5

I was reading about dependency grammars on Wikipedia, and then, following up on the term "(non-)projectivity", was lead to the page about discontinuity. Now, the concept is quite easy to understand, but what I don't get is how

[a]ll theories must have a means of addressing discontinuities.

Now, I am a native German speaker (and not a linguist by training), so all kinds of scrambling words and phrases, with non-projective (naive?) parses, seem to me like a perfectly natural thing (if a language allows them). But it appears that there is a thrive for syntactic theories to avoid ascribing ascribing discontinuous parses (e.g., here is a whole section dedicated to how projectivity might be "violated", and this problem be "addressed"). I don't really get why this should be.

  • One reason I can think of is purely practical: some (automata-based?) parsers might have a harder time with non-projective structures.
  • Also, in a certain sense, continuous structures are more "aesthetic": the hierarchical and the linear sentence align.

But these are both quite extra-scientific rationalizations. I do not see a reason why a theory should not freely make use of non-projective structures to describe languages -- avoiding them is a purely topological constraint, as opposed to, say, requirements involving heads and government, for which there exist (IMHO) quite natural linguistic explanations.

So, what am I missing? Am I just misinterpreting things and "addressing discontinities" is not a major thing at all?

2
  • I don’t deal much with dependency grammars so forgive me if this is mistaken. But many discontinuous dependencies (as defined in dependency grammars) represent things that are accounted for via syntactic movement in constituency grammars. Many dependency grammars don’t use movement; they prefer to generate everything “in situ”. So if you can’t use movement, your theory needs some other way of indicating that two distant elements share an underlying affinity of some sort.
    – Khove
    Jun 16 at 15:21
  • You're right, the reasons are technical. (Context-free) phrase grammars admit only non-tangled trees so discontinuous phrases are somewhat of a problem.
    – Atamiri
    Jun 22 at 11:14
3

Addressing discontinuities is a big deal in the field of theoretical syntax. It may be of less importance in natural language processing (NLP) because one is more interested there in processing texts that are already given rather than generating new texts from scratch. In any case, my answer here pertains to theoretical syntax since I am a theoretician.

When one explores natural language syntax, one of the primary questions concerns word order. One is interested in identifying the principles and aspects of the grammar that allow one to predict which word orders are and are not possible. Projectivity is fundamental in this regard, since it allows one to posit a basis for beginning the exploration. The assumption is that projectivity violations are not normal, and if they occur, they occur for a reason. It is quite apparent in all of this that projectivity restricts the possible word orders in a given language substantially. The role it plays is particularly great in languages with strict word order such as English, and to a lesser extent in languages with freer word order such as German.

We can consider a German example, one with an embedded clause:

enter image description here

Projectivity is a straightforward (but only partial) means of accounting for the fact that a majority of the possible ways to scramble the words in this sentence result in ungrammaticality, e.g.

enter image description here

Many more of the potential scramblings are ungrammatical. All told, there are 36 (=8+7+6+5+4+3+2+1) possible permutations of the words in this sentence, but only two or three of them are actually grammatical. In addition to (1a), we also have:

enter image description here

The order in (1g) actually changes the speech act from a statement to a question, so it should probably not count as one of the possible permutations of the words in the starting sentence (1a). In any case, projectivity is the means by which a significant majority of the 33 or 34 possible permutations are predictably bad. Certainly, much more is needed beyond projectivity, but projectivity is the obvious basis that one can build on.

I do not think that the right way to understand the issue is in terms of theories of syntax making "use of non-projective structures", but rather theories of syntax must acknowledge the existence of non-projective structures. They should then endeavor to predict which non-projective structures can occur and why they can occur, and which non-projective structures do not occur and why they do not occur.

One of the primary goals of scientific endeavor is to understand the field of inquiry better. When it comes to the word order of natural languages, projectivity is an indispensable foundation for the exploration.

2
  • Ah, so the real question is really more about constraints of linearization, and projectivity just (empirically) turns out to be a frequently useful notion there ("this analysis would require a non-projective tree, so it's less likely, but maybe the language has those"), rather than being an end in itself ("our theory would analyse sentences as non-projective, so we must revise it until it doesn't")? Jun 21 at 8:38
  • I think your point is correct that some theories choose to address non-projective structures in terms of movement, or in terms of feature passing, or in terms of... Theories that do not assume movement, or feature passing, or what have you must still address those non-projective structures, so the challenge is present in any case. Jun 21 at 14:11

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

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