004) cites Eastment (1996)
agreeing that the rise of the Internet ‘reshaped the uses of computers’ (1996:
187) for L2 learning at the end of the 20th century meaning ‘learners of a
language can now communicate…with other learners or speakers of the target
language all over the world.’ (Crystal 2004: 232).
It is clear an L1 plays a significant role in L2 learning. The more similar
that a feature is in a learner’s L1, the more likely they are to acquire with
few errors. In the case of this report, German differences and Swedish
similarities to English final devoicing tendencies mean the Swedish speakers
make more target-like productions than the German speakers.