Dear list,
I am using binomial GLMMs (random intercept models) to determine the influences
on determiner omission in PPs in German. Examples are "ohne groessere
Wanderung" (without bigger, i.e. longer hike), "mit blutiger Spritze" (with
bloody syringe). According to the grammar of count nouns, we should always find
a determiner here, i.e ohne *eine* groessere Wanderung, mit *einer* blutigen
Spritze.
For the purposes of this question, I'll focus on the models for the
prepositions *unter* (under) and *mit* (with).
The GLMMs I am using consist of categorial features, i.e. factors, one of which
is called nominal_dep_mod (aggregated nominal dependency). This factor
describes possible extensions of the noun in the PP, i.e. complements and
modifiers, such as postnominal genitives, PPs, relative clauses, and complement
clauses. The factor nominal_dep_mod has seven levels, and I am contrasting
possible extensions with the reference level *na* (no extension).
The GLMMs for *unter* and *mit* share the factor nominal_dep_mod, but they
differ in the other fixed effects (partly because other fixed effects are
relevant, and partly, as in the case of the interpretation of the prepositions,
because the interpretations of the prepositions simply differ, see the models
below).
I would like to compare the odds ratios for nominal_dep_mod. If I just compare
them, then the results are what I would expect from looking into the data: for
one of the prepositions one level of nominal_dep_mod has a much stronger
influence than for the other preposition: an odds ratio of 64 vs. an odds ratio
of 4, while the other levels are on a par. This level is rc, i.e. extension by
a relative clause. The odds ratio of 64 holds for *unter*, so that the presence
of a relative clause makes the absence of a determiner very unlikely. For
*mit*, it is only 4.
My perhaps somewhat basic question is: Given the two models, can I conclude
from comparing the odds ratios across models that the influence of
nominal_dep_mod == rc is much stronger for *unter*, and much weaker for *mit*,
although the other factors (random and fixed) differ? From my understanding of
GLM(M)s, nothing speaks against this interpretation, but you'll never know ...
I have added the two calls as well as the xtabs for nominal_dep_mod for
reference (the feature names are a crude mix of German and English).
Thanks a lot.
With kind regards
Tibor
unter.050315.glmm@call
glmer(formula = determiner ~ nominal_dep_mod + adja_in_hit +
TN_LEX_nominalisierung + prep_meaning + (1 | target_noun_lemma),
data = unter.050315.data, family = binomial(), contrasts = list(prep_meaning
= contr.treatment(levels(unter.050315.data$prep_meaning), 2),
nominal_dep_mod = contr.treatment(levels(unter.050315.data$nominal_dep_mod),
4)))
> xtabs(~nominal_dep_mod, unter.050315.data)
nominal_dep_mod
ag app mnr na oc op other rc
1765 9 412 2832 116 578 260 92
mit.050315.glmm@call
glmer(formula = determiner ~ adja_in_hit + nominal_dep_mod +
TN_LEX_nominalisierung + prep_meaning + TN_LEX_GN_Kommunikation +
TN_LEX_GN_Besitz + TN_LEX_GN_Attribut + TN_LEX_GN_Geschehen +
(1 | target_noun_lemma), data = mit.050315.data, family = binomial(),
contrasts = list(prep_meaning =
contr.treatment(levels(mit.050315.data$prep_meaning), 14), nominal_dep_mod =
contr.treatment(levels(mit.050315.data$nominal_dep_mod), 4)))
> xtabs(~nominal_dep_mod, mit.050315.data)
nominal_dep_mod
ag app mnr na oc op other rc
2866 254 3413 8395 299 1068 1253 350
Prof. Dr. Tibor Kiss, Sprachwissenschaftliches Institut
Ruhr-Universität Bochum D-44780 Bochum
Office: +49-234-322-5114
