bastien!de!clercq! the!productionof!nonnative!vowels...
TRANSCRIPT
Faculteit Letteren & Wijsbegeerte Academiejaar 2011-‐2012
Bastien De Clercq
The production of nonnative vowels by early learners: A study on phonological elicitation methods for children
Master in Advanced Studies in Linguistics
Masterpaper voorgedragen tot het bekomen van de graad van
Master in Advanced Studies in Linguistics
Promotor: Dr. Ellen Simon
Vakgroep Taalkunde
Acknowledgements
I would like to express my gratitude for the help and support the following people
have given me while writing my thesis. First and foremost, I would like to thank my
supervisor, Dr. Ellen Simon, for her continuous theoretical and practical advice,
without which this thesis would not have been possible.
I am also indebted to the administration, the teaching staff and, last but not least, the
pupils of the Sint-‐Janscollege, who were so kind as to help me set up the
phonological experiments and gather the data for my thesis.
Many thanks are also due to the native speakers of English for their cooperation to
my study and for their enthusiasm for my research.
Finally, I would like to thank my parents and Elien for their keen interest in my
thesis.
3
Table of Contents
1. INTRODUCTION ........................................................................................................................... 5
2. BACKGROUND .............................................................................................................................. 8
2.1 ENGLISH IN THE EUROPEAN UNION .............................................................................................. 8 2.1.1 The inner, outer and expanding circles of English .................................................................... 8 2.1.2 English in Belgium and Flanders ...................................................................................................... 9
2.2 ACQUIRING A SECOND LANGUAGE ............................................................................................... 11 2.2.1 Language-‐dependent factors in SLA .............................................................................................. 11 2.2.2 Learner-‐dependent factors in SLA ................................................................................................. 13 2.2.3 Foreign accent, comprehensibility and intelligibility ............................................................ 14
2.3 DUTCH AND ENGLISH PHONOLOGY ............................................................................................. 17 2.3.1 The Dutch and English vowel systems ......................................................................................... 17 2.3.2 The perception and production of the English vowels /ɪ,ɛ,æ/ by Dutch speakers ... 19
2.4 METHODOLOGICAL CONSIDERATION .......................................................................................... 21
3. EXPERIMENT 1: READING ...................................................................................................... 23
3.1 METHODOLOGY ........................................................................................................................... 23 3.1.1 Participants .............................................................................................................................................. 23 3.1.2 Materials .................................................................................................................................................... 24 3.1.3 Procedure .................................................................................................................................................. 25 3.1.4 Analysis and evaluation ...................................................................................................................... 25
3.2 RESULTS ....................................................................................................................................... 26 3.2.1 Acoustic analysis .................................................................................................................................... 26 3.2.2 Native speaker evaluation ................................................................................................................. 27 3.2.3 Statistical analysis ................................................................................................................................. 29
3.3 DISCUSSION .................................................................................................................................. 30
4. EXPERIMENT 2: REPETITION ............................................................................................... 31
4.1 METHODOLOGY ........................................................................................................................... 31 4.1.1 Participants .............................................................................................................................................. 31 4.1.2 Materials .................................................................................................................................................... 31 4.1.3 Procedure .................................................................................................................................................. 31 4.1.4 Analysis and evaluation ...................................................................................................................... 32
4.2 RESULTS ....................................................................................................................................... 32 4.2.1 Acoustic analysis .................................................................................................................................... 32
4
4.2.2 Native speaker evaluation ................................................................................................................. 33 4.2.3 Statistical analysis ................................................................................................................................. 35
4.3 DISCUSSION .................................................................................................................................. 35
5. EXPERIMENT 3: PICTURE-‐NAMING .................................................................................... 36
5.1 METHODOLOGY ........................................................................................................................... 36 5.1.1 Participants .............................................................................................................................................. 36 5.1.2 Materials .................................................................................................................................................... 36 5.1.3 Procedure .................................................................................................................................................. 37 5.1.4 Analysis and evaluation ...................................................................................................................... 39
5.2 RESULTS ....................................................................................................................................... 39 5.2.1 Acoustic analysis .................................................................................................................................... 39 5.2.2 Native speaker evaluation ................................................................................................................. 41 5.2.3 Statistical analysis ................................................................................................................................. 42
5.3 DISCUSSION .................................................................................................................................. 43
6. GENERAL DISCUSSION ............................................................................................................. 43
6.1 PARTICIPANTS’ REALIZATION OF /ɪ, Ɛ, Æ/ ................................................................................. 44 6.2 NATIVE SPEAKERS’ PERCEPTION OF /ɪ, Ɛ, Æ/ ............................................................................ 46 6.3 DATA GATHERING METHODS FOR RESEARCH IN SLA ................................................................ 50
7. CONCLUSION .............................................................................................................................. 51
8. APPENDIX .................................................................................................................................... 54
8.1 APPENDIX A: ENGLISH CONTEXT-‐DEPENDENT WORDS ............................................................. 54 8.2 APPENDIX B: DUTCH CONTEXT-‐DEPENDENT WORDS ............................................................... 54
BIBLIOGRAPHY .............................................................................................................................. 55
5
1. Introduction
In the last couple of decades, phonological research has revealed much about first
and second language acquisition. One of the main insights has been that the degree
of success in learning non-‐native sounds can be partially predicted by the distance
between a learner’s native language (L1) and the second language (L2) (e.g. Best,
McRoberts & Goodell, 2001; Flege, 1995). Furthermore, numerous studies have
emphasized the importance of the age of learning in second language acquisition
(SLA). It is striking that the literature has mainly focused on inexperienced and
experienced adult learners and that the speech of inexperienced young learners has
thus remained under-‐investigated. Not only is the study of the learning process of
these younger groups of scientific importance, the social and pedagogical
implications for this age group are also considerable, since there is a growing trend
to start L2 teaching at an early age, but little research on the positive or negative
effects of this for children’s (language) development.
The role of English as a lingua franca has been established in a growing number of
countries in the course of the 20th and 21st centuries. Knowledge of English has
become indispensable in the professional and private life, which is now reflected in
the educational system of many European countries. In Belgium, for instance,
English is often taught from the age of 12-‐13 years on.
This study aims to shed more light on the early stages of language learning. To this
end, we will examine the relation between foreign accentedness and
comprehensibility. Though both these notions refer to perceptual phenomena, there
are some obvious differences. Whereas foreign accentedness is speech perceived as
non-‐native, this does not imply that it is incomprehensible, i.e. that the listener
cannot identify the speech sounds (Munro, 2008). It has, for instance, been found
that the English of L1 speakers of Chinese may be highly comprehensible to English
native speakers, even though these same native speakers found their English
strongly accented (Munro & Derwing, 1995). On the contrary, speech does not have
to be foreign accented in order to be incomprehensible, as reduced speech clarity
may also impair L1 comprehensibility. However, in some cases, ‘ill fits’ between the
phonological systems of an L1 and an L2 may result in incomprehensible, foreign
accented speech. Recent studies (e.g. Escudero, Benders & Lipski, 2009) have also
6
emphasized the influence of acoustic cues, such as vowel duration, on
comprehensibility and foreign accentedness.
The first aim of this paper is of a methodological nature. Considering the gap in the
literature on child L2 acquisition, there has been no formal methodological
comparison of phonological elicitation methods. The choice of a methodology
depends on theoretical and practical considerations. Reading, repetition and
picture-‐naming tasks may all be designed in order to elicit data, but they do not do
this in the same way. With reading tasks, for instance, one has to take into account
the potential influence of orthography. Furthermore, some of these tests may be too
abstract for children and thus prove impractical.
Our first research question is thus which elicitation methods are practical for
children and which are not. The notion of practicality has different facets in this
question. First, we will consider the feasibility of the tasks. Some tasks may be too
cognitively challenging for the children. Secondly, the tests may or may not yield
different results. On the one hand, the theoretical differences between the tasks,
such as orthographic or aural input, may not surface in the results. On the other
hand, these differences may still influence the results. In this case, we will ask
ourselves which of the methods is most suitable for the research question. Finally, it
may also be that the most reliable results are to be obtained through a
complementary approach.
The second aim of this paper is to study the accentedness and comprehensibility of
the pronunciation of the English vowels /ɪ, ɛ, æ/ by Flemish children before they
have any formal instruction in English. A previous study by Flege (1992) showed
that speakers of (Netherlandic) Dutch tend to assimilate English /æ/ to /ɛ/ because
Dutch lacks a phonetic and phonological category for the former sound. Based on
these findings and the predictions of the Perceptual Assimilation Model (Best et al.,
2001) and the Speech Learning Model (Flege, 1995) (see Section 2.2.1) we
hypothesize that Flemish children will comprehensibly produce the vowels /ɪ/ and
/ɛ/, but that they will tend to assimilate English /æ/ and /ɛ/ to Dutch /ɛ/. Since this
study is concerned with foreign accent and comprehensibility, special attention will
be devoted to the influence of acoustic cues (Escudero, 2001). Specifically, we will
examine which acoustic cues weigh more on accentedness or comprehensibility.
Since the scope of this study is too restricted for an elaborate investigation into
7
these topics, these secondary questions only serve to indicate tendencies that should
be explored more deeply in further research. In this way, we hope to pave the way
for more methodologically consistent descriptions of the pronunciation of children
before formal instruction of the L2.
Thus, we will compare three different elicitation techniques, a reading task, a
repetition task and a picture-‐naming task to examine the pronunciation of the
English vowels /ɪ, ɛ, æ/ by 10 Flemish children. Then, the vowel closeness (F1),
vowel frontness and liprounding (F2) and vowel duration will be measured using
PRAAT (Boersma & Weenink, 2012). This acoustic analysis will be complemented by
a categorization task, in which 4 native speakers of British English will categorize
the vowels produced by the children and rate their accentedness. These
categorizations and ratings will provide an insight in comprehensibility and foreign
accentedness, respectively, since the categorization is believed to reflect
comprehensibility and the goodness rating the degree of foreign accent. The acoustic
analyses and categorization task will then be statistically analysed in order to
examine the relationship between the acoustic cues and the comprehensibility and
accentedness of the vowels. Finally, since these analyses will be made for the three
tasks independently, their results will be compared. This comparison will reveal
whether the choice of elicitation technique significantly influences the results or not.
In Section 2, we will present a brief overview of the sociolinguistic situation of
English in Europe, Belgium and Flanders (Section 2.1), a summary of some of the
insights of research in SLA (Section 2.2), previous studies on the acquisition of
English by speakers of Dutch (Section 2.3) and a few comments on the choice of
methodology in previous research (Section 2.4). Our main hypotheses will be
elaborated in Sections 2.3.2 and 2.4. We will describe the procedure, analyses and
results of the three experiments separately in Sections 3 to 5. These results will be
compared to each other in Section 6 and will be followed by some concluding
remarks in Section 7.
8
2. Background
2.1 English in the European Union
2.1.1 The inner, outer and expanding circles of English
As English came to attain different statuses in the European Union, it has also had to
fulfil different roles. Kachru (1992) describes these roles on the basis of three
concentric circles. The first or ‘inner’ circle of English consists of countries where
English is used as the primary language, as in the USA or Ireland. The second circle is
the ‘outer circle’ or ‘extended circle’. In countries belonging to this circle, “English is
only one of two or more codes in the linguistic repertoire of such bilinguals or
multilinguals, and [it] has acquired an important status in the language policies in
most of such multilingual nations” (Kachru 1985 in Kachru, 1992: 38). Finally, the
third circle is the expanding circle and contains countries where English is used as
an international language, without there necessarily being colonial ties with inner
circle countries. Countries belonging to the expanding circle are equally
characterized by their dependency on the linguistic norms provided by the inner
circle.
Applied to the European Union, the concentric circle model mainly consists of an
inner and expanding circle. Berns (1995: 4) suggests a regional grouping for the
countries of the European Union (of 1994), in which English functions as a “language
of wider communication” within Europe. Moreover, Berns (1995) argues that this
grouping would be more representative of a culturally unified and economically
open European Union. English, then, seems to be the primary mode of
communication between the members of the EU. Of course, countries such as
England and Ireland belong to the inner circle of world Englishes, where English is
spoken as a primary language. Many other countries, however, belong to the non-‐
norm providing expanding circle according to Berns (1995). Since none of the
European countries outside of the inner circle have “indigenous, non-‐native
varieties” (Berns, 1995: 8), there is no actual outer circle. Instead, in most countries
“English has the role of international language, is generally taught as a foreign
language, and develops as performance varieties among speakers. Learners are
expected to acquire the norms of behavior appropriate to the users of English in the
inner circle” (Berns, 1995: 8).
9
However, Berns (1995) argues that some countries, i.e. the Netherlands,
Luxembourg and Germany, do not solely belong to the expanding circle. Especially in
the Netherlands and Luxembourg, English is essential for economic purposes.
Moreover, these countries are more exposed to English media influences than other
expanding circle countries, such as Japan. The use of English in national professional
settings would also characterize them as in-‐between the outer and the expanding
circle. In these countries, however, the use of English is not institutionalized. Berns
(1995) thus characterises these countries as ‘expanding/outer circle’ countries.
While Berns (1995) effectively proves that there is a certain variability in the
expanding circle, it seems unnecessary to resort to an in-‐between category. Several
objections can be put forth. First, it is unclear why countries such as Belgium, in
which the use of English seems similar to that in the Netherlands, belong exclusively
to the expanding circle. The grounds for differentiation thus seem somewhat
inconsistent. Secondly, it seems natural that there is some variability in the use of
English in the expanding circle and that English is more integrated in some countries
than others. However, that this should imply that the more integrated countries
belong in-‐between the outer and expanding circle is unconvincing. Contrary to
actual outer circle countries, the Netherlands, Germany and Luxembourg do not
have nativized varieties of English any more than Belgium or France do. The need
for such an intermediate circle thus seems unnecessary.
2.1.2 English in Belgium and Flanders
In order to understand the use of English in Flanders, we will first briefly sketch the
linguistic situation in Belgium. The Belgian constitution acknowledges the official
status of Dutch, French and German, though this does not imply that Belgians are all
multilingual. Rather, the Belgian situation is one of territorial multilingualism, where
each language is spoken in a more or less fixed region, i.e. Dutch in Flanders, French
in Wallonia, and German in the Eastern region of Eupen-‐Malmedy. Brussels is a
bilingual region where both French and Dutch are spoken. In 1997, about 61% of the
Belgian population was Dutch-‐speaking, 38% was French-‐speaking and 1% was
German-‐speaking (Goethals, 1997). Goethals (1997) stresses that French is, as much
as English, a foreign language to the Flemish. Without elaborating too much on this
topic, English probably has as much claim to the main mode of interlingual
10
communication as French or Dutch. A survey of the European Union1, including
1000 Belgian participants, is equally revealing in this respect. The most important
languages in professional and personal domains (apart from the mother tongue) are
English and French for 82% and 53% of the Belgians, respectively2. 88% of the
Belgians want their children to learn English and 59% think the ideal age for
learning a second language is around 6 to 12 years. Though 77% had learned
languages at school, only 44% believes this is the most efficient way to acquire a
new language. Of the people who report to know English, 24% claims to speak it
very well, 59% speaks it well and 29% has a basic knowledge of English. Since these
figures do not treat the Dutch and French speaking Belgians as separate groups, it is
difficult to evaluate to which extent this situation also separately applies to Flanders
and Wallonia.
Goethals (1997) states that most 13-‐year-‐old Flemish children know about 400
English words without haven taken formal classes, though this has been shown to
depend on the type of secondary education chosen (Lippens, 2010). It seems that
children in the BSO (Vocational Secondary Education) trail behind the TSO
(Technical Secondary Education) and ASO (General Secondary Education) children
in terms of English lexical knowledge before formal instruction. The ASO and TSO
pupils have a more comparable lexical knowledge, though the former are better in
terms of spelling than the latter (Lippens, 2010). The media plays an important role
in acquiring the basics of English. In secondary school, Flemish children get about 2
to 4 hours of English every week in General Secondary Education, depending on the
year and option chosen by the students (Goethals, 1997). After secondary school,
adults can still take language courses in evening classes. Goethals (1997) also notes
that schools (and their teachers) are becoming increasingly embedded in a more
international network.
As for target models of English in secondary education, Goethals (1997) states that
there is no fixed pronunciation standard, though norms are often still referred to.
Although, traditionally, the norm was British Received Pronunciation (RP), the
increasing contact with the USA often leads to an American inflected English
(Goethals, 1997). Even though Goethals’s (1997) account of English in Flanders may
1 Special Eurobarometer 243: Europeans and their Languages, European Commission, February 2006. 2 Note that 38% of the sample were French-‐speaking, 56% were Dutch-‐speaking. It thus seems that French is more important for the Dutch speakers than Dutch for the French speakers.
11
be somewhat dated, the general situation has not changed radically since. If
anything, the influence of American English has arguably increased over the passed
few years. It has been argued that this has given rise to a Mid-‐Atlantic English
pronunciation, which introduces General American features in RP (see Simon,
2005). Personal experience has also shown that Flemish children are less familiar
with formal British English (RP) than General American pronunciation. De Meyere
(2010) confirms that pupils in secondary education no longer believe RP to be their
target pronunciation for professional and everyday communication.
2.2 Acquiring a second language
In phonology, Second Language Acquisition has long held a central place in the
literature for a number of reasons. Researchers have tried to address topics such as
phonological maturation (see Best et al., 2001), foreign accent and
comprehensibility (e.g. Munro, 2008). The fundamental questions underlying these
topics pertain to the possibility of second language acquisition: what allows us to
acquire a new language’s phonology? What hurdles have to be overcome? To which
degree can we expect to master a second or third language’s phonology?
2.2.1 Language-‐dependent factors in SLA
The field of SLA has given rise to two important models that try to systematically
describe and predict the dynamics of phonological acquisition, namely the
Perceptual Assimilation Model (henceforth PAM; Best et al., 2001) and the Speech
Learning Model (Flege, 1995).
Best et al.’s (2001) model supposes that speakers consciously and subconsciously
detect information regarding the articulatory gestures that generate sounds. This
information concerns articulatory organs (e.g. glottal stops and nasals), constriction
locations (e.g. lips in labial consonants) and constriction degree (e.g. friction in
fricative consonants) (Best et al., 2001). When perceiving a non-‐native sound, a
speaker may assimilate the sound to his/her L1 phonology in a number of ways: (1)
a non-‐native phone is categorized if it is perceived as an exemplar of a native
phoneme; (2) it is uncategorized if it falls in between two L1 sounds; (3) it is non-‐
assimilable if it does not correspond in any way to native phonemes.
In this model, the perceived distance between the L1 and L2’s phonologies is thus
key in predicting the learnability of L2 sounds. If two L2 sounds are assimilated to
12
two different categories (i.e. Two Category assimilation), the speaker is very likely to
perceive the L2 contrast. If one L2 sound is categorized but the other is
uncategorized (or non-‐assimilable), the learner will also be likely to discriminate
between the two sounds. If, however, both sounds are assimilated to the same L1
phoneme (i.e. Single Category assimilation), the speaker’s L1 will prevent the
learner to perceive the contrast. Even if both L2 sounds are perceived as exemplars
of a single L1 phoneme, there may, however, be a difference in ‘category goodness’:
one L2 sound may be a better exemplar of the L1 sound than the other. Finally, both
L2 sounds may also not be assimilable to the L1 system. In this case, the speaker
may still be able to perceptually perceive the non-‐native contrast, since the sounds
are perceived as non-‐speech sounds. For instance, Best et al. (2001) cite the ability
of English speakers to discriminate Zulu clicks, even if these cannot be assimilated to
any English phonemes.
The aforementioned assimilation patterns in PAM (Best et al., 2001) are mostly
indicative of the perceptual assimilation of naïve (or inexperienced) L2 listeners.
The predictions made by PAM may thus be more accurate for short-‐term than for
long-‐term learning. A revised version of PAM, called PAM-‐L2 (Best & Tyler, 2007)
was developed in order to be more applicable to more experienced L2 learners and
hence make more accurate predictions about long-‐term learning. According to PAM-‐
L2, a difference in category goodness may lead to the creation of a new phonological
category. Two L2 sounds may, for instance, be assimilated to the same L1 category
in early stages of L2 learning but differ in category goodness. In this case, PAM-‐L2
predicts that a new phonological and phonetic category may eventually develop for
the most deviant sound. In this way, the L2 contrast is made in later stages of
language learning, even though it was not accurately perceived in earlier stages.
An earlier model that predicts non-‐native speech perception and L2 sound
acquisition is the Speech Learning Model (henceforth SLM; Flege, 1995). The
postulates of SLM are similar to those of PAM-‐L2 in their focus on experienced L2
learners. The model also resembles PAM-‐L2 by stressing the role of the perceptual
distance between native and non-‐native sounds, though it is not as explicit about
assimilation patterns. Flege & Hillenbrand (1984) state that “[j]udging acoustically
different phones to be members of the same category is a fundamental aspect of
human speech perception.” Moreover, Flege (1995) dismisses any strict views on a
phonological (and cognitive) maturation period that would obstruct the creation of
13
new L2 phonological categories. On the contrary, he states that “the mechanisms and
processes used in learning the L1 sound system, including category formation,
remain intact over the life span, and can be applied to L2 learning” (Flege, 1995:
239). Instead, problems in the acquisition of non-‐native sounds are more likely to be
caused by perceptual issues. Furthermore, Flege (1995) stresses the role of
experience and age of learning as influences on perception.
However, even if the influence of the L1 is generally accepted, studies have also
attempted to look beyond mother tongue interference in the L2. In a study on the
intelligibility of the English speech of Vietnamese speakers, Cunningham (2009)
remarks that, apart from transfer-‐related issues, the speech of some speakers is
subject to a more general, spontaneous variability. In a similar vein, Bohn (1995:
279) stresses that learners, and especially adults, are “not always language-‐specific
perceivers” (emphasis in original). Contrary to Cunningham (2009), however, Bohn
(1995) focuses on systematic, rather than spontaneous errors. For instance, a
speaker’s interlanguage may be influenced by certain universal phonotactic
preferences that cannot be attributed to the L1 or L2. With regard to vowel
contrasts, they found that Chinese learners of English are able to discriminate vowel
length differences, even if they do not have experience with this feature in their L1.
Bohn (1995) claims that, in contexts where learners are unable to differentiate non-‐
native spectral vowel differences, these spectral cues may be overridden by
durational cues. In other words, Bohn (1995: 300) hypothesizes that “listeners will
increasingly attend to duration differences between vowel pairs as spectral
differences between vowels become smaller.” Still, Bohn (1995) does not disregard
L1 influences, but argues that these tendencies complement any mother tongue
transfer that may occur.
2.2.2 Learner-‐dependent factors in SLA
From the above discussion, it should be apparent that the result of SLA does not only
depend on the dynamics between the L1 and L2 but also on the experience with the
L2. The influence of experience has been accounted for in different ways. For those
situations where a speaker lives in the country where the L2 is spoken, studies have
pointed out the roles of the length of residence (LOR) and the age of arrival (AOA)
(e.g. McAllister, 2001; Meador, Flege & Mackay, 2000). Of these two factors, the
latter has been corroborated the most. McAllister (2001), for instance, found that
14
experienced Spanish learners of Swedish with an average LOR of 18 years
pronounced Swedish vowels less accurately than Spanish learners with an average
LOR of 3.6 years. Meador, Flege & Mackay (2000) found that the AOA of learners was
more significant: the earlier the AOA of the speaker, the more accurate his/her L2
pronunciation. Moreover, Flege, Munro & Mackay (1995), found that the effects of
LOR were overruled by the age of learning (AOL) of speakers. Speakers with a
comparable LOR had various proficiency levels depending on their AOL.
Though there is a general consensus that the earlier SLA occurs, the better, the
grounds for this fact have been much debated. Earlier phonological frameworks
have claimed that after a certain age, a so-‐called ‘critical period’, the sensorineural
mechanisms used in attuning our ability to discriminate native speech sounds
mature (Eimas (1975), Aslin and Pisoni (1980) in Best et al., 2001). More recently,
however, it has been acknowledged that new contrasts can be learned during
adulthood. Meador et al. (2000) note the inconsistency of a theory where
neurological maturation prevents some contrasts to be learned but others not. Flege
(1995: 239) even explicitly states that “the mechanisms and processes used in
learning the L1 sound system, including category formation, remain intact over the
life span, and can be applied to L2 learning.” An alternative explanation, that does
not reject the notion of a critical period, is that after this period, the L2 sounds are
more likely to be perceptually assimilated to the L1 phonological system because the
latter is more developed (Meador et al., 2000). This explanation is equally
compatible with the previously discussed PAM (Best et al., 2001) and SLM (Flege,
1995).
2.2.3 Foreign accent, comprehensibility and intelligibility
As indicated above, phonological research in SLA often entails dealing with the
question of foreign accent. The concept of foreign accent has been ill defined (see
Munro, 2008) and is thus often taken for granted. Speech can be said to be produced
with a foreign accent when it deviates from native (though not necessarily standard)
norms. Alternatively, speech with a foreign accent can be defined as speech
containing characteristics that lead listeners to identify it as non-‐native.
The concept thus refers to a perceptual phenomenon, where speech is only
considered to be produced with a foreign accent if listeners perceive it as such. Thus,
15
any study attempting to shed a light on the nature of foreign accent, should ideally
complement an acoustic analysis with a perceptual evaluation task.
Views on foreign accent have been varying. Munro (2008), for instance, cites
prescriptivist interpretations of foreign accent, which claim it is “of the nature of
imperfect or defective speech” (Greene & Welles, 1927 in Munro, 2008: 193).
According to Griffen (1980 in Munro, 2008), the goal of second language acquisition
is to attain an accent “free of any indication that the speaker is not a clinically
normal native” (Munro, 2008: 193). However, Munro (2008) states that it is now a
commonly held view that a complete eradication of a foreign accent is not always
possible or necessary. Nevertheless, foreign-‐accented speech has been and still is
extensively discussed in both phonetic-‐phonological and pedagogical literature and
remains one of the key concerns of both language teachers and learners.
As views on foreign accent have become more tolerant – at least in academic circles
–, the discussion has been enriched by the concepts of intelligibility and
comprehensibility. Munro & Derwing (1995: 76) define intelligibility as “the extent
to which a speaker’s message is actually understood by a listener.”
Comprehensibility, then, pertains to whether listeners can identify specific speech
sounds and to how they judge the pronunciation (Floccia, Butler, Goslin & Ellis,
2009). It is not surprising that foreign accent, intelligibility and comprehensibility
are, to a certain extent, interrelated. Munro & Derwing (1995), for instance, found
that native English speech fragments were generally perceived as less accented and
more intelligible than non-‐native fragments. However, the speech of one of their
native speakers was less intelligible than that of their non-‐native participants, even
if it was less accented. In this case, intelligibility may have been impaired by factors
such as speech rate and clarity (Munro & Derwing, 1995). Interestingly, Munro &
Derwing (1995) also found that, while the speech of the non-‐native, Chinese
speakers was often perceived as having a strong foreign accent by native speakers of
English, their productions remained easy to understand (Munro & Derwing, 1995).
Thus, foreign accentedness and intelligibility can also operate independently.
An important difference between comprehensibility and intelligibility are the stages
of lexical processing to which they relate. Comprehensibility pertains to the pre-‐
lexical processing stage, while intelligibility pertains to the post-‐lexical processing
stage (Floccia et al., 2009). As a consequence, impaired comprehensibility may not
16
surface in intelligibility ratings, since these are also subject to a top-‐down
interpretation of speech. A listener’s communicational expectations may thus allow
him to understand speech that may be less comprehensible on a purely phonological
level. According to Floccia et al. (2009), comprehensibility may also be influenced by
listener expectations. In their experiment, listeners’ judgements of
comprehensibility were delayed if the speakers were (unexpectedly) changed. A
change from native speech to non-‐native speech impaired comprehensibility, though
this effect was not noted for regional variation (Floccia et al., 2009). Floccia et al.
(2009) found that this disturbance of comprehensibility does not habituate in the
short or long term, though the post-‐lexical processing of speech (and thus of
intelligibility) may improve, “perhaps by applying a specific phonological accent-‐
filter onto the outcome of lexical activation” (Floccia et al., 2009: 402).
More recently, the discussion about intelligibility and comprehensibility has been
extended by a focus on acoustic cues and their weight (e.g. Escudero, 2001;
Escudero et al., 2009; Idemaru & Guion-‐Anderson, 2010). Acoustic cues allow
listeners to perceive sound contrasts. For instance, the contrast between English /ɪ/
and /iː/ is signalled by both spectral cues and durational cues (Escudero, 2001).
These cues may be exploited to different extents in different regional varieties.
Escudero (2001) notes that in Scottish English, for instance, the contrast is made
almost exclusively through the spectral cue. In perception, these differences also
occur as various cues have different ‘weights’. Scottish speakers of English thus
attribute more weight to the spectral cues when discriminating the above-‐
mentioned contrast than Southern speakers of English. The importance of cue
weighting for SLA should not be understated. During L1 acquisition, speakers learn
to attribute weight to the relevant cues of their L1. If the cue weight is differently
distributed in an L2, speakers may have problems correctly identifying L2 contrasts
(Escudero, 2001).
As the focus on foreign accent moves to intelligibility, this is reflected by a change in
attitude towards normative targets. The goal of L2 acquisition emphasizes the
importance of cross-‐linguistic intelligibility, involving varieties of English from all
the concentric circles (see Section 2.1.1). The typical interlocutor of the L2 speaker
of English is thus not necessarily a native speaker of English, since the language is
used “to serve international functions among L2 speakers in international contexts”
(Jenkins, 2000: 16). Jenkins (2000) thus rejects the role of Inner Circle countries as
17
norm dictating and argues in favour of a model based on international
comprehension.
Views regarding the attainment of a native-‐like accent or an intelligibility-‐grounded
target also vary among teachers of English in higher education in Flanders. Some
lecturers believe students should aim for a native-‐like pronunciation. For others,
intelligibility is the main target, since the interlocutors of the students are not
necessarily native speakers themselves (Simon, 2005). These views are also echoed
in the lecturers’ views on the specific target of pronunciation. According to the
former view, the students’ choice of accent ought to be consistent. The latter view,
however, does not stress the need for a ‘pure accent’ (Simon, 2005). Among
university students of English, the target of pronunciation is commonly thought to
be (almost) native-‐like. Simon (2005) argues that this high standard is a way for the
students of English to distinguish themselves from an already high proficiency level
of English among students who do not study English.
2.3 Dutch and English phonology
2.3.1 The Dutch and English vowel systems
Within Europe, Dutch is mainly spoken in the Netherlands (Northern Dutch) and
Belgium (Southern Dutch). Between these Northern and Southern varieties of Dutch,
there are a number of phonological, grammatical and lexical differences not unlike
those observed between the varieties of English. Within Belgium, Dutch is mainly
spoken in the Flemish region and in Brussels. Again, the different regions of
Flanders and Brussels are home to many different accents and dialects. There is a
standard of Belgian Dutch, but often this standard is coloured by the regional
varieties (Booij, 1999). Because of the regional variation in Dutch, a uniform
description of its phonology is often practically impossible. In studies involving
Dutch, it is thus of paramount importance to take into account that the results may
not be generalizable beyond this variety (see also Escudero, Simon & Mitterer, 2012
in Section 2.3.2).
Figure 1 gives an overview of the Standard Northern Dutch vowels. With regard to
the short vowels, the Northern and Southern varieties correspond by and large.
Dutch contains 5 short vowels, /ɪ, ɛ, ɔ, ʏ, ɑ/, and 7 long vowels /i, y, u, e, ø, o, a/. The
Northern and Southern standard varieties also contain 3 diphthongs, /ɛi, œy, ɔu/,
that are not pictured in Figure 1.
18
Figure 1 The Standard Dutch vowels (Booij, 1999: 5)
The difference between the short and long vowels is not merely one of length, but
also of vowel quality. For example, ‘bom’ /bɔm/ (‘bomb’) and ‘boom’ /boːm/ (‘tree’)
differ both in terms of vowels length and vowel quality. Booij (1999) notes vowel
lengths of about 100ms for the short vowels and about 200ms for the long vowels.
Adank, van Hout & Smits (2004) report the formant values and vowel durations in
Table 1 for the Southern Standard Dutch front vowels /ɪ/ and /ɛ/.
Male talkers Female talkers Mean
Vowel F1 F2 Duration F1 F2 Duration F1 F2 Duration /ɪ/ 364 1745 76 455 2115 88 410 1930 82 /ɛ/ 475 1616 86 581 1932 101 528 1774 94 Table 1 F1 and F2 values (in Hz) and duration (in ms) of Southern Standard Dutch vowels
(adapted from Adank et al., 2004)
The situation for English is comparable to that of Dutch in terms of geographical
variation. We shall thus restrict ourselves to a brief overview of the General British
English (see Figure 2) and General American English vowels (see Figure 3).
Figure 2 The British English vowels (Dillon, 2009)
19
Figure 3 The American English vowels (Dillon, 2009)
Since this study will be concerned with the pronunciations of the English front
vowels /ɪ, ɛ, æ/, let us remark that these have a comparable phonemic value in both
varieties.
Hawkins & Midgely (2005) report formant values of the three front vowels /ɪ, ɛ, æ/
for male speakers of RP of different ages (see Table 2).
65+ 50-‐55 35-‐40 20-‐25 Mean
Vowel F1 F2 F1 F2 F1 F2 F1 F2 F1 F2
/æ/ 644 1678 693 1579 696 1574 917 1473 738 1576
/ɛ/ 454 1962 489 1920 512 1888 600 1914 514 1921
/ɪ/ 382 2024 341 2074 374 2115 393 2174 373 2097
Table 2 F1 and F2 values of RP vowels in Hz (Hawkins & Midgley, 2005)
2.3.2 The perception and production of the English vowels /ɪ,ɛ,æ/ by Dutch speakers
A comparison of the vowel diagrams in the previous section (Figures 1, 2 and 3)
reveals an asymmetry between the English and Dutch front vowels. Both varieties of
English have an open front vowel /æ/ where Dutch has no such vowel. Previous
studies have identified this asymmetry as a potential problem for speakers of Dutch
learning English (Escudero et al., 2012; Flege, 1992; Wang & van Heuven, 2006).
Flege (1992), for instance, found that the Dutch pronunciation of English /æ/ is
often perceived as /ɛ/ or /ʌ/ by native speakers of both American and British
English. The American and British native speakers, however, were able to identify
the /æ/ correctly 67% and 72% of the time, respectively. Interestingly, the native
speakers’ evaluations of the British English pronunciations of /æ/ were not perfect
either. American native speakers identified the vowel correctly in 85% of the cases.
Surprisingly, native speakers of British English only identified these native vowel
productions correctly 73% of the time. Thus, the Dutch pronunciation of /æ/ was
almost as comprehensible as the British pronunciation for speakers of British
20
English. Flege (1992) puts this down to a methodological difference. The British
participants had a wider choice of vowels than the American participants. A closer
look at the relatively high accuracy of correctly identified Dutch pronunciations of
/æ/ also reveals an effect of degree of foreign accent. The pronunciations of /æ/ by
speakers with a strong foreign accent were significantly less comprehensible than
those by speakers with moderate or mild foreign accents (Flege, 1992).
Even though Dutch has no equivalent /æ/, speakers of Dutch are thus in some cases
able to distinguish /æ/ from other vowels. This observation is equally noted by
Wang & van Heuven (2006: 242), who found that there was a “fair degree of
separation” between the /æ/ and /ɛ/ in the English pronunciation of Dutch speakers
who had not studied English after secondary school. However, these speakers had a
significantly shorter pronunciation of /æ/ than the American pronunciation of the
vowel (Wang & van Heuven, 2006).
It should be noted, however, that the studies by Flege (1992) and Wang & van
Heuven (2006) are concerned with Northern Dutch, not Southern Dutch. The
distinction is relevant, as has been discussed in Section 2.3.1. Though comparative
studies of the perception of English vowels by speakers of different varieties of
Dutch have been few, Escudero et al. (2011) note that different (regional) vowel
pronunciations of Dutch influence the perception of English vowels. More
specifically, speakers of Northern Dutch perceived English /æ/ more often as /ɑ/
than Flemish listeners, while Flemish listeners perceived /ɛ/ more often as /ɪ/ than
their Northern neighbours. Both groups also tended to perceive the English vowel
/æ/ as /ɛ/. Escudero et al. (2011: 8) conclude that “the acoustic (and hence
auditory) properties are important in determining similarity across languages.”
Based on the results of the studies by Flege (1992), Wang & van Heuven (2006) and
Escudero et. al (2011) and on the predictions of the PAM (Best et al., 2001), we
hypothesize that inexperienced Flemish learners of English will assimilate the
English vowels /ɛ/ and /æ/ to one category, namely /ɛ/ (Hypothesis 1a, henceforth
H1a). The English /æ/, however, may differ perceptually from the /ɛ/ in terms of
category goodness. Should this be the case, PAM-‐L2 (Best & Tyler, 2007) predicts
that, as they become more experienced, the learners may come to differentiate
between both vowels, as may have been the case in the studies by Flege (1992),
Wang & van Heuven (2006) and Escudero et al. (2011). Of course, for the children to
21
develop a new phonological category for /æ/, they need to be exposed to a more or
less accurate English contrast between /æ/ and /ɛ/.
The pronunciation of the English vowels /ɪ/ and /ɛ/, then, should be less
problematic, since both vowels have a Dutch counterpart. We thus hypothesize that
these English vowels will be assimilated to two different categories (Hypothesis 1b,
henceforth H1b). In sum, the children should be able to distinguish between the
English words ‘bit’ and ‘bet’, but not between ‘bet’ and ‘bat’.
2.4 Methodological consideration
SLA studies on the production of L2 sounds generally gather data using one of three
methods. A survey of the literature on SLA reveals that data is typically elicited
through a reading task, a repetition task or a picture-‐naming task. Though the basic
design of these elicitation methods is quite self-‐explanatory, the different uses of
these tasks suggest that the choice of a methodology is relevant. We will thus outline
some of the basic elements of these methods.
Reading tasks involve a subject reading a word, sentence, or longer utterance (e.g.
Flege & Hillenbrand, 1984; Kennedy & Trofimovich, 2008). Because the intonation
pattern of a sentence may significantly influence a word’s articulation, neutral
carrier phrases (e.g. ‘I say the word ___ again’) are often used in order to neutralize
the effect of prosody (e.g. Major & Kim, 1999). Another important choice in these
studies is whether the elicited words are real words (e.g. Mah & Archibald, 2003) or
not (e.g. Motohashi-‐Saigo & Hardison, 2009), since the use of real words may
influence participants’ pronunciation if the stimuli are frequently occurring words.
Non-‐words allow participants to produce words that have not yet been lexically
encoded. The drawback of this method is that, instead of relying on their own
phonological representation of a word, participants may be influenced by its
orthography. Furthermore, not all phonological possibilities may appear in the same
context. In English, for instance, it would be difficult do elicit the vowels /u/ and /ʊ/
in the same consonantal context, since they often appear in mutually exclusive
contexts (e.g. <boot> /but/, <put>/pʊt/). These issues can be overcome by adding
similar sounding words in the margin, so that the participant can rely on both words
for the intended pronunciation (e.g. Kewley-‐Port, Akhane-‐Yamada & Aikawa, 1996).
For instance, /bʊt/ could then be elicited by adding <put> in the margin.
22
Another way of eliciting data is through a repetition task, in which a word or a
sentence containing the target stimuli is repeated. As with reading tasks, this
method is often carried out using a carrier phrase. The choice between real words
(e.g. Flege et al., 1995) or non-‐words (e.g. Whalen, Best & Irwin, 1997) is also
relevant. Psycholinguistic research has also been interested in the repetition of non-‐
words (e.g. Summers et al., 2010; Windsor, Kohnert, Lobitz & Pham, 2010). Some
studies combine elements from different methods. Flege et al. (1995), for instance,
use a repetition task in which the stimuli are accompanied by a written
representation. If one wishes to study the phonological representation of sounds,
however, the danger exists that the participants imitate a sound they have not yet
encoded. In order to prevent this, some studies have worked with a delay between
the stimulus and the repetition (e.g. Flege & Yeni-‐Komshian, 1999). This ensures
that the speaker relies less on the imitation of the stimuli and more on his/her own
phonological representation. A related issue is that it is difficult to work with accent-‐
neutral stimuli. If one wishes to examine whether learners have an English accent
that is closer to British or American English, for instance, the aural stimuli may guide
the speakers’ pronunciations.
A third common way of eliciting words is through a picture-‐naming task. In this task,
a word is prompted by using a visual representation of the word. The task is
interesting because it directly prompts the participant’s lexical or phonological
representation. While it thus circumvents the potential influence of aural and
orthographical representation, this task also has limitations. In order for the
participant to name a picture, the prompted word should of course be actively
known by the participant. While this especially presents difficulties for the use of
non-‐words (e.g. Matsumoto-‐Shimamori, Ito, Fukuda & Fukuda, 2011), new learners
of a language may not always know the real words either. When using real words,
these should ideally also be intuitively picturable. For this reason, these tasks are
sometimes preceded by a training period, in which the participants acquaint
themselves with the words and their visualizations. Alternative methods include
displaying the picture multiple times. The first time the picture would then be
accompanied by an aural stimulus, so that the participants may learn which word is
intended (e.g. Oh et al., 2011).
While these methods are frequently used, there has been no methodological study
comparing the results obtained by these different tasks. Therefore it is not always
23
clear to what degree the orthography or aural input significantly influences
speakers’ productions. Since few SLA studies have focused on children, it is
especially important to carry out such a methodological comparison for this age
group. This may reveal how to best approach future research in this domain and
avoid methodological pitfalls. Hence, we will compare the results of a reading,
repetition and picture-‐naming task in this study.
If we consider the limitations and advantages of each method, we predict that the
results of each task will differ accordingly. In the reading task, the participants may
try to distinguish English /æ/ from /ɛ/ on the basis of the graphemes <a> (for /æ/)
and <e> (for /ɛ/). We hypothesize that in this case the competing Dutch vowel for
the grapheme <a>, i.e. /ɑ/, may influence the speakers’ pronunciations of the
English vowel /æ/ (Hypothesis 2a, henceforth H2a). Since the repetition task is the
only task in which the speakers are confronted with a native speaker’s
pronunciation, we hypothesize that this pronunciation will influence the learners’
realization of the /æ/ -‐/ɛ/ contrast and allow them to distinguish both sounds
(Hypothesis 2b, henceforth H2b). With regard to the picture-‐naming task, finally,
we hypothesize that the cognitively more challenging design of the task will also
have an effect on the participants’ pronunciation (Hypothesis 2c, henceforth H2c).
3. Experiment 1: Reading
3.1 Methodology
3.1.1 Participants
A group of 10 native speakers of Dutch, 4 male and 6 female, voluntarily participated
in this study. The participants were selected on the basis of a consent form filled out
by their parent(s). The participants were all 12-‐13 years old. They were in their first
year of secondary school and had thus not had English classes at school. The
children all came from the same school in the East-‐Flemish region, around Ghent.
They were all monolingual native speakers of Dutch, apart from a basic knowledge
of French as a foreign language. None of the participants reported having hearing
problems. The three experiments were carried out individually. The children were
told that the experiments would last about 15 minutes. They were also told not to
tell their peers about the procedure of the experiments. They were assured that
none of the gathered data would be used for grades. The experiments were
24
conducted in the same order as they are described here and all took place in the
same setting.
A small questionnaire was carried out in order to gauge the children’s knowledge
and use of English. This questionnaire pointed out that only one of the participants
had previously been to an English-‐speaking country. Three participants had an
English-‐speaking relative, though they met with these at most once a month. All of
the participants’ parents could speak English with various levels of proficiency. One
child’s father, for instance, was an interpreter of English and French and her mother
had studied linguistics. Mostly, however, the parents’ contact with English was
work-‐related. Only four participants had older siblings who could speak English, but
not better than their parents.
Five of the participants claimed to sometimes speak English with their friends or
relatives. For most children, this use of English was limited to some words or
sentences. One participant, however, did have more elaborate conversations once a
month with an aunt who was a native speaker of English.
Unsurprisingly, most of the English input happened through popular media. In
general, most participants sometimes sang songs in English and, in some cases, even
knew them by heart. Almost all of the children watched English television programs
every day, but almost always subtitled. Half of the participants also played English
video games, without Dutch subtitles. Some children had noticed people speaking
English in the bus, or on the street, but not on a regular basis.
3.1.2 Materials
Though this study will only examine the productions of the English vowels /ɪ, ɛ, æ/,
the vowels /iː, uː, ʌ/ were also included in filler words. Previous studies (see Section
2.3.2) have shown that the vowel set /ɪ, ɛ, æ/ may prove difficult for speakers of
Dutch, since the vowel /æ/ is not part of the Dutch phonological inventory and is
often merged with /ɛ/ in perception and production. These six vowels were
embedded in a /bVt/ context. This context was selected, because plosives have
sharp boundaries with the following or preceding vowel and thus facilitate vowel
analysis. The resulting words (‘bit, bet, bat’) are context-‐neutral, i.e. they are existing
words with the same consonantal context, and the native-‐speaker evaluations (see
Section 3.1.4) will thus not be guided by lexico-‐semantic knowledge. Finally, these
25
words were embedded in the carrier sentence ‘I say ___ again’ in order to neutralize
effects of sentence prosody.
3.1.3 Procedure
For this experiment, the participants were seated in front of a computer screen and
had to wear a microphone (Rode HS1-‐P) plugged into a recorder (Marantz
PMD661). They were instructed, in Dutch, to read a number of sentences that
appeared on the screen at a normal speaking rate. The participants read the
sentences containing the vowels /ɪ, ɛ, æ, iː, uː, ʌ/ twice, yielding 120 stimuli in total,
of which 60 were target items. The written stimuli were embedded in the carrier
sentence ‘I say ___ again’. The sentences were displayed in randomized order and
separated by a 5 second interval.
Before the experiment, the children were asked some questions in English so as to
create a more English environment. These questions were asked by a native speaker
of Dutch and concerned the child’s name, age and hobbies. Even though the children
were not required to give elaborate answers, these conversations were assumed to
facilitate their switch from Dutch to English. Only one participant had trouble
answering the questions in English. The recordings were made in a classroom at the
children’s school during their lunch break. This way, the corridors were empty and
background noise was kept to a minimum.
3.1.4 Analysis and evaluation
Afterwards, the context-‐neutral words were cut out of the sentence-‐long stimuli,
analysed acoustically and used in a categorization task. The length, closeness (F1)
and frontness and liprounding (F2) of the vowels /ɪ, ɛ, æ/ were then measured
acoustically in PRAAT (Boersma & Weenink, 2012).
For the categorization task, the accentedness and comprehensibility of 30 of the
stimuli containing the vowels /ɪ, ɛ, æ/ (10 productions per vowel) were evaluated
by four native speakers, so that each vowel was evaluated 40 times. This task also
included the stimuli gathered in Experiments 2 and 3 (see Sections 4 and 5). Four
female native speakers of English participated in the evaluation phase of the
experiments. These native speakers had lived in Belgium for approximately 9
months to 8 years, but did not speak Dutch. Two of them came from the UK, from the
South and the Midlands; the other two participants came from Scotland and the US.
26
All of the participants had travelled frequently and claimed to get in contact with
different (native) accents of English quite often.
The native speakers were instructed to evaluate the pronunciation of a number of
English vowels. First, they had to select the vowel they heard out of a grid containing
the free and checked English vowels /æ, ɑ, ɒ, ɔ, ʌ, ɪ, ɛ, ei, ɜ/. Then, they had to select
how accented the pronunciation of the vowel was on a 9-‐point Likert scale ranging
from ‘no foreign accent’ (1) to ‘very strong foreign accent’ (9). The vowel choice is
interpreted as a rating of comprehensibility. Thus, if the target vowel is selected, this
vowel is believed to be comprehensible; if a non-‐target vowel is selected, the vowel
is coded as incomprehensible. It may, however, also be that the target vowel was
selected, but that the listeners found it difficult to decide which vowel to choose. In
these cases, a high foreign accent may indicate the native speaker’s hesitation in
selecting a vowel. The difference between comprehensibility and accentedness can
thus not fully be conceptualized as the difference between the selection of a vowel
and its goodness rating.
3.2 Results
3.2.1 Acoustic analysis
The reading task yielded 20 instances of each vowel, leading up to a total of 60
vowels. The mean formant values (F1 and F2) and duration for each vowel are
summarized in Table 3.
Vowel F1 (in Hz) F2 (in Hz) Duration (in ms)
/æ/
(N=20)
Mean 697 1629 124
Std. Deviation 101 302 39
/ɛ/
(N=20)
Mean 670 1750 108
Std. Deviation 70 255 30
/ɪ/
(N=20)
Mean 473 1915 90
Std. Deviation 54 281 23
Table 3 Reading: mean values and standard deviations of vowels
While the mean F1 values of the vowels /æ/ (F1: 6973) and /ɛ/ (F1: 670) are fairly
similar, the standard deviation is the highest for the /æ/. It may be that this higher
variability reflects the hesitation some of the participants displayed when reading
the grapheme <a>. The vowels /ɛ/ (F2: 1750) and /æ/ (F2: 1629) to differ slightly in 3 Formant values are given in Hz.
27
terms of F2 values. Still, it still seems that the formant values of the /æ/ are not
differentiated enough in order to clearly contrast with those of /ɛ/. The small
acoustic distance between the /ɛ/ and /æ/ productions of the children can also be
seen in Figure 4.
Compared to the formant values for RP /æ/ (F1: 738; F2: 1576) reported by
Hawkins & Midgely (2005), the participants’ pronunciation of /æ/ is surprisingly
similar to the English /æ/. The F1 and F2 values of /ɛ/, on the contrary, are
respectively higher and lower than those of RP /ɛ/ (F1: 514; F2: 1921). In fact, the
participants’ /ɛ/ shows properties more similar to those of /æ/, as can be seen in
Figure 4. The /ɪ/, finally, has a slightly lower mean F2 value but higher F1 value than
the RP /ɪ/ (F1: 373; F2: 2097) reported by Hawkins & Midgely (2005). In fact, Figure
4 shows how the participants’ productions of /ɪ/ are acoustically closer to RP /ɛ/
than /ɪ/.
Figure 4 Reading: vowel diagram for participants (circled) and Hawkins & Midgely (2005)
We find similar data when comparing the participants’ English pronunciation of /ɛ/
and /ɪ/ to the Data for the Standard Dutch vowels. In both cases, the children’s
realizations of /ɛ/ (F1: 670) and /ɪ/ (F1: 473) are more open than the Standard
Dutch realizations of /ɛ/ (F1: 528) and /ɪ/ (F1: 410), as is reflected in the F1 values.
3.2.2 Native speaker evaluation
Table 4 presents the native speakers’ evaluations of the Flemish vowel productions.
æɛ
ɪ
æ
ɛɪ
200
400
600
800
1000
120050010001500200025003000
F2 (Hz)
F1 (H
z)
28
Response Target /ɑ/ /æ/ /ɛ/ /ei/ /ɪ/ /ɒ/ /ʌ/ /ɜ/ /æ/
5 (4.0) 12.5%
19 (3.9) 47.5%
4 (5.8) 10%
0 0 1 (6) 2.5%
10 (4) 25%
1 (6) 2.5%
/ɛ/
2 (5.0) 5%
22 (3.5) 55%
9 (4.4) 22.5%
1 (7) 2,5%
0 0 4 (3.5) 10%
2 (6) 5%
/ɪ/ 0 1 (7) 2.5%
10 (6) 25%
1 (7) 2.5%
23 (3.6) 57.5%
0 3 (6.3) 7.5%
2 (4.5) 5%
Table 4 Reading: count and percentages of native speaker responses and mean accent
ratings (most frequent choices in bold)
Table 4 shows that the native speakers selected a large number of different vowels.
This is not very surprising, since the high standard deviation in the acoustic analysis
suggested that the acoustic properties of the vowels differed significantly for each
target. as was predicted, the vowels /æ/and /ɛ/ are perceived similarly. What is
surprising is that they are both perceived as /æ/ rather than /ɛ/ in half of the cases.
However, we already noted that the formant values for these two vowels were more
similar to RP /æ/ than /ɛ/.
For /æ/, the second most common response is /ʌ/ (25%). It is possible that, because
of the high variation in F1 and F2 values, some tokens were more central variants,
resembling the vowel /ʌ/. The target vowel /æ/ is less frequently perceived as /ɑ/
(12.5%) and /ɛ/ (10%). In terms of accentedness, it seems that the three most
frequent vowel choices were perceived as less accented than the less frequent
choices. However, the target /æ/ is never perceived as a native English vowel
production, regardless of the native speakers’ responses.
The /ɛ/ was most frequently perceived as /æ/ (55%), and this is equally the least
accented vowel, together with /ʌ/ (10%). The acoustic analysis showed that there
was a slight difference between the vowels /æ/and /ɛ/. This is also reflected in the
vowel choices for these two vowels. Apart from the most frequent choice for /æ/,
target /æ/ is perceived more often as /ɑ/, while target /ɛ/ tends to be perceived as
/ɛ/ (22.5%) more often. For those cases where the opposite is true, the vowels have
a higher accentedness rating.
Finally, the /ɪ/ is most frequently perceived as /ɪ/ (57.5%), as expected. This is at
the same time also the least accented choice. The /ɪ/ is also commonly perceived as
/ɛ/ (25%), but this vowel is already perceived as very accented. As with the
previous target vowels, the participants sometimes also perceived the more central
vowel /ʌ/ (7.5%).
29
In sum, the evaluations of the vowels from the reading task are clustered in
important ways. While both /æ/and /ɛ/ are clustered around /æ/, each seems to
deviate in its own way. The target vowel /ɪ/, then, is the only vowel that is perceived
accurately in more than half of the cases.
Table 5 summarizes the mean accentedness ratings for the target vowels. The
accentedness ratings for the /æ/ and /ɪ/ are both higher for the comprehensible
vowels than for the incomprehensible ones, though the difference is clearly more
marked for the /ɪ/. This suggests that the /ɛ/ and, to a lesser extent, the /æ/ have
more competing vowels, since the accentedness ratings of the incomprehensible
choices are similar or lower to those of the comprehensible choices. This is the most
striking for /ɛ/, where incomprehensible choices are less accented than
comprehensible ones. From Table 4, we could already surmise that the target vowel
/ɛ/ has a tendency to be perceived as /æ/.
Target vowel Mean accent rating N %
/æ/ Incomprehensible 4.52 21 52.5% Comprehensible 3.89 19 47.5%
/ɛ/ Incomprehensible 3.90 31 77.5% Comprehensible 4.44 9 22.5%
/ɪ/ Incomprehensible 6.00 17 42.5% Comprehensible 3.57 23 57.5%
Table 5 Reading: mean accent ratings for target vowels
A statistical correlation analysis equally confirmed the strong relation between
comprehensibility and accentedness for the /ɪ/. Comprehensible targets tend to be
significantly less accented than incomprehensible targets (r(40) = -‐0.635, p < 0.001).
No such correlation was observed for /ɛ/ and /æ/.
3.2.3 Statistical analysis
Next, we will examine the influence of the acoustic weight of the F1 and F2 values,
and the duration of the vowels on their comprehensibility and accentedness. A
statistical correlation analysis revealed that the comprehensibility of the target
vowel /æ/ was only significantly influenced by the F2 values of the vowels (r(40) =
0.388, p < 0.05). Specifically, the higher the F2 value of the /æ/, the more
comprehensible it was.
For the vowel /ɛ/, vowel duration was found to influence the comprehensibility of
the vowel (r(40) = 0.332, p < 0.05). Longer instances of the vowel were thus found to
30
be more comprehensible. Moreover, the F1 value of this vowel also influenced
accentedness (r(40) = -‐0.355, p < 0.05), but not comprehensibility. Vowels with
higher F1 values were less accented but not necessarily more comprehensible.
The F2 values of the /ɪ/, finally, significantly influenced both accentedness and
comprehensibility. Vowels with higher F2 values were both more comprehensible
(r(40) = 0.332, p < 0.05) and less accented (r(40)= -‐0.313, p < 0.05).
The influence of the acoustic properties on vowel accentedness should always be
interpreted on the basis of the vowels’ comprehensibility. In Table 4, it was revealed
that the least accented vowels for /ɛ/ were not the most comprehensible ones. Thus,
because the /æ/ is the least accented vowel choice for /ɛ/, the influence on
accentedness bears on the perceived vowel /æ/, not /ɛ/. In this case, instances of
/ɛ/ with a higher F1 value were perceived as /æ/ with a lower accentedness.
3.3 Discussion
The data from the reading task indicate several important tendencies. First, the
acoustic analysis of the vowels showed that the children’s vowel productions of
English /æ/ and /ɛ/ were acoustically more similar to RP /æ/ than /ɛ/. On first
sight, this is surprising, since descriptions of Dutch tend to agree that the Dutch
vowel in ‘bed’ is an /ɛ/, rather than an /æ/ (e.g. Booij, 1999). However, the Dutch
variety spoken in East-‐Flanders is well known for having a more open pronunciation
of the vowel /ɛ/ (see Collins & Vandenbergen, 2000).
A comparison with the data from Adank et al. (2004) revealed that this vowel was
indeed realized with a higher F1 value. Moreover, the data suggested that the
participants’ realization of /ɛ/ was acoustically closer to RP /æ/ than /ɛ/. These
findings were corroborated by the native speaker evaluation. In the majority of the
cases, both vowels were perceived as /æ/. Note, however, that both vowels were
comprehensible in less than half of the cases and received relatively high accent
ratings.
Secondly, the participants’ productions of the vowel /ɪ/ were found to be
acoustically close to English /ɛ/. Compared to the Standard Dutch pronunciation, the
children’s pronunciation would also have been more open in Dutch. In the native
speaker evaluations, this vowel was also often perceived as /ɛ/, although it received
high accent ratings. In consequence, the vowel was only comprehensible in 57% of
31
the cases. Furthermore, a statistical analysis revealed that instances of /ɪ/ with
higher F2 values were more comprehensible and less accented. This is not
surprising, since these instances are acoustically more distant from RP /ɛ/.
In sum, the data suggested that there is indeed a merger of the English vowels /æ/
and /ɛ/ in the pronunciation of the children. While the vowel /ɪ/ was
comprehensible in the majority of the cases, it is also often perceived as a more
accented realization of /ɛ/. Moreover, the vowels /ɛ/ and /ɪ/ were realized with
higher F1 values than in RP vowels. Possibly, this could be a transfer from the
children’s regional accent. Even though the reading task did not seem challenging
for the children, we have noted a significant degree of variation in the pronunciation
of the vowels that was also reflected in the evaluation task.
4. Experiment 2: Repetition
4.1 Methodology
4.1.1 Participants
The participants in Experiment 2 were the same as those in Experiment 1.
4.1.2 Materials
This experiment was designed to elicit the same vowels as in Experiment 1, i.e. the
vowel set /ɪ, ɛ, æ/ and the filler vowels /iː, uː, ʌ/ in a /bVt/ context.
4.1.3 Procedure
A female native speaker of (Southern) British English was instructed to read the list
of carrier phrases from the reading task twice at a normal speaking rate. Thus, in
total, 12 sentences containing the words ‘bit’, ‘bet’, ‘,bat’, ‘beat’, ‘boot’, ‘but’ were
recorded for the repetition task. The native speaker had lived in Belgium for about 2
years but had not taken any courses in Dutch and used only English both at work
and at home. The recordings were made in a silent room, using a portable recorder
(Marantz PMD661) and a headset microphone (Rode HS1-‐P).
In Experiment 2, the 12 stimuli were played to the 10 participants through
headphones (B&W P-‐5) that were superimposed on the microphone. The children
were told that they would hear some English sentences and had to repeat them after
a two-‐second pause. This pause was used in order to reduce the possible effects of
imitation. The stimuli were played in randomized order. The instructions were
32
provided in Dutch in order to ensure that the participants understood the task. In
those cases where a participant’s answer was unclear and was told to repeat his/her
answer, the participant was told only his/her last answer would be analysed.
Experiment 2 yielded 120 stimuli in total, of which 60 were target words, or 20
productions per vowel. The repetition task was carried out immediately after the
reading task.
4.1.4 Analysis and evaluation
The 60 target sentences were processed, analysed and evaluated in the same way as
in Experiment 1. As in the Experiment 1, half of the stimuli, i.e. 10 per vowel, were
evaluated by the native speakers.
When repeating the stimuli, some of the participants tended to pronounce the final
consonant /t/ differently. In some cases, this consonant resembled a fricative. One
participant pronounce the stimuli as /bɪχt/, /bɛχt/ and /bæχt/. Probably, the
participant interpreted the period of voicelessness following as a /h/, but
pronounced is as /χ/. Since the native speakers were asked to focus only on the
vowel, these stimuli were included in the categorization task.
4.2 Results
4.2.1 Acoustic analysis
In the repetition task, 60 target words were elicited, yielding 20 instances of each
vowel. Table 6 presents the mean F1 and F2 values and the durations of these
vowels.
Vowel F1 (in Hz) F2 (in Hz) Duration (in ms)
/æ/ (N=20)
Mean 754 1513 123 Std. Deviation 105 245 21
/ɛ/ (N=20)
Mean 701 1758 114 Std. Deviation 86 277 16
/ɪ/ (N=20)
Mean 461 2063 85 Std. Deviation 45 263 15
Table 6 Repetition: mean values and standard deviations of vowels
The vowels differ in some respects from those elicited in the reading task. The most
striking difference is the lower mean F2 value for /æ/. This is reflected in the clearer separation of /æ/ and /ɛ/ in Figure 5, compared to Figure 4. The standard
deviation of the /æ/ is also lower than in the reading task.
33
Compared to Hawkins & Midgely’s (2005) data, the Flemish participants’ /æ/ is
more or less similar to the native speakers’ realization of this vowel (F1: 738; F2:
1576). The vowel /ɛ/ has higher mean F1 values and lower F2 values and is
acoustically more similar to English /æ/, as can be seen in Figure 5. The analysis of
/ɪ/, finally, shows higher F1 values but similar F2 values in comparison to its RP
pronunciation (F1: 373; F2: 2097). The participants’ productions of /ɪ/ are
acoustically in-‐between the native speakers’ realizations of /ɪ/ and /ɛ/, but closer to
/ɪ/ than in the reading task.
Figure 5 Repetition: vowel diagram for participants (circled) and Hawkins & Midgely
(2005)
4.2.2 Native speaker evaluation
The native speaker evaluations for the repetition task vowels can be seen in Table 7.
Response Target /ɑ/ /æ/ /ɛ/ /ei/ /ɪ/ /ɔ/ /ɒ / /ʌ/ /ɜ/ /æ/ 15 (4.1)
37.5% 10 (5.5)
25% 3 (5.7) 7.5%
1 (7.0) 2.5%
0 4 (7.5) 10%
1 (4.0) 2.5%
4 (3.8) 10%
2 (6.0) 5%
/ɛ/ 2 (7.0) 5%
20 (3.6) 50%
7 (5.9) 17.5%
2 (8.0) 5%
0 0 0 3 (4.3) 7.5%
6 (4.9) 15%
/ɪ/ 0 0 9 (5.1) 22.5%
1 (9.0) 2.5%
29 (3.9) 72.5%
1 (8.0) 2.5%
0 0 0
Table 7 Repetition: count and percentages of native speaker responses and mean accent
ratings (most frequent choices in bold)
These evaluations show that the clearer differentiation between /æ/ and /ɛ/ that
was revealed in the acoustic analysis is also reflected here. The resulting split is,
æɛ
ɪ
æ
ɛɪ
200
400
600
800
1000
120050010001500200025003000
F2 (Hz)
F1 (H
z)
34
however, not more accurate: /ɛ/ is again often perceived as /æ/ (50%) and target
/æ/ tends to be perceived as /ɑ/ (37.5%).
This is not entirely surprising, since the acoustic analysis also showed that it was
mostly the target vowel /æ/ that was pronounced differently in the repetition task.
The evaluation of the reading task vowels already indicated that the /æ/ also tended
to be perceived as /ɑ/ (12.5%) and we observe this same tendency in the repetition
task. It is also worth noting that those times where the target vowel /æ/ was
perceived as /æ/, it is also more accented (5.5) than in the reading task (3.9).
The evaluation of the vowel /ɛ/, then, is similar to that of the reading task. The most
frequent vowel choice was again /æ/ (50%), and this was equally perceived as the
least accented one.
The /ɪ/ was perceived more accurately (72.5%) than in the reading task evaluations
(57.5%). Again, the /ɛ/ is the second most frequent (22.5%), but more accented
choice.
Table 8 presents the mean accentedness ratings for the comprehensible and
incomprehensible vowels.
Target vowel Accentedness N %
/æ/ Incomprehensible 4.87 30 75.0% Comprehensible 5.50 10 25.0%
/ɛ/ Incomprehensible 4.33 33 82.5% Comprehensible 5.86 7 17.5%
/ɪ/ Incomprehensible 5.73 11 27.5% Comprehensible 3.89 29 72.5%
Table 8 Repetition: mean accent ratings for target vowels
For both /æ/ and /ɛ/, the comprehensible vowel choices are perceived as more
accented than the incomprehensible ones. This is not entirely surprising,
considering the data from Table 7, which shows that these vowels were most
commonly perceived as /ɑ/ and /æ/, respectively. The /ɪ/, however, is still
perceived as less accented when it is comprehensible. As a consequence, we only
found a correlation between comprehensibility and accentedness for the /ɪ/ (r(40) =
-‐0.408, p < 0.05).
35
4.2.3 Statistical analysis
A statistical analysis of the acoustic data and the evaluations revealed that the
comprehensibility of the vowel /ɛ/ was influenced by the vowel’s duration (r(40) =
0.345, p < 0.05). Longer vowels were more comprehensible than shorter ones.
Vowel duration also had a significant influence on the perceived accentedness of the
vowel /ɪ/ (r(40) = 0.343, p < 0.05). In this case, the longer the vowel was, the more
accented it was.
4.3 Discussion
The acoustic analysis and categorization task revealed that the data from the
repetition task were different from those of the reading task. Most markedly, the
vowel merger of /æ/ and /ɛ/ observed in the reading task is not as clear in the
repetition task. Instead, we find a split in both the acoustic properties of the vowels
and their evaluations. However, this split is not the target one, as it is realized (or
perceived) as a contrast between /ɑ/ and /æ/.
One of the drawbacks of a repetition task is that the participants may be inclined to
imitate the stimuli, rather than rely on their own pronunciations. Yet, these data
suggest that the participants did not merely imitate the stimuli. If this were the case,
one would have expected values for /ɛ/ that were different from the reading task,
while these were the most similar to those of the reading task. It thus seems that the
participants relied on the same phonemic category as in the reading task for this
vowel. It may be that the participants perceived a difference between the stimuli for
/ɛ/ and /æ/ and tried to differentiate these vowels by pronouncing the latter as /ɑ/.
This would be plausible, since the /ɑ/ is equally part of their L1 (Dutch) inventory. It
thus appears that there is an influence from the aural stimuli of the repetition task,
but this manifested itself differently than expected.
Still, the /ɛ/-‐/æ/ contrast is not necessarily inherent to the children’s phonemic
system, since the split was not as strong in the reading task stimuli. Rather, it may be
the result of the children’s reliance on their L1 to perceive the vowels. In other
words, the observation that the participants perceived and reproduced a distinction
between /ɛ/ and /æ/, does not imply that they would have encoded this difference
in their phonological representations of <bet> and <bat>.
36
With regard to the pronunciation of the vowel /ɪ/, the data revealed that it was
acoustically more similar to the RP pronunciation of the vowel than in the reading
task. Possibly, the native speaker’s realization of /ɪ/ influenced the children’s
pronunciation. In consequence, the vowel was more comprehensible in the
repetition task.
The data from the repetition task were also similar to that of the reading task in a
number of respects. First, even the least accented vowel choices were never
perceived as native. The vowels /ɛ/ and /æ/ elicited in the repetition task were even
less comprehensible than those from the reading task, though the /ɪ/ was more
comprehensible. Secondly, the data also suggest a high degree of variation in the
pronunciation. In the acoustic analysis, this was reflected by the variation of the
vowels’ acoustic properties. In consequence, the native speakers perceived a high
number of different vowels. The vowel /æ/, however, was produced more
consistently in terms of F2 value and duration than in the reading task. If the greater
variation of these properties in the reading task may be put down to linguistic
insecurity caused by the orthographic input, then the smaller variation in the
repetition task may be explained in the same way. Without the potential influence of
the ambiguous grapheme <a>, the participants may have been more able to imitate
the speaker’s pronunciation more accurately.
We can thus conclude that, in contrast to the reading task, the aural stimuli from the
repetition task may have brought about a vowel contrast in the children’s
pronunciation of /ɛ/ and /æ/. The nature of this contrast further confirmed the
tendency observed in the reading task analysis, that if a contrast between /ɛ/ and
/æ/ is produced, the vowels are realized as /æ/ and /ɑ/, respectively.
5. Experiment 3: Picture-‐naming
5.1 Methodology
5.1.1 Participants
The participants for Experiment 3 were the same as those in Experiments 1 and 2.
5.1.2 Materials
Since the design of Experiment 3 differs from the previous two experiments, other
stimuli were elicited. First, the experiment was restricted to the three vowels /ɪ, ɛ,
37
æ/, because including filler vowels would unnecessarily complicate and lengthen the
experiment. Secondly, apart from ‘bit, bet, bat’, six other words containing the same
vowels but different consonants were used (i.e. ‘pig, pink, red, bed, cat, hat’; see
Appendix A). These words were monosyllabic words that the children were likely to
know on the basis of a previous study (Simon, Sjerps & M., to appear). Since this
experiments also had a Dutch training phase, 10 monosyllabic Dutch words were
used for this part (e.g. ‘ster, zon’, see Appendix B). The cognates of these Dutch
words were avoided for the English part.
5.1.3 Procedure
The goal of Experiment 3 was to probe the phonological representations of the
participants without potential influences from orthographical input (as in
Experiment 1) or phonetic input (as in Experiment 2). This, however, presented two
problems. First, the target words had to be actively known by the children. Secondly,
the target words also had to be intuitively picturable. Alternatively, these issues
could be overcome by using a pre-‐test training period. For practical reasons, this
was not possible in the framework of the present study4.
One solution to these problems is to work with a delayed repetition (see Tsukada et
al., 2005). In such a task, each target word would be pictured and elicited several
times with an aural stimulus being played during the first elicitation. The
subsequent elicitations are then assumed to be less reliant on this aural stimulus
and thus more representative of the participant’s phonological representation. For
the present study, however, delayed repetition does not prove very practical. First,
this kind of test resembles the repetition task, since both tasks require the repetition
of an aural stimulus. Secondly, there may be an asymmetry between words that the
participants know, and for which the aural stimulus is superfluous, and words for
which the stimulus is necessary.
Hence, a third and hitherto unexplored method, which consists of two parts, is used
in this experiment. First, the participant has to name a picture s/he sees on a
computer screen. The pictured word is a basic English word (see Section 5.1.2)
containing the vowel /ɪ/, /ɛ/ or /æ/. Then, the participant has to transfer the vowel
of this word to the context-‐neutral /bVt/. For instance, upon seeing a picture of a 4 A picture-‐naming task with a training period has the advantage of being less cognitively challenging during the experiment. Still, if such a training period is not possible, this experiment may prove an interesting alternative to delayed repetition tasks.
38
cat, the participant would say ‘cat sounds like bat’. This way, the participant
produces the context-‐neutral target without aural or orthographical stimuli.
Because the design of the experiment was more complex than that of Experiments 1
and 2, the children first completed a similar test in Dutch. This test used 10 basic
Dutch words, containing the Dutch vowels /u, a, ɑ, ɪ, o, ɔ, ɛ/ (see Appendix B). The
participants were instructed that they would see a number of pictures and had to
repeat the vowel in the sentence ‘___ klinkt als bVt’ (‘___ sounds like bVt’). The
experimenter then went over the pictures with the children in order to ensure that
they knew the corresponding words. In general, the participants had little problem
understanding the test and made few mistakes by the second or third illustration.
After completing the experiment in Dutch, the children were told that they now had
to repeat the exercise in English. In order to ensure that they knew the English
words, the experiment went over each picture with them. In many cases, the
children knew the pictured words. Some children did not know the word ‘pig’,
others only had to be reminded of some of the other words. Only one participant did
not know the majority of the words. In these cases, the experimenter told the
participants the target words and asked them to repeat them until they were certain
they knew them. Once the actual experiment started, the participants had no
problem remembering the words.
For the English part, the number of vowels was reduced to the three target vowels
/ɪ, ɛ, æ/. Each vowel was represented by two different illustrated words. This
yielded 6 picturable, context-‐dependent words and 6 context-‐neutral words per
participant, for a total of 60 context-‐dependent and 60 context-‐neutral words, or 40
productions per vowel. They were told that they had to use the English sentence ‘___
sounds like bVt’. If the participants hesitated over an answer, they were asked to
repeat that sentence, though this was rarely necessary. The experiment was
conducted in the same setting as the previous experiments.
Even though this experiment is more cognitively challenging than Experiments 1
and 2, it is worth testing because it provides an alternative to the use of a delayed
repetition. The complexity of the task is also reduced by using a Dutch training test.
It seemed that the participants easily understood the design of the experiment.
Furthermore, the participants were already more familiar with the target words
since they had already come across them in Experiments 1 and 2.
39
5.1.4 Analysis and evaluation
The 60 context-‐neutral and 60 context-‐dependent words were extracted from the
sentences, processed and acoustically analysed in the same way as for Experiments
1 and 2. The native speakers evaluated 30 of the context-‐neutral words.
In total, 240 words were elicited and analysed in the three experiments. Of these
240 elicitations, 90 were also evaluated by native speakers.
5.2 Results
5.2.1 Acoustic analysis
Table 9 presents the acoustic properties of the vowels of the 60 context-‐neutral and
60 context-‐dependent words.
Taking both groups of words together, the /æ/ and /ɛ/ are again quite similar in
terms of formant values and duration, though the mean F1 and F2 of /æ/ are slightly
higher. Consequently, the contrast between both vowels is minimal (see Figure 6).
Turning to the standard deviations of these values, it seems that the /æ/ and /ɛ/ are
realized with a similar consistency. The F2 values of the /ɪ/, however, show more
variation.
Vowel F1 (in Hz) F2 (in Hz) Duration (in ms) /æ/ (N=40)
Mean 730 1791 145 Std. Deviation 79 224 28
/ɛ/ (N=40)
Mean 708 1754 144 Std. Deviation 75 235 31
/ɪ/ (N=40)
Mean 512 2056 119 Std. Deviation 52 351 26
Table 9 Picture-‐naming: mean values and standard deviations of vowels
The most important difference between these results and those from the previous
tasks are the vowel durations. The vowels elicited in this task are at least 20 ms
longer than those from the other tasks. This may be partly explained by the sentence
context in which the words were elicited. In the picture-‐naming task, the target
words were either in initial or final position. This different prosodic context might
have influenced the duration of the vowels. The design of the task could also have
influenced the vowel durations. Since this task was more cognitively taxing than the
previous two tasks, it is possible that the participants hesitated more when
pronouncing the target words, leading to a slower speech rate and longer vowel
40
durations. However, even if the complexity of the task influenced the vowel
durations, the vowel durations of /æ/ and /ɛ/ are comparable and longer than the
/ɪ/, as in the other tasks. Moreover, the standard deviation of these durations (26-‐31
ms) is comparable to that of the reading task (23-‐39 ms). If hesitation were the main
cause of longer durations, one would expect more diverging results.
Figure 6 Picture-‐naming: vowel diagram for participants (circled) and Hawkins & Midgely
(2005)
In comparison to Hawkins & Midgely’s (2005) data, the formant values for the
vowels /æ/ and /ɛ/ are again more similar to English /æ/ (F1: 738; F2: 1576) than
to /ɛ/ (F1: 514; F2: 1921), though the F1 and F2 values are slightly higher. The
values for /ɪ/, then, are comparable in terms of F2 values, but the mean F1 value is
again higher, so that it is acoustically closer to RP /ɛ/ than /ɪ/. As in the reading task,
the vowels /ɛ/ and /ɪ/ were also realized with higher F1 values than in Standard
Dutch.
Since the design of the picture-‐naming task requires the vowels realized in the
context-‐dependent words to be transferred to the context-‐neutral words, we will
also compare these two categories (Table 10).
æɛ
ɪ
æ
ɛɪ
200
400
600
800
1000
120050010001500200025003000
F2 (Hz)
F1 (H
z)
41
Vowel F1 (in Hz) F2 (in Hz) Duration (in ms)
Context-‐dependent
/æ/ (N=20)
Mean 749 1824 140 Std. Deviation 80 222 27
/ɛ/ (N=20)
Mean 719 1729 132 Std. Deviation 78 216 33
/ɪ/ (N=20)
Mean 520 2157 124 Std. Deviation 56 340 33
Context-‐neutral
/æ/ (N=20)
Mean 712 1758 152 Std. Deviation 74 227 29
/ɛ/ (N=20)
Mean 697 1778 156 Std. Deviation 72 257 23
/ɪ/ (N=20)
Mean 503 1955 114 Std. Deviation 48 341 16
Table 10 Comparison: mean values and standard deviations of vowels
While the mean formant values vary slightly between both groups (F1 values tend to
be higher for the pictured words), it seems that the difference between /æ/ and /ɛ/
on the one hand and /ɪ/ on the other hand is similar. Even the small difference
between /æ/ and /ɛ/ is maintained. The durations of the vowels are slightly higher
for context-‐neutral /æ/ and /ɛ/, but not for /ɪ/. A potential reason for this might
again be the prosodic influence of the sentence. Since the difference between the
different vowels is maintained, it thus seems that the participants were indeed able
to transfer the vowel from the pictured word to the context-‐neutral word.
5.2.2 Native speaker evaluation
Table 11 presents the native speakers’ evaluations of the vowels from the picture-‐
naming task. As in the reading task, the /æ/ is rarely perceived as /ɑ/ (2.5%), but
more frequently perceived as /ɛ/ (17.5%). It is still most often perceived as /æ/
(60%), though, and is then the least accented vowel.
For /ɛ/, the most commonly perceived vowel is again /æ/ (45%) and this vowel also
received the lowest accentedness rating. Those times when it is perceived as /ɛ/
(22.5%), it was also judged to be more accented.
The /ɪ/, finally, is comprehensible in about half of the cases (52.5%) and received a
remarkably low accentedness rating (2.9). The second most commonly perceived
vowel here is the /ɛ/ (32.5%), though it is already deemed more accented in this
case (5.2).
42
Response Target /ɑ/ /æ/ /ɛ/ /ei/ /ɪ/ /ɔ/ /ɒ/ /ʌ/ /ɜ/ /æ/ 1 (5.0)
2.5% 24 (4.0)
60% 7 (5.0) 17.5%
1 (8.0) 2.5%
0 0 0 1 (6.0) 2.5%
6 (4.7) 15%
/ɛ/ 3 (6.0) 7.5%
18 (3.6) 45%
9 (5.0) 22.5%
1 (7.0) 2.5
0 1 (6.0) 2.5%
1 (5.0) 2.5%
2 (7.5) 5%
5 (4.0) 12.5%
/ɪ/ 0 1 (6.0) 2.5%
13 (5.2) 32.5%
0 21 (2.9) 52.5%
0 0 4 (7.0) 10%
1 (4.0) 2.5%
Table 11 Picture-‐naming: count and percentages of native speaker responses and mean
accent ratings (most frequent choices in bold)
The mean accentedness ratings in Table 12 reveal a similar pattern to that in the
reading task, except that the /æ/ is slightly more comprehensible and the /ɪ/
slightly less comprehensible in the picture-‐naming task.
Target vowel Accentedness N %
/æ/ Incomprehensible 5.13 16 40.0%
Comprehensible 4.04 24 60.0%
/ɛ/ Incomprehensible 4.39 31 77.5%
Comprehensible 5.00 9 22.5%
/ɪ/ Incomprehensible 5.58 19 47.5%
Comprehensible 2.86 21 52.5%
Table 12 Picture-‐naming: mean accent ratings for target vowels
The accentedness ratings for this task are higher for both /æ/ and /ɛ/, though the
/ɪ/ is the least accented in this task. A statistical correlation analysis showed that
comprehensible instances of /æ/ and /ɪ/ tended to be less accented. This correlation
was stronger for /ɪ/ (r(40) = -‐0.657, p < 0.01) than for /æ/ (r(40 = -‐0.317, p < 0.05).
This suggest that, for these two vowels, the target vowel is the most valid choice,
while for /ɛ/, the /æ/ seems to be the most valid (or unaccented) choice.
5.2.3 Statistical analysis
A statistical correlation analysis revealed a number of patterns of influence for the
vowels from the picture-‐naming task. The F1 value of the /æ/ significantly
influenced vowel comprehensibility (r(40) = 0.472, p < 0.01). Higher F1 values thus
increased the comprehensibility of the vowel.
43
A similar effect was noted for the accentedness of /ɛ/ (r(40) = -‐0.373, p < 0.05). In
this case, the vowel was less accented if its F1 value was higher. Longer instances of
the vowel, however, were judged more accented (r(40) = 0.381, p < 0.05).
For the /ɪ/, then, higher F1 values also made the vowel less accented to the native
speakers (r(40) = -‐0.313, p < 0.05).
5.3 Discussion
One of the presuppositions that lay behind the design of the picture-‐naming task is
that the children should be able to transfer the vowel from the context-‐dependent
words to the context-‐neutral words. The similar acoustic properties of these two
groups of words revealed that the vowels were indeed similar and that it can thus be
assumed that the vowel was transferred from one word to the other. We also
noticed that the vowel durations were slightly longer than in the previous tasks.
Most importantly, however, the contrast between /æ/ and /ɛ/ in the repetition task
is not found in the picture-‐naming task, as both vowels are perceived as /æ/.
As in the reading task, the participants’ realization of /ɪ/ was acoustically similar to
RP /ɛ/ and this was reflected in the native speakers’ categorization of the vowel.
Since the participants’ productions of the English vowels /ɛ/ and /ɪ/ are more open
than the corresponding Dutch vowels, it is possible that the children’s East-‐Flemish
accent influenced their English productions.
It thus seems that the data gathered for this task are more similar to those of the
reading task than to those of the repetition task. Except that the /æ/ was slightly
more comprehensible in this task, and the /ɪ/ slightly less comprehensible, the /æ/
and /ɛ/ were generally more accented and the /ɪ/ was significantly less accented.
6. General discussion
In the previous sections, we have described the data from the reading, repetition
and picture-‐naming tasks in order to answer our main research questions. First, we
shall address the question whether the participants were able to realize the English
vowels /ɪ/, /ɛ/ and /æ/ by examining the data from the acoustic analyses (Section
6.1) and the native speaker evaluations (Section 6.2). Secondly, we shall turn to the
issue of which of the data elicitation methods might be most suited for research into
accentedness and comprehensibility (Section 6.3).
44
6.1 Participants’ realization of /ɪ, ɛ, æ/
The acoustic analyses of the data gathering methods revealed two important points.
First, the pronunciation of /ɪ/ and /ɛ/ in the reading and picture-‐naming task was
more open than in RP and Standard Dutch. In fact, the vowels /ɪ/ and /ɛ/ were
acoustically more similar to the RP vowels /ɛ/ and /æ/, respectively. We have
already suggested that the children relied on these more open pronunciations in
their regional Dutch accent. If this should be the case, there is a transfer of their L1
vowels to the English vowels. In order to verify this claim, the regional
pronunciation of children of the Dutch vowels /ɪ/ and /ɛ/ should be examined. Still,
evidence for an open pronunciation of the /ɛ/ is, for instance, noted by Collins and
Vandenbergen (2000: 47), who state that “[s]ome Flemish dialects have a very open
vowel /ɛ/ which sounds more like [English] /æ/”.
Evidence for such a transfer was not as clear in the data from the repetition task.
While the /ɛ/ from this task is also acoustically more similar to RP /æ/ than /ɛ/, the
participants’ /ɪ/ was acoustically closer, though not identical, to RP /ɪ/ than in the
other tasks. These different results could be interpreted as an influence of the aural
stimuli on the children’s pronunciation. Upon repeating the aural stimuli, the
children may have imitated the slightly more closed pronunciation of the native
speaker’s /ɪ/.
Secondly, the participants’ realizations of /æ/ were acoustically similar to their
realizations of /ɛ/ in the reading and picture-‐naming tasks. Thus, their
pronunciation of the English vowel /æ/ was similar to RP /æ/, since their
realization of /ɛ/ was also similar to RP /æ/. In other words, the English phonemes
/ɛ/ and /æ/ are merged in the children’s pronunciation and are acoustically more
similar to RP /æ/ than /ɛ/.
The exact nature of this merger, however, is less clear. From a phonetic perspective,
the vowels are closer to [æ] than [ɛ], but descriptions of Dutch (e.g. Booij, 1999)
generally agree that Dutch phonology does not contain a phoneme /æ/. If we
consider that, because of their regional pronunciation, the participants’ /ɛ/ is
realized as [æ] (or [ɛ̝]) in Dutch and English, the phonemic value of this vowel
remains /ɛ/. From a phonological point of view, it seems the vowels are merged to
/ɛ/. We shall return to this matter in Section 6.2.
45
While the data from the repetition task did not reveal a merger of /ɛ/ and /æ/, they
did not contradict the findings from the reading and picture-‐naming tasks either.
The children’s realizations of the /ɛ/ in this task remained comparable to those in
the other task, that is, they were acoustically similar to RP /æ/. Their realizations of
English /æ/, however, were more open than those of /ɛ/, so that the vowels were
contrasted. Since Escudero et al. (2011) have pointed out that, to some extent, adult
speakers of Dutch are able to perceptually distinguish between English /ɛ/ and /æ/,
it is possible that, in this study, the participants also perceived a difference between
these vowels and tried to produce this difference.
If the children perceived English /ɛ/ as Dutch /ɛ/ and realized it as [æ] and English
/æ/ was perceived as different from /ɛ/, the speakers could not rely on their [æ]
realization of this vowel. If the learners (unconsciously) interpreted English /æ/ as
more open than /ɛ/, they would thus have to make their pronunciations [æ] more
open. A study on the perception of English vowels is needed to verify whether
children are able to perceive the /ɛ/-‐/æ/ contrast.
The fact that the data of the repetition task differ from those of the reading and
picture-‐naming tasks might be explained by the nature of the repetition task. The
influence of the aural stimuli seems to overrule the participants’ encoded
phonological system, as it allows the children to imitate the contrast. Alternatively,
the aural stimuli might activate an already encoded contrast. This would entail that
the children did not imitate the stimuli but relied on their own phonological
representations once the stimuli made them aware of this distinction. However, we
have found no proof for this explanation. On the contrary, if the contrast had been
activated in the children’s phonology, one would expect that it would remain
activated in the following task. Yet, the data for the picture-‐naming task suggest a
merger, rather than a contrast.
These findings are consistent with the predictions of the PAM-‐L2 (Best & Tyler,
2007). We have, for instance, found that because of the lack of a Dutch /ɛ/ -‐ /æ/
contrast, the learners tend to assimilate both vowels to one L1 category, namely, /ɛ/.
The fact that the participants tended to distinguish /æ/ and /ɛ/ in the repetition
task may indicate that they considered English /æ/ as a bad exemplar of /ɛ/, though
evidence of a perception task is needed here. In the long term, the learners might
also be able to encode this difference, as they were also able to imitate the contrast.
46
Finally, PAM-‐L2 also accounts for the fact that learners may consciously or
unconsciously gather information regarding the articulatory gestures at the basis of
sounds, in this case, the openness of the vowel.
Overall, the data from the reading and picture-‐naming task are thus more similar
than that of the repetition task in that they both reveal a merger of the vowels /ɛ/
and /æ/, as well as a more open pronunciation of /ɪ/. This is not to say that there
were no differences in the data of these tasks, but these were not as pronounced as
the differences between these two tasks and the repetition task. One such difference
is that the vowel realizations from the picture-‐naming task were longer than those
from the reading task. The different results in the repetition task can be explained by
the influence of the aural stimuli on the participants’ productions.
6.2 Native speakers’ perception of /ɪ, ɛ, æ/
The native speaker evaluations corroborated the findings from the acoustic
analyses. Since the results of the reading and picture-‐naming tasks are comparable,
we shall discuss these together and mention any differences where necessary. The
data of these two tasks revealed that the vowel /ɪ/ tended to be perceived as /ɪ/ or
/ɛ/ and that /ɛ/ and /æ/ were often perceived as /æ/.
The acoustic analyses already indicated that the pronunciation of the /ɪ/ was
acoustically close to that of /ɛ/ in the reading and picture-‐naming tasks, so it is
unsurprising that in some cases it is also perceived as such. However, in the majority
of the cases the /ɪ/ was perceived as /ɪ/. Moreover, when it is perceived as /ɛ/, the
vowel received higher accentedness ratings.
The data from the repetition task revealed that the /ɪ/ was more similar to the RP
pronunciation in this task. This is reflected in the categorization of this vowel, as it
was more comprehensible than in the reading and picture-‐naming tasks (see Figure
7).
The picture-‐naming and reading tasks also revealed that the acoustically similar
vowels /ɛ/ and /æ/ were also both perceived as /æ/ in the majority of the cases.
The native speaker evaluations also reflected an equal amount of variation in the
pronunciation. In both tasks, the vowel evaluations clustered around a number of
vowels in a similar way: the target vowels /ɛ/ and /æ/ were often perceived as /æ/,
/ɛ/ or /ɑ/.
47
The analysis of the realizations of /ɛ/ and /æ/ in the repetition task indicated that
the vowel /æ/ was realized differently from the other tasks. The lower
comprehensibility of the /æ/ in the repetition task (see Figure 7) reveals that this
pronunciation is not more target-‐like. In fact, the evaluations indicated that the /æ/
was perceived as /ɑ/, rather than /æ/.
There are also some differences in the data of the reading and picture-‐naming tasks.
We have already mentioned that the vowels elicited in the picture-‐naming task were
longer than those from the reading task. In the case of the /ɛ/, this resulted in a
higher accentedness rating. Furthermore, the accentedness ratings and
comprehensibility of the vowels differed slightly. These differences, however, are
not one-‐sided. If the vowels /æ/ and /ɛ/ elicited in the picture-‐naming task are
more accented, then the opposite is true for the /ɪ/. Similarly, the /æ/ from the
picture-‐naming task was more comprehensible than that from the reading task, but
its /ɪ/ was less comprehensible, as can be seen in Figure 7.
Figure 7 Comprehensibility of the vowels /ɪ, ɛ, æ/ in the reading, repetition and picture-‐
naming tasks
At this point, we would also like to mention that the design of the categorization task
had some important implications in those cases where the vowel was
incomprehensible. In these cases, the vowel’s accentedness was judged on the basis
of the perceived vowel, not the target vowel. Thus, if a native speaker perceives a
target /ɛ/ as /æ/, the accentedness rating will pertain to the /æ/. This has for
instance revealed that in certain cases target /ɛ/ and /æ/ were both perceived as
similarly accented instances of /æ/. Alternatively, the task might have been
designed so that the accentedness rating would be made on the basis of the target
vowel, by displaying the target vowel after a vowel is selected. Thus, if a native
0%
20%
40%
60%
80%
100%
/æ/ /ɛ/ /ɪ/
Reading task
Repetition task
Picture-‐naming task
48
speaker perceives a target /ɛ/ as /æ/, the accentedness rating would still bear on
the target vowel /ɛ/.
A closer consideration of the East-‐Flemish situation reveals a number of
complexities. According to the data, East-‐Flemish learners of English pronounce
English /ɛ/ and /æ/ as [æ]. In Dutch, this presents no immediate problems, since the
East-‐Flemish [æ] or the more standard pronunciation [ɛ] are both instances of the
same phoneme /ɛ/.
If the speaker relies on this [æ] in order to realize an English /æ/, the learner will be
understood correctly, but the situation is ambivalent. If the learner knows that
English has a vowel /æ/ and is also aware that his/her /ɛ/ is realized as [æ], then
there are no immediate problems. If the learner ignores this fact, s/he will have still
have a target /ɛ/ in his/her mind, since [æ] and [ɛ] remain allophones of (Dutch)
/ɛ/. The accurate pronunciation is in this case unintentional.
In the case of English /ɛ/, the situation becomes more complex. A speaker ignorant
of the fact that his/her phoneme /ɛ/ is perceived as /æ/ and realized as [æ] will rely
on this vowel to realize English /ɛ/. In this case the vowel will be incomprehensible,
but the speaker might remain unaware of this. Instead, the learner might believe
s/he correctly identified the English phoneme /ɛ/ as equivalent to his/her Dutch
phoneme /ɛ/.
If the speaker is aware that English has a contrast between these vowels and that
his/her pronunciation resembles [æ], s/he may attempt to correct this
pronunciation by relying on the more standard Dutch pronunciation of /ɛ/. In this
case, the speaker would have more or less accurately contrasted English /ɛ/ and
/æ/ by relying on, respectively, the regional and standard pronunciations of Dutch
/ɛ/. Alternatively, the speaker may try to correct his/her regional /ɛ/ without
reference to Standard Dutch /ɛ/.
However, it is more likely that a speaker aware of the English contrast, is not aware
of his/her own realization of /ɛ/ as [æ]. In this case, the learner may attempt to
make the contrast by changing his/her realization of the English /æ/, instead of that
of the English /ɛ/, because s/he may think that his/her realization of /ɛ/, i.e. [æ], is
accurate.
It should be noted that the actual situation is not as clear-‐cut as presented here. In
some cases, the learners were able to accurately contrast the vowels. In the
49
repetition task, the participants sometimes merged the vowels as /æ/, instead of
producing an /æ/ -‐ /ɑ/ contrast. This suggests that there is still a significant amount
of variation in the participants’ pronunciations and that in some cases they might
perceive the contrast, while in other cases they do not. Furthermore, the
accentedness ratings showed that none of the vowels were pronounced with native-‐
like accuracy and that the learners’ [æ] is thus not perceived as a native /æ/. This is
not surprising, since the learners were not necessarily trying to produce an /æ/.
However, we take the comprehensibility ratings as the most important guide to the
learners’ pronunciation5.
What remains to be studied is the development of the English phonological system
of these learners. Returning to the above explanations, three important situations
should be considered. In a first situation, the learner would not be aware of his/her
regional pronunciation, nor of the English contrast. We hypothesize that, in this case,
the learner would continue to realize both English vowels as /æ/. In a second
situation, the learner would be able to perceive the contrast and be aware of his/her
pronunciation. In this case, the learner might be able to adjust his/her pronunciation
of /ɛ/ and possibly acquire the contrast accurately. In the third case, the learner
might perceive the difference but not be aware of his/her own regional
pronunciation. The learner might in a first instance rely on a /æ/ -‐ /ɑ/ contrast, as
was the case in our data. The learner may then encode this contrast or, if s/he senses
the contrast is inaccurate, reject the contrast. It is possible that the learner would
then rely on his/her regional variant of /ɛ/ for both sounds, unsure of how to make
the contrast. Although these cases are hypothetical, the participants to this study
would most likely fit in the first or third situation. If a learner of English thus wishes
to accurately learn the contrast, phonological instruction should not only point out
the existence of an English phoneme /æ/, but should also indicate that the English
phoneme /ɛ/ is not necessarily the same as the learner’s realization of this phoneme
in Dutch.
5 For this reason, we have also chosen to notate the learners’ pronunciation as [æ] rather than [ɛ̝].
50
6.3 Data gathering methods for research in SLA
From the discussion in Sections 6.1 and 6.2, it transpires that the data elicitation
methods influenced the productions of the children. Nonetheless, the data from the
reading and picture-‐naming tasks were more similar than those from the repetition
task.
The data from the reading task seemed the least affected by the data elicitation
method. We hypothesized that the grapheme <a> would, in some cases, be
interpreted as the vowel /ɑ/. The acoustic analysis revealed that, although the
productions of the target vowel /æ/ were less consistent in the reading task than in
the repetition and picture-‐naming tasks, the /æ/ was only perceived as /ɑ/ in 12.5%
of the cases in the reading task.
The productions elicited in the picture-‐naming task, then, were longer than those
from the repetition and reading tasks, though this only had a significant influence on
the accentedness of the /ɪ/. Nonetheless, we have verified that the vowels from the
context-‐dependent words were transferred to the context-‐neutral words. This
revealed that the task was not too cognitively challenging for the children.
While the reading and picture-‐naming tasks yielded similar results and were
practically feasible for the target group, the reading task may still prove the most
practical method, since it requires no training period and is less demanding for the
participants. Still, the focus of the study may also be relevant. Since not all
phonological distinctions can be made in writing, a reading task may prove
impractical in some cases. The distinction between /but/ and /bʊt/, for instance,
might be difficult to elicit in a reading task. Of course, one might rely on different
words, such as <boot> and <book>, to elicit the two vowels, but then the neutral
consonantal context would be lost. If one wishes to obtain this neutral context, the
picture-‐naming task may prove more practical.
With regard to the repetition task, the influence of the task design on the data was
more noticeable. Most importantly, the data reflected the participants’ ability to
imitate non-‐native vowels rather than their phonological system, since they
contrasted the vowels /ɛ/ and/æ/ by pronouncing the latter more open. The
influence of the aural stimuli might also explain the more comprehensible
realizations of the vowel /ɪ/. This influence, however, was not limited to the vowel.
A number of participants, for instance, realized the final /t/ as a fricative. One
51
participant interpreted the period of voicelessness after the vowel of the stimuli as a
phonemic consonant /χ/. While the native speakers were told to focus on the vowel
in their ratings, it is possible that their accentedness ratings reflected some of these
factors. It should also be noted that, even if these influences may obscure the
participants’ phonological system, the influences might nevertheless be revealing.
Even if the vowel contrast found in the repetition task data was not encoded in the
participants’ phonological systems, the contrast at least indicates that the
participants were able to perceptually distinguish between the sounds. It is possible
that this perceptual difference may later be encoded as a phonological difference.
At the very least, we would like to suggest that, in the context of this paper, the
reading task and the picture-‐naming task can be advanced as valid alternatives,
since there were no significant differences between the tasks that surfaced in the
data. On the basis of the data from the repetition task, this task has some
disadvantages when studying the phonological system of learners, although it is
practical for studies focusing on the ability of learners to imitate non-‐native
contrasts.
7. Conclusion
In this paper, we compared three data gathering methodologies in order to describe
the acquisition of the English vowels /ɪ, ɛ, æ/ by Flemish children who had not yet
received formal instruction of English. Thus, we aimed to address two issues in the
Second Language Acquisition literature. First, the literature has primarily focused on
adults, while it has also been shown that the earlier a language learned, the more
successfully it will be acquired. Secondly, since this age group is under-‐investigated,
there is a lack of knowledge on research methodologies influence data and this study
provides a methodological basis for studies on phonological acquisition.
On the basis of the PAM-‐L2 (Best & Tyler, 2007), we hypothesized that the
participants would assimilate the two English vowels /ɛ/ and /æ/ to the Dutch
vowel /ɛ/, because Dutch lacks a vowel /æ/ (H1a), but that they would be able to
assimilate the vowels /ɛ/ and /ɪ/ to two different categories (H1b). We have found
evidence for a merger of /ɛ/ and /æ/, as was predicted by H1a, but the situation
proved more complicated than that described by Flege (1992) for Northern Dutch.
In the case of East-‐Flemish Dutch, both vowels tend to be perceived as /æ/ by native
speakers of English. Moreover, the acoustic analysis indicated that the vowel /ɛ/
52
was more open than in RP and Standard Dutch. Possibly, this more open
pronunciation is caused by a transfer from the participants’ East-‐Flemish accent,
although this should be verified by a study of children’s productions of Dutch
vowels. Because it was perceived as /æ/, the target vowel /ɛ/ was less
comprehensible than the vowel /æ/. Our data also suggested that, when imitating
the English vowels /ɛ/ and /æ/, these tended to be produced as /æ/ and /ɑ/,
respectively. While the vowel /ɪ/ was also pronounced more open than in RP, this
was the most comprehensible and least accented vowel. If the children’s Dutch
pronunciation of the vowels /ɛ/ and /ɪ/ are also more open than the Standard Dutch
pronunciation, the English vowels /ɛ/ and /ɪ/ were assimilated to their Dutch
counterparts, as was predicted by H1b. However, because of their regional
pronunciation, the vowels were less comprehensible than expected.
We have also found that, throughout all the tasks and for all the vowels, the
pronunciation of the participants displayed a significant amount of variation.
Moreover, the vowel productions were all perceived as at least moderately or even
strongly accented. There thus appears to be a significant difference between the
vowels of both languages (or both varieties), even in cases where the L2 vowel can
be mapped on the L1 vowel, as with the /ɪ/. In order to verify some of the claims
made in this study, future research should examine children’s perception of English
vowels.
With regard to the data gathering methods, we have found an important influence of
the method on the data. We found evidence for the influence of the grapheme <a> in
the reading task, as was predicted by H2a. The target vowel /æ/ was produced with
more variation in the reading task than in the other tasks and was also perceived as
/ɑ/ in some cases. For the repetition task, the participants’ productions were
influenced by the aural stimuli they repeated, as was predicted by H2b. Specifically,
the children tended to produce a contrast between /ɛ/ and /æ/ that was not
observed in the other tasks. With regard to the picture-‐naming task, we
hypothesized that the cognitive load of the task would influence the vowel
productions (H2c) and found that the vowels were longer than in the other tasks.
Nonetheless, the data from the reading task and picture-‐naming task were similar
and a picture-‐naming task is thus still a viable method in the absence of a learning
phase. Using similar-‐sounding words, the participants were able to transfer the
vowels from real words to the target words.
53
On a final note, this study covered only a small portion of the Dutch and English
phonology and it would be interesting to see what the effect of regional variation in
Dutch is on the acquisition of the whole English phonological system. Since the
native speakers of English in this study were not from a homogeneous geographical
background, future research could also examine the effect of geographical
background on vowel perception. This might also lead to a more thorough
investigation into the acoustic weight of formant values and vowel duration.
In sum, we hope that the methodological comparison will provide a basis for more
consistent research methods in future research. We have revealed that, although the
predictions of the PAM-‐L2 (Best & Tyler, 2007) were correct, special attention
should be devoted to regional variation in order to correctly predict the acquisition
of sounds.
54
8. Appendix
8.1 Appendix A: English context-‐dependent words
/ɪ/ Pig
Kiss
/ɛ/ Red
Bread
/æ/ Hat
Cat
8.2 Appendix B: Dutch context-‐dependent words
Zwart Bal
Paard Bril
Boom Noot
Groen Boek
Zon Ster
55
Bibliography
Adank, P., R. van Hout & R. Smits. 2004. An acoustic description of the vowels of Northern and Southern Standard Dutch. Acoustical Society of America 116, 1729-‐38.
Berns, M. 1995. English in the European Union. English Today 11, 3-‐11.
Best, C. T., G. W. McRoberts & E. Goodell. 2001. Discrimination of non-‐native consonant contrasts varying in perceptual assimilation to the listener's native phonological system. The Journal of the Acoustical Society of America 109, 775-‐94.
Best, Catherine T. & Michael D. Tyler. 2007. Nonnative and second-‐language speech perception. Linguistic Experience in Second Language Speech Learning,
Boersma, P. & D. Weenink. 2012. Praat: doing phonetics by computer. Computer program.
Bohn, O.S. 1995. Cross-‐language speech perception in adults: First language transfer doesn't tell it all. In W. Strange (ed.), Speech perception and linguistic experience: Issues in cross-‐language research, 275-‐300. Timonium, MD: York Press.
Booij, Geert. 1999. The Phonology of Dutch (The phonology of the world's languages). New York: Oxford University Press Inc.
Collins, B. & A.M. Vandenbergen. 2000. Modern English pronunciation, 2 edn. Gent: Academia Press.
Cunningham, U. 2009. Phonetic correlates of unintelligibility in Vietnamese-‐accented English. Paper presented at Fonetik, Stockholm.
De Meyere, E. 2010. English Pronunciation Learning in Flanders: A research on pupils’ attitudes of one secondary school in Flanders. dissertation, Ghent University.
Dillon, G. 2009. The Vowel Quadrilaterals for Stereotypical National Accents.
Escudero, P. 2001. The role of the input in the development of L1 and L2 sound contrasts: language-‐specific cue weighting for vowels. Paper presented at Annual Boston University Conference on Language Development, Boston.
Escudero, P., T. Benders & S.C. Lipski. 2009. Native, non-‐native and L2 perceptual cue weighting for Dutch vowels: The case of Dutch, German, and Spanish listeners. Journal of Phonetics 37, 452-‐65.
Escudero, Paola, Ellen Simon & Holger Mitterer. 2012. The perception of English front vowels by North Holland and Flemish listeners: Acoustic similarity predicts and explains cross-‐linguistic and L2 perception. Journal of Phonetics 40, 280-‐8.
Flege, J. E. 1995. Second Language Speech Learning: Theory, Findings, and Problems. In W. Strange (ed.), Speech Perception and Linguistic Experience: Issues in Cross-‐language research, 229-‐73. Timonium, MD: York Press.
Flege, J. E. & J. Hillenbrand. 1984. Limits on phonetic accuracy in foreign language speech production. The Journal of the Acoustical Society of America 76, 708-‐21.
56
Flege, J. E. & G. H. Yeni-‐Komshian. 1999. Age Constraints on Second-‐Language Acquisition. Journal of Memory and Language 41, 78-‐104.
Flege, J.E. 1992. The intelligibility of English vowels spoken by British and Dutch talkers. In R. D. Kent (ed.), Intelligibility in Speech Disorders: Theory, Measurement,and Management, 157-‐232. Amsterdam: John Benjamins.
Flege, James, Murray J. Munro & Ian R. A. Mackay. 1995. Effects of age of second-‐language learning on the production of English consonants. Speech Communication 16, 1-‐26.
Floccia, C., J. Butler, J. Goslin & L. Ellis. 2009. Regional and Foreign Accent Processing in English: Can Listeners Adapt? Journal of Psycholinguistic Research 38, 379-‐412.
Goethals, M. 1997. English in Flanders (Belgium). World Englishes 16, 105-‐14.
Hawkins, S. & J. Midgley. 2005. Formant frequencies of RP monophthongs in four age groups of speakers. Journal of the International Phonetic Association 25, 183-‐99.
Idemaru, K. & S. Guion-‐Anderson. 2010. Relational Timing in the Production and Perception of Japanese Singleton and Geminate Stops. Phonetica 67, 25-‐46.
Jenkins, J. 2000. The Phonology of English as an International Language. Oxford: Oxford University Press.
Kachru, B. 1992. The other tongue: English across cultures. Champaign: University of Illinois Press.
Kennedy, S. & P. Trofimovich. 2008. Intelligibility, Comprehensibility, and Accentedness of L2 Speech: The Role of Listener Experience and Semantic Context. The Canadian Modern Language Review 63, 459-‐89.
Kewley-‐Port, D., R. Akhane-‐Yamada & K. Aikawa. 1996. Intelligibility and acoustic correlates of Japanese Accented English vowels. Paper presented at ICSLP.
Lippens, Charlotte. 2010. The English productive vocabulary of secondary school children in Flanders prior to English instruction. dissertation, Ghent University.
Mah, Jennifer & John Archibald. 2003. Acquisition of L2 Length Contrasts. Paper presented at Proceedings of the 6th Generative Approaches to Second Language Acquisition Conference (GASLA 2002), Somerville, MA.
Major, Roy C. & Eunyi Kim. 1999. The Similarity Differential Rate Hypothesis. Language Learning 49, 151-‐83.
Matsumoto-‐Shimamori, S., T. Ito, S. E. Fukuda & S. Fukuda. 2011. The transition from the core vowels to the following segments in Japanese children who stutter: The second, third and fourth syllables. Clinical Linguistics & Phonetics, 25, 804–13.
McAllister, Robert. 2001. Experience as a factor in L2 phonological acquisition. Working Papers 49, 116-‐9.
57
Meador, Diane, James Flege & Ian R. A. Mackay. 2000. Factors affecting the recognition of words in a second language. Bilingualism: Language and Cognition 3, 55-‐67.
Motohashi-‐Saigo, Miki & Debra M. Hardison. 2009. Acquisition of L2 Japanese Geminates: Training with Waveform Displays. Language Learning and Technology 13, 29-‐47.
Munro, M. J. & T. M. Derwing. 1995. Foreign Accent, Comprehensibility and Intelligibility in the Speech of Second Language Learners. Language Learning 45, 73-‐97.
Munro, M.J. 2008. Foreign accent and speech intelligibility. In Jette G. Hansen Edwards & Mary L. Zampini (eds.), Phonology and second language acquisition, 199-‐224. Amsterdam: John Benjamins Publishing Company.
Oh, G. E., et al. 2011. A one-‐year longitudinal study of English and Japanese vowel production by Japanese adults and children in an English-‐speaking setting. Journal of Phonetics 39, 156-‐67.
Simon, E., M. Sjerps & Fikkert. M. to appear. Phonological representations in children’s native and non-‐native lexicon. Bilingualism: Language and Cognition,
Simon, Ellen. 2005. How native-‐like do you want to sound? A study on the pronunciation target of advanced learners of English in Flanders. Moderna Sprak 99, 12-‐21.
Summers, C., et al. 2010. Bilingual performance on nonword repetition in Spanish and English. International Journal of Language and Communication Disorders 45, 481-‐93.
Tsukada, K., et al. 2005. A developmental study of English vowel production and perception by native Korean adults and children. Journal of Phonetics 33, 263-‐90.
Wang, H. & V.J. van Heuven. 2006. Acoustical analysis of English vowels produced by Chinese, Dutch and American speakers. Linguistics in the Netherlands 2006, 237-‐48.
Whalen, D.H., C. T. Best & J. R. Irwin. 1997. Lexical effects in the perception and production of American English /p/ allophones. Journal of Phonetics 25, 501-‐28.
Windsor, J., K. Kohnert, K. Lobitz & G Pham. 2010. Cross-‐language nonword repetition by bilingual and monolingual children. American Journal of Speech-‐Language Pathology 19, 298-‐310.