Assessing Corpus Evidence for Formal and Psycholinguistic Constraints on Nonprojectivity

Abstract

Formal constraints on crossing dependencies have played a large role in research on the formal complexity of natural language grammars and parsing. Here we ask whether the apparent constraints on crossing dependencies observable in treebanks might in fact result from other, independent constraints on syntactic trees, such as low arity and dependency length minimization. We address this question using two sets of experiments. In Experiment 1, we compare the distribution of formal properties of crossing dependencies, such as gap degree, between real trees and baseline trees matched for rate of crossing dependencies and various other properties. In Experiment 2, we model whether two dependencies cross given certain psycholinguistic properties of the dependencies. We find no evidence for constraints on gap degree and well-nestedness beyond what can be explained by constraints on rate of crossing dependencies and topological properties of the trees. However, edge degree, end-point crossings, and heads' depth difference differ significantly between real and random trees. Further, modeling results show that information locality and working-memory limitations affect whether two dependencies cross or not, but we do not find evidence that these pressures can fully explain the distribution of crossing dependencies in natural languages.