Íslenskt mál og almenn málfræði - 01.01.2023, Side 210

‘be cold’ (i.e., these are experiencer subjects), and on objects of the novel verbs
meaning ‘throw’, ‘push’, ‘love’ and the indirect object of ‘give’. The preference for
producing dative case on objects of ‘throw’ and ‘push’ is consistent with the gen-
eralization that dative is particularly associated with motion theme objects.
Taken together, Nowenstein’s comprehension and production experiments illus-
trate children’s highly regular patterns in associating particular cases with partic-
ular syntactic and semantic roles, even when applying dative case to subjects and
direct objects. This sets the stage for answering how children form these associ-
ations on the basis of their input.
The corpus analysis (chapter 4) enables a careful investigation of the notion
of ‘productivity’ in grammar, an essential concept within theories of grammatical
representation and learning. The case marking system in Icelandic provides an
interesting test case for Yang’s Tolerance Principle (Yang 2016) because of its
combination of canonical mapping patterns and significant range and amount of
exceptions, which historically have been thorns in the side of rule-based learning
models. The Tolerance Principle, crucially, allows for exceptions within a pro-
ductive rule, as long as the number of exceptions is below a certain threshold.
And the threshold is dependent on the size of the particular class, such that small-
er classes tolerate a larger percentage of exceptions. Importantly, the Tolerance
Principle is not about applying raw frequencies in the input to rules: what mat-
ters is the proportion of exceptions, relative to the size of the category.
Applying the Tolerance Principle to the carefully coded corpus reveals the
following patterns: Pairing the syntactic role of direct object with the thematic
role of patient yields productivity for accusative case for the whole set of verbs in
the corpus, whereas without that combination the syntactic role of direct object
only predicts accusative case for the most frequent verbs. Nevertheless, there is
no combination of the features of direct object, motion, and theme that yields
productivity for dative case. In terms of deriving thematic role from case (com-
prehension), nominative subjects can hold a variety of thematic roles, so even the
combination of case and syntactic position is not very informative about the the-
matic interpretation of a nominative subject argument. On the other hand, non-
nominative subjects are invariably experiencers. This is true despite the fact that
the majority of experiencer subjects bear nominative case. For objects, combining
the direct object syntactic role with dative case predicts a theme role only for the
more frequent verbs in the corpus, and combining accusative case with the direct
object syntactic role predicts either patients or themes, so there is a less straight-
forward relationship between case and meaning for objects (in that direction),
even when incorporating syntactic role information.
Thus, the overall claim is that syntactic role is a better predictor of case than
thematic role, but that syntactic role plus thematic role is an even better predictor
in some cases. Moreover, the combination of syntactic role (form) and thematic
role (meaning) is necessary to form productive rules for case marking patterns that
deviate from the most basic, robust alignment, for example for dative subjects.
Misha Becker210