1.

df <- read.csv("http://goo.gl/aE3yVg")
fit <- ctree(v.elision~., data = df)
plot(fit)

print(fit)
## 
##   Conditional inference tree with 4 terminal nodes
## 
## Response:  v.elision 
## Inputs:  group, word, position 
## Number of observations:  591 
## 
## 1) group == {third}; criterion = 1, statistic = 87.011
##   2) position == {strong}; criterion = 0.999, statistic = 14.01
##     3)*  weights = 150 
##   2) position == {weak}
##     4)*  weights = 86 
## 1) group == {first, second}
##   5) group == {first}; criterion = 1, statistic = 27.204
##     6)*  weights = 172 
##   5) group == {second}
##     7)*  weights = 183
predict(fit, df[45,-1], response = TRUE)
## [1] yes
## Levels: no yes
fit2 <- cforest(v.elision~., data = df, controls=cforest_unbiased(ntree=1000, mtry=2))
predict(fit2, df[45,-1],OOB=TRUE)
## [1] no
## Levels: no yes
varimp(fit2)
##      group       word   position 
## 0.04977419 0.01377419 0.02137788

2.

df <- read.csv("https://goo.gl/F9m4O2")
fit <- ctree(s.deletion~phon.cont+social, data = df)
plot(fit)

print(fit)
## 
##   Conditional inference tree with 5 terminal nodes
## 
## Response:  s.deletion 
## Inputs:  phon.cont, social 
## Number of observations:  8846 
## 
## 1) phon.cont == {pause, vowel}; criterion = 1, statistic = 344.156
##   2) social <= 1; criterion = 1, statistic = 92.762
##     3)*  weights = 182 
##   2) social > 1
##     4) phon.cont == {vowel}; criterion = 1, statistic = 61.559
##       5)*  weights = 1806 
##     4) phon.cont == {pause}
##       6)*  weights = 1258 
## 1) phon.cont == {consonant}
##   7) social <= 1; criterion = 1, statistic = 118.054
##     8)*  weights = 397 
##   7) social > 1
##     9)*  weights = 5203
predict(fit, df[1,-c(1:2)], response = TRUE)
## [1] no
## Levels: no yes
fit2 <- cforest(s.deletion~phon.cont+social, data = df, controls=cforest_unbiased(ntree=100, mtry=2))
varimp(fit2)
##     phon.cont        social 
## -2.150538e-05  3.702304e-02
predict(fit2, df[45,-1],OOB=TRUE)
## [1] no
## Levels: no yes