Sentences like the following occur frequently as examples for cross-serial dependencies in Dutch.

Ik dacht dat Jan Piet Marie zag helpen lezen.

Often the dependencies are depicted only as arcs between Jan and zag, Piet and helpen, and Marie and lezen. I found the following non-projective dependency structure for the above sentence in http://www.mitpressjournals.org/doi/pdf/10.1162/COLI_a_00125:

enter image description here

This figure does not contain syntactic functions at the edges.

(In my conception the figure depicts a graph (in the mathematical sense) whose nodes are word forms. An arrow-shaped edge indicates a dependency between two words, where the direction of the arrow is from head to dependent. This edge can be equipped with a label specifying the type of argument or adjunct (direct object, subject, etc.) the descendent is for its head; which is not the case in the above figure. A dashed edge indicates the position of the word in the sentence.)

In some publications I found constituent structures for this sentence, where Piet and Marie are attached as direct object in a sibling position to the verb complements helpen and lezen, respectively. However, if Marie is attached below lezen, then labeling this edge with direct object does not seem plausible to me (as linguistic layman).

Is the depicted dependency structure plausible at all?

What would be appropriate edge labels?

  • I don't know what "edge" means in your question. It is common to think of verbs as syntactic functions, and then using usual notation, the diagram you give could be written saw( Jan, help( Piet, read( Marie))), so if "edge" means where the form of a function starts, diagrammatically a left paren, then there are syntactic functions at the three "edges".
    – Greg Lee
    Aug 5, 2015 at 1:41
  • By “edge” I mean an arrow (a dependency) between two words, and by “edge label” a description of this dependency similarly as the green labels in this graphic.
    – Kilian
    Aug 5, 2015 at 7:52
  • Well, then, edge means "is an argument of", and that is expressed in the diagram. If you want something to better express your feelings about the examples, perhaps you could use different colored pens.
    – Greg Lee
    Aug 5, 2015 at 16:30

2 Answers 2


To the first question "Is the depicted dependency structure plausible at all?", one would have to answer: "That depends on the type of dependency grammar one employs". Kuhlmann, the author of the cited paper, uses a ("mildly") non-projective dependency grammar.
"Non-projective" (aka "discontinuous") is any (sub-)tree structure when a dependency edge (the arrows that connects nodes as in the picture above) crosses over a projection edge (the vertical dotted edge in picture above). This property has been discussed at length within the dependency grammar literature. Please see the Wikipedia page on this topic [https://en.wikipedia.org/wiki/Discontinuity_%28linguistics%29].
Usually non-projectivity (aka discontinuity) arises when the dependency grammar views dependency basically as "government". In the next example

 Wem  hat er geholfen?
 whom has he helped 

the question word Wem is governed by geholfen, rather than by hat. A government-based dependency grammar will then connect Wem and "geholfen*, which results in a non-projective structure because the dependency edge connecting Wem and geholfen will cross the projection edge of hat (and that of er).
A number of dependency grammarians regard the limitation on government as too restrictive, and they view the apparent non-projective dependent node (Wem in the above example) as somehow displaced. My own take on this phenomenon can be found here. That paper also contains references to other authors within the dependency grammar framework.

In a projective dependency grammar, such as the one I prefer, the depicted structure is not possible. Rather the nodes Piet and Marie would attach, i.e. be connected via a dependency edge, to the highest verb node, namely zag. Such a constellation is licit under the following condition:

A node A may be immediately dependent to another B, even though B doesn't govern A, if B dominates the node C, the governor of A.

In my example, Wem = A, hat = B, and geholfen = C. Since B hat dominates geholfen, Wem may rise in order to connect to hat. There are a number of constructions that elicit rising, but they have different syntactic properties.
[I'd also like to emphasize that I believe that the depicted tree structure is inaccurate even on a purely non-projective approach. Piet is governed by zag, in a specific construction that requires an infinitive verb, here helpen. Hence Piet is not a dependent of helpen. The same holds for Marie, which must be a dependent of helpen, rather than of lezen. In non-projective systems, semantics sometimes occludes the appropriate syntactic relationships.]

The answer to the second question, i.e. "What would be appropriate edge labels?", must first decide on whether the apparatus is projective or non-projective. If it is non-projective, standard (i.e. government) labels are in order. In the Dutch example, Marie would be the indirect object of helpen, and Piet the direct object of zag (rather than the subject of helpen).
A projective account can choose whether it wants the labels to reflect the actual dependency connections, or the underlying government relationships (which for the most part are identical; only in examples as the ones used here does a difference appear).
Yet another aspect is whether the grammar is mono- or multistratal. If the grammar is multistratal, i.e. employs more than one level of description, then labels may be transferred from one level to the next higher one. Or such a grammar may choose to use different sets of labels depending on the strata it uses at a specific moment.
If the depicted tree structure were correct (which it isn't as I argued above), the labels for all three nominals, i.e. Jan, Piet, and Marie, would have to be [subject], and obviously that cannot be true because the verbs helpen and lezen are infinitive forms, and infinitives cannot command subjects.

I hope that helps a bit.

  • Thank you for the comprehensive answer. For now I assume a non-projective structure. (I have not read your paper yet and currently lack the linguistic background to enter a discussion on which kind of dependency model is better.)
    – Kilian
    Aug 7, 2015 at 8:53
  • I understand the argument that helpen and lezen are infinite forms and therefore cannot have a subject. Still, Piet and Marie are in a certain sense the actors in some process called helpen and lezen. Is this a level of semantics? Did you address this point with semantics sometimes occludes the appropriate syntactic relationships?
    – Kilian
    Aug 7, 2015 at 8:59
  • 1
    @Kilian Yes, semantic relationships (Piet and Marie indeed have a semantic relationship to the verbs helpen and lezen respectively) sometimes cause linguists to assume syntactic relationships (dependency is a syntactic relationship) when such an assumption is not warranted by the surface data. Non-projective approaches within dependency grammar are often employed by computer linguists. In the theories these linguists use, semantics is central in order to provide the conversion from one language to another one. My doubts concerning such approaches are too numerous to be explained here. Aug 7, 2015 at 12:02

"Ik dacht dat Jan Piet Marie zag helpen lezen." This is not a correct sentence in Dutch. Jan Piet Marie seems to be one name.

This is Dutch grammar.

Ik dacht dat ik zag Jan Piet, Marie helpen lezen.

  • 2
    You should make it clear that you are not a native speaker. In all your answers on this site so far, you have mistakes in Dutch.
    – Keelan
    Aug 22, 2018 at 5:42
  • Be specific, point out the mistakes. Do you know how many mistakes a normal Dutch makes on a daily basis? There are even mistakes in 200-year- old names. Point out my mistakes. Aug 22, 2018 at 14:13

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.