Recent Advances in Dependency Parsing

Data-driven (statistical) approaches have been playing an increasingly prominent role in parsing since the 1990s. In recent years, there has been a growing interest in dependency-based as opposed to constituency-based approaches to syntactic parsing, with application to a wide range of research areas and different languages. Graph-based and transition-based methods are the two dominant data-driven approaches to dependency parsing. In a graph-based model, it defines a space of candidate dependency trees for a given sentence. Each candidate tree is scored via a local or global scoring function. The parser (usually uses dynamic programming) outputs the highest-scored tree. In contrast, in a transition-based model, it defines a transition system for mapping a sentence to its dependency tree. It induces a model for predicting the next state transition, given the transition history. Given the induced model, the output parse tree is built deterministically upon the construction of the optimal transition sequence.