I believe it is often cited that dependency grammars are more adequate for free word order languages. How actually, is dependency grammar more descriptive or less limited, than constituency grammar, in describing free word order language texts? can you provide few common cases where constituency grammar would fail to capture the grammar of a sentence in a free word order language?!

  • 3
    It isn’t. The two approaches are complementary. Also it’s better to distinguish between configurational and nonconfigurational languages rather than free/strict word order. Any serious grammar formalism has to account for both constituency and headship.
    – Atamiri
    Nov 4, 2017 at 15:12
  • 1
    The idea isn't really "odd", it looks plausible at the first sight. The point is that phrase-based grammars aren't very good at expressing heads whereas dependency grammars aren't good at expressing constituency, so viable theories must somehow mix both approaches together. Once you do that you can adequately account for both strict and free word order languages. There's no such thing as pure dependency grammars (they'd be less descriptive).
    – Atamiri
    Nov 9, 2017 at 0:29
  • Well I'm no longer sure why it even seems plausible at first sight... @Atamiri any resource/s you might recommend on this notion of pure dependency grammars that you just mentioned? I've typically encountered only superficial accounts of that, judging from your last comment. Feeling mildly theoretically confused :)
    – matanox
    Nov 9, 2017 at 5:36
  • You might want to look at XDG and how incomplete it would be without the topological component which accounts for word order.
    – Atamiri
    Nov 9, 2017 at 12:37
  • 1
    The text is ill-formulated, it mixes unrelated notions, so I’m not surprised it might have confused you.
    – Atamiri
    Nov 23, 2017 at 15:34

1 Answer 1


Dependency structures tend to be flatter than phrase structures, which produces fewer opportunities for discontinuities to occur in the syntax. These flatter structures are the reason why many dependency grammar (DG) people claim that dependency is better than phrase structure (i.e. constituency), at addressing free word order. However, dependency is actually no better or worse than phrase structure at addressing the syntax of languages with relative free word order. The two approaches are actually equivalent in their potential to address word order. This equivalence is due to the ability to translate any dependency structure to the corresponding phrase structure -- Note though, that there are phrase structures that cannot be translated to corresponding dependency structures, which means dependency and phrase structure are NOT notational variants of the same thing.

Typical phrase structures are more layered, i.e. taller, than the corresponding dependency structures. Consider an example in this regard, a simple German sentence with a transitive verb (N = nominal, S = sentence, V = verb, VP = verb phrase):

enter image description here

The phrase structure analysis assumes a VP constituent, which is not unusual for many phrase structure approaches to German sentence structure. The presence of the VP constituent means that the phrase structure shown has three layers. In contrast, dependency does not and really cannot acknowledge a finite VP constituent, so the dependency analysis has just two layers. The dependency structure is hence flatter than the phrase structure.

The advantage that the flatter analysis has becomes apparent when one alters the word order:

enter image description here

The hierarchy of words in the dependency analysis remains consistent despite the change in word order. There is one change only: the two nominals N1 and N2 have simply swapped positions. In contrast, what the corresponding phrase structure analysis for the sentence should be is not immediately evident. To achieve the word order shown, the phrase structure account would need at least two changes: the VP constituent would have to swap positions with N1, and then within the VP and the V and the nominal N2 would have two switch positions.

The difficulty facing the phrase structure account in this case is due to the presence of three layers instead of just two. Phrase structure syntax, however, does not in itself necessitate such layered analyses. If one converts the dependency analysis in (1a) directly to phrase structure, one gets the following tree:

enter image description here

This tree is a manifestation of phrase structure because the number of nodes (four) outnumbers the number of words (three). But there are only two levels now, which means the ability to address the alternative word order is the same as for the dependency approach:

enter image description here

Just one change is necessary, just like for the dependency analysis in (2a). The problem for many phrase structure grammars in this area is that they build on the presence of a VP constituent. If in fact no such constituent exists, then a crucial pillar of support for phrase structure syntax in general disappears and there is less reason to choose phrase structure over dependency.

In sum, the reason why DG people claim that DG is better for addressing the syntax of free word order has to do with the flatter dependency structures. Flatter structures provide less opportunity for difficulties to arise when confronted with alternative word orders. The issue of flat vs. layered structures is addressed in more detail in Wikipedia:


  • 1
    I feels like DG proponents are saying "If we do this then we don't have to account for free word order, therefore we have a better account of free word order"... still this is a helpful answer
    – curiousdannii
    Mar 15, 2020 at 3:28
  • I don't see why you used '?' for the Phrase Structure in (2). The cases of the pronouns that allow free word order also tell the analyst how to 'blow on the Calder mobile' to orient the tree.
    – amI
    Mar 15, 2020 at 4:17
  • @aml, Yes, the pronoun forms tell one how to interpret the sentence. The message is that the phrase structure mobile needs two pivot points to get the right word order, whereas the dependency mobile needs just one. Mar 15, 2020 at 6:35
  • @curiousdannii I agree with your point insofar as many DGs are not concerned with predicting which word orders can and cannot occur. This has to do in part, I believe, with the practical goals of the NLP community. From a theoretical standpoint, though, there is no reason why dependency cannot be as, if not more, effective in accounting for free word order than phrase structure. Certainly my efforts with DG focus directly on the word orders that can and cannot occur. Mar 15, 2020 at 6:47

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.