You are on page 1of 12

Language, Cognition and Neuroscience

ISSN: 2327-3798 (Print) 2327-3801 (Online) Journal homepage: http://www.tandfonline.com/loi/plcp21

Facilitation and interference of phoneme


repetition and phoneme similarity in speech
production

Elisa Monaco, Pauline Pellet Cheneval & Marina Laganaro

To cite this article: Elisa Monaco, Pauline Pellet Cheneval & Marina Laganaro (2017) Facilitation
and interference of phoneme repetition and phoneme similarity in speech production, Language,
Cognition and Neuroscience, 32:5, 650-660, DOI: 10.1080/23273798.2016.1257730

To link to this article: http://dx.doi.org/10.1080/23273798.2016.1257730

View supplementary material

Published online: 21 Nov 2016.

Submit your article to this journal

Article views: 78

View related articles

View Crossmark data

Full Terms & Conditions of access and use can be found at


http://www.tandfonline.com/action/journalInformation?journalCode=plcp21

Download by: [BCU/KUB Fribourg - University of Fribourg] Date: 19 May 2017, At: 03:27
LANGUAGE, COGNITION AND NEUROSCIENCE, 2017
VOL. 32, NO. 5, 650660
http://dx.doi.org/10.1080/23273798.2016.1257730

REGULAR ARTICLE

Facilitation and interference of phoneme repetition and phoneme similarity in


speech production
Elisa Monaco, Pauline Pellet Cheneval and Marina Laganaro
Faculty of Psychology and Educational Science, University of Geneva, Geneva, Switzerland

ABSTRACT ARTICLE HISTORY


Speakers know that it is hard to produce tongue-twister-like utterances involving repeated and Received 11 May 2016
similar phonemes (She sells socks in a small shoe shop). While utterances involving Accepted 25 October 2016
phonologically similar phonemes consistently give rise to speech errors, repeated phonemes
KEYWORDS
have been shown to interfere as well as to facilitate production. Here we try to shed light on this Speech errors; tongue-
paradox by investigating brain processes underlying planning utterances involving similar and twister; language production;
repeated phonemes (tongue-twister-like) and utterances with only repeated phonemes facilitation; interference
elicited with picture naming during electroencephalographic/event-related potentials (EEG/ERP)
recording. Tongue-twister-like utterances increased error rates while repeated onset phonemes
decreased reaction times. Both types of utterances modulated ERPs after 370 ms following
picture onset by shortening a period of stable electrophysiological pattern. Another microstate
was lengthened for tongue-twister-like utterances only, wiping out the timing difference of the
previous microstate. These results indicate priming of repeated phonemes and cost of encoding
similar phonemes, both phenomena co-occurring in tongue-twisters.

Introduction involving both similar word onset phonemes and


repeated phonemes and comparing them to the pro-
Every speaker has experienced the production of phono-
duction of utterances involving repeated onset pho-
logical errors while trying to recite speech sequences
nemes. In addition, we get rid of reading and repetition
involving similar and repeated phonemes such as She
paradigms and elicit the production of tongue-twister-
sells six socks in a small shoe shine shop. Besides the
like sequences with a picture naming task.
funny tongue-twister experience, the observation that
error rate increases in the production of such sequences
questions the limits of our speech planning and pro-
Error elicitation with similar and repeated
duction system. It also provides a unique opportunity to
phonemes
understand the mechanisms underlying speech planning.
Tongue-twister-like sequences have been widely used in Spontaneous and elicited speech errors (slips of the
psycholinguistic research to study the processes involved tongue) have been analysed in (psycho)linguistic
in encoding the form of the utterances to be produced. In research to draw inferences on the representations and
the different paradigms used to elicit speech errors par- processes involved in speech production. Whereas the
ticipants are asked to read aloud or to repeat speech initial systematic investigations of speech errors were
sequences involving both similar phonemes and based on the observation of spontaneous errors
repeated phonemes. However, a paradox exists in the lit- (Fromkin, 1973; Meringer & Mayer, 1896), Baars and
erature regarding the planning and production of Motley (1974) were the first to try to elicit speech
sequences with similar and repeated phonemes: while errors in experimental settings. In their original paradigm
phonologically similar phonemes consistently resulted (SLIP, spoonerism of laboratory induced predisposition)
in increased production difficulty (Wilshire, 1999), the rep- they asked the participants to silently read sequences
etition of identical phonemes in consecutive words has of word pairs. Every few trials a target word pair was pre-
been shown to facilitate production (Damian & Dumay, sented with a signal asking the participants to produce
2009). Here we try to shed light on this paradox by inves- the words overtly. The word pairs preceding the target
tigating errors and event-related potentials (ERPs) during pair had a fixed sequence of phonologically similar
the production of tongue-twister-like utterances word onset phonemes which was inversed in the

CONTACT Marina Laganaro marina.laganaro@unige.ch


Supplemental data for this article can be accessed at 10.1080/23273798.2016.1257730.
2016 Informa UK Limited, trading as Taylor & Francis Group
LANGUAGE, COGNITION AND NEUROSCIENCE 651

target pair (e.g. the target pair bad goof preceded by It should be noted that in these frameworks spon-
give book, go back and gap boot). In such cases taneous or experimentally induced slips of the tongue
subjects are likely to produce an onset exchange error have been considered phonemic (categorical) substi-
(gad boof, Baars, Motley, & MacKay, 1975). Sub- tutions. Slightly different interpretations are provided
sequently other paradigms have been developed to by studies applying acoustic analyses to speech errors.
elicit segmental errors. In the tongue-twister paradigm For instance, Frisch and Wright (2002) carried out acous-
multiple repetitions of tongue-twister-like sequences tic analyses on /s/-/z/ errors produced in an experimental
are elicited (Shattuck-Hufnagel, 1992). Participants are tongue-twister task and reported both, intermediate
often asked to follow a rhythm imposed by a metronome phonetic errors and phonemic errors. Hence, in addition
(Schwartz, Saffran, Bloch, & Dell, 1994) or encouraged to to the error mechanisms described above for categorical
repeat sequences as fast as possible (Sevald & Dell, 1994). substitutions, McMillan and Corley (2010) suggested that
Wilshire (1999) carried out a detailed analysis of the con- a simultaneous activation of target and non-target pho-
ditions inducing the production of errors in the tongue- nemes can lead either to articulatory variations (phonetic
twister tasks, in which participants repeated sequences transformations) or to whole phoneme substitution
of four words visually displayed on screen. Higher error errors. Note however that this interpretation holds for
rates were observed when the four words in the sequences sequences of competing similar phonemes, not for
involved two repeated onset phonemes which were sequences of repeated onset phonemes.
phonologically similar (e.g. dirt bus boot dose). Errors
increased from the second repetition of the word
Facilitation with repeated phonemes
sequences and were also observed when the four words
in the quadruplets were not recited sequentially but in As summarised above, tongue-twister-like sequences
random order (Wilshire, 1999, Exp. 2). The influence of involve both, similar phonemes and repeated phonemes
the presence of repeated phonemes in error elicitation and the two seem to contribute to errors. Whereas
has also been shown by Dell (1984) in a delayed pro- feature similarity between subsequent word onset con-
duction task of two word sequences. Error rates were sonants appears to be a key condition to elicit slips of
higher when the sequences involved repeated phonemes the tongue, producing sequences of words with
than when they only involved different phonemes (e.g. repeated phonemes has been shown to interfere with
mad back vs. mad bake) whether the repeated pho- (Stemberger, 2009; see also Breining, Nozari, & Rapp,
nemes were adjacent to the similar word onset phonemes 2015) but also to facilitate production (Meyer, 1992;
or not (e.g. boom comb vs. boom coat). Roelofs, 1999). For instance, in a different vein and for
Hence, both phoneme repetition and presence of different goals than those pursued by tongue-twister
similar onset phonemes seem to play a role in the pro- studies, Damian and Dumay (2009) showed that partici-
duction of speech errors in tongue-twisters and similar pants were faster to name pictures with adjectival
reciting tasks. Increased error rates while repeating the noun phrases when the adjective and the noun con-
same utterances involving repeated phonemes is inter- tained repeated onset phonemes (e.g. blue bag). The
preted as the consequence of re-activating the same same effect was replicated when phonemes were
phonemes. This difficulty arises either because pho- repeated at different word positions (e.g. green chain).
nemes are recalled from a previous selection/activation Of interest for our purpose here is also the observation
which decayed in the meantime (Dell, 1988), or that phoneme repetition did not affect error rates in
because two units (e.g. syllables) containing the same any direction in the studies reported by Damian and
phoneme are activated via the shared phoneme and Dumay. In addition, facilitation was not observed when
compete for selection (Dell, 1984) or because a unit the participants produced the sequence after a delay,
is temporary inhibited after activation/selection to suggesting that facilitation was not due to articulation,
prevent perseveration (Stemberger, 2009). On the but likely to the repeated activation/selection of the
other hand, interference by phoneme similarity same phonemes. Fact is that the facilitation reported
suggests that phonologically similar phonemes when the two consecutive words involve repeated pho-
compete for activation/selection due to shared fea- nemes seems at odds with the interpretation that
tures. This observation favours phonological encoding repeated phonemes contribute to errors in tongue-twis-
models claiming that features are represented and ters. However, one should take the comparison of results
activated along with segmental information (Dell, from tongue-twister studies with those of facilitation
1986, 1988). If shared features become activated, mis- using repeated phonemes with caution, as both the
selection of similar phonemes is more plausible than paradigms and the linguistic materials diverge. First,
errors on distant phonemes. whereas reading and repetition are used to elicit
652 E. MONACO ET AL.

production in tongue-twister tasks, the phoneme rep- similar phonemes, we should observe two distinct ERP
etition priming paradigms use picture naming tasks to modulations, one being similar to the modulation in
elicit utterance production. Second, the utterances the repeated phoneme task.
used in most tongue-twister studies are composed of
sequences of juxtaposed words, that is, of ungrammati- Method
cal utterances; this is not the case in the repeated pho-
nemes priming tasks, where the participants produce Participants
adjectival noun phrases, that is, grammatical utterances. Twenty-two undergraduate students from the University
Finally, whereas multiple repetitions of the same of Geneva (six men) were recruited for this study. They
sequences are elicited in the tongue-twister paradigms, were all native French speakers, right-handed (as con-
the utterance is not repeated (at least not immediately) firmed by the Edinburgh Handedness Scale, Oldfield,
in picture naming paradigms. 1971), aged between 20 and 34 (mean: 24.5). They did
There are thus core differences between the tongue- not report hearing or language disorders. The partici-
twister paradigms and the repeated onset phoneme pants filled a written informed consent and were paid
priming paradigms which do not allow us to further for their participation.
discuss the discordant results and interpretations on
planning and producing repeated phonemes relative
Material
to similar phonemes. In the present study we tried to dis-
entangle facilitation from interference of repeated Tongue-twister task
phonemes by having the same subjects producing Twenty-four nouns and their corresponding line drawings
tongue-twister-like adjectival noun phrases (involving were selected from two French databases (Alario &
both similar phonemes and repeated phonemes) and Ferrand, 1999; Bonin, Peereman, Malardier, Mot, &
sequences involving repeated onset phonemes and Chalard, 2003). The words were three to eight phoneme
their respective control utterances in a picture naming long (one to three syllables), all with a consonant onset.
task. In addition to the behavioural analysis we take Each noun was associated with two out of six pre-
advantage of ERP electrophysiological analyses and nominal adjectives (petit small-, grand big-, trois
track the time-course of the modulations related to three-, quatre four-, demi half-, vieux old-) to create
repeated and to similar phonemes. If encoding repeated tongue-twister-like adjectival noun phrases with phonolo-
phonemes facilitates production, we should observe gically similar onset phonemes and the respective control
reduced production latencies as in Damian and Dumay sequences. The six pre-nominal adjectives were chosen to
(2009), which should reflect in ERP modulations in the meet the condition of being easily represented in a
time window associated with phonological encoding. picture naming task and of creating a possible tongue-
On the other hand, tongue-twister-like sequences twister-like sequence by manipulating the phoneme simi-
should result in increased error rates and production larity with the nouns. Accordingly, the original line-
latencies. Regarding electrophysiological modulations drawing images (350 350 pixels) were modified with
related to the production of tongue-twister-like GIMP (GNU Image Manipulation Program, http://www.
sequences, a very few ERP studies used the SLIP para- gimp.org) to elicit adjectival noun phrases for each of
digms to elicit production errors with written word the six adjectives (see examples in the Appendix).
sequences. Acheson and Hagoort (2014) were interested The 24 tongue-twister-like adjectival noun phrases
in the error-related negativity rather than in the planning were built such that the noun onset and the adjective
processes and therefore analysed the post-response onset differed on one phonological feature and at least
ERPs. To our knowledge only Mller, Jansma, Rodri- one phoneme was shared between the noun and the
guez-Fornells, and Mnte (2007) analysed the pre- adjective (e.g. la grande cravate /lagRdkRavat/ -the big
response ERPs in a SLIP task. The authors reported diver- tie-). Every two tongue-twister-like sequences, two
ging ERPs between erroneous and error-free trials, which control sequences were created by rotating the two
they interpreted as reflecting conflict during phonetic adjectives. In this way, a noun was used both in the
planning. In our paradigm, if similar onset phonemes experimental condition and in the control condition
and repeated phonemes both contribute to speech with a different adjective from the same set of six pre-
errors, we should observe a similar ERP signature as nominal adjectives and all adjectives were repeated an
Mller et al. for tongue-twister-like and for repeated equal amount of times in the experimental and the
onset phoneme utterances. Finally, if tongue-twister- control utterances. The matched control utterances did
like sequences involve both facilitation due to repeated not involve similar or identical phoneme onsets. An
phonemes and interference due to phonologically example of quadruplet is given in Table 1(a).
LANGUAGE, COGNITION AND NEUROSCIENCE 653

Repeated onset phoneme task presented without the written utterance and participants
Twenty-four additional nouns and their corresponding were asked to name the pictures with the expected
line drawings were selected from the same databases. adjectival noun phrase with an increasingly pressing
The 24 items were matched with those used for timing. This procedure guaranteed that each participant
tongue-twisters on the following psycholinguistic vari- knew and used the expected adjectives and nouns for
ables: visual complexity of the picture, concept famili- each pictures during the experiment. An experimenter
arity, image agreement, name agreement, lexical ensured that each subject performed the task as
frequency, age of acquisition, word length (as number intended before the actual experiment was run.
of phonemes and as number of syllables). Each noun During the experimental phase, each stimulus was
was associated with two out of six pre-nominal adjec- presented on the screen for 1500 ms, preceded by a
tives (same adjectives as for tongue-twister sequences) cross in the middle of the screen lasting 500 ms and a
such that the noun and the adjective shared the same blank screen of 200 ms. Inter-stimuli interval (ISI) lasted
onset consonant. For every two repeated phoneme 1000 ms. Participants were instructed to respond (to
sequences, two control sequences were constructed by name each picture with the adjectival noun phrase) as
rotating the adjectives in order to create adjective fast and accurately as possible. Each block of 96 stimuli
noun sequences with different onsets (control condition, started with 3 filler trials and a short break was taken
see examples in Table 1(b)). between the 2 blocks. E-prime (Psychology Software
Five additional nouns were selected for the examples Tools, Inc., Pittsburgh, PA, USA; http://www.pstnet.com/
and as filler items. eprime) was used for stimulus presentation and timing
of stimuli.

Procedure
Analysis of behavioural data
The participants sat approximately 60 cm away from the
Every produced utterance was digitised and the Check-
computer screen. Pictures were presented in constant
Vocal software (Protopapas, 2007) was used to manually
size of 9.26 9.26 cm (approximately 4.4 of visual angle)
extract production latencies (response times or RTs, i.e.
on a grey screen. The stimuli from the tongue-twister
the time separating the onset of the picture and the
task and the repeated phoneme task were mixed up to
vocal onset) based on the acoustic waveforms. Every pro-
increase variation and presented twice in 2 blocks of 96
duction which did not correspond to the expected
pseudo-randomised stimuli. Four different orders were
correct sequence was considered as error. The following
used and counterbalanced across participants.
error categories were identified: lexical errors (e.g. other
Before the experiment, participants were familiarised
nouns or adjectives than expected), segmental errors
with each original line-drawing and with the modified
(e.g. /lakRdkRavat/ for /lagRdkRavat/, or /letRwabiRb/
pictures eliciting the adjectival noun phrases. At first
for /letRwabibR/) and other errors (e.g. lack of answer,
each original drawing appeared on the screen with the
determiner omission or gender error, interrupted or
expected noun. The stimuli appeared in a random
incomplete answer, laughter, guttural sounds, dysfluen-
order and the participants could proceed at their pace
cies). Production errors were transcribed and coded by
by pressing a key after having named the picture accord-
the first author and cross-checked by the second
ing to the label provided. Then two object pictures (not
author. In case of disagreement between the two
belonging to the experimental set) appeared combined
judges, a consensus was found with a third rater.
with each of the six possible modifications to elicit
the six adjectivenoun combinations. Pictures were pre-
EEG acquisition and pre-processing
sented with the expected adjectival noun phrases and
A high density continuous electroencephalographic (EEG)
the participants were asked to read them aloud. During
(128 channels covering the entire scalp) was acquired at
the second familiarisation phase, pictures were
512 Hz (filters: DC to 104 Hz, 3 dB/octave slope) using
the ActiveTwo Biosemi system (Biosemi V.O.F. Amsterdam,
Table 1. Example of stimuli in the tongue-twister (a) and Netherlands, http://www.biosemi.com). The common
repeated onset phoneme (b) tasks.
mode sense (CMS; active electrode) driven right leg
(a) Tongue-twister task (b) Repeated phonemes
(CMS-DRL) was the online reference. Afterwards, EEG
Experimental Control Experimental Control
epochs with amplitudes exceeding 100 V and contain-
La grande cravate La demie cravate Le vieux vlo Les trois vlos
(the big tie) (the half tie) (the old bike) (the three bikes ing eye blinks or other noise were rejected before further
La demie ttine La grande ttine Les trois tambours Le vieux tambour analyses and averaging, using the CarTool software
(the half dummy) (the big dummy) (the three drums) (the old drum)
(Brunet, Murray, & Michel, 2011). Band-pass filters
654 E. MONACO ET AL.

between 0.2 and 30 Hz and the notch filter at 50 Hz were stable global electric fields, likely corresponding to par-
applied and stimulus-aligned epochs and response- ticular periods in mental information processing (Chan-
aligned epochs corresponding to correct productions or geux & Michel, 2004; Koukkou & Lehmann, 1987;
to production classified as segmental errors were Lehmann, Strik, Henggeler, Koenig, & Koukkou, 1998)
extracted. Stimulus-aligned epochs covered the time and to compare them across tasks. The advantages of
period from picture onset to 500 ms (forward direction); tracking global topographic configurations are related
response-aligned epochs were locked 100 ms before the to their reference-independence and to their direct link
vocal onset of each trial and covered from 300 to to changes in the configuration of the intracranial
100 ms before the onset of the response acoustic sources (Michel, Koenig, Brandeis, & Gianotti, 2009).
signal (backward direction). Only trials for which both Thus, the analyses of global electrophysiological patterns
stimulus-aligned and response-aligned ERPs were artefact provide insights into how tasks differ in terms of likely
free were retained, then averaged separately per partici- underlying neurophysiological mechanisms (Michel &
pant, task (tongue-twister-like and repeated onset pho- Murray, 2012; Murray, Brunet, & Michel, 2008), in addition
nemes) and per condition (experimental and matched to the temporal information about ERP differences.
control utterances). A minimum of 30 epochs were aver- At first, a non-parametric randomisation test to the
aged per participant and condition (mean = 38.2 epochs, global dissimilarity between two electric fields was
SD = 4.1). Bad channels were interpolated with a 3D carried out. This analysis, called TANOVA was carried
spline interpolation method (Perrin, Pernier, Bertnard, out to identify time point by time point periods of sig-
Giard, & Echallier, 1987). For the spatio-temporal analysis nificant topographic differences between conditions
the overlapping signal between stimulus-locked and using the CarTool software (Brunet et al., 2011). The
response-locked ERP according to participants mean pro- global dissimilarity is a quantification of topographic
duction latencies in each condition was removed (see differences between two electric fields independent of
details in the next section). their strength ranging from 0 to 2 (Lehmann & Skran-
dies, 1980, see an example of its computation in
ERP analyses Murray et al., 2008). The permutation of the data is
In order to identify time periods in which amplitude accomplished by re-assigning randomly the topo-
differences were present between the tongue-twister- graphic maps of single subjects to the different con-
like condition and its control condition, and between ditions. The global dissimilarity of these random
the phoneme repetition condition and its control con- group-averaged ERPs is compared time point by time
dition, a standard waveform analysis was performed on point with the values of topographic dissimilarity of
the averaged ERPs of the participants. At each electrode the actual tasks. A time period criterion of 10 ms of con-
and for each time point, the amplitudes of the evoked secutive significant differences was applied.
potentials were compared across conditions with a para- Subsequently, a spatio-temporal segmentation was
metric repeated measures ANOVA using STEN (Statistical performed on the participants grand-averaged ERPs
Toolbox for Electrical Neuroimaging1). This analysis was per task and condition to determine topographic differ-
performed on the stimulus-aligned time periods and ences across conditions and statistically validate them
on the response-aligned time periods independently. in the responses of single participants. In order to deter-
The experimental conditions and their corresponding mine the most dominant electric field configurations at
control sequences for the tongue-twister-like and the the scalp (topographic ERP maps), we used a modified
repeated phoneme utterances were analysed separately hierarchical clustering algorithm (Murray et al., 2008),
and a correction of the alpha criterion to 0.025 was the agglomerative hierarchical clustering and a combi-
applied in addition to a spatial criterion of five neigh- nation of a cross-validation and the KrzanovskiLai cri-
bouring electrodes displaying ERP differences and a tem- terion (see Murray et al., 2008) to select the optimal
poral criterion of five consecutive time frames (10 ms). number of ERP topographic maps that best explain the
If any difference on amplitudes is indeed to be found group-averaged data across conditions. Statistical
across conditions, the waveform amplitude analysis does smoothing was applied to remove temporally isolated
not indicate per se if this difference would be due to topographic maps with low explanatory power. The
simple modulations in amplitudes, to different under- same time criterion used for the ERP waveform analyses
lying brain functional states or to shifts of the ERP com- was adopted (minimum duration of 10 ms). The pattern
ponents across conditions (Michel & Murray, 2012). To of topographic map templates observed in the group-
gain additional information on the underlying electro- averaged data was statistically tested by comparing
physiological differences, a topographic analysis was each of these map templates with the moment-by-
run. The aim of this analysis is to identify periods of moment scalp topography of individual ERPs in each
LANGUAGE, COGNITION AND NEUROSCIENCE 655

Table 2. Mean production latencies (RTs) and proportion of phoneme experimental as compared to the control
errors in the experimental and control conditions for the sequences (t(1453) = 2.07, p = .038).
tongue-twister and the repeated phoneme tasks. In the error analysis there was no significant difference
Tongue-twister Repeated phonemes
across conditions on overall error rates (tongue-twisters: z
Experimental Control Experimental Control
= 1.51, p = .13; repeated phonemes: z = 1.38, p = .17),
RTs (SD) in ms 742 (83) 747 (69) 732 (72) 747 (72)
Overall errors 11.0% 10.7% 7.4% 8.8% but the rate of segmental errors was significantly larger
Segmental errors 2.3% 0.4% 0.9% 1.3% on the tongue-twister stimuli than on the control stimuli
(z = 3.06, p < .01), with no difference across conditions
condition. The aim of this procedure (called fitting) was for the repeated phonemes (z < 1).
to establish how well a topographic template map
explains single participant responses in each condition. ERP results
For this analysis, stimulus and response-aligned ERPs
were combined on the basis of each participants mean Waveform analyses
production latency in each condition, by removing the The comparison between waveforms in the tongue-
overlapping signal from the response-aligned ERPs. twister-like and the matched control utterances revealed
This procedure was designed to obtain ERPs covering significant differences in amplitudes on a large cluster of
the exact time window of encoding, from picture onset central electrodes in four time windows: between 20
to 100 ms before the initiation of articulation (Laganaro, and 40 ms, from 375 to 390 ms and from 410 to
2014) for each participant and condition. Each data 430 ms after stimulus onset, and between 190 and
sampling point in each condition was labelled according 155 ms before articulation onset (see Figure 1(a), top).
to the template map with which it best correlated Significant differences are also found on the global dis-
spatially, yielding a measure of map presence in each similarity (TANOVA) in the same time windows, with
individual data, its duration (number of time frames) the exception of the very first one.
and the global explained variance (GEV). These measures Across repeated phoneme experimental and control
are then used to statistically test topographic differences utterances significant amplitude differences were
across experimental conditions. observed in the time windows from 95 to 115 ms and
between 460 and 490 ms on a cluster of central-posterior
electrodes. A short time window of significant TANOVA
Results was observed only in the 100120 ms time window
Four participants were excluded from the analyses due (see Figure 1(b)).
to high amounts of lexical errors, and/or to lack of suffi-
cient uncontaminated ERP epochs. The following ana- Global topographic ERP pattern analysis (spatio-
lyses were performed on 18 participants. temporal segmentation)
A spatio-temporal segmentation analysis was run on the
four grand-averaged ERPs corresponding to tongue-
Behavioural results
twister and to repeated phoneme sequences and their
For latency and ERP analyses, RTs beyond three standard respective control sequences. The same sequence of
deviations with respect to the average of the subject seven topographic maps was revealed for each con-
across all the utterances (1.6% of the data) and errors dition of the two tasks (see Figure 2), accounting for
(9.5% of the data) were excluded from the analyses. the 98% of the explained variance. Based on the
The average naming latency and mean error rate per results of the waveform analysis and of the spatio-tem-
task and condition are displayed in Table 2. Mixed poral segmentation, a fitting procedure was run from
effects models were computed separately on each task 310 ms until the end of the response-locked ERP
on RTs and on errors (generalised mixed effect model) signals (100 ms before articulation onset) with map
with experimental conditions as fixed effect and by templates E, F and G. The period of quasi-stable
subject and by-item adjustments to the intercept for global electrophysiological pattern at scalp labelled
the random structure (conducted with the statistical soft- map E was present in the ERPs of 16 out of 18 partici-
ware R, R Development Core Team, 2007, version 3.2.4., pants, and map F appeared in all participants ERPs,
and the packages lme4 and lmerTest, Bates, Maechler, whereas the last period (map G) was less consistent
Bolker, & Walker, 2014). No significant differences were across participants. The mean duration of each period
observed on RTs between the tongue-twister and the of topographic stability is displayed in Figure 2. A sig-
control conditions (t(1402) = 1.21, p = .23), but RTs nificant interaction between map and condition was
were significantly shorter in the repeated onset observed for the tongue-twister task on duration and
656 E. MONACO ET AL.

Figure 1. (a) Time points of significant amplitude differences (at p < .025, p < .01, p < .001, on at least five consecutive time frames and
five adjacent electrodes) between the tongue-twister-like utterances and their matched control sequences on each electrode for stimu-
lus-aligned and response-aligned ERPs with display of the electrodes yielding the significant differences and results of the topographic
TANOVA analysis under each graph. An example of waveform for each condition is presented for Cz from 0 to 500 ms following picture
onset and from 300 to 100 ms before articulation. The highlighted time windows correspond to the significant differences on ampli-
tudes. (b) Time points of significant amplitude differences (at p < .025, p < .01, p < .001, on at least five consecutive time frames and five
adjacent electrodes) between the repeated onset phoneme sequences and their control sequences on each electrode as a function of
time on stimulus-locked and response-locked ERPs with display of the electrodes yielding the significant differences and results of the
topographic TANOVA analysis under each graph. An example of waveforms for each condition is presented for Oz. The highlighted time
windows correspond to the significant differences on amplitudes.

GEV (duration: F(1, 17) = 8.71, p < .01; GEV: F(1,17) = 8.82, = 1.69, p = .1), while no significant difference appeared
p < .01). The period of stability corresponding to map E on map F across experimental and control sequences
was less present (had shorter duration and lower GEV) (duration: t(17) = 1.14, p = .27; GEV: t(17) = 1.62, p = .12).
in the experimental than in the control sequences (dur-
ation: t(17) = 2.9, p < .01; GEV: t(17) = 3.07, p < .01).
Discussion
The difference across conditions was in the opposite
direction for duration of map F which was more The experimental manipulation of tongue-twister-like
present in the tongue-twister sequences relative to adjectival noun phrases in a picture naming task actually
their control sequences (duration: t(17) = 2.14, p = .047; elicited a higher rate of segmental errors relative to the
GEV: t(17) = 1.5, p = .15). For the repeated phoneme control utterances, without differences in production
task the interaction between map and condition was latencies. On the other hand, utterances with repeated
significant on GEV only (duration: F(1, 17) = 3.15, p onset phonemes were produced 15 ms faster than
= .09; GEV: F(1, 17) = 8.82, p < .01). The period of topo- their matched control sequences but did not elicit
graphic stability starting around 270 ms after picture additional errors. In both tasks the experimental manipu-
onset (map E in Figure 2) was more present in the lation modulated waveform amplitudes and duration of
control sequences than in the repeated phoneme periods of topographic stability in late time windows,
sequences (duration: t(17) = 2.45, p = .025; GEV: t(17) after 375 ms post picture onset. Crucially, the period of
LANGUAGE, COGNITION AND NEUROSCIENCE 657

Figure 2. Stimulus and response-aligned grand-average ERPs for tongue-twister experimental and control conditions (a) and for
repeated phoneme experimental and control conditions (b) covering the average time period from picture onset to 100 ms before
the vocal response and temporal distribution of the periods of topographic stability presented with different colours and labelled
with the corresponding map templates. Vertical lines are presented every 100 ms. Scalp distribution of the electrophysiological
signal is displayed on a colour scale from dark blue (negative values) to magenta (positive values). On the right: the mean duration
in ms of the periods of topographic stability E, F and G from the fitting procedure in the individual participants.

stable electrophysiological pattern starting around create a general interfering condition in tongue-
270 ms after picture onset (map E in Figure 2) lasted twister-like adjectival noun phrases. Only segmental
shorter for both, tongue-twister and repeated phonemes errors were more frequent on the tongue-twister-like
relative to their control utterances, whereas the following utterances than on their corresponding control utter-
period (map F) was lengthened for tongue-twisters ances. The rate of segmental errors (2.3%) was neverthe-
only. In the following we will first discuss interference less lower than in the standard SLIP or tongue-twister
and facilitation separately before attempting an inte- paradigms. This is likely due to different eliciting tasks
gration of the two phenomena. (picture naming vs. reading or repetition), different utter-
ances (adjectival noun phrases vs. ungrammatical
sequences in the SLIP paradigms) and different pro-
Tongue-twister interference
duction conditions (iterations of the same utterance at
A similar global error rate was observed on tongue- high production speed in the SLIP and in previous
twister-like utterances and on their respective control tongue-twister tasks). Thus, adjectival noun phrases con-
utterances, indicating that the experiment did not taining both similar onsets and at least one repeated
658 E. MONACO ET AL.

phoneme elicited with a picture naming task increase the onset phonemes to be encoded might not slow
segmental errors but to a lesser extent than the standard down the encoding process per se, but it might require
SLIP tasks. additional monitoring time to ensure that the production
Despite increased error rates, the similar response is error-free. The two interpretations are not exclusive,
latencies between tongue-twister-like utterances and and we will not be able to disentangle them with the
control utterances indicate that overall the participants present data. On the other hand, the same interpretation
did not take more time to plan the tongue-twisters. cannot be extended to the results on map E, as a similar
However, if based on the behavioural results one pattern was observed for the phoneme repetition task. A
would claim that the planning time is similar for different interpretation has to be searched for the shorter
tongue-twister-like and control sequences, a different period of topographic stability between 270 and 400 ms
interpretation can be made based on the ERP spatio- which holds for both tasks. These two parallel results will
temporal results, where the experimental manipulation be discussed in the final section.
resulted in both lengthening and shortening of two con-
secutive periods of stable topographic patterns. This
Facilitation by phoneme repetition
period of stable global electrophysiological pattern E
started around 270280 ms in both conditions and The repeated onset phonemes decreased production
lasted longer in the control than in the experimental con- latencies, thus replicating previous behavioural results
dition, causing a shift between maps E and F around by Damian and Dumays (2007, 2009). Planning two con-
400 ms, that is, in the time window of different waveform secutive words starting with the same phoneme is
amplitudes between experimental and control utter- speeded up possibly because phoneme repetition
ances. The subsequent microstate (map F) displayed primes the selection/activation during phonological
the opposite pattern, longer duration for the tongue- encoding (Meyer, 1992; Roelofs, 1999). Repeated pho-
twister-like condition than for the control utterances. nemes did not increase error rates in the current exper-
These two microstates had different duration across con- iment, which seems to contradict previous results on
ditions, but in opposite directions, which seems to equal increased error rates for sequences involving repeated
out the overall processing time across conditions. These phonemes (Dell, 1984, exp. 2). It should be noted that
results thus seem to indicate that the tongue-twister in the present study repeated phonemes were all on
manipulation speeds up the brain processes engaged word onset positions and this is also the word position
between 270 and 400 ms, but it lengthens the follow- in which errors mostly occurred in previous studies.
ing processing. The latter likely reflects increased plan- However, whereas previous studies involved mainly
ning cost due to conflict between similar and/or monosyllabic words, our French stimuli were composed
repeated phonemes and is consistent with Mller of plurisyllabic nouns, thus enabling errors on other
et al.s (2007) results and with usual interpretations of word positions.
the SLIP effect. Mller et al. reported a frontocentral Besides amplitude differences on a small cluster of
negativity from 350 ms after stimulus onset when the channels in the P100 time window, possibly due to
word pairs elicited speech errors. The authors interpreted more complex pictures in one condition, amplitudes dif-
their results as a conflict during phonetic encoding. fered between experimental and control utterances on a
Whereas the Mller et al. results refer to the time- large cluster of centro-posterior-right channels from 460
course in a word pair reading task, the time period of to 500 ms after picture onset. This time window corre-
map F in the present study (after 400 ms) has been sponds to the end of the quasi-stable topographic
associated with word form encoding and internal moni- pattern E which was shorter in the repeated
toring in previous estimates of the time-course in picture phoneme adjectival noun phrases relative to the corre-
naming tasks (Indefrey, 2011; see also Laganaro, Python, sponding control utterances.2 The result on the duration
& Toepel, 2013). There are at least two plausible expla- of periods of stable electrophysiological pattern closely
nations for the observed effect on duration of map F. matches the shorter RTs in the experimental condition.
The first interpretation is related to the timing of phono- As reminded above, the time period corresponding to
logical-phonetic encoding, which might be lengthened map E has been associated with word form encoding
in case of similar phonemes either because of a conflict in picture naming (Indefrey, 2011). Here it suggests
between phonemes sharing common features or that encoding the same word onset phoneme twice
because non-target phonemes have to be inhibited. An speeds up phonological encoding. The time window of
alternative interpretation for a longer lasting period of ERP modulation by repeated phonemes in the present
topographic stability in this late time window can be study is later than the one observed for repeated onset
associated with internal monitoring. The similarity of phonemes in picture naming by Yu, Mo, and Mo
LANGUAGE, COGNITION AND NEUROSCIENCE 659

(2014). The authors reported diverging ERP waveforms that RTs are not increased in the tongue-twister con-
between repeated onset phonemes and the control con- dition. This seems to be due to a compensation of facili-
dition from about 180 to 300 ms. However, the linguistic tation by the cost of conflict due to similar onset
stimuli are different from ours. In the study by Yu et al. phonemes as identified in the longer duration for the
(2014), two nouns corresponding to two different pic- tongue-twister-like condition of the period of electro-
tures were named consequently, while in the present physiological stability labelled map F. Fact is that the
study phoneme repetition is manipulated within the present results provide evidence that interference is
adjectival noun phrase elicited from a single picture. not the only mechanism at play in the production of
Although Yu, Mo and Mo also interpreted their early tongue-twister-like sequences, which is rather a situation
effect as reflecting phonological encoding, it is possible in which the speaker is submitted to both, facilitation
that the underlying mechanisms of onset phonemes and interference. It is possible that both contribute to
priming and the overall dynamic of speech planning is speech errors, priming of identical phonemes increasing
different for two consecutive object names relative to the conflict with similar sounding phonemes.
adjectival noun phrases. More crucially for the purpose
of the present research, similar late waveform modu-
lations after 400 ms on central-posterior electrodes and Notes
similar shortening of the same period of topographic 1. STEN toolbox: http://www.unil.ch/fenl/home/menuguid/
stability were observed in the tongue-twister and in infrastructure/software--analysis-tools.html.
the phoneme repetition condition, although the first 2. Notice that the ERP amplitude modulation by the exper-
imental conditions at the end of the stimulus-aligned
manipulation interfered with the correct production
ERP signal appears later than in the tongue twister like
and the second manipulation facilitated production by task. In both cases the divergences in amplitudes in
reducing naming latencies. this late time window likely reflect the delay in the tran-
sitions from maps E to F across conditions. This delay
seems to happen earlier for the tongue-twister like task
Linking facilitation and interference than for the phoneme repetition task. We statistically
compared the onset of map F in the experimental con-
The crucial issue we wished to address is related to the dition across tasks and it actually happens to be earlier
usual presence of both similar and repeated phonemes for tongue-twister like than for repeated phonemes
in tongue-twister sequences. The similarity in the (t(17) = 2.01, p = .05).
results on the shortening of a shared period of stable
topographic configuration in both tasks allows us to
propose an interpretation of the processes involved Disclosure statement
in the production of tongue-twister sequences. The No potential conflict of interest was reported by the authors.
period corresponding to map E which is shortened in
both, repeated phonemes and tongue-twister-like
sequences, likely reflects phoneme priming across adjec- Funding
tive and noun during phonological encoding, whichever This research was supported by Swiss National Science Foun-
the position of the repeated phonemes. dation [grant number 105319_146113].
In the tongue-twister task the priming is unlikely due
to shared features between phoneme onsets as similar
onset phonemes do not seem to be a valid priming con- References
dition in word production (see Roelofs, 1999). The effect Acheson, D. J., & Hagoort, P. (2014). Twisting tongues to test for
observed on map E being similar between tongue- conflict-monitoring in speech production. Frontiers in Human
twister and repeated onset phoneme sequences, it is Neuroscience, 8, 4671. doi:10.3389/fnhum.2014.00206
likely due to repeated phonemes, although in other pos- Alario, F.-X., & Ferrand, L. (1999). A set of 400 pictures standar-
dized for French: Norms for name agreement, image agree-
itions than word onset in the tongue-twister-like utter-
ment, familiarity, visual complexity, image variability, and
ances. Behavioural facilitation by repeated phonemes age of acquisition. Behavior Research Methods, Instruments,
in other word positions in adjectival noun phrases has & Computers, 31(3), 531552. doi:10.3758/BF03200732
been reported previously (see Exps. 4 and 5 in Damian Baars, B. J., & Motley, M. T. (1974). Spoonerisms: Experimental eli-
& Dumay, 2009). Hence, the ERP results indicate some citation of human speech errors. American Psychological
Association, Journal Supplement Abstract Service.
amount of facilitation on processing speed in the
Baars, B. J., Motley, M. T., & MacKay, D. G. (1975). Output editing
tongue-twister utterances, which cannot be observed for lexical status in artificially elicited slips of the tongue.
in the behavioural results, where only interference in Journal of Verbal Learning and Verbal Behavior, 14(4), 382
terms of errors is observed. However, we also observed 391. doi:10.1016/S0022-5371(75)80017-X
660 E. MONACO ET AL.

Bates, D., Maechler, M., Bolker, B., & Walker, S. (2014). lme4: Lehmann, D., & Skrandies, W. (1980). Reference-free
Linear mixed-effects models using Eigen and S4. Retrieved identification of components of checkerboard-evoked multi-
from http://CRAN.R-project.org/package=lme4 channel potential fields. Electroencephalography and Clinical
Bonin, P., Peereman, R., Malardier, N., Mot, A., & Chalard, M. Neurophysiology, 48(6), 609621. doi:10.1016/0013-4694
(2003). A new set of 299 pictures for psycholinguistic (80)90419-8
studies: French norms for name agreement, image agree- Lehmann, D., Strik, W. K., Henggeler, B., Koenig, T., & Koukkou,
ment, conceptual familiarity, visual complexity, image varia- M. (1998). Brain electric microstates and momentary con-
bility, age of acquisition, and naming latencies. Behavior scious mind states as building blocks of spontaneous think-
Research Methods, Instruments, & Computers, 35(1), ing: I. Visual imagery and abstract thoughts. International
158167. doi:10.3758/BF03195507 Journal of Psychophysiology, 29(1), 111.
Breining, B., Nozari, N., & Rapp, B. (2015). Does segmental McMillan, C. T., & Corley, M. (2010). Cascading influences on the
overlap help or hurt? Evidence from blocked cyclic naming production of speech: Evidence from articulation. Cognition,
in spoken and written production. Psychonomic Bulletin & 117(3), 243260. doi:10.1016/j.cognition.2010.08.019
Review, 23(2), 500506. doi:10.3758/s13423-015-0900-x Meringer, R., & Mayer, K. (1896). Versprechen und Verlesen.
Brunet, D., Murray, M. M., & Michel, C. M. (2011). Spatiotemporal Stuttgart: Goschen.
analysis of multichannel EEG: CARTOOL. Computational Meyer, A. S. (1992). Investigation of phonological encoding
Intelligence and Neuroscience, 2011, 2:12:15. doi:10.1155/ through speech error analyses: Achievements, limitations
2011/813870 and alternatives. Cognition, 42, 181211.
Changeux, J.-P., & Michel, C. M. (2004). Mechanisms of neural Michel, C. M., Koenig, T., Brandeis, D., & Gianotti, L. R. R. (2009).
integration at the brain scale level: The neuronal workspace Electrical neuroimaging. Cambridge: Cambridge University
and microstate models. In S. Grillner & A. M. Grabyel (Eds.), Press.
Microcircuits: The Interface between neurons and global Michel, C. M., & Murray, M. M. (2012). Towards the utilization of
brain function (pp. 347370). Cambridge, MA: MIT Press. EEG as a brain imaging tool. NeuroImage, 61(2), 371385.
Damian, M. F., & Dumay, N. (2007). Time pressure and phonolo- doi:10.1016/j.neuroimage.2011.12.039
gical advance planning in spoken production. Journal of Mller, J., Jansma, B. M., Rodriguez-Fornells, A., & Mnte, T. F.
Memory and Language, 57(2), 195209. doi:10.1016/j.jml. (2007). What the brain does before the tongue slips.
2006.11.001 Cerebral Cortex, 17(5), 11731178. doi:10.1093/cercor/bhl028
Damian, M. F., & Dumay, N. (2009). Exploring phonological Murray, M. M., Brunet, D., & Michel, C. M. (2008). Topographic
encoding through repeated segments. Language and ERP analyses: A step-by-step tutorial review. Brain
Cognitive Processes, 24(5), 685712. doi:10.1080/ Topography, 20(4), 249264. doi:10.1007/s10548-008-0054-5
01690960802351260 Oldfield, R. C. (1971). The assessment and analysis of handed-
Dell, G. S. (1984). Representation of serial order in speech: ness: The Edinburgh inventory. Neuropsychologia, 9(1), 97
Evidence from the repeated phoneme effect in speech 113. doi:10.1016/0028-3932(71)90067-4
errors. Journal of Experimental Psychology: Learning, Perrin, F., Pernier, J., Bertnard, O., Giard, M. H., & Echallier, J. F.
Memory and Cognition, 10, 222233. (1987). Mapping of scalp potentials by surface spline interp-
Dell, G. S. (1986). A spreading-activation theory of retrieval in olation. Electroencephalography and Clinical Neurophysiology,
sentence production. Psychological Review, 93(3), 283321. 66(1), 7581. doi:10.1016/0013-4694(87)90141-6
doi:10.1037/0033-295X.93.3.283 Protopapas, A. (2007). Check vocal: A program to facilitate
Dell, G. S. (1988). The retrieval of phonological forms in pro- checking the accuracy and response time of vocal responses
duction: Tests of predictions from a connectionist model. from DMDX. Behavior Research Methods, 39(4), 859862.
Journal of Memory and Language, 27(2), 124142. doi:10. doi:10.3758/BF03192979
1016/0749-596X(88)90070-8 Roelofs, A. (1999). Phonological segments and features as plan-
Frisch, S. A., & Wright, R. (2002). The phonetics of phonological ning units in speech production. Language and Cognitive
speech errors: An acoustic analysis of slips of the tongue. Processes, 14(2), 173200. doi:10.1080/016909699386338
Journal of Phonetics, 30(2), 139162. doi:10.1006/jpho.2002. Schwartz, M. F., Saffran, E. M., Bloch, D. E., & Dell, G. S. (1994).
0176 Disordered speech production in aphasic and normal speakers.
Fromkin, V. (1973). Slips of the tongue. San Francisco: WH Brain and Language, 47(1), 5288. doi:10.1006/brln.1994.1042
Freeman. Sevald, C. A., & Dell, G. S. (1994). The sequential cuing effect in
Indefrey, P. (2011). The spatial and temporal signatures of word speech production. Cognition, 53(2), 91127. doi:10.1016/
production components: A critical update. Language 0010-0277(94)90067-1
Sciences, 2, 255. doi:10.3389/fpsyg.2011.00255 Shattuck-Hufnagel, S. (1992). The role of word structure in seg-
Koukkou, M., & Lehmann, D. (1987). An information-processing mental serial ordering. Cognition, 42(13), 213259. doi:10.
perspective of psychophysiological measurements. Journal 1016/0010-0277(92)90044-I
of Psychophysiology, 1(2), 109112. Stemberger, J. P. (2009). Preventing perseveration in language
Laganaro, M. (2014). ERP topographic analyses from concept to production. Language and Cognitive Processes, 24(10), 1431
articulation in word production studies. Language Sciences, 5, 1470. doi:10.1080/01690960902836624
493. doi:10.3389/fpsyg.2014.00493 Wilshire, C. E. (1999). The Tongue Twister paradigm as a tech-
Laganaro, M., Python, G., & Toepel, U. (2013). Dynamics of pho- nique for studying phonological encoding. Language and
nologicalphonetic encoding in word production: Evidence Speech, 42(1), 5782. doi:10.1177/00238309990420010301
from diverging ERPs between stroke patients and controls. Yu, M., Mo, C., & Mo, L. (2014). The role of phoneme in Mandarin
Brain and Language, 126(2), 123132. doi:10.1016/j.bandl. Chinese production: Evidence from ERPs. PLOS One, 9(9),
2013.03.004 e106486. doi:10.1371/journal.pone.0106486

You might also like