I have been trying to learn linguistics, mainly English. Recently I have been studying clauses and dependency trees. I have been wondering -

Whether can we assume a rough tree structure for every kind of clause. For example, if we talk about independent clauses, there are many types of clauses like declarative, interrogative, etc. Now I was wondering whether can we make some sort of fixed(but a little bit flexible) dependency tree??

For example:- considering a declarative clause, we can have a 'nsubj ROOT dobj' dependency tree. Now I know this is not gonna be the case every time, it can be flexible in these ways:

  • det nsubj ROOT dobj
  • nsubj ROOT det dobj
  • det nsubj ROOT aux dobj

and so on....

but still, the basic dependency tree is 'nsubj ROOT dobj'. My question now - is it okay to assume this kind of dependency tree for different types of clause or I might be wrong and if I am wrong, what are the chances.

1 Answer 1


Basic clause patterns can indeed be captured in dependency grammars (DGs) along the lines suggested in the question. However, much rides of course on the language under investigation, since the basic clause patterns in the one language are often distinct from those in the next language. The particular annotation scheme one assumes is also important. Consider in this regard the third pattern listed in the question, i.e. det nsubj ROOT aux dobj. That pattern is apparently assuming that when an auxiliary verb is present, it is NOT the root of the sentence, but rather the content verb is the root. This aspect of the pattern is controversial, since DGs traditionally view the finite verb as the root of the clause, be this finite verb a content verb or an auxiliary verb.

A couple of examples from German can demonstrate the importance of getting the basic annotation scheme right. If the finite verb is the root of the clause as the DG tradition has it, then the V2 trait (verb second) of German matrix clause structure is captured in a straightforward way by the insight that the root verb necessarily has one and only one pre-dependent. Observe the following two trees next; they are of a sentence for which the word order is variable:

enter image description here

Two more possible word orders are shown with the next trees:

enter image description here

As long as the second position in the clause is occupied by the finite verb (i.e. habe ‘have’), there is much flexibility concerning the constituent that appears in the first position. This insight is captured in DG terms by acknowledging that the root must have one and only one pre-dependent. If the content verb getan ‘done’ were the clause root instead of the auxiliary habe ‘have’, this V2 trait of matrix clause structure would be more difficult to discern because there would be no way to know how many constituents could precede the root. Consider in this regard that in the fourth tree given, i.e. the d-tree, the content verb is in first position.

Let’s consider examples from English next. Embedded finite clauses in English have a predictable basic structure. They are almost always SV (subject + finite verb) -- exceptions do occur, although they are rare, for instance in the event of negative inversion and regarding some dialectal variation. This is true, for instance, of object clauses introduced by that, e.g.

(2) They said that FrankS sentV Jill chocolates.

(3) They said that yesterday FrankS sentV Jill chocolates.

The bolded S and V are marking the subject and finite verb. Embedded interrogative clauses that lack an auxiliary verb also consistently have SV order, e.g.

(4) I wonder whoS sentV chocolate to Jill.

(5) I wonder what FrankS sentV to Jill.

(6) I wonder who FrankS sentV chocolate to.

(7) I wonder to whom FrankS sentV chocolate.

It remains true if an auxiliary verb is present, e.g.

(8) I wonder whoS willV send chocolate to Jill.

(9) I wonder what FrankS willV send to Jill.

(10) I wonder who FrankS willV send chocolates to.

(11) I wonder to whom FrankS willV send chocolates.

(Note that I view modal verbs like will as finite because among the verbs, they appear leftmost in a position where the finite verb generally appears.) It’s also true of relative clauses, e.g.

(12) Jill, who FrankS sentV chocolates to

(13) Jill, who FrankS willV send chocolates to

In all of these examples, the subject in the embedded clause immediately precedes the finite verb, be this finite verb an auxiliary verb or a content verb. Therefore, it does indeed appear that the sequence SV is the most distinctive trait of embedded clauses in English. This insight can be easily captured in a DG approach, but in order to do so, one needs to assume that the finite verb dominates the content verb, not vice versa, for only in this manner is there a direct dependency linking the subject to the finite verb. This one dependency is, then, the core dependency responsible for the most distinctive word trait of embedded clauses.

To sum up, the answer to the question is that the patterns suggested in the question can indeed serve as templates for establishing basic aspects of clause structure in a DG approach to syntax. However, for such an approach to work well, one has to get the basic dependency analysis correct. In particular, one has to be clear and confident about which word order qualifies as the root of the clause.

  • Thanks for an elaborate response, really this response is a gold mine for me. As for the question, the situation is that for a particular sentence I am already getting the dpendency tree structure, pos tag, lemmatized token. So there is plenty of info from where I can check the conditions you mentioned. Here I am not concerned much about how down far in the DG, verb succeeds subject, I just wanna confrim that verb do succeeds subject....that you verified....and with other info at disposal, I think its good for me....thanks a lot again Jun 5, 2020 at 17:13
  • @AdityaRustagi Note though that the auxiliary-as-root approach isn’t so traditional after all, in the Prague school of linguistic (Prague Dependency Treebank) auxiliaries depend on the main verb (in both Czech and English). Just for completeness.
    – Atamiri
    Jun 5, 2020 at 17:45
  • 1
    @TimOsborne Why is the object a dependent of the verb in (1a,b) but not (1c,d)? The semantic relation between the verb and its object is the same in each case, right? Also, English embedded clauses don't always exhibit the SV order. Inversion can occur in embedded clauses, giving surface forms like "It was obvious that never before had-V anyone-S seen a sight quite like it". You can find many further examples with a Google search for "that never before had" in quotation marks. This occurs in contexts besides negative inversion too; some dialects even allow it in embedded questions. Jun 5, 2020 at 19:15
  • @Atamari, what you write is not fully accurate. It's just the copula in the Prague Tree Bank that is subordinated to the predicative expression. Modal verbs, in contrast, are positioned as the root. Furthermore, most DG theoreticians have long arrived at the conclusion that the finite verb is the root, e.g. Hudson, Mel'cuk, Heringer, Eroms, Gerdes, Kahane, Gross, etc. More importantly, there are many concrete linguistic reasons for taking the finite verb as the root. I have written much about this directly: glossa-journal.org/articles/10.5334/gjgl.537. Jun 6, 2020 at 2:29
  • @one-off-post Yes, I omitted the analysis of discontinuities in my trees in order to not overwhelm the OP. My preferred account actually augments the tree analyses in terms of what my coauthors and I call rising. Doing this for German is crucial in part because of the prevalence of scrambling; the word order is quite free. In this case in particular, the object viel rises to attach higher in the tree. We develop this aspect of discontinuities in detail here: linguistics.fi/julkaisut/SKY2009/Gross_Osborne_NETTI.pdf. Jun 6, 2020 at 2:40

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.