Complex Topic-Comment Structures in HPSG

Based on Krifka (1992) and de Kuthy (2000), this paper develops an architecture for complex topic-comment structures in HPSG and applies it to predicate fronting in English with the goal of capturing the insights of Ward (1988) on this construction. We argue that predicate fronting is a distributed constructional form consisting of an auxiliary occurring in a predicate preposing phrase. The use of predicate preposing is a function of a combination of simultaneous constraints on its theme structure, its background-focus distribution, and its presuppositional structure. It is shown that these constraints can be made explicit within the HPSG architecture developed here. 1 Non-canonical Syntactic Constructions Höhle (1982) has argued that non-canonical syntactic constructions in German typically have fewer information structural options compared to canonical sentence patterns. The same has been argued for English. Ward (1988) concludes that the preposing constructions in (1)-(2) require the (meaning of the) preposed constituent to be a backward looking center. Similarly, Birner (1996) shows that inversion constructions like (3) are felicitous only if the preposed constituent is at least as discourse-familiar as the postposed NP: (1) One of these rugs Chambers delivered TO HARRY DEXTER WHITE. (2) (It was necessary to pass the exam and) pass I DID. (3) On the desk was A BIG LAMP. There is a generalization that cuts across these English constructions and others:1 in their prototypical use 1. the italicized constituent is the leftmost constituent of its predicate-argument complex, and 2. it is followed by another constituent of the same predicate-argument complex which is prosodically more prominent than it; 3. each sentence is more “about” the meaning of the italicized constituent than the meaning of the constituent in small caps (backgrounded, contrastive topic). I would like to thank Dorothee Beermann, Betty Birner, Regine Eckardt, Lars Hellan, Kordula de Kuthy, Detmar Meurers, Ivan Sag, Manfred Sailer, Gautam Sengupta, the members of the CoGETI research network and the audience at HPSG 2007 for discussions and comments at various stages during the development of the theory presented in this article. The responsibility for all errors remains with me. The generalization extends to German as well. 307 Previous work on information structure in HPSG (e.g. Vallduvi (1992)) has not derived this generalization. The work reported here is part of a larger research project which aims at developing a theoretical architecture that makes it possible to state this generalization in HPSG while also accounting for the properties specific to each non-canonical sentence pattern. The present article will only deal with a small portion of this subject matter, namely predicate preposing. 2 Case Study: Predicate Preposing Ward (1988) provides the following attested examples of predicate topicalization sentences:2 (4) As members of the Gray Panthers committee, WE WENT TO CANADA TO LEARN, and learn we did. [Philadelphia Inquirer, 6/16/85] (5) THE KING HAS INSTRUCTED ME TO BE BRIEF, and since I am His Majesty’s loyal subject, brief I will be. [A Man for All Seasons, Messenger] He arrives at the following conclusion concerning the felitous use of this sentence form: Ward’s Generalization Ward (1988) Predicate preposing is associated with the function of proposition affirmation. Proposition affirmation serves to affirm a proposition explicitly evoked in the discourse. The contrast in (6) serves to illustrate this analysis. (6c) cannot felicitously follow (6a), since the proposition affirmed by (6c), namely that I have enough money is not explicitly introduced into the discourse by (6a). That there is nothing wrong with this sequence of meanings in principle is shown by (6b) which can felicitously follow (6a). The difference is that unlike predicate preposing the emphatic dosupport construction (the verum focus of Höhle (1992)) does not require the proposition it affirms to have been explicitly evoked in the previous discourse: (6) a. I want to buy a car. b. And I DO have enough money. c. # And have enough money I DO. As predicted by Ward’s assumptions, if (6c) is put into a discourse context where the affirmed proposition has been introduced explicity, its use becomes felicitous: (7) They said I WOULDN’T HAVE ENOUGH MONEY TO BUY A CAR, but have enough money I DO. Most of the examples in this article that relate to predicate preposing are taken from Ward (1988), by far the most careful and sophisticated study of the construction. 308 3 Towards an Architecture for Information Structure in HPSG The theory of the relationship between syntax, semantics, and information structure developed below is based on the assumption that there are dependencies and interrelations between meanings and context that can only be expressed by a semantic representation language that makes reference to objects of specific semantic types, in particular properties (or their extensions). Moreover, it should be possible to impose discourse-anaphoric requirements on semantic pieces of constructions (and likely also rhetorical relations). A natural choice for this purpose is Discourse Representation Theory (Kamp and Reyle (1993)), in particular Lambda-DRT, because the latter is typed. Another important question that arises concerns the degree of articulation of the information structure. Krifka (1992) and Jacobs (2001) draw a four-way distinction between topic-comment and background-focus. de Kuthy (2000), in essence following Vallduvi (1992), distinguishes between background and focus and adds a (contrastive) topic in the sense of Büring (1997). For the purposes of this paper, the three-way distinction appears to be sufficient and I will consequently adopt it. Borrowing from the Prague School, Halliday (1967) develops the concepts theme which for him is the starting point of an utterance, its leftmost constituent. A related concept is proposed in Jacobs (2001): semantic subjecthood is one dimension in Jacobs’ multidimensional conception of topichood. According to Jacobs, the semantic subject of a clause is the highest term that specifies a variable in the meaning of the clause’s main predicate. As a consequence of the syntax-semantics interface, a sentence-initial constituent will frequently (but not always) contribute the semantic subject to the clause’s logical form. I will adopt Jacobs’ idea of theme3 as a configurational notion in logical form and even generalize it to the case where a predicate itself is topical in the sense under discussion. Overall, then, the architecture that is developed in this paper, consists on the one hand of the information structural triad background-focus-(contrastive) topic and on the other hand of the notion of theme. I believe that these two dimensions of information structure have different functions in the system of choices that a natural language grammar represents. This is stated in the following hypothesis. Hypothesis Syntactic non-canonicality is strongly associated with the choice of theme. On the other hand, prosody is more concerned with the information structural triad of background-focus-(contrastive)topic. Of course, elements which appear in syntactically non-canonical positions may also be prosodically prominent, so that the two concepts will often interact. Crucial I prefer Halliday’s name theme to Jacobs’ own name semantic subject for Jacobs’ concept. 309 evidence for the relevance of the notion theme comes from inversion constructions. Birner (1996) has shown that this sentence type serves an information packaging function in the sense of Chafe (1976). The following data show, however, that this function is independent of the triad background-focus-topic and needs to be characterized in some other fashion. In (8), the initial PP is in the background, given the context question: (8) a. Witness, when you walked into the office, what was on the desk? b. [bg On the desk was] [foc a KNIFE]. It is also possible for the inverted PP to be a contrastive topic: (9) a. Witness, you told us that was on the shelf, but what was on the desk? b. [top On the DESK] [bg was] [foc a KNIFE]. And, finally, inversion sentences can be all-focus, as is shown by (10): (10) a. Witness, when you walked into the office, what did you see? b. [foc On the desk was a KNIFE]. Discourses like the last one thus show that the preverbal and the postverbal constituents of inversion sentences can be in focus at the same time. Yet, even those sentences are felt to be more about the meaning of the initial PP than the meaning of the final NP. I would like to argue that what underlies this intuition is that inversion sentences are characterized by the following combination of information structural constraints: (11) 1. The preverbal constituent of an inversion is the theme (in Jacobs’ sense). 2. The postveral constituent of an inversion is part of the focus. Furthermore, I postulate the following preference principles (which could be seen in terms of harmonic alignment in Optimality Theory): (12) 1. Preferably, themes are unfocused. 2. Preferably, themes are discourse-familiar. This combination of assumptions derives the observation in Birner (1996) that the initial constituent in inversions prefers to be discourse-familiar over being discourse-new by a ratio of about 10:1. Assuming that information foci typically are discourse-new, this is compatible with Birner’s finding that the ratio for the postverbal NP in inversions is practically the reverse. We anticipate that it will be useful to have a notion of relative aboutness that is more general than that of a theme, e.g. in order to capture the typical informationstructural differences between the two objects in the double object construction of English discussed in detail by Bresnan in recent years (e.g. Bresnan et al. (2007)) and the effects typically associated with scrambling in languages like German (e.g. Webelhuth (1992), Haider and Rosengren (1998)). To this end, we define a relation more thematic than in terms of logical form configurations, as follows:4 The symbol ⊳ represents the relation that holds between two LF terms iff the first one is a (not necessarily proper) subconstituent of the second one. 310 (13) Definition of ”more thematic than” (≪) α ≪ β in LF Λ =df ∃γ ⊳ Λ such that 1. γ =   appl FUNC β’ ARG α’ 