10

I am confused regarding the distinction dependency- vs. constituency-based tree: to me they look like they encompass the same information but presented differently.

E.g. in the Wikipedia example the constituency-based tree and dependency-based tree seem to contain the same information:

enter image description here

Same in Klein, Dan, and Christopher D. Manning. "Corpus-based induction of syntactic structure: Models of dependency and constituency." Proceedings of the 42nd Annual Meeting on Association for Computational Linguistics. Association for Computational Linguistics, 2004.:

enter image description here

In what situation a constituency-based tree cannot be constructed from a dependency-based tree, and vice-versa?

hippietrail
  • 14,687
  • 7
  • 61
  • 146
user48665
  • 743
  • 1
  • 7
  • 16

2 Answers2

6

All headed constituency-based structures (i.e. endocentric structures) can be easily translated to the corresponding dependency-based structures. One need merely collapse all the projections (minimal, intermediate, and maximal) of a word down to one node.

A non-headed constituency-based structure (i.e. it has exocentric structures), however, cannot be translated to a dependency-based structure, because dependency by its very nature views all syntactic structures as headed (i.e. endocentric). See the distinction between endocentric and exocentric structures here: https://en.wikipedia.org/wiki/Endocentric.

All dependency-based structures can also easily be automatically translated into corresponding constituency-based structures. But here's the difference: the constituency-based structures that result from the translation are necessarily quite flat, much flatter than most constituency-based grammars want to assume. Above all, the resulting flat structures do not acknowledge a finite VP constituent. Without the finite VP constituent, there is no way to do much of standard Chomskyan syntax. For instance, all explanations of syntactic phenomena that build on c-command are not possible without the finite VP constituent.

The distinction between dependency and constituency is quite profound. Dependency is a strict one-to-one ratio (words to nodes), whereas constituency is a one-to-one-or-more ratio (words to nodes). If one operates with a dependency-based model, the amount of syntactic structure that one can posit is very limited. Whereas if one is operating with a constituency-based model, one has the ability to assume much more layered (i.e. taller) syntactic structures. One has the ability to acknowledge many more groupings of words (i.e. constituents) in the constituency-based model.

As an experiment, count the number of nodes and edges in the two trees in the question. The dependency-based tree contains 7 words and 7 nodes with 6 edges. The constituency tree, in contrast, has 13 nodes and 12 edges. These numbers bear witness to drastically different ways to approach the syntax of natural language. Dependency-based models are minimal, whereas constituency-based models are maximal.

The pertinent question is whether the extra structure that constituency enables is warranted. My personal view is that it is not, but each grammarian has to make up his or her own mind about that.

Tim Osborne
  • 5,747
  • 1
  • 17
  • 40
1

I think you can go the other way too (From Dependency Grammar(DG) to Constituent Grammar(CG)), but it is more complicated, and unlike going from CG to DG there is no one kind of step that can do it all. To change the ones that are exocentric in CG, need a different procedure from the one in which both types of grammar are headed. You may have to introduce an entirely new symbol, such as S, which has nothing corresponding to it in DG, unless you recognize S as a projection of V. And different CG people have different opinions about the number of extensions a given node has. So you have to make a choice in each case and hope that it works for your purposes.

Dan Maxwell

Dan Maxwell
  • 61
  • 3
  • Yes, constituency trees vary significantly from one constituency grammar to the next. Much of the MP now assumes strictly binary branching trees that branch almost exclusively rightward, yet more traditional constituency trees assumed left branching VPs. This sort of variation is less of a concern in DG, since there is not that much room for alternative analyses. The dependency structures are transparent for most. – Tim Osborne Apr 01 '14 at 15:39
  • @TimOsborne What does MP stand for? – user48665 Apr 01 '14 at 21:50
  • 1
    @user48665, MP stands for "Minimalist Program". It is the version of Chomskyan syntax that is now dominant, since about 1995. Before that was GB (government and binding), and before that was TG (transformational grammar). These are all constituency grammars, also called phrase structure grammars (PSGs). The acronyms are terrible, I know. – Tim Osborne Apr 01 '14 at 22:29