![Example waveform, spectrogram, and F1 trajectory (the critical difference between diphthongal and monophthongal realizations) for the SDS realization of /e:n/ as [ein]. Toward the end of the vowel, the F1 falls.](https://www.researchgate.net/publication/333793268/figure/fig1/AS:11431281247854478@1716945053583/Example-waveform-spectrogram-and-F1-trajectory-the-critical-difference-between_Q320.jpg)
![Example waveform, spectrogram, and F1 trajectory (the critical difference between diphthongal and monophthongal realizations) for the non-SDS realization of /e:n/ as [e:n]. The F1 stays stable throughout the vowel.](https://www.researchgate.net/publication/333793268/figure/fig2/AS:11431281247801276@1716945053816/Example-waveform-spectrogram-and-F1-trajectory-the-critical-difference-between_Q320.jpg)
Access to this full-text is provided by Frontiers.
Content available from Frontiers in Neuroscience
This content is subject to copyright.
ORIGINAL RESEARCH
published: 14 June 2019
doi: 10.3389/fnins.2019.00546
Frontiers in Neuroscience | www.frontiersin.org 1June 2019 | Volume 13 | Article 546
Edited by:
Mary Rudner,
Linköping University, Sweden
Reviewed by:
Alexander Francis,
Purdue University, United States
Erin Ingvalson,
Florida State University, United States
*Correspondence:
Cesko C. Voeten
c.c.voeten@hum.leidenuniv.nl
Specialty section:
This article was submitted to
Auditory Cognitive Neuroscience,
a section of the journal
Frontiers in Neuroscience
Received: 05 March 2019
Accepted: 13 May 2019
Published: 14 June 2019
Citation:
Voeten CC and Levelt CC (2019) ERP
Responses to Regional Accent Reflect
Two Distinct Processes of Perceptual
Compensation.
Front. Neurosci. 13:546.
doi: 10.3389/fnins.2019.00546
ERP Responses to Regional Accent
Reflect Two Distinct Processes of
Perceptual Compensation
Cesko C. Voeten 1,2
*and Clara C. Levelt 1,2
1Leiden University Center for Linguistics, Leiden University, Leiden, Netherlands, 2Leiden Institute for Brain and Cognition,
Leiden University, Leiden, Netherlands
Humans possess a robust speech-perception apparatus that is able to cope with
variation in spoken language. However, linguists have often claimed that this coping ability
must be limited, since otherwise there is no way for such variation to lead to language
change and regional accents. Previous research has shown that the presence or absence
of perceptual compensation is indexed by the N400 and P600 components, where the
N400 reflects the general awareness of accented speech input, and the P600 responds
to phonological-rule violations. The present exploratory paper investigates the hypothesis
that these same components are involved in the accommodation to sound change, and
that their amplitudes reduce as a sound change becomes accepted by an individual. This
is investigated on the basis of a vowel shift in Dutch that has occurred in the Netherlands
but not in Flanders (the Dutch-speaking part of Belgium). Netherlandic and Flemish
participants were presented auditorily with words containing either conservative or novel
vowel realizations, plus two control conditions. Exploratory analyses found no significant
differences in ERPs to these realizations, but did uncover two systematic differences.
Over 9 months, the N400 response became less negative for both groups of participants,
but this effect was significantly smaller for the Flemish participants, a finding in line with
earlier results on accent processing. Additionally, in one control condition where a “novel”
realization was produced based on vowel lengthening, which cannot be achieved by any
rule of either Netherlandic or Flemish Dutch and changes the vowel’s phonemic identity,
a P600 was obtained in the Netherlandic participants, but not in the Flemish participants.
This P600 corroborates a small number of other studies which found phonological
P600s, and provides ERP validation of earlier behavioral results that adaptation to
variation in speech is possible, until the variation crosses a phoneme boundary. The
results of this exploratory study thus reveal two types of perceptual-compensation
(dys)function: on-line accent processing, visible as N400 amplitude, and failure to recover
from an ungrammatical realization that crosses a phoneme boundary, visible as a P600.
These results provide further insight on how these two ERPs reflect the processing
of variation.
Keywords: accent processing, perceptual compensation, language change, language variation, N400, P600
Voeten and Levelt ERP Responses to Regional Accent
1. INTRODUCTION
It has been successfully argued by many historical linguists that
one of the key factors responsible for language variation and
change, particularly when it relates to phonetics and phonology,
is a poor ability of the human perceptual system to deal
with unintentional variation in the speech signal, leading to
misperception of a speaker by a listener (e.g., Boyd-Bowman,
1954; Hyman, 1976, 2013; Ohala, 1981, 2012; Blevins, 2004;
Bermúdez-Otero, 2015). However, multiple decades of research
on speech processing by psycholinguists show that, in fact,
the human speech system is very capable of handling non-
meaningful variation, such as variation due to anatomical
differences between speakers or the use of a regional accent.
Processes such as perceptual learning (Norris et al., 2003), rate
normalization (Bosker and Reinisch, 2015), compensation for
coarticulation (Mann and Repp, 1980), and many other innate
or acquired perceptual skills (cf. Cutler, 2012) enable the listener
to accurately make the link between the forms speakers intend
vs. the sounds they actually produce. If historical linguists are
correct that the driving force behind linguistic change (and
particularly sound change) is misperception, then the question is
when and how these perceptual-compensation processes found
in psycholinguistics “give way”, i.e. fail to correctly compensate
for variation, thus enabling historical sound change to actuate.
In empirical terms: under what conditions can we detect
psycho- or neurolinguistic correlates of unsuccessful perceptual
compensation for variation? The present paper provides a
starting point in answering this question using evidence from
ERP data.
Evans and Iverson (2007) have shown that it is possible
to detect long-term accommodation to variation in speech,
by investigating the speech production and perception of 19
English first-year university students. These students hailed
from different dialect regions of the United Kingdom, and
were shown behaviorally to adapt their speech production to
the Standard-Southern-British-English university norms. In
addition, a correlation was found with participants’ perception
of accented speech, but the latter did not reliably change on
its own over time. The present study takes a similar approach,
but focuses on the processing aspect. The language used
for the investigation is Dutch. Dutch is spoken both in the
Netherlands and in the northern part of Belgium (henceforth:
Flanders), but due to thorough standardization processes that
took place in the Netherlands but not in Flanders (Grondelaers
and van Hout, 2011), there are significant differences in the
phonological systems of these two varieties. The Netherlandic
variety (henceforth: Standard Dutch), has undergone changes in
its distribution of the tense mid vowels (/e:,ø:,o:/), diphthongs
(/Ei,œy,Au/), and rhotic (/r/). Specifically, Standard Dutch has
diphthongal realizations of /e:,ø:,o:,Ei,œy,Au/ (thus realizing
these vowels as [ei,øy,ou,Ei,œy,Au]) and a glided coda /r/
(realized [ô]), whereas the Belgian variety (henceforth “Flemish
Dutch”) has monophthongal realizations of /e:,ø:,o:/ (yielding
realizations [e:,ø:,o:]), markedly less diphthongization in
/Ei,œy,Au/, and does not glide the coda rhotic (thus realizing
it [r]). These differences have all arisen via sound changes that
have taken place in Standard Dutch but not in Flemish Dutch
(Van de Velde, 1996; Sebregts, 2015). This makes the present-day
variation between Standard Dutch and Flemish Dutch a useful
proxy for historical sound change.
The study reported here investigates the perception of these
speech sounds in speakers of Flemish Dutch who have migrated
to the Netherlands. Ten Flemish-Dutch speakers (henceforth:
FDS), all first-year university students who migrated to the
Netherlands, are compared to 10 Standard-Dutch speakers
(henceforth: SDS). Participants are tested multiple times to
test for possible longitudinal adaptation on the part of the
Flemish students. Using an exploratory extension of the violation
paradigm (Friederici et al., 1993) to phonological processing,
the objective of the investigation is to find behavioral and
electrophysiological correlates of the processing of the type
of variation under discussion. While it will turn out that
this endeavor will be unsuccessful, two robust differences
between the two groups provide new information about the
types of phonological variation whose processing neurolinguistic
methods can detect. It will be shown that the FDS are less
able to “take in” SDS speech, which is reflected by a smaller
N400 decrease over two repetitions of the same experiment
compared to the SDS (cf. behavioral findings by Floccia et al.,
2009). In addition, it will be shown that there are ERP-detectible
differences in the processing of SDS speech between the two
groups. Specifically, in words where the vowel /E/ is realized
as [E:]—an ungrammatical realization that cannot be achieved
by any known phonological rule of Standard Dutch—a P600 is
obtained in the SDS, but not in the FDS. This is in line with
previous research (Pater et al., submitted; Domahs et al., 2009)
about the role of the P600 in the processing of phonological rules.
The remainder of the paper is structured as follows. Section 2
discusses the psycholinguistic and neurolinguistic correlates of
accent (violation) processing that have been identified in the
prior literature. Due attention is paid to the well-known N400
component, and to the P600 which is a relatively new, but
not unknown, component in this field. Section 3 details the
methodology used in the present experiment. Section 4 provides
the results, which are discussed in section 5, first separately for the
two findings (N400 and P600) and then together. Finally, section
6 concludes the paper.
2. ACCENT PROCESSING
At its core, the present study is about accent accommodation,
a subfield intersecting psycholinguistics and phonetics. Previous
research in this field has shown that listeners are very adept at
compensating for linguistic variation, particularly in the vowel
system. Maye et al. (2008), for instance, show that listeners
are able to accommodate to a completely novel vowel shift in
English (all vowels lowered by one degree, so “wicked witch”
becomes “weckud wetch”) after only a few minutes of exposure.
At the same time, however, Floccia et al. (2006) found that a
notable regional accent incurs a slowdown in lexical decision
tasks of about 30 ms. This effect accumulated over time, i.e., with
longer words this delay increased more than proportionally. This
Frontiers in Neuroscience | www.frontiersin.org 2June 2019 | Volume 13 | Article 546
Voeten and Levelt ERP Responses to Regional Accent
suggests an interference effect starting from the very beginning
of lexical processing, which persists even as the listener receives
more exposure to the accented speech (Floccia et al., 2009).
These results suggest that while participants in behavioral
tasks are able to accommodate successfully in order to fulfill the
task, their processing is still somehow impaired when confronted
with accentual variation. This processing difficulty has been
measured directly in ERP investigations. Goslin et al. (2012)
found that accented realizations reduced the amplitude of both
the phonological-mapping negativity (otherwise known as the
N280) and the N400. The relationship between accent and the
phonological-mapping negativity is obvious, but the involvement
of the N400 might be considered surprising, given that this
component is normally connected to semantic processing, or
more generally to lexical predictions (Dambacher et al., 2006).
The results to be presented in section 4 will give reason
to postulate two different ways in which the N400 may be
modulated by accentual variation. On the one hand, the findings
by Goslin et al. (2012) suggest that a persistent regional accent
reduces the strength of the lexical predictions made by a listener,
resulting in a reduced overall N400 amplitude due to simple
parsing difficulty, which causes the listener to predict more
cautiously. On the other hand, the results from the present
paper will give reason to postulate an N400-increasing effect for
regionally-accented words, due to their decreased consolidation
in the lexicon over time.
Recent studies have intimated another ERP component in
accented-speech processing: the P600. Originally known from
syntax (Osterhout and Holcomb, 1992), it was observed by Liu
et al. (2011) that a P600 could also be elicited by phonology.
Specifically, Liu et al. (2011) observe a P600 in Chinese
participants who read well-known poems in which some words
were replaced with synonyms that only differed in orthographical
and phonological form. They argue that the P600 that was
elicited by these “deviant” words must be due to phonological
processing, since other sources of integration difficulties such
as semantic violations were absent. Kung et al. (2014) found
a similar effect in a lexical-decision task in Chinese. In this
task, Chinese words with a low lexical tone were embedded as
the final word in a sentence that carried the intonation pattern
of a question. Because questions in Chinese end with rising
intonation, the pitch contour of such words is very similar to
the pitch contour of words with a high lexical tone in a regular
statement sentence. The resulting processing difficulty (“is this
a word with a low lexical tone that rises because the sentence
is a question, or is the sentence a statement and does the word
simply carry a high lexical tone?”) manifested as a P600, which
Kung et al. (2014) interpret as being caused by reanalysis when
the listener resolves this conflict by choosing (in these cases) for
the question interpretation.
Phonological P600 effects have also been found beyond
prosody, viz. in the domain of segmental phonology. Domahs
et al. (2009) obtained a P600 in a lexical-decision task, and
Pater et al. (submitted) found a P600 in a phonological artificial-
language-learning task. Both of these studies investigated a
specific subset of accented speech, viz. violations of allophonic
rules: Pater et al. (submitted) violated an artificially-learned
voicing-agreement rule, and the Domahs et al. (2009) study
found a P600 when two stop consonants followed each other
in a way that violated the phonotactics of German (the native
language of their participants). It is important to mention that
both violations were neutralizing: the violating consonant was
already present on its own in the phoneme inventory. Behavioral
findings by Witteman et al. (2015) suggest that this matters: they
show that adaptation to an accent in general is possible, but not to
individual sounds that cross a phoneme boundary. Furthermore,
there is more interference when a single sound in a word is
replaced by a realization that does not occur normally (e.g., Dutch
/œy/ replaced by German [OI], which does not exist in Dutch;
Witteman et al., 2014).
It thus appears that, even though adaptation is possible, there
are multiple behavioral and electrophysiological correlates of
problems faced by listeners when processing accented speech. In
reaction-time experiments, they are slower. In ERP studies, the
N280, N400, and P600 play a role. Two of the four phonological
P600 studies discussed found the effect in the context of
allophonic-rule violations. The present study integrates these
results and attempts to extend them by investigating reaction
times and ERP responses to Standard Dutch speech by Flemish
Dutch students. The approach is similar to that taken by
Witteman et al. (2015) and used in the P600 studies by Domahs
et al. (2009) and Pater et al. (submitted): only a single sound is
manipulated in an otherwise normal Standard Dutch word.
Given the above, the aim of the present study is to investigate
two things. Behaviorally, it can be expected that the well-
known effect of identity priming (participants being faster to
read a word aloud if they have just been presented the same
word auditorily) will be less strong for realizations that do not
conform to participants’ phonological grammars. Specifically,
the expectation is that unmanipulated identity primes facilitate
word reading, but manipulated identity primes incur the
same RT slowdown reported by Floccia et al. (2006, 2009)
on top of this identity-priming effect. Electrophysiologically,
the manipulations are expected to specifically elicit a P600
ERP when they result in ungrammatical allophones, and
possible across-the-board N280 and N400 effects may arise
in general.
The hypotheses presented above are evaluated for Standard-
Dutch speakers (henceforth: SDS) and Flemish-Dutch speakers
(henceforth: FDS), who have only just moved to the Netherlands
to start their university studies there (parallelling Evans and
Iverson’s 2007 study). The expectation is that there will be
differences, but that they will reduce over months of time as the
FDS participant receive more exposure; the present experiment
takes 9 months divided into three sessions. Differences are
expected in terms of RTs and in terms of ERPs. Concerning RTs,
the expectation is that the hypothesized difference in identity-
priming effects will be smaller for the FDS than for the SDS,
as the FDS will have remaining difficulty parsing also the non-
manipulated segments of the words. In terms of ERPs, it is
expected that the FDS have different N400 responses (regardless
of task or type of violation), in line with findings by Goslin et al.
(2012). In addition, the P600 effect in response to allophonic
violations (Pater et al., submitted; Domahs et al., 2009) is
Frontiers in Neuroscience | www.frontiersin.org 3June 2019 | Volume 13 | Article 546
Voeten and Levelt ERP Responses to Regional Accent
expected to be smaller for the FDS than for the SDS, as the FDS
should have less robust prior expectations due to having had less
exposure to Standard Dutch speech.
The task used in the present study, explained in section 3, has
not been used in a violation paradigm before, and the research is
therefore of an exploratory nature. The results will show that the
task is sensitive to phonological violations in individual speech
sounds, but that this effect must have a deeper source than surface
allophones: it will be shown that the task cannot detect contextual
violations, but is instead sensitive only to realizations that lie
outside the set of possible realizations of a phoneme, further
precising the behavioral findings by Witteman et al. (2014, 2015).
3. MATERIALS AND METHODS
3.1. Participants
Participants consisted of 10 FDS participants (seven female, three
left-handed; mean age =22.71 years, SD =3.54 years) and
10 SDS controls (seven female, one left-handed; mean age =
20.53 years, SD =2.48 years). The FDS participants were all in
their first year of study at a Dutch university in the Randstad
(either Leiden University or the University of Amsterdam)
and were speakers of a variety of Flemish Dutch. The control
participants were Standard-Dutch students, not necessarily in
their first year, who were also studying in the Randstad and
had grown up in a Randstad-Dutch environment. The FDS
participants were tested as soon as possible after the beginning
of the academic year (mean number of days past September 1st
=21.5 days; SD =7.93 days).This restriction was not applied to
the control group (mean number of days past September 1st =
104.30 days; SD =54.40 days).
To find out about possible longitudinal adaptation processes,
participants were tested over the course of three sessions. The
mean interval between the sessions was 129.29 days (SD =23.19)
for session 1–2, and 112.75 days (SD =22.94) for session 2–
3. The experimental procedure and tasks, which are described
below, were the same for all three sessions. In the end, 23 FDS
datasets were collected and 28 SDS datasets; the discrepancy with
the expected 3×10 =30 datasets per group is due to drop-outs
(session 1–2: two left-handed FDS and one right-handed SDS;
session 2–3: one additional right-handed FDS) and equipment
failure (one left-handed FDS in session 1 and one right-handed
FDS in session 2). Table 1 lists the final set of participants present
in the sample.
3.2. Stimuli
A phonetically-trained female speaker from the Randstad area
of the Netherlands produced 309 prime words embedded in a
carrier sentence. The experiment was an exploratory part of a
larger battery of both neurolinguistic and non-neurolinguistic
tests; for purposes of the present paper, 160 of these words are
relevant and the remainder are fillers. The 160 experimental
words comprised 8 groups of 20 words containing one of
the phonemes /e:,ø:,o:,Ei,œy,Au,a:ö,E/ (the last vowel does not
differ between Standard Dutch and Flemish Dutch and was
TABLE 1 | Overview of the final population from which data were obtained.
Session
Participant 1 2 3
FDS-0 ✓ ✓
FDS-1 ✓ ✓
FDS-2 ✓
FDS-3 ✓ ✓ ✓
FDS-4 ✓ ✓ ✓
FDS-5 ✓ ✓ ✓
FDS-6 ✓ ✓ ✓
FDS-7 ✓
FDS-8 ✓ ✓
FDS-9 ✓ ✓ ✓
SDS-0 ✓ ✓ ✓
SDS-1 ✓ ✓ ✓
SDS-2 ✓ ✓ ✓
SDS-3 ✓ ✓ ✓
SDS-4 ✓ ✓ ✓
SDS-5 ✓ ✓ ✓
SDS-6 ✓ ✓ ✓
SDS-7 ✓
SDS-8 ✓ ✓ ✓
SDS-9 ✓ ✓ ✓
included as a control) in stressed1position. An equal number of
fillers was used, containing the same phonemes separated into
the same conditions, but positioned in a different phonotactic
environment, namely preceding coda /l/. In this environment,
the Standard-Dutch realization of these vowels is the same as the
Flemish-Dutch realization. An obvious exception was made for
consonantal control /a:ö/, which cannot be followed by coda /l/;
this condition was simply included as a target twice, in order to
retain the balance of the conditions presented to the participants.
The same held for the /Aul/ condition, as a lexical gap in Dutch
prevents the vowel /Au/ from being followed by coda /l/. For
reasons of convenience, 3×3 words beginning with one of the
point vowel phonemes /i,u,a:/ were also included as fillers, both
before /l/ and before non-/l/.
The 309 prime words thus present in the design were selected
on the basis of frequency: for each cell, the 20 words selected
are the 20 most-frequent words according to CELEX (Baayen
et al., 1995) starting with the relevant phoneme(s) (mean log
frequency =6.41; SD =2.07). The critical phonemes were
always located at the beginning of words to maximize any
possible priming effects on participants’ reaction times, and
to enable time-locking of ERPs to the onset of the critical
manipulations. There were two exceptions: the requirement of
word-initiality was dropped for the vowel /ø:/, as no words
beginning with stressed /ø:/ were available in the corpus,
presumably as a result of this vowel’s general low frequency
1Unstressed vowels in Dutch may optionally undergo reduction (Booij, 1995),
which could result in the elision of the crucial upgliding diphthongization.
Frontiers in Neuroscience | www.frontiersin.org 4June 2019 | Volume 13 | Article 546
Voeten and Levelt ERP Responses to Regional Accent
in Dutch. The initiality requirement was also rescinded for
the filler items.
All primes, both target and filler with the exception of the
nine point-vowel tokens, were recorded in two different variants.
One of these variants was typical for SDS phonology and one
of these variants was atypical of SDS phonology (and, in all
non-filler non-control items, typical of FDS phonology). These
two variants will henceforth be referred to as “SDS realizations”
and “non-SDS realizations.” Table 2 summarizes the design. For
the SDS realizations, the vowels /e:,ø:,o:,Ei,œy,Au/ were realized
as [ei,øy,ou,Ei,œy,Au] in the experimental items. This is typical
of SDS phonology, but atypical for FDS phonology. For the
filler items where these vowels were followed by a coda /l/,
they were realized as [e:,ø:,o:,E:,ø:,A:], which is typical for both
SDS and FDS phonology. For the /E/ control, the vowel was
realized as [E] (typical for both SDS and FDS), and for the
/a:ö/ control, the sequence was realized [a:ô] (typical of SDS
only). For the non-SDS realizations of the experimental items,
the vowels /e:,ø:,o:,Ei,œy,Au/ were realized as [e:,ø:,o:,E:,ø:,A:],
which is typical of FDS but not permissible in SDS given the
lack of a following coda /l/ in the experimental items. For the
filler items where these vowels were followed by a coda /l/, they
were realized as [ei,øy,ou,Ei,œy,Au], which is not grammatical
in FDS (where these realizations simply do not occur) or
SDS (because these realizations are not permitted before coda
/l/). The non-SDS /E/ control was realized as [E:], which is a
phonologically illicit realization in either SDS or FDS speech,
and the /a:ö/ control was realized as [a:ö], which does not
apply the SDS-typical but FDS-atypical rule gliding /ö/ to [ô]
in coda position.
Figures 1 and 2show spectrograms of the word /e:n/
realized as [ein] and [e:n] that demonstrate the difference
under discussion. A crucial property of the experiment is
that only the target phoneme (or phoneme sequence) was
realized in a specific way; the remainder of the word was
produced naturally. This prevents confounding the specific
effect of allophone pronunciation with a global effect of
regional accent, which is precisely the distinction that this study
aims to tease apart. A reviewer additionally asks if there are
orthographical differences between Standard Dutch and Flemish
Dutch that could influence FDS’ performance on the task; this is
not the case.
Each prime was presented auditorily followed by a target
presented visually. The targets were selected from the same set
of 309 words as the primes. The pairing of targets to primes is
as follows. In three conditions (19.5% of the experiment), viz.
/e:,Ei,a:ö/, the prime word and target word were the same; in
the other conditions (80.5% of the experiment), the word on
the screen was a random selection (without replacement) of the
non-/e:,Ei,a:ö/ words in the experiment.
FIGURE 1 | Example waveform, spectrogram, and F1 trajectory (the critical
difference between diphthongal and monophthongal realizations) for the SDS
realization of /e:n/ as [ein]. Toward the end of the vowel, the F1 falls.
TABLE 2 | Overview of the allophone variants used in the experimental items.
Realization (SDS vs. Non-SDS) used in prime items
Before non-/l/ (Target) Before /l/ (Filler)
Phoneme SDS Typical for Non-SDS Typical for SDS Typical for Non-SDS Typical for
e: ei SDS e: FDS e: FDS ei Neither
ø: øy SDS ø: FDS ø: FDS øy Neither
o: ou SDS o: FDS o: FDS ou Neither
Ei Ei SDS E: FDS E: FDS Ei Neither
œy œy SDS œ: FDS œ: FDS œy Neither
Au Au SDS A: FDS A: FDS Au Neither
a:ö a:ô SDS a:ö FDS
E E SDS+FDS E: neither ESDS+FDS E: Neither
i i SDS+FDS iSDS+FDS iSDS+FDS iSDS+FDS
u u SDS+FDS uSDS+FDS uSDS+FDS uSDS+FDS
a: a: SDS+FDS a: SDS+FDS a: SDS+FDS a: SDS+FDS
Frontiers in Neuroscience | www.frontiersin.org 5June 2019 | Volume 13 | Article 546
Voeten and Levelt ERP Responses to Regional Accent
FIGURE 2 | Example waveform, spectrogram, and F1 trajectory (the critical
difference between diphthongal and monophthongal realizations) for the
non-SDS realization of /e:n/ as [e:n]. The F1 stays stable throughout the vowel.
3.3. Procedure and Data Acquisition
Participants were seated in front of a computer screen,
loudspeakers, and a microphone, in a sound-attenuated and
electrically-shielded booth. The experiment consisted of 618
trials with three breaks in between, spaced evenly throughout
the experiment. All trials were presented on the computer
using PsychoPy version 1.83.04. Before the experiment started,
participants were presented instructions on the computer screen,
which were also read aloud by a male Standard-Dutch speaker
with a neutral (i.e., Randstad) accent.
Each trial started with a black screen, followed by auditory
presentation of the prime. When this prime had finished
playing, the target word appeared on the screen (presented
orthographically), which participants had been instructed to read
aloud. Between two trials, a fixation cross was presented for 1 s.
The data collected from the task are ERP responses to the prime
words and vocal reaction times (i.e., speaking latencies) to the
target words. A diagram showing an example trial and the data
recorded from it is shown in Figure 3.
The manipulation took place in the primes: for each of the
309 words recorded, participants heard both the Standard-Dutch
variant and, in a different trial, the non-Standard-Dutch variant.
Which of these two variants was presented first for each word was
randomized and counterbalanced.
During the whole task, continuous-time EEG activity was
recorded using 32 Ag/AgCl electrodes with a sampling rate of
512 Hz. Two flat electrodes at the mastoids provided a reference
signal, which was subtracted from the EEG signal in off-line
processing; an additional four flat electrodes recorded horizontal
and vertical extra-oculograms. In addition to the EEG activity,
the speech of the participant was recorded while they realized the
target word. Recording was started immediately after the prime
word was presented, right when the target word became visible
on the screen.
4. RESULTS
4.1. Reaction Times
The words realized by the participants were aligned to their
phonetic transcriptions (which were obtained from CELEX)
using the Viterbi forced aligner present in HTK (Young et al.,
2002). Forced alignment of speech sounds to phonemes is a
more principled measure of speech onset time than thresholding
raw acoustic energy, as the procedure uses speech-specific
information in the signal and can hence do a better job at
separating speech from background noise. For every word, the
program produced a list of start and end points of the individual
consonants and vowels present in the speech stream. Vocal RTs
were obtained by extracting the time index of the first phoneme
following the word-initial silence. RTs were obtained with a
granularity of 10 ms.
The effects of the various factors in the design on these
reaction times were analyzed by means of a generalized linear
mixed-effects model with identity link and gamma errors2,
following Lo and Andrews (2015). The fitting engine used for
the model was function glmer from R(R Core Team, 2019)
package lme4 (Bates et al., 2015). Fixed effects were added for
“Group” (treatment-coded: 0 =SDS, 1 =FDS), “Allophone”
(treatment-coded: 0 =non-Standard-Dutch; 1 =Standard-
Dutch), “Session” (coded for linear and quadratic trends using
orthogonal polynomials), “Identity” (treatment-coded: 1 =the
prime and target word were the same, which was the case
in the /a:ö,e:,Ei/ conditions; 0 =the prime and target words
differed), “Condition” (sum-coded), and all possible interactions.
Using Rpackage buildmer (Voeten, 2019), random slopes by
participants and words were included over all terms as long as
the model would still converge; these terms were entered in the
order of their contribution to the log-likelihood, such that when
the model eventually failed to converge, the most information-
rich random slopes had been included. From this maximal model,
terms were excluded in backward stepwise order based on the
change in BIC. (Given the large number of interaction parameters
present in the maximal model, BIC is a more natural elimination
criterion than the likelihood-ratio test.) The raw data for these
models are available in the Supplementary Materials as file
RTdata.csv. The code used to fit the models is available in the
Supplementary Materials as file RTcode.R.
Results of the analysis are shown in Table 3. Because the
model used an identity link, the resulting model coefficients
are directly interpretable as milliseconds of response latency.
The intercept is placed at 827 ms ( ˆ
β=827.16, SE =2.22,
t=372.06, p<0.001). This reflects the temporal onset of
the first phoneme in the participant’s response, for the Dutch
control participants when they were presented with non-identity,
Standard Dutch targets. Participants became slightly slower over
the three sessions ( ˆ
β=54.22, SE =1.86, t=29.13, p<0.001),
although the speed loss between sessions 2 and 3 was not as large
as the speed loss between sessions 1 and 2 ( ˆ
β=–10.59, SE =
2.90, t=–3.65, p<0.001). Overall, the FDS were slower than
2Inverse-Gaussian (Wald) errors were also considered, but provided a worse fit to
the data (higher AIC) compared to the gamma-errors model.
Frontiers in Neuroscience | www.frontiersin.org 6June 2019 | Volume 13 | Article 546
Voeten and Levelt ERP Responses to Regional Accent
FIGURE 3 | Example trial for the production task.
TABLE 3 | Fixed-effect coefficients of the reaction-times analysis.
Factor Estimate (SE) t p Sig.
(Intercept) 827.16 (2.22) 372.06 <0.001 ***
Session (Linear) 54.22 (1.86) 29.13 <0.001 ***
Session (Quadratic) –10.59 (2.90) –3.65 <0.001 ***
Group =FDS 41.32 (1.93) 21.37 <0.001 ***
Prime =Identity –46.64 (2.03) –23.02 <0.001 ***
Allophone =Incorrect Dutch –0.15 (1.71) –0.09 0.93
Group =FDS ×Prime =Identity 11.61 (2.28) 5.10 <0.001 ***
Prime =Identity ×Allophone =Incorrect Dutch –6.00 (1.67) –3.59 <0.001 ***
The model additionally includes random intercepts by participants and by words, and random slopes for the factor “Session” by participants and by words. ***p<0.001.
the SDS ( ˆ
β=41.32, SE =1.93, t=21.37, p<0.001). Identity
primes incurred faster RTs than non-identity primes ( ˆ
β=–46.64,
SE =2.03, t=–23.02, p<0.001), although this advantage was
smaller for the FDS ( ˆ
β=11.61, SE =2.28, t=5.10, p<0.001).
Overall, the non-Standard-Dutch allophones incurred slightly
faster RTs than the Standard-Dutch allophones, although the size
of this effect was smaller than the 10 ms granularity with which
HTK had provided the reaction times ( ˆ
β=–6.00, SE =1.67,
t=–3.59, p<0.001).
4.2. ERP Results
The ERP data were detrended and an off-line bandpass filter
was applied passing a frequency domain of 1–30 Hz. Epochs
were time-locked to the onset of the prime words and were
extracted 0–800 ms post-stimulus-onset, after subtracting a
–100 ms baseline3. Epochs contaminated by eyeblinks or other
movement-related artifacts were rejected. The resulting grand-
average waveforms (averaged over all participants, items, and
electrodes), are shown in Figure 4.
In order to determine the precise temporal window and ROI
to be used in the statistical analyses and data plots, permutation
testing (Maris and Oostenveld, 2007) was applied to identify
the locations of significant differences between the conditions.
Repeated ANOVAs were run on each htimepoint,electrodeipair
3The rather short span of this baseline was necessary due to divergence of the
baselines earlier than –100 ms.
in the data using Rpackage permutes (Voeten, 2018). The
design of the test was a 2×2 ANOVA with fixed factors for
“Allophone” (encoding the type of allophone presented to the
participant), “Group” (FDS or SDS), and the interaction between
the two. Because it was conceivable that the experimental
items /e:,ø:,o:,Ei,œy,Au/ and the control items /E,a:ö/ might
be differentially sensitive to the manipulation of the prime
allophones, the permutation tests were run twice: once on the
full dataset (to identify global differences between the groups
of participants and allophones) and once for each of the eight
conditions separately (to identify possible differences between
the experimental items, the control vowel /E/, and the control
consonant condition /a:ö/). The analysis of the whole dataset,
plotted in Figure 5, identified a global effect of “Group,” ranging
from 390 to 470 ms at frontal, central, and parietal sites.
Figure 6 shows the grand-average waveforms corresponding to
this global difference between the groups. The analyses of the
individual vowels failed to identify meaningful windows in the
allophonic conditions [ei∼e:, øy∼ø:, ou∼o:, Ei∼E:, œy∼œ:,
Au∼A:, a:ô∼a:ö], but for the [E∼E:] condition, an effect of
“Allophone×Group” was observed at essentially all electrode
sites within a temporal window of 560 to 660 ms. This effect is
plotted in Figure 7;Figure 8 shows the corresponding grand-
average waveforms. Reasons why only this condition elicited a
significant ERP are discussed in section 5.
The effects found in Figures 5 and 7appear to correspond,
respectively, to the classic N400 and P600 effects. To analyze
Frontiers in Neuroscience | www.frontiersin.org 7June 2019 | Volume 13 | Article 546
Voeten and Levelt ERP Responses to Regional Accent
the N400 effect, each response was averaged over the 390–470
ms window and over the frontal and central electrodes. The
resulting data were analyzed by means of a mixed ANOVA
with fixed effects for “Group”, “Allophone”, and “Session”,
and a complete random-effects structure for participants
and items using the same procedure as in section 4.1.
Terms were selected for inclusion in the model by means
of backward stepwise elimination, using the significance of
FIGURE 4 | Grand-average waveforms calculated over the full dataset, averaged over all participants, electrodes, and the three sessions.
FIGURE 5 | Permutation tests performed on the whole dataset, showing the effect of the factor “Group.” A significant difference can be observed, which reaches
permutation-based significance between 390 and 470 ms.
TABLE 4 | Fixed-effect coefficients for the N400 effect.
Factor Estimate (SE) t df p Sig.
Intercept 1.43 (0.74) 1.94 17.78 0.07
Group =FDS –0.89 (1.04) –0.85 17.97 0.41
Session (Linear) 0.45 (0.18) 2.55 12,650.91 0.01 *
Session (Quadratic) 0.79 (0.17) 4.55 12,672.79 <0.001 ***
Group =FDS ×Session (Linear) –1.17 (0.29) –3.97 12,250.75 <0.001 ***
Group =FDS ×Session (Quadratic) 0.21 (0.27) 0.78 12,665.72 0.44
*p<0.05, ***p<0.001.
Frontiers in Neuroscience | www.frontiersin.org 8June 2019 | Volume 13 | Article 546
Voeten and Levelt ERP Responses to Regional Accent
the change in log-likelihood as the inclusion criterion. The
resulting model is shown in Table 4; the raw data for this
model are available in the Supplementary Materials as file
N400data.csv. The code used to fit the models is available in
the Supplementary Materials as file N400code.R.
The results show significant effects for “Session (Linear)” ( ˆ
β
=0.45, SE =0.18, t12,650.91 =2.55, p=0.01) and “Session
(Quadratic)” ( ˆ
β=0.79, SE =0.17, t12,672.79 =4.55, p<
0.001). The linear component shows that the N400 became
less pronounced over the three sessions, whereas the quadratic
component implies that the N400 shrinkage between sessions
2 and 3 was larger than the reduction between sessions 1 and
2. The linear component additionally entered into a significant
FIGURE 6 | Grand-average waveforms calculated over the full dataset,
averaged over the three sessions and the two allophone conditions. A
difference in amplitude can be observed between the FDS and the controls,
which reaches permutation-based significance in the 390–470 ms window.
interaction with “Group =FDS” ( ˆ
β= −1.17, SE =0.29, t12,250.75
= −3.97, p<0.001). This suggests that the linear trend of
decreasing N400 amplitudes was significantly less pronounced
for the FDS than it was for the SDS.
To analyze the P600 effect, the data were averaged over
the 560–660 ms window and all electrodes. The data were
analyzed in the same way as for the N400 effect. The model
containing the terms that remained after stepwise elimination is
shown in Table 5; the raw data for this model are available in
the Supplementary Material as file P600data.csv. The code
used to fit the models is available in the Supplementary Material
as file P600code.R.
The results show a significant effect for “Session (Linear)” ( ˆ
β
= −0.76, SE =0.31, t1,179.68 = −2.47, p<0.01). This suggests
that the average amplitude in this window became slightly
smaller in magnitude over the three sessions. A significant main
effect was found for “Allophone =Non-SDS” ( ˆ
β=1.70, SE
=0.44, t1,394.70 =3.86, p<0.001), indicating that the non-
SDS allophone elicited a much larger P600 response than the
SDS allophone. However, this factor interacted significantly with
“Group =FDS” ( ˆ
β= −2.51, SE =0.67, t1,395.54 = −3.76, p<
0.001), such that the P600 was completely negated in the FDS and
in fact only showed up in the SDS group.
4.3. Topographical Distribution
The topographical distribution for the two effects is shown in
Figure 9 for the N400 and Figure 10 for the P600 effect. For the
N400, both the FDS and the SDS showed the lowest activity in
central-parietal areas. The difference between the two was that
the FDS’s activity was lower than the SDS’s at especially the frontal
and frontal-central sites. Since this between-group difference was
persistent throughout the whole experiment, it also shows up in
the topographical plots of the P600 effect. In those plots, the FDS
do not show any interpretable differences between the [E] and the
[E:] allophones, other than the aforementioned frontal activity
FIGURE 7 | Permutation tests performed on the data for the [E∼E:] contrast, showing the effect of the factor “Allophone×Group.” A significant difference can be
observed, which reaches permutation-based significance between 560 and 660 ms.
Frontiers in Neuroscience | www.frontiersin.org 9June 2019 | Volume 13 | Article 546
Voeten and Levelt ERP Responses to Regional Accent
FIGURE 8 | Grand-average waveforms calculated over the [E∼E:] condition only, averaged over all participants, electrodes, and the three sessions. A difference in
amplitude can be observed between the two allophone conditions, in the SDS group only, which reaches permutation-based significance in the 560–660 ms window.
TABLE 5 | Fixed-effect coefficients for the P600 effect.
Factor Estimate (SE) t df p Sig.
Intercept 0.81 (0.51) 1.60 24.60 0.12
Session (Linear) –0.76 (0.31) –2.47 1,179.68 0.01 *
Session (Quadratic) –0.17 (0.29) –0.59 1,400.81 0.56
Allophone =Non-SDS 1.70 (0.44) 3.86 1,394.70 <0.001 ***
Group =FDS 0.49 (0.74) 0.66 27.79 0.51
Allophone =Non-SDS ×Group =FDS –2.51 (0.67) –3.76 1,395.54 <0.001 ***
*p<0.05, ***p < 0.001.
being more negative for the latter allophone. The SDS, however,
show significantly more activity in parietal-occipital areas for the
[E:] allophone compared to the [E] allophone, which corresponds
to the classic ROI of the P600.
5. DISCUSSION
The reaction-time data do not show any meaningful results for
the research question. An expected effect of identity priming,
with a plausible effect size of 47 ms facilitation, was found,
but no significant difference in this facilitatory effect was
found between the SDS and Non-SDS allophones. What was
found, however, was that the FDS were in general slower
responders than the SDS, by approximately 41 ms, an effect
that is of similar magnitude to the identity-prime effect. This
is in line with similar findings on accented-speech perception
by Floccia et al. (2006, 2009).
One of the two main findings of this study is the difference
in N400 amplitudes between the FDS and the SDS over the three
sessions of the experiment. The magnitude of the N400 decreased
in both groups over the three sessions, but did less so for the
FDS than it did for the SDS, with an effect size of −1.17µV.
The aforementioned findings by Dambacher et al. (2006) relating
the N400 to general familiarity can explain this result: the FDS
had less experience with Standard Dutch speech than the SDS
did, and therefore were not as strongly facilitated in sessions 2
and 3 by their previous experiences with session 1. This result
mirrors the behavioral findings by Floccia et al. (2009), who show
that the processing impairment incurred by accented stimuli does
not improve with more exposure. The present study extends this
finding, by showing that it has an electrophysiological correlate
in the N400.
The second main finding of this study was the P600 found
when the phoneme /E/ was realized as [E:], which is an impossible
realization of this phoneme (cf. Witteman et al., 2015). This
phonological P600 is in line with recent papers, particularly those
by Domahs et al. (2009) and Pater et al. (submitted). The effect
was only found in the [E∼E:] condition, which differed from the
other conditions in one way, namely that the [E:] is not just
phonologically illegal, but also does not exist as an allophone
of any phoneme in either SDS or FDS, making this condition
most similar to Witteman et al. (2015). This sheds new light
on the phonological P600 found by Domahs et al. (2009) and
Pater et al. (submitted). They obtained P600s for allophonic
violations, but their critical conditions were phonologically
neutralizing. The artificial rule violated in Pater et al. (submitted)
Frontiers in Neuroscience | www.frontiersin.org 10 June 2019 | Volume 13 | Article 546
Voeten and Levelt ERP Responses to Regional Accent
FIGURE 9 | Topographical distribution of the N400 effect. The left head shows the FDS, the middle head shows the SDS, and the right head shows the difference
between the two.
FIGURE 10 | Topographical distribution of the P600 effect. The top three heads show the FDS, and the bottom three heads show the SDS. From left to right, both
rows display, respectively: the SDS-allophone words; the Non-SDS-allophone words; the difference between the two.
was a voicing-agreement rule between stop consonants; the
Domahs et al. (2009) study investigated phonotactic violations
by, again, stop consonants. In both studies, participants’ native
languages (English and German, respectively) contained a full
stop system, making these specific violations cross phoneme
boundaries. The present study’s finding of the same P600
effect in the [E∼E:] condition, but not the allophonic-violation
conditions, implies that the phonological P600 is restricted to
these neutralizing violations.
What remains to be discussed is the finding that the P600
was only observed in the SDS, and not in the FDS. The
most likely explanation is that this is due to the FDS being
less familiar with Standard Dutch, and therefore being less
disturbed (or not significantly more than their baseline levels)
by the [E:] realizations. This interpretation is supported by
the finding that their N400 amplitudes decreased less steeply
over the course of the three sessions. Note however that,
while it was smaller, the FDS group still showed a decrease in
N400, just as the SDS group did. This suggests that long-term
accommodation is possible, and that the FDS may simply require
more exposure. The present study cannot shed any light on
possible future long-term accommodation by the FDS to the
Standard Dutch accent in general, because the present set-up
cannot distinguish between accommodation to the differences
in accent and accommodation to this specific experiment.
Further research, with diverse stimuli over the multiple sessions,
is necessary.
The differences between the SDS and the FDS have
implications for our knowledge of the neural processing of
on-going historical phonological change. The FDS, who serve
as a proxy for a more conservative stage of Dutch, showed
increased N400 amplitude by the three sessions and did not
show significant P600 modulation in the condition where it
was found for the SDS. These findings suggest that the present
study successfully managed to elude the robust perceptual
compensation mechanisms discussed in the Introduction. It
additionally supports the logical assumption that this elusion is
not permanent: while the FDS’s N400s did not shrink as much
as the SDS’s did, they did shrink nonetheless. It is conceivable
that eventually, the two groups’ N400s would come to coincide,
which may be the point at which an on-going change can be
considered to have been acquired. The finding of the P600 in the
[E∼E:] condition for the SDS only further specifies the conditions
under which this phonological P600 can be elicited. Of historical
phonological change, this result implies that on-going sound
change in the form of new allophonic variation is processed
more subtly by the human perceptual apparatus than a phoneme
merger or split.
Frontiers in Neuroscience | www.frontiersin.org 11 June 2019 | Volume 13 | Article 546
Voeten and Levelt ERP Responses to Regional Accent
The present study is not without its limitations. One
difference between the manipulation in the present study vs. the
manipulation used by Witteman et al. (2015) is that they used
cross-spliced speech, while the present study used natural speech
produced by a trained phonetician. A small but critical difference
between the present study and Witteman et al. (2015) means that
this study is not as critically reliant on splicing as theirs was.
This is the fact that, in contrast to Witteman et al. (2015), the
present study used entirely native Dutch material: even the [E:]
realizations correspond to perfectly fine Dutch phones, found in
normal Dutch speech as the realizations of /Ei/ before coda /l/.
Nonetheless, the gain in naturalness of the speech material due
to the absence of splicing artifacts might have been offset by a
loss in naturalness of the manipulated stimuli realized by the
speaker. While none of the participants commented on certain
stimuli sounding artificial, and the speaker was phonetically
trained and hence used to the task of realizing particular
stimuli, this can be seen as a point of criticism in the design.
In addition, the paradigm used in the present experiment—
listening to single words and reading words aloud as a cover
task—was one chosen out of convenience, this study being an
exploratory part of a larger project for which this set-up was
advantageous. Finally, the sample size—23 vs. 26 EEG recordings,
but only 2×10 participants—in the present experiment was
comparatively small. In sum, the effects found in this pilot
experiment are in need of independent replication. I recommend
that these findings be re-investigated in different languages
using different paradigms, to ascertain whether the effects are
cross-linguistic reflections of the processing of phonological
differences between accentual varieties, or whether they are
specific to these varieties of Dutch, or even to this specific
task.
6. CONCLUSION
This study identified two electrophysiological correlates of
accent processing and the processing of on-going phonological
change in Standard Dutch and Flemish Dutch listeners
to (experimentally-controlled) Standard Dutch speech. The
amplitude of the N400, measuring listeners’ familiarity with the
general accent in which the experimental stimuli were spoken,
was decreased for the Flemish Dutch group compared to the
Standard Dutch group, indicating that they were more cautious
in applying their predictive processing abilities to the unfamiliar
accent (pace Goslin et al., 2012). This effect decreased over
the three measurement sessions, indicating that over the course
of nine months, the Flemish Dutch participants became more
familiar with Standard Dutch speech, or at least the Standard
Dutch speech of these experimental stimuli. In addition, a P600
effect was found for a very specific violation, viz. the realization
of /E/ as the illicit realization [E:], in the Standard Dutch
group of listeners only. This shows that the brain is capable
of detecting this specific type of violations (viz. violations that
cross a phoneme boundary, pace Witteman et al., 2015), but only
after sufficient familiarity with the general accent is achieved
(as implied by the significant difference in sensitivity to this
violation between the Flemish Dutch group and the Standard
Dutch group).
Inherent limitations to this exploratory study,
particularly concerning the number of participants and
the way in which the stimuli were created, mean that
the results of this experiment need to be subjected to
independent replication using different languages and
paradigms before any definitive conclusions should
be drawn. The present pilot experiment, however, has
taken the first steps toward an electrophysiological
investigation of the processing of on-going historical
phonological change.
DATA AVAILABILITY
The datasets analyzed for this study are available on request to the
corresponding author. The datasets used in the statistical analyses
are also included in the Supplementary Materials.
ETHICS STATEMENT
This study was carried out in accordance with the
recommendations of the Ethics Code for linguistic research
in the faculty of Humanities at Leiden University with written
informed consent from all subjects. All subjects gave written
informed consent in accordance with the Declaration of
Helsinki. The protocol was approved by the Leiden University
Center for Linguistics.
AUTHOR CONTRIBUTIONS
CV and CL conceived the experiment, designed the stimuli and
edited the manuscript, CV performed the experiments, analyzed
the data and wrote the manuscript.
FUNDING
This work is part of the research programme Watching
Dutch Change with project number PGW-15-15, which is
(partly) financed by the Netherlands Organisation for Scientific
Research (NWO).
ACKNOWLEDGMENTS
We thank Dr. Vya Chen for valuable discussions on the design of
the experiment and comments on the manuscript.
SUPPLEMENTARY MATERIAL
The Supplementary Material for this article can be found
online at: https://www.frontiersin.org/articles/10.3389/fnins.
2019.00546/full#supplementary-material
Frontiers in Neuroscience | www.frontiersin.org 12 June 2019 | Volume 13 | Article 546
Voeten and Levelt ERP Responses to Regional Accent
REFERENCES
Baayen, R. H., Piepenbrock, R., and Gulikers, L. (1995). The CELEX Lexical
Database (Release 2). Philadelphia, PA: Linguistic Data Consortium; University
of Pennsylvania.
Bates, D., Mächler, M., Bolker, B., and Walker, S. (2015). Fitting linear mixed-
effects models using lme4. J. Stat. Softw. 67, 1–48. doi: 10.18637/jss.v0
67.i01
Bermúdez-Otero, R. (2015). Amphichronic Explanation and the Life Cycle of
Phonological Processes. Oxford, PA: Oxford University Press.
Blevins, J. (2004). Evolutionary Phonology: The Emergence of Sound Patterns.
Cambridge: Cambridge University Press.
Booij, E. E. (1995). The Phonology of Dutch. Oxford: Oxford University Press.
Bosker, H. R., and Reinisch, E. (2015). “Normalization for speechrate in native and
nonnative speech,” in Proceedings of the 18th International Congress of Phonetic
Sciences 2015 (ICPhSXVIII), eds M. Wolters, J. Livingstone, B. B. R. Smith,
M. MacMahon, J. Stuart-Smith, and J. Scobbie (Glasgow; Paris: International
Phonetic Association).
Boyd-Bowman, P. (1954). From Latin to Romance in Sound Charts. Washington,
DC: Georgetown University Press.
Cutler, A. (2012). Native Listening. Cambridge, MA: MIT Press.
Dambacher, M., Kliegl, R., Hofmann, M., and Jacobs, A. M. (2006).
Frequency and predictability effects on event-related potentials
during reading. Brain Res. 1084, 89–103. doi: 10.1016/j.brainres.2006.
02.010
Domahs, U., Kehrein, W., Knaus, J., Wiese, R., and Schlesewsky, M.
(2009). Event-related potentials reflecting the processing of phonological
constraint violations. Lang. Speech 52, 415–435. doi: 10.1177/0023830909
336581
Evans, B. G., and Iverson, P. (2007). Plasticity in vowel perception and production:
a study of accent change in young adults. J. Acoust. Soc. Am. 121, 3814–3826.
doi: 10.1121/1.2722209
Floccia, C., Butler, J., Goslin, J., and Ellis, L. (2009). Regional and foreign accent
processing in English: can listeners adapt? J. Psycholinguist. Res. 38, 379–412.
doi: 10.1007/s10936-008-9097-8
Floccia, C., Goslin, J., Girard, F., and Konopczynski, G. (2006). Does a regional
accent perturb speech processing? J. Exp. Psychol. Hum. Percept. Perform.
32:1276. doi: 10.1037/0096-1523.32.5.1276
Friederici, A. D., Pfeifer, E., and Hahne, A. (1993). Event-related brain
potentials during natural speech processing: effects of semantic,
morphological and syntactic violations. Cogn. Brain Res. 1, 183–192.
doi: 10.1016/0926-6410(93)90026-2
Goslin, J., Duffy, H., and Floccia, C. (2012). An ERP investigation of
regional and foreign accent processing. Brain Lang. 122, 92–102.
doi: 10.1016/j.bandl.2012.04.017
Grondelaers, S., and van Hout, R. (2011). The standard language situation
in the low countries: top-down and bottom-up variations on a diaglossic
theme. J. German. Linguist. 23, 199–243. doi: 10.1017/S14705427110
00110
Hyman, L. M. (1976). “Phonologization,” in Linguistic Studies Offered to Joseph
Greenberg on the Occasion of his Sixtieth Birthday, Vol. 2, ed A. Juilland,
Phonology (Saratoga, NY: Anima Libri), 407–418.
Hyman, L. M. (2013). “Enlarging the scope of phonologization,” in Origins of
Sound Change: Approaches to Phonologization, ed A. C. L. Yu (Oxford: Oxford
University Press), 3–28.
Kung, C., Chwilla, D. J., and Schriefers, H. (2014). The interaction
of lexical tone, intonation and semantic context in on-line
spoken word recognition: an ERP study on Cantonese Chinese.
Neuropsychologia 53, 293–309. doi: 10.1016/j.neuropsychologia.2013.
11.020
Liu, B., Jin, Z., Wang, Z., and Xin, S. (2011). An ERP study on whether the P600
can reflect the presence of unexpected phonology. Exp. Brain Res. 212, 399–408.
doi: 10.1007/s00221-011-2739-3
Lo, S., and Andrews, S. (2015). To transform or not to transform: using generalized
linear mixed models to analyse reaction time data. Front. Psychol. 6:1171.
doi: 10.3389/fpsyg.2015.01171
Mann, V. A., and Repp, B. H. (1980). Influence of vocalic context on
perception of the [R]-[s] distinction. Percept. Psychophys. 28, 213–228.
doi: 10.3758/BF03204377
Maris, E., and Oostenveld, R. (2007). Nonparametric statistical testing
of EEG- and MEG-data. J. Neurosci. Methods 164, 177–190.
doi: 10.1016/j.jneumeth.2007.03.024
Maye, J., Aslin, R. N., and Tanenhaus, M. K. (2008). The weckud wetch
of the wast: lexical adaptation to a novel accent. Cogn. Sci. 32, 543–562.
doi: 10.1080/03640210802035357
Norris, D., McQueen, J. M., and Cutler, A. (2003). Perceptual learning in speech.
Cogn. Psychol. 47, 204–238. doi: 10.1016/S0010-0285(03)00006-9
Ohala, J. J. (1981). “The listener as a source of sound change,” in Papers From the
Parasession on Language and Behavior (Chicago Linguistics Society), 178–203.
Ohala, J. J. (2012). “The listener as a source of sound change: an update,” in The
Initiation of Sound Change. Perception, Production and Social Factors, eds M.
-J. Solé and D. Recasens (Amsterdam: John Benjamins), 21–35.
Osterhout, L., and Holcomb, P. J. (1992). Event-related brain
potentials elicited by syntactic anomaly. J. Mem. Lang. 31, 785–806.
doi: 10.1016/0749-596X(92)90039-Z
Pater, J., Sanders, L., Moore-Cantwell, C., Staubs, R., and Zobel, B. H.
(submitted). Event-Related Potential Evidence of Abstract Phonological Learning
in the Laboratory. Available online at: http://www.clairemoorecantwell.org/
phonolearn/moore-cantwell- et-al-2018- ERP-ALL.pdf
R Core Team (2019). R: A Language and Environment for Statistical Computing.
Vienna: R Foundation for Statistical Computing.
Sebregts, K. (2015). The Sociophonetics and Phonology of Dutch /r/. Utrecht: LOT.
Van de Velde, H. (1996). Variatie en Verandering in het Gesproken Standaard-
Nederlands. Nijmegen: Katholieke Universiteit Nijmegen.
Voeten, C. C. (2018). permutes: Permutation Tests for Time Series Data. R
package version 0.1. Available online at: https://CRAN.R-project.org/package=
permutes
Voeten, C. C. (2019). buildmer: Stepwise Elimination and Term Reordering for
Mixed-Effects Regression. R package version 1.1. Available online at: https://
CRAN.R-project.org/package=buildmer
Witteman, M. J., Bardhan, N. P., Weber, A., and McQueen, J. M. (2015).
Automaticity and stability of adaptation to a foreign-accented speaker. Lang.
Speech 58, 168–189. doi: 10.1177/0023830914528102
Witteman, M. J., Weber, A., and McQueen, J. M. (2014). Tolerance for
inconsistency in foreign-accented speech. Psychon. Bull. Rev. 21, 512–519.
doi: 10.3758/s13423-013-0519-8
Young, S., Evermann, G., Gales, M., Hain, T., Kershaw, D., Liu, X., et al. (2002).
The HTK Book. Cambridge: Cambridge University Press.
Conflict of Interest Statement: The authors declare that the research was
conducted in the absence of any commercial or financial relationships that could
be construed as a potential conflict of interest.
Copyright © 2019 Voeten and Levelt. This is an open-access article distributed
under the terms of the Creative Commons Attribution License (CC BY). The
use, distribution or reproduction in other forums is permitted, provided the
original author(s) and the copyright owner(s) are credited and that the original
publication in this journal is cited, in accordance with accepted academic practice.
No use, distribution or reproduction is permitted which does not comply with these
terms.
Frontiers in Neuroscience | www.frontiersin.org 13 June 2019 | Volume 13 | Article 546
Available via license: CC BY
Content may be subject to copyright.
