0

My goal is to build semantic representation of Russian sentences, i.e. to extract verb predicates and fill in the actant words. The tool I have is some kind of a shallow syntactic parser which works with analytically built CFGs.

My question is, what are these predicates? Is there a typology or classification behind them? I need some theoretical ground for defining this predicate-argument structure for myself properly.

  • 4
    The question is mixing concepts from competing fields of grammar and syntax. The expression "actant words" suggests a dependency grammar approach to parsing, since the term "actant" is firmly anchored in Lucien Tesniere's theory. But the term "CFG" stems from early Chomskyan syntax, i.e. from early phrase structure grammar. Thus there is a mixing of theoretical assumptions that may be difficult to unify. I suggest commenting further on these notions. With more information, I might be able to make a suggestion. – Tim Osborne Aug 5 '14 at 10:12
  • 1
    Yeah, well, if you wanna build a computer program you're probly not gonna worry about theoretical consistency, as long as the program works. I would start by asking whether the OP's familiar with Wordnet and Framenet. Lists of Russian verbs is another matter altogether. – jlawler Aug 5 '14 at 16:51
  • @TimOsborne, actually, from the syntax perspective I am limited to 2 approaches (corresponding to 2 tools): either building dependency syntax using statistical parsing, or defining context-free grammar rules by myself and using them for extracting relations from text (I'm not aiming to produce full parses though). And I'm just asking for the very basic typology of predicates to help me select the most important cases in my texts and cover them. If it differs for these two approaches, please provide me with starting points for both:) – Igor Shalyminov Aug 5 '14 at 17:11
  • 1
    @jlawler, I need basic theoretical consistency mostly for myself, to know what I'm doing:) WordNet can be useful for building lexical semantic relations, while I'm interested more in building predicate-argument structures (you can think of my goal as of a very rough approximation of formal semantics). I'm not familiar with FrameNet (anyway, I don't know any Russian alternative to it:) ), but maybe if I can learn how to build semantic frames by myself, it can help me. Could you please suggest anything as a starting point? – Igor Shalyminov Aug 5 '14 at 17:21
  • There well may be Russian versions of or applications of Framenet; I'm not a Slavicist, but I know Russians that are passionately involved in CL/NLP at all levels, for all purposes. If you want verb lists, start with the semantactically categorized sets in Levin (which is limited to lexical verbs that govern cyclic alternation rules) – jlawler Aug 5 '14 at 19:19
1

I am not aware that anyone has produced a typology of predicates in the sense of semantic predicate-argument structures. Of course loads and loads is written about semantic predicate-argument structures, but any attempt to produce a typology of these predicates is going to be fraught with much difficulty, since determining exactly what should and should not count as a predicate is difficult. I myself have written about how predicates are manifest in syntax, so perhaps I can help by illustrating the difficulty facing the intended endeavor.

But first a couple of points of opinion: I suggest ejecting the "CFG" part of the question. Attempting to produce a typology of predicates based on a CFG backbone is going to be difficult and convoluted. Predicate-argument structures are much easier to deal with using a dependency-based model. I can back this claim up with my own research (if anyone is interested).

Assuming a dependency-based model of syntax and semantics, Igor Melʹc̆uk's works are a good place to start. Two of his most known works are listed here:

 Melʹc̆uk, Igor A. (1987). Dependency syntax : theory and practice. Albany: State University Press of New York. ISBN 978-0-88706-450-0. Retrieved 24 August 2012.

 Melʹc̆uk, I. (2003). Levels of dependency in linguistic description: Concepts and problems. In Ágel et al., 170-187.

Melʹc̆uk also has a more recent book that focuses on semantics. I haven't looked at it personally, but I imagine it is going to have a detailed account of semantic predicate-argument structures. Note as well that Melʹc̆uk is Russian. But also beware, Melʹc̆uk's works can be difficult to access for those who do not have a firm basis in the theory of grammar.

If a typology of lexical verbs is actually what you are interested in, many dictionaries provide sentence frames indicating how verbs are used. For instance, I have Wahrig's Woerterbuch der deutschen Sprache here in front of me now. It classifies verbs into about 100 different types according to their valency. Perhaps this is what you are actually interested in. If so, I imagine that there are dictionaries of Russian that also provide sentence frames indicating the valency of verbs. If the term "valency" does not call up clear associations, try reading this article here.

Now on to actual predicate-argument structures. A typology of predicates would have to be able to accommodate the following data:

 1a.  The house is new.

 1b.  the new house

The predicate-argument structure across these two utterances should be closely similar. Using simplified predicate-calculus-style representations, this is what one gets:

 1a'.  is new [the house]

 1b'.  new [the...house]

In both cases, one views (is) new as a predicate taking the house as its argument. Yet from a syntactic point of view, the two utterances are obviously quite different structurally. Thus if one is interested in producing a typology of predicates, the typology should accommodate this sort variation in syntactic structures.

A second example should further illustrate the challenge of establishing a typology of predicates:

 2a.  The people are laughing at him.

 2b.  The people laughing at him are mean.

Again using simplified predicate-calculus-style representations, one might assume the following predicate-argument analyses:

 2a'. are laughing [the people, at him]

 2b'. are mean [laughing [the people, at him]]

The thing to note in this case is the manner in which one predicate-argument structure serves as the argument of another predicate. In other words, the theory of predicate-argument structures that one employs should be able to recognize the embedding of predicate-argument structures.

There are many many further challenges facing any attempt to produce a typology of predicates. A third example:

 3a.  Fred showered.

 3b.  Fred took a shower.

These sentences are almost synonymous, which means they should have almost identical predicate-argument structures:

 3a'.  showered [Fred]

 3b'.  took a shower [Fred]

The challenge that such data present should be apparent. The theory of predicate-argument structures that one assumes should be in a position to recognize when the object is part of the predicate (as opposed to being an argument of the predicate). In this case, a shower is not an argument of took, but rather it forms the predicate with took.

In sum, I think the endeavor is facing major challenges. There is no consensus about what does and does not count as a predicate in theories of semantics and syntax. The difficulties sketched here are going to require a decision after every turn.

| improve this answer | |
  • Thank you for the comprehensive answer! Although I'd like to hear why dependency syntax is better for predicate semantics (this claim is language-independent, right?). – Igor Shalyminov Aug 6 '14 at 8:43
  • Yes, it is language independent. Part of the reason that dependency syntax is more suited to serve as a backbone for establishing a theory of predicate-argument structures stems for the directness of the syntactic links. Dependency syntax puts the subject argument on equal footing with object arguments; it thus rejects the binary division of the clause that most phrase structure assume. This positions dependency syntax much closer to predicate-argument structures from the get-go, since the subject is judged to be an argument just like the object. – Tim Osborne Aug 6 '14 at 9:03
  • There are a few dependency trees in the Wikipedia article on predicates (en.wikipedia.org/wiki/Predicate_%28grammar%29). You can get an idea about suitability by pondering these trees. – Tim Osborne Aug 6 '14 at 9:07
2

If you are going for a computational approach to predicates, I think you may wish to consider consulting VerbNet. It deals a lot with knowing just what arguments a (verb) predicate can take and also classifies verbs in many different semantically categories. I do not think there is a Russian form of VerbNet yet, but I think it is a relevant resource to your problem.

You may also wish to read about Beth Levin's work on verb classes. This could prove helpful in creating your own ontology of Russian (verb) predicates, but I suspect that such a resource may already exist or at least someone has worked on it. I know there has been work for an Arabic VerbNet.

| improve this answer | |

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.