The Simple Truth about Dependency and Phrase Structure Representations: An Opinion Piece

Human Language Technologies: The 2010 Annual Conference of the North American Chapter of the ACL, pages 337–340, Los Angeles, California, June 2010. c⃝2010 Association for Computational Linguistics The Simple Truth about Dependency and Phrase Structure Representations An Opinion Piece Owen Rambow CCLS, Columbia University New York, NY, USA rambow@ccls.columbia.edu Abstract There are many misconceptions about de- pendency representations and phrase structure representations for syntax. They are partly due to terminological confusion, partly due to a lack of meta-scientific clarity about the roles of representations and linguistic theories. This opinion piece argues for a simple but clear view of syntactic representation. 1 Introduction To the machine learning community, treebanks are just collections of data, like pixels with captions, structural and behavioral facts about genes, or ob- servations about wild boar populations. In contrast, to us computational linguists, treebanks are not nat- urally occurring data at all: they are the result of a very complex annotation process. While the text that is annotated (usually) is naturally occurring, the annotation itself is already the result of a scientific activity. This opinion piece argues that the level of discourse about treebanks often found in our com- munity does not reflect this fact (presumably due to the influence of the brute machine learning per- spective). We, as a community of computational lin- guists, need to be very precise when talking about treebanks and syntactic representations in general. So let’s start with three very important concepts which we must always distinguish. The representa- tion type: what type of mathematical object is used to represent syntactic facts? In this opinion piece, I only consider dependency trees (DTs) and phrase structure trees (PSTs) (Section 2). The represented syntactic content: the morphological and syntactic facts of the analyzed sentence (Section 3). The syn- tactic theory: it explains how syntactic content is represented in the chosen representation type (Sec- tion 4). A crucial confusing factor is the fact that the terms dependency and phrase structure both have both a mathematical and a linguistic meaning. The math- ematical meaning refers representation types. The linguistic meaning refers to syntactic content. I dis- cuss this issue in Section 3. I discuss the issue of converting between DTs and PSTs in Section 5, as an example of how my proposed conceptualization of syntactic representation throws light on a compu- tational problem. This opinion piece will be a success if after read- ing it, the reader concludes that actually he or she knew this all along. In fact, this opinion piece does not advocate for a controversial position; its mission is to make its readers be more precise when talking about syntactic representations. This opinion piece is intentionally polemical for rhetorical reasons. 2 DTs and PSTs as Representation Types Assume we have two disjoint symbol sets: a set of terminal symbols which contains the words of the language we are describing; and a set of nontermi- nal symbols. A Dependency Tree (DT) is a tree in which all nodes are labeled with words (elements of …

Content truncated to 500 words