Abstract

The notion of dependency has come to play an increasingly central role in natural language parsing in recent years. On the one hand, lexical dependencies have been incorporated in statistical models for a variety of syntactic representations. On the other hand, dependency relations extracted from such representations have been exploited in many practical applications. Given these developments, it is not surprising that there has also been a growing interest in parsing models that map sentences directly to dependency trees, an approach that may be called “bare-bones dependency parsing” to distinguish it from parsing methods where dependencies are embedded into or extracted from other types of syntactic representations. In this talk, I will survey recent advances in bare-bones dependency parsing, covering all major approaches but focusing on transition-based methods for highly efficient parsing. I will specifically address the question of how such systems can handle long-distance dependencies and other phenomena that have been argued to require richer representations, and I will discuss recent work that attempts to evaluate bare-bones dependency parsers in relation to other methods for producing dependency trees.