A Computational Evaluation of Case-Assignment Algorithms

This project seeks to evaluate how successfully a new theory based on syntactic structure models the distribution of morphological case in Icelandic. This algorithm for case assignment will be tested against a traditional, grammatical-function-based theory on a large number of sentences from a corpus. Morphological case is a noun’s syntactic license to appear in its environment based on dependencies with other constituents, expressed as morphology on the noun. Traditionally, it has been held that case on a noun phrase corresponds to a grammatical function. This correspondence induces an algorithm for assigning case to a noun in a syntactic tree, based on its function. This account, however, has failed to account for the distribution of cases observed in Icelandic. A new theory, based on the structural relations of heads rather than grammatical functions, has been devised to model the Icelandic irregularities while still correctly predicting the cross-linguistic data that appears to be function-based. The theory claims that case is assigned based on lexical properties of heads and syntactic relations among constituents. While its algorithm for assigning case has been well motivated in theory and has succeeded on isolated examples, it has not been widely studied on large quantities of data. This new structure-based algorithm is operationalized as a computer program and compared to the function-based one. Each algorithm is applied to syntax trees from a tree bank of over a million words. Disregarding the cases listed in the tree bank, the program marks the nouns in the trees for case (according to the algorithm at hand) and compares its assignment against the attested case in the corpus. Along the way, it keeps track of how many nouns the given algorithm has marked correctly. The relative scores of each algorithm will answer the question of how successful as a theory of case distribution the structural theory is compared to the traditional account.