Data, documentation and analysis scripts for Articulatory differences between L1 and L2 speakers of English
Martijn Wieling (University of Groningen, Haskins Laboratories), Pauline Veenstra (University of Groningen), Patti Adank (University College London), and Mark Tiede (Haskins Laboratories)
1 Abstract
In this study, we investigate differences between native English speakers and the English pronunciation of Dutch and German speakers. We focus on the articulatory trajectories obtained using electromagnetic articulography and particularly investigate two sound contrasts: /t/-/θ/ and /s/-/ʃ/. Our results show that while German speakers make both sound contrasts adequately, the Dutch speakers do not distinguish them clearly. To further evaluate these results, both a human Dutch listener as well as an automatic speech recognition (ASR) system classified the pronounced words on the basis of the acoustic recording. Both classifications lined up with the articulatory results. For Dutch speakers, /θ/-words (and /s/-words) were more frequently recognized as /t/-words (and /ʃ/-words). However, the intended utterance was still recognized in the majority of cases for the Dutch speakers. The perceptual results therefore do not support a complete merger of the sounds in Dutch.
Journal: Submitted to Proceedings of ISSP 2017Submitted
Preprint: http://www.martijnwieling.nl/files/ISSP-Wieling.pdf
Keywords: Generalized additive modeling; Tutorial; Articulography; Second language acquisition
## Generated on: August 30, 2017 - 23:39:252 Libraries and functions
The following commands load the necessary functions and libraries and show the version information.
# install packages if not yet installed
packages <- c("mgcv","itsadug","lme4")
if (length(setdiff(packages, rownames(installed.packages()))) > 0) {
install.packages(setdiff(packages, rownames(installed.packages())))
}
# load required packages
library(mgcv)
library(itsadug)
library(lme4)
# version information
R.version.string## [1] "R version 3.4.1 (2017-06-30)"cat(paste('mgcv version:',packageVersion('mgcv')))## mgcv version: 1.8.18cat(paste('itsadug version:',packageVersion('itsadug')))## itsadug version: 2.2.4cat(paste('lme4 version:',packageVersion('lme4')))## lme4 version: 1.1.123 Production datasets
The following shows the columns of the full dataset and their explanation.
if (!file.exists('datth.rda')) {
download.file('http://www.let.rug.nl/wieling/ISSP2017/datth.rda', 'datth.rda')
}
if (!file.exists('datsh.rda')) {
download.file('http://www.let.rug.nl/wieling/ISSP2017/datsh.rda', 'datsh.rda')
}
load('datth.rda')
load('datsh.rda')3.1 Column names
The dataset datsh consists of 265599 rows and 10 columns, whereas the dataset datth consists of 223954 rows and 10 columns. Both datasets have the following column names:
colnames(datth)## [1] "Speaker" "Lang" "Sensor" "Axis" "Trial" "Word" "Sound"
## [8] "Loc" "Time" "Pos"3.2 Data description
- Speaker – ID of the speaker
- Lang – Native language of the speaker (
"NL"for Dutch,"DE"for German, or"EN"for English) - Sensor – The sensor (in this case only `“TT”, the tongue tip sensor)
- Axis – The Axis (in this case only
"X", the anterior-posterior position) - Trial – The trial number of the word
- Word – The label of the word
- Sound – The sound contrast (
"TH"for words with the dental fricative,"T"for words with the stop ; or"SH"for words with the post-alveolar fricative, and"S"for words with the alveolar fricative in datasetdatsh) - Loc – The location where in the word the sound contrasts occurs (
"START"when it occurs at the beginning of the word or"END"when it occurs at the back of the word - Time – The normalized (between 0: beginning of the word, to 1: end of the word)
- Pos – The standardized (mean 0, standard deviation 1) position for each speaker of the T1 sensor in the anterior-posterior direction (higher values, more anterior)
4 Contrasting /θ/ and /t/ in production
datth <- start_event(datth,event=c("Speaker","Trial"))
datth$LangLoc <- interaction(datth$Lang, datth$Loc)
datth$IsENTHStart <- (datth$Lang == "EN" & datth$Sound == "TH" & datth$Loc == "Start")*1
datth$IsNLTHStart <- (datth$Lang == "NL" & datth$Sound == "TH" & datth$Loc == "Start")*1
datth$IsDETHStart <- (datth$Lang == "DE" & datth$Sound == "TH" & datth$Loc == "Start")*1
datth$IsENTHEnd <- (datth$Lang == "EN" & datth$Sound == "TH" & datth$Loc == "End")*1
datth$IsNLTHEnd <- (datth$Lang == "NL" & datth$Sound == "TH" & datth$Loc == "End")*1
datth$IsDETHEnd <- (datth$Lang == "DE" & datth$Sound == "TH" & datth$Loc == "End")*1
datth$SpeakerSoundLoc <- interaction(datth$Speaker, datth$Sound, datth$Loc)
system.time(th1 <- bam(Pos ~ LangLoc + s(Time,by=LangLoc) + s(Time,by=IsENTHStart) + s(Time,by=IsENTHEnd) + s(Time,by=IsNLTHStart) + s(Time,by=IsNLTHEnd) + s(Time,by=IsDETHStart) + s(Time,by=IsDETHEnd) + s(Time,SpeakerSoundLoc,bs="fs",m=1) + s(Time,Word,bs="fs",m=1), data=datth, discrete=TRUE, rho=0.999, nthreads=8, AR.start=datth$start.event))## Warning in gam.side(sm, X, tol = .Machine$double.eps^0.5): model has
## repeated 1-d smooths of same variable.## user system elapsed
## 872.532 6.548 150.889acf_resid(th1)
(smryth1 <- summary(th1))##
## Family: gaussian
## Link function: identity
##
## Formula:
## Pos ~ LangLoc + s(Time, by = LangLoc) + s(Time, by = IsENTHStart) +
## s(Time, by = IsENTHEnd) + s(Time, by = IsNLTHStart) + s(Time,
## by = IsNLTHEnd) + s(Time, by = IsDETHStart) + s(Time, by = IsDETHEnd) +
## s(Time, SpeakerSoundLoc, bs = "fs", m = 1) + s(Time, Word,
## bs = "fs", m = 1)
##
## Parametric coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) -0.2872 0.1038 -2.768 0.00564 **
## LangLocEN.End 0.1297 0.1701 0.762 0.44601
## LangLocNL.End 0.3559 0.1563 2.277 0.02280 *
## LangLocDE.Start 0.2498 0.1676 1.491 0.13597
## LangLocEN.Start 0.1647 0.1790 0.920 0.35751
## LangLocNL.Start 0.2462 0.1773 1.389 0.16493
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Approximate significance of smooth terms:
## edf Ref.df F p-value
## s(Time):LangLocDE.End 1.002 1.003 7.464 0.00623 **
## s(Time):LangLocEN.End 3.017 3.368 3.357 0.03084 *
## s(Time):LangLocNL.End 4.531 4.872 2.160 0.10285
## s(Time):LangLocDE.Start 7.318 7.474 4.999 7.34e-06 ***
## s(Time):LangLocEN.Start 6.906 7.096 2.609 0.00530 **
## s(Time):LangLocNL.Start 6.835 7.019 1.431 0.23196
## s(Time):IsENTHStart 7.338 7.592 8.109 1.04e-10 ***
## s(Time):IsENTHEnd 6.053 6.354 2.200 0.05283 .
## s(Time):IsNLTHStart 4.724 5.054 1.249 0.23463
## s(Time):IsNLTHEnd 6.228 6.526 0.529 0.83424
## s(Time):IsDETHStart 8.601 8.755 25.994 < 2e-16 ***
## s(Time):IsDETHEnd 8.558 8.715 17.439 < 2e-16 ***
## s(Time,SpeakerSoundLoc) 2037.213 2481.000 15.669 < 2e-16 ***
## s(Time,Word) 150.959 176.000 133.483 < 2e-16 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## R-sq.(adj) = 0.468 Deviance explained = 47.3%
## fREML = -2.1637e+05 Scale est. = 3.7864 n = 223954par(mfrow=c(3,2),mar=c(5.1, 5.1, 4.1, 2.1))
plot(th1,select=7,shade=T,rug=F, ylim=c(-0.6,2.1), main='TH vs T: English (start)', cex.lab=2.5, cex.axis=2.5, cex.main=2.5, cex.sub=2.5, ylab='Position difference')
abline(h=0)
plot(th1,select=8,shade=T,rug=F, ylim=c(-0.6,2.1), main='TH vs T: English (end)', cex.lab=2.5, cex.axis=2.5, cex.main=2.5, cex.sub=2.5, ylab='Position difference')
abline(h=0)
plot(th1,select=11,shade=T,rug=F, ylim=c(-0.6,2.1), main='TH vs T: German (start)', cex.lab=2.5, cex.axis=2.5, cex.main=2.5, cex.sub=2.5, ylab='Position difference')
abline(h=0)
plot(th1,select=12,shade=T,rug=F, ylim=c(-0.6,2.1), main='TH vs T: German (end)', cex.lab=2.5, cex.axis=2.5, cex.main=2.5, cex.sub=2.5, ylab='Position difference')
abline(h=0)
plot(th1,select=9,shade=T,rug=F, ylim=c(-0.6,2.1), main='TH vs T: Dutch (start)', cex.lab=2.5, cex.axis=2.5, cex.main=2.5, cex.sub=2.5, ylab='Position difference')
abline(h=0)
plot(th1,select=10,shade=T,rug=F, ylim=c(-0.6,2.1), main='TH vs T: Dutch (end)', cex.lab=2.5, cex.axis=2.5, cex.main=2.5, cex.sub=2.5, ylab='Position difference')
abline(h=0)
5 Contrasting /ʃ/ and /s/ in production
datsh <- start_event(datsh, event=c("Speaker","Trial"))
datsh$LangLoc <- interaction(datsh$Lang, datsh$Loc)
datsh$IsENSHStart <- (datsh$Lang == "EN" & datsh$Sound == "SH" & datsh$Loc == "Start")*1
datsh$IsNLSHStart <- (datsh$Lang == "NL" & datsh$Sound == "SH" & datsh$Loc == "Start")*1
datsh$IsDESHStart <- (datsh$Lang == "DE" & datsh$Sound == "SH" & datsh$Loc == "Start")*1
datsh$IsENSHEnd <- (datsh$Lang == "EN" & datsh$Sound == "SH" & datsh$Loc == "End")*1
datsh$IsNLSHEnd <- (datsh$Lang == "NL" & datsh$Sound == "SH" & datsh$Loc == "End")*1
datsh$IsDESHEnd <- (datsh$Lang == "DE" & datsh$Sound == "SH" & datsh$Loc == "End")*1
datsh$SpeakerSoundLoc <- interaction(datsh$Speaker, datsh$Sound, datsh$Loc)
system.time(sh1 <- bam(Pos ~ LangLoc + s(Time,by=LangLoc) + s(Time,by=IsENSHStart) + s(Time,by=IsENSHEnd) + s(Time,by=IsNLSHStart) + s(Time,by=IsNLSHEnd) + s(Time,by=IsDESHStart) + s(Time,by=IsDESHEnd) + s(Time,SpeakerSoundLoc,bs="fs",m=1) + s(Time,Word,bs="fs",m=1), data=datsh, discrete=TRUE, rho=0.999, , nthreads=8, AR.start=datsh$start.event))## Warning in gam.side(sm, X, tol = .Machine$double.eps^0.5): model has
## repeated 1-d smooths of same variable.## user system elapsed
## 1239.356 10.288 216.458acf_resid(sh1)
(smrysh1 <- summary(sh1))##
## Family: gaussian
## Link function: identity
##
## Formula:
## Pos ~ LangLoc + s(Time, by = LangLoc) + s(Time, by = IsENSHStart) +
## s(Time, by = IsENSHEnd) + s(Time, by = IsNLSHStart) + s(Time,
## by = IsNLSHEnd) + s(Time, by = IsDESHStart) + s(Time, by = IsDESHEnd) +
## s(Time, SpeakerSoundLoc, bs = "fs", m = 1) + s(Time, Word,
## bs = "fs", m = 1)
##
## Parametric coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) 0.01990 0.09964 0.200 0.842
## LangLocEN.End -0.10295 0.14271 -0.721 0.471
## LangLocNL.End -0.15537 0.14255 -1.090 0.276
## LangLocDE.Start 0.17203 0.14287 1.204 0.229
## LangLocEN.Start 0.16462 0.14636 1.125 0.261
## LangLocNL.Start -0.01689 0.15045 -0.112 0.911
##
## Approximate significance of smooth terms:
## edf Ref.df F p-value
## s(Time):LangLocDE.End 7.394 7.534 12.784 < 2e-16 ***
## s(Time):LangLocEN.End 7.530 7.669 6.548 1.68e-08 ***
## s(Time):LangLocNL.End 5.816 6.055 1.855 0.096879 .
## s(Time):LangLocDE.Start 6.069 6.311 3.629 0.000718 ***
## s(Time):LangLocEN.Start 6.526 6.739 6.669 2.74e-07 ***
## s(Time):LangLocNL.Start 5.588 5.846 0.678 0.675705
## s(Time):IsENSHStart 6.180 6.509 4.055 0.000187 ***
## s(Time):IsENSHEnd 5.075 5.420 2.061 0.090637 .
## s(Time):IsNLSHStart 3.747 4.062 0.807 0.562415
## s(Time):IsNLSHEnd 5.620 5.949 1.025 0.417433
## s(Time):IsDESHStart 8.497 8.666 20.351 < 2e-16 ***
## s(Time):IsDESHEnd 7.074 7.339 7.621 1.81e-05 ***
## s(Time,SpeakerSoundLoc) 2040.295 2481.000 16.271 < 2e-16 ***
## s(Time,Word) 166.613 194.000 139.157 < 2e-16 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## R-sq.(adj) = 0.452 Deviance explained = 45.7%
## fREML = -2.9479e+05 Scale est. = 2.8639 n = 265599par(mfrow=c(3,2),mar=c(5.1, 5.1, 4.1, 2.1))
plot(sh1,select=7,shade=T,rug=F, ylim=c(-2.1,0.6), main='SH vs S: English (start)', cex.lab=2.5, cex.axis=2.5, cex.main=2.5, cex.sub=2.5, ylab='Position difference')
abline(h=0)
plot(sh1,select=8,shade=T,rug=F, ylim=c(-2.1,0.6), main='SH vs S: English (end)', cex.lab=2.5, cex.axis=2.5, cex.main=2.5, cex.sub=2.5, ylab='Position difference')
abline(h=0)
plot(sh1,select=11,shade=T,rug=F, ylim=c(-2.1,0.6), main='SH vs S: German (start)', cex.lab=2.5, cex.axis=2.5, cex.main=2.5, cex.sub=2.5, ylab='Position difference')
abline(h=0)
plot(sh1,select=12,shade=T,rug=F, ylim=c(-2.1,0.6), main='SH vs S: German (end)', cex.lab=2.5, cex.axis=2.5, cex.main=2.5, cex.sub=2.5, ylab='Position difference')
abline(h=0)
plot(sh1,select=9,shade=T,rug=F, ylim=c(-2.1,0.6), main='SH vs S: Dutch (start)', cex.lab=2.5, cex.axis=2.5, cex.main=2.5, cex.sub=2.5, ylab='Position difference')
abline(h=0)
plot(sh1,select=10,shade=T,rug=F, ylim=c(-2.1,0.6), main='SH vs S: Dutch (end)', cex.lab=2.5, cex.axis=2.5, cex.main=2.5, cex.sub=2.5, ylab='Position difference')
abline(h=0)
6 Perception dataset
The following shows the columns of the full dataset and their explanation.
if (!file.exists('perc.rda')) {
download.file('http://www.let.rug.nl/wieling/ISSP2017/perc.rda', 'perc.rda')
}
load('perc.rda')6.1 Column names
The dataset perc consists of 6468 rows and 13 columns with the following column names:
colnames(perc)## [1] "Speaker" "Gender" "BirthYear"
## [4] "Lang" "Trial" "WordActual"
## [7] "WordRecognized" "WordRecognizedASR" "Sound"
## [10] "SoundRecog" "SoundRecogASR" "Correct"
## [13] "CorrectASR"6.2 Data description
- Speaker – ID of the speaker
- Gender – Gender of the speaker
- BirthYear – Year of birth of the speaker (age is calculated by subtracting the year of birth from the recording year: 2014)
- Lang – Native language of the speaker (
"NL"for Dutch,"DE"for German, or"EN"for English) - Trial – The trial number of the word
- WordActual – The label of the word
- WordRecognized – The recognized word by a native Dutch speaker with good English proficiency
- WordRecognizedASR – The recognized word by the Google Cloud Speech API (at the end of 2016)
- Sound – The sound of interest (
"TH"for words with the dental fricative,"T"for words with the stop,"St"for words with an s instead of the stop or dental fricative,"SH"for words with a post-alveolar fricative,"S"for words with an alveolar fricative - SoundRecog – The sound recognized by the native Dutch speaker
- SoundRecogASR – The sound recognized by the Google Cloud Speech API
- Correct –
1if the Dutch speaker recognized the word which was pronounced by the speaker,0if not - CorrectASR –
1if the Google ASR system recognized the word which was pronounced by the speaker,0if not
6.3 Demographics
round(mean(2014-perc[perc$Lang=='EN',]$BirthYear),1)## [1] 25table(unique(perc[perc$Lang=='EN',c("Speaker","Gender")])$Gender)##
## F M
## 14 8round(mean(2014-perc[perc$Lang=='NL',]$BirthYear),1)## [1] 20.7table(unique(perc[perc$Lang=='NL',c("Speaker","Gender")])$Gender)##
## F M
## 8 12round(mean(2014-perc[perc$Lang=='DE',]$BirthYear),1)## [1] 23table(unique(perc[perc$Lang=='DE',c("Speaker","Gender")])$Gender)##
## F M
## 16 117 Perceptual analysis: human listener
7.1 /θ/ incorrectly recognized as /t/
# TH slechter in NL herkend
percth = droplevels(perc[perc$Sound %in% c('TH'),])
m = glmer(Correct ~ Lang + (1|Speaker), data=percth, family='binomial', control = glmerControl(optimizer = "bobyqa"))
summary(m)## Generalized linear mixed model fit by maximum likelihood (Laplace
## Approximation) [glmerMod]
## Family: binomial ( logit )
## Formula: Correct ~ Lang + (1 | Speaker)
## Data: percth
## Control: glmerControl(optimizer = "bobyqa")
##
## AIC BIC logLik deviance df.resid
## 1082.4 1102.9 -537.2 1074.4 1241
##
## Scaled residuals:
## Min 1Q Median 3Q Max
## -3.5770 0.2608 0.3328 0.4291 1.1997
##
## Random effects:
## Groups Name Variance Std.Dev.
## Speaker (Intercept) 0.7665 0.8755
## Number of obs: 1245, groups: Speaker, 69
##
## Fixed effects:
## Estimate Std. Error z value Pr(>|z|)
## (Intercept) 2.2043 0.2485 8.871 < 2e-16 ***
## LangDE -0.1513 0.3327 -0.455 0.649330
## LangNL -1.2735 0.3410 -3.734 0.000188 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Correlation of Fixed Effects:
## (Intr) LangDE
## LangDE -0.728
## LangNL -0.721 0.532round(prop.table(with(percth[percth$Lang=='NL',],table(Sound,SoundRecog))),2)## SoundRecog
## Sound St T TH
## TH 0.14 0.17 0.70round(prop.table(with(percth[percth$Lang=='DE',],table(Sound,SoundRecog))),2)## SoundRecog
## Sound St T TH
## TH 0.07 0.07 0.86round(prop.table(with(percth[percth$Lang=='EN',],table(Sound,SoundRecog))),2)## SoundRecog
## Sound St T TH
## TH 0.09 0.04 0.887.2 /s/ incorrectly recognized as /ʃ/
percs = droplevels(perc[perc$Sound %in% c('S'),])
m = glmer(Correct ~ Lang + (1|Speaker), data=percs, family='binomial')
summary(m)## Generalized linear mixed model fit by maximum likelihood (Laplace
## Approximation) [glmerMod]
## Family: binomial ( logit )
## Formula: Correct ~ Lang + (1 | Speaker)
## Data: percs
##
## AIC BIC logLik deviance df.resid
## 880.6 901.6 -436.3 872.6 1436
##
## Scaled residuals:
## Min 1Q Median 3Q Max
## -5.0750 0.1493 0.2343 0.3490 1.0307
##
## Random effects:
## Groups Name Variance Std.Dev.
## Speaker (Intercept) 1.053 1.026
## Number of obs: 1440, groups: Speaker, 69
##
## Fixed effects:
## Estimate Std. Error z value Pr(>|z|)
## (Intercept) 2.5928 0.2912 8.904 <2e-16 ***
## LangDE 0.8431 0.4187 2.014 0.0440 *
## LangNL -0.8972 0.3951 -2.271 0.0232 *
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Correlation of Fixed Effects:
## (Intr) LangDE
## LangDE -0.630
## LangNL -0.711 0.473round(prop.table(with(percs[percs$Lang=='NL',],table(Sound,SoundRecog))),2)## SoundRecog
## Sound S SH
## S 0.81 0.19round(prop.table(with(percs[percs$Lang=='DE',],table(Sound,SoundRecog))),2)## SoundRecog
## Sound S SH
## S 0.95 0.05round(prop.table(with(percs[percs$Lang=='EN',],table(Sound,SoundRecog))),2)## SoundRecog
## Sound S SH
## S 0.9 0.17.3 /ʃ/ incorrectly recognized as /s/
percsh = droplevels(perc[perc$Sound %in% c('SH'),])
m = glmer(Correct ~ Lang + (1|Speaker), data=percsh, family='binomial')
summary(m)## Generalized linear mixed model fit by maximum likelihood (Laplace
## Approximation) [glmerMod]
## Family: binomial ( logit )
## Formula: Correct ~ Lang + (1 | Speaker)
## Data: percsh
##
## AIC BIC logLik deviance df.resid
## 457.5 478.0 -224.7 449.5 1256
##
## Scaled residuals:
## Min 1Q Median 3Q Max
## -5.4273 0.1450 0.1836 0.2206 0.4708
##
## Random effects:
## Groups Name Variance Std.Dev.
## Speaker (Intercept) 0.6346 0.7966
## Number of obs: 1260, groups: Speaker, 69
##
## Fixed effects:
## Estimate Std. Error z value Pr(>|z|)
## (Intercept) 3.6008 0.3561 10.112 <2e-16 ***
## LangDE -0.5327 0.4186 -1.273 0.203
## LangNL -0.1490 0.4698 -0.317 0.751
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Correlation of Fixed Effects:
## (Intr) LangDE
## LangDE -0.734
## LangNL -0.641 0.539round(prop.table(with(percsh[percsh$Lang=='NL',],table(Sound,SoundRecog))),2)## SoundRecog
## Sound S SH
## SH 0.04 0.96round(prop.table(with(percsh[percsh$Lang=='DE',],table(Sound,SoundRecog))),2)## SoundRecog
## Sound S SH
## SH 0.06 0.94round(prop.table(with(percsh[percsh$Lang=='EN',],table(Sound,SoundRecog))),2)## SoundRecog
## Sound S SH
## SH 0.03 0.978 Perceptual analysis: ASR
8.1 /θ/ incorrectly recognized as /t/
percth[percth$SoundRecogASR %in% c('S','SH'),]$SoundRecogASR = NA # other word recognized
m = glmer(CorrectASR ~ Lang + (1|Speaker), data=percth, family='binomial')
summary(m)## Generalized linear mixed model fit by maximum likelihood (Laplace
## Approximation) [glmerMod]
## Family: binomial ( logit )
## Formula: CorrectASR ~ Lang + (1 | Speaker)
## Data: percth
##
## AIC BIC logLik deviance df.resid
## 1409.1 1429.6 -700.6 1401.1 1241
##
## Scaled residuals:
## Min 1Q Median 3Q Max
## -1.0977 -0.5992 -0.5002 1.1029 2.8116
##
## Random effects:
## Groups Name Variance Std.Dev.
## Speaker (Intercept) 0.2989 0.5467
## Number of obs: 1245, groups: Speaker, 69
##
## Fixed effects:
## Estimate Std. Error z value Pr(>|z|)
## (Intercept) -0.7867 0.1595 -4.933 8.09e-07 ***
## LangDE -0.1526 0.2181 -0.700 0.484210
## LangNL -0.9084 0.2501 -3.632 0.000282 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Correlation of Fixed Effects:
## (Intr) LangDE
## LangDE -0.727
## LangNL -0.625 0.460round(prop.table(with(percth[percth$Lang=='NL',],table(Sound,SoundRecogASR))),2)## SoundRecogASR
## Sound St T TH
## TH 0.67 0.14 0.19round(prop.table(with(percth[percth$Lang=='DE',],table(Sound,SoundRecogASR))),2)## SoundRecogASR
## Sound St T TH
## TH 0.62 0.07 0.31round(prop.table(with(percth[percth$Lang=='EN',],table(Sound,SoundRecogASR))),2)## SoundRecogASR
## Sound St T TH
## TH 0.61 0.06 0.338.2 /s/ incorrectly recognized as /ʃ/
percs = droplevels(perc[perc$Sound %in% c('S'),])
percs[percs$SoundRecogASR %in% c('T','St','TH'),]$SoundRecogASR = NA # other word recognized
m = glmer(CorrectASR ~ Lang + (1|Speaker), data=percs, family='binomial')
summary(m)## Generalized linear mixed model fit by maximum likelihood (Laplace
## Approximation) [glmerMod]
## Family: binomial ( logit )
## Formula: CorrectASR ~ Lang + (1 | Speaker)
## Data: percs
##
## AIC BIC logLik deviance df.resid
## 1041.2 1062.3 -516.6 1033.2 1436
##
## Scaled residuals:
## Min 1Q Median 3Q Max
## -4.2568 0.1821 0.2455 0.3982 1.1722
##
## Random effects:
## Groups Name Variance Std.Dev.
## Speaker (Intercept) 0.99 0.995
## Number of obs: 1440, groups: Speaker, 69
##
## Fixed effects:
## Estimate Std. Error z value Pr(>|z|)
## (Intercept) 2.64434 0.29105 9.085 < 2e-16 ***
## LangDE 0.02867 0.38225 0.075 0.940218
## LangNL -1.34533 0.38586 -3.487 0.000489 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Correlation of Fixed Effects:
## (Intr) LangDE
## LangDE -0.703
## LangNL -0.735 0.529round(prop.table(with(percs[percs$Lang=='NL',],table(Sound,SoundRecogASR))),2)## SoundRecogASR
## Sound S SH
## S 0.84 0.16round(prop.table(with(percs[percs$Lang=='DE',],table(Sound,SoundRecogASR))),2)## SoundRecogASR
## Sound S SH
## S 0.98 0.02round(prop.table(with(percs[percs$Lang=='EN',],table(Sound,SoundRecogASR))),2)## SoundRecogASR
## Sound S SH
## S 0.93 0.078.3 /ʃ/ incorrectly recognized as /s/
percsh = droplevels(perc[perc$Sound %in% c('SH'),])
percsh[percsh$SoundRecogASR %in% c('T','St','TH'),]$SoundRecogASR = NA # other word recognized
m = glmer(CorrectASR ~ Lang + (1|Speaker), data=percsh, family='binomial')
summary(m)## Generalized linear mixed model fit by maximum likelihood (Laplace
## Approximation) [glmerMod]
## Family: binomial ( logit )
## Formula: CorrectASR ~ Lang + (1 | Speaker)
## Data: percsh
##
## AIC BIC logLik deviance df.resid
## 1161.6 1182.1 -576.8 1153.6 1256
##
## Scaled residuals:
## Min 1Q Median 3Q Max
## -3.5596 0.2156 0.3561 0.4816 1.6034
##
## Random effects:
## Groups Name Variance Std.Dev.
## Speaker (Intercept) 1.243 1.115
## Number of obs: 1260, groups: Speaker, 69
##
## Fixed effects:
## Estimate Std. Error z value Pr(>|z|)
## (Intercept) 1.9637 0.2893 6.788 1.13e-11 ***
## LangDE -0.5200 0.3792 -1.371 0.170
## LangNL -0.2110 0.4108 -0.514 0.608
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Correlation of Fixed Effects:
## (Intr) LangDE
## LangDE -0.748
## LangNL -0.686 0.520round(prop.table(with(percsh[percsh$Lang=='NL',],table(Sound,SoundRecogASR))),2)## SoundRecogASR
## Sound S SH
## SH 0.09 0.91round(prop.table(with(percsh[percsh$Lang=='DE',],table(Sound,SoundRecogASR))),2)## SoundRecogASR
## Sound S SH
## SH 0.18 0.82round(prop.table(with(percsh[percsh$Lang=='EN',],table(Sound,SoundRecogASR))),2)## SoundRecogASR
## Sound S SH
## SH 0.14 0.869 Replication
To replicate the analysis presented above, you can just copy the following lines to the most recent version of R. Please note that you first need to install Pandoc.
download.file('http://www.let.rug.nl/wieling/ISSP2017/analysisISSP.Rmd', 'analysisISSP.Rmd')
if (length(setdiff('rmarkdown', rownames(installed.packages()))) > 0) {
install.packages('rmarkdown')
}
library(rmarkdown)
render('analysisISSP.Rmd') # generates html file with results
browseURL(paste('file://', file.path(getwd(),'analysisISSP.html'), sep='')) # shows result