In phrase structure grammars, discourse functions (topic, focus) have structural positions (cf. topicalization in English, right-edge focus in Russian, clause-initial focus in Welsh, preverbal focus in Hungarian). What do dependency grammars employ to capture information structure? There are no nonterminals so one can't have an FP, for example, or a TopP.


One can't account for information structure at the level of surface syntax. Consider the sentence илсьірбоит (I'm showing it to her; Abkhaz). It has three topical arguments (only topics can be pro-dropped) and since a sentence without focus wouldn't be felicitous, the predicate is focal. But the syntax tree consists of only one node. It's useless. The LFG people suggest to ignore the categorial structure and introduce discourse functions into the functional structure. But it was Tracy Holloway King I think who pointed out that this approach doesn't work for subconstituent focus. It's easy to see that any dependency grammar would have this problem. The solution within LFG was to add a new layer of representation for discourse functions to the theory.

In FGD Sgall et al. reorder their tectogrammatical trees with respect to information structure. But at the same time they require that the trees be projective. I don't think that their solution is particularly elegant but if they weren't able to come up with a nice solution, then there's probably none within the framework.

One possible "solution" is to say that information structure isn't part of syntax sensu stricto and capture it at the level of pragmatics. This is what Jerry Hobbs did in his framework and it makes sense, but it also moves us from linguistics to computational logic and/or automated reasoning.

  • I don't think what's dropped in your Russian example qualify as topics. The dropped pronouns are old information, and old information and topics do not necessarily overlap. Consider in this regard that according to your reasoning, clauses would often contain more than one topic. In other words, your analysis is claiming that in a sentence such as "He gave her it", three topics are present: "He", "her", and "it". Is that right? – Tim Osborne Mar 20 '15 at 7:13
  • "One can't account for information structure at the level of surface syntax." I disagree. Topological theories (Duchier and Debusmann, Gerdes and Kahane) actually work very well. – Thomas Gross Mar 20 '15 at 8:11
  • @TimOsborne It's Abkhaz and it's topic because Abkhaz has a rule that pro-dropped arguments are always topical. It's language specific. Old information can be focal it it's contrastive. BTW clauses often contain more than one topical phrase/argument. The terminology usually is that sentences are divided into topic and focus and in the example above the topic would consist of three nominal arguments. – Atamiri Mar 20 '15 at 14:58
  • @ThomasGross It may work in some languages, but in a polysynthetic language like Abkhaz one gets into difficulties (in Debusmanns theory the topological tree would have only one node, right?). I'm not familiar with MTT so maybe someone from that camp could comment on the example. – Atamiri Mar 20 '15 at 15:02
  • Your use of terminology is debatable. You are using the term "topical" to mean old information. Sentences have one topic; my linguistics dictionaries agree about that. Crystal writes: "The topic of a sentence is the entity (person, thing, etc.) about which somehting is said, whereas the further statement made about this entity is the comment." Note that he does not write "...the entities about which...". – Tim Osborne Mar 21 '15 at 0:22

There are two types of answers to this question, depending upon whether a multistratal syntax (deep syntax and surface syntax) or monostratal syntax (surface syntax only) is assumed.

In monostratal syntax, information structure is encoded by linear order alone. This is the same for both grammar types, dependency and constituency. For instance, a topic can be construed as the constituent that appears at the left edge of an utterance.

Note that I have used the term "constituent" to describe the syntactic unit at the left edge of the utterance in a dependency grammar. For example:

 (1)   a. You still have much to learn.

       b. Much to learn you still have.

In a dependency grammar, complete subtrees are constituents. In both (1a) and (1b), the string much to learn is a constituent, i.e. a complete subtree. In (1b) it can be labeled the "topic". As a constituent, it has a concrete status in the dependency hierarchy. The following trees (taken from the article on topicalization in Wikipedia) illustrate this state of affairs in both a constituency and a dependency grammar:

enter image description here

These trees illustrate the extent to which notions such as topic and focus are going to receive the same sort of explanation in both dependency- and constituency-based systems. These notions are taken care of by linear order alone, and the distinction between dependency and constituency has no bearing on topics and foci.

On a multistratal approach, however, things become more complex, since dependency grammars that posit deep syntax also acknowledge information structure in deep syntax. The main dependency grammars that do this are Meaning-Text Theory (MTT) and Functional Generative Description (FGD). However,these DGs can nevertheless also assume that constituents (complete subtrees) are marked in deep syntax as topic or focus. In the mapping that occurs fron deep syntax to surface structure, a topic should then appear at the left periphery of the utterance.

Now if the question is more concerned with how notions of FocP and TopP in Chomskian syntax translate to dependency grammar, that question is perhaps more difficult to anwer in a coherent manner. The difficulty stems from the fact that modern Chomskian syntax assumes all syntactic branching to be binary. This can lead to a large number of various functional categories, all located above the lexical domain in the syntactic hierarchy. What one has to ask oneself, though, is whether these functional categories can exist without linear order. That is, one has to ask oneself whether one can have a topic that does not appear at the left edge of the utterance. If no such topics are possible, the notion of topic cannot be separated from linear order, so there is no reason to locate the notion of topic in the syntactic hierarchy. But if one can have a topic that is not at the left periphery of the utterance, then one has to ask oneself what the Chomskian syntax tree would look like that acknowledges such a topic. For instance, can a topic follow the finite verb? I think many syntacticians would answer this question with "no".

  • "information structure is encoded by linear order" That wouldn't work for languages that signal information structure morphologically. "constituents (complete subtrees) are marked in deep syntax as topic or focus" The information-structural status of a head might differ from that of its children. – Atamiri Mar 20 '15 at 6:22
  • OK, but it wouldn't need to mark it with linear order, because it marks it morphologically. The redundancy would be unnecessary. But the issue is not so clear. I'm thinking of the topic marker "-wa" in Japanese. The extent to which "-wa" actually marks a topic is debatable. – Tim Osborne Mar 20 '15 at 6:26
  • "For instance, can a topic follow the finite verb? I think many syntacticians would answer this question with "no"." Yes, it can, in many languages. – Atamiri Mar 20 '15 at 6:29
  • For instance? Examples? And are these topics marked morphologically? – Tim Osborne Mar 20 '15 at 6:31
  • Focalization in Welsh is a nice (configurational) example. Or (anti)topics in French. But it's possible in most (if not all) free word order languages. I don't think there's one syntactician who'd seriously believe that there's a language that doesn't allow for topics following verbs. – Atamiri Mar 20 '15 at 7:01

Chomsky proposed that linguistics should be explanatory rather than taxonomic, and mostly, modern linguists have agreed with him. Taxonomic linguistics endeavors to develop a framework of terminology that is sufficient to describe facts of natural language, as we encounter them. This is the sort of thing that preoccupied linguists in the midst of the 20th century, and led to, e.g., tagmemics, or stratificational grammar.

Under Chomsky's influence, we don't do that anymore. It appears to me, you are asking a taxonomical question. Not wishing to sound mean, but my answer is: Who cares?

Your Answer

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