Question/Answer Congruence and the Semantics of wh-Phrases∗

This paper is about the semantics of wh-phrases. It is argued that wh-phrases should not be analyzed as indefinites as, for example, Karttunen (1977) and many others have done, but as functional expressions with an indefinite core —their function being to restrict possible focus/background structures in direct or congruent answers. This will be argued for on the basis of observations made with respect to the distribution of term answers in well-formed question/answer sequences. This claim having been established, it will then be integrated into a variant of Schwarzschild’s (1999) information-theoretical approach to F-marking and accent placement, and —second— its consequences with respect to the focus/background structure of wh-questions will be outlined. 1 Direct Answers, Focus, Background Deletion Since the work of Hermann Paul (1920) and M.A.K. Halliday (1967) it has been commonly assumed that in well-formed, i.e., congruent, question/answer sequences (Q/Asequences) there is a systematic correlation between the wh-question Q and the focus/background structure (F/B-structure) of its direct sentential answers A, cf. (1). (1) If A is a direct/congruent answer to Q, then every constituent in A that corresponds to a wh-phrase in Q is focused (i.e., F-marked). This generalization can be illustrated by the Q/A-sequences given in (2). (2) a. Who likes John? [MARY]F likes John, ... b. Who likes whom? [MARY]F likes [JOHN]F, ... c. What did Sandra say? Sandra said [that Mary kissed [JOHN]F]F, ... In (2a) the constituent Mary corresponds to the wh-phrase who, and Mary must be focused; in (2b) Mary corresponds to who, John corresponds to whom, and both must be focused. Given that the generalization in (1) is in fact basically correct, then (2c) shows that the property of being focused does not coincide with the property of being accented in a strict sense, but that a focused and accented constituent may license an abstract focus (F-marking) on a larger constituent containing it. Dynamically speaking, the focus on John (the ‘focus exponent’) in (2c) ‘projects up to the that-clause’ in a way to be specified. ∗ I would like to thank Marga Reis, Manfred Krifka, and two anonymous reviewers of “Theoretical Linguistics” for valuable comments and suggestions. 1 As is usual, pitch accents are indicated by capitals. It should be emphasized that the generalization given in (1) does not entail that every focused constituent in a direct answer A needs to correspond to a wh-phrase in the respective wh-question Q . Actually, such a claim would be far too strong, for one always has to reckon with the presence of so-called ‘contrastive topics,’ cf. (3). (3) a. Whom do John and Mary like? b. [MARY]F likes [JOHN]F and [JOHN]F likes [SANDRA]F. For reasons of space, however, the possibility of contrastive topics will be almost completely ignored in the remainder of this paper. As presented here, the generalization in (1) is intended in the first instance as a generalization about sentential answers. Typically, however, questions are not answered by sentential answers but by ‘short’ or so-called ‘term answers,’ cf. (4) and (5). (4) a. Who likes John? Mary. b. Who likes whom? Mary, John; ... c. What did Sandra say? That Mary kissed John. (5) a. Whom do John and Mary like? b. Mary, John and John, Sandra. This immediately raises the question of whether, and —if so— in what way, sentential answers and term answers are related to each other. Apart from the obvious parallel between the F/B-structures of sentential answers in (2) and (3) and the term sequences in (4) and (5), there are good reasons to assume that the latter are derived from the former by some kind of elliptical process. To mention just two arguments, term answers and the respective wh-phrases must agree in case, cf. (6), and term answers may occur in the form of reciprocals, cf. (7). Both phenomena, however, are known to be strictly local, confined more or less to the minimal clause they are contained in. (6) Wen traf Hans? *Ein Mann. / *Eines Mannes. / *Einem Mann. / Einen Mann. Who met Hans? *[A man]-nom / *[A man]-gen / *[A man]-dat / [A man]-acc ‘Who did Hans meet? A man.’ (7) Wem vertrauen Schröder und Blair? Einander. Whom trust Schröder and Blair? Each other. ‘Who do Schröder and Blair trust? Each other.’ The way term answers are derived from sentential ones seems to be quite straightforward: starting from a well-formed sentential answer everything is phonologically reduced (p-reduced) that is not embedded in an F-marked node. Thus, this kind of elliptical process has to be conceived of as an instance of background deletion, and can be stated in a maximally theory neutral (and descriptive) manner as indicated in (8). 2 For further discussion, cf. e.g. Büring (1997), Krifka (1998), Reich (2001). 3 For further evidence, cf. e.g. Schwabe (1994), Reich (2001). 4 In the following, I will always switch to German data if the point to be made can be better illustrated using German examples, or if the data is rather subtle. (8) Background deletion in Q/A-sequences (optional) Let 〈Q, A〉 be a well-formed Q/A-sequence and let the F/B-structure of sentential A be of the form α0 [β0]F α1 [β1]F α2 ... [βn -1]F αn (where n ≥ 1, αi, 0 ≤ i ≤ n, possibly null), then p-reduce αi for 0 ≤ i ≤ n: α0 [β0]F α1 [β1]F α2 ... [βn -1]F αn. As recent research on ellipsis has shown, background deletion plays a crucial role in presumably all kinds of elliptical processes, and may thus be considered as a general strategy underlying elliptical phenomena in general. Typically, this process is further restricted by additional syntactic and/or semantic requirements like, for example, ‘directionality requirements’ in RNR-constructions (cf. e.g. Klein 1993, Hartmann 1999) or ‘correspondence requirements’ in VP-ellipsis phenomena (cf. e.g. Fiengo & May 1994, Merchant 1999). However, apart from the implemented maximality condition, background deletion in Q/A-sequences seems to be fairly —though not completely— unrestricted (cf. Kuno 1982). 2 The Problem Keeping this in mind, consider the following discourse (cf. Schwarzschild 1999:161). (9) (John drove Mary’s red conVERtible.) a. What did he drive before that? b. He drove her [BLUE]F convertible. As I will show below in some detail, ‘standard’ projection theories on F-marking like, for example, that in Selkirk (1996), as well as information-theoretical approaches like that developed in Schwarzschild (1999), predict —first— that the prenominal adjective blue in (9b) is F-marked, and —second— that no other constituent is. However, given that the assumptions about the derivation of term answers made above are basically 5 Cf. Rooth (1992b), Tancredi (1992), Klein (1993), Romero (1998), and Schwabe & Zhang (2000) among others. 6 There are at least two restrictions on background deletion in Q/A-sequences that should be mentioned here. First, term answers of category VP need to contain the uninflected part of the verbal predicate, cf. (ia) vs. (ib); as a consequence, term answers of category VP are confined to the perfective forms of tense in German. (i) a. Was machte Peter? *Peter kauftei [Anna ein FAHRrad ti]F What did Peter? *Peter boughti [Anna a bike ti]F ‘What did Peter do? Peter bought a bike for Anna.’ b. Was hat Peter gemacht? Peter hat [Anna ein FAHRrad gekauft]F What has Peter done? Peter has [Anna a bike bought]F ‘What has Peter done? Peter has bought a bike for Anna.’ Second, as an example shows to which my attention was drawn by one of the anonymous reviewers, there are certain cases in which a constituent can be phonologically reduced without all the non-focused parts of the answer having to be phonologically reduced, cf. (ii). (ii) What did John drive to Martha’s funeral? a. He drove [a CAbrio]F to Martha’s funeral. b. He drove [a CAbrio]F. c. [A CAbrio]F. These cases, however, seem to be restricted to a certain class of adjuncts, and thus are not crucial for the argument presented here. correct, the F/B-structure of the answer in (9b) together with the generalization in (8) predict that (10b) is a well-formed term answer in the context of (10a). But in fact it is not. The correct term answer is that given in (10c) —it is the whole constituent corresponding to the wh-phrase. (10) a. What did he drive before that? b. *[BLUE]F. c. Her [BLUE]F convertible. Actually, it turns out that this contrast is not restricted to the nominal domain, but can be observed with respect to the sentential and the verbal domain, too, cf. (11) and (12). (11) (John said that he likes to drive conVERtibles.) a. What else did he say? b. *[OLDtimers]F. c. That he likes to drive [OLDtimers]F. (12) (Peter hat Anna ein CAbrio gekauft ‘Peter bought a conVERtible for Anna’) a. Und was hat er sonst noch gemacht? And what has he else Part done ‘And what else did he do?’ b. Er hat [SANdra]F ein Cabrio gekauft. he has Sandra a convertible bought ‘He bought a convertible for SANdra.’ c. *SANdra. d. SANdra ein Cabrio gekauft. Again, it is the constituent corresponding to the wh-phrase that constitutes the term answer and not the constituent in focus. Thus, this data together with the generalization about the derivation of term answers stated in (8) strongly suggests that it is not only the prenominal adjective that is F-marked, but in fact the whole constituent corresponding to the wh-phrase. Moreover, it suggests that this effect is due to some property of the wh-phrases involved. This is what I will call the functional character of wh-phrases. The major claim I wish to argue for in this paper is that this property has to be located in the semantics of wh-phrases. Obviously, it may be immediately objected that this data merely shows that the above assumptions behind the derivation of term answers are too simplistic and must be revised or restricted in one way or another. The crucial point, however, is that there would seem to be no straightforward way of doing so without merely stating the facts; and even if someone came up with a proposal, (8) still seems to be the null hypothesis 7 Following a different line of argumentation, Drubig (1994) draws similar (although not identical) conclusions with respect to the F/B-structure of so-called ‘negative contrastive constructions’ like not ..., but ... in English or nicht ..., sondern ... in German. For further discussion, cf. Reich (2001). 8 Examples like (10) suggest that the derivation of term answers has to respect the ‘minimal functional complex’ containing the focus. This restriction may in fact lead to correct results in examples like (10), but it won’t do so in more complex cases like (11) —cf. *Her BLUE convertible. vs. That he likes to drive her BLUE convertible.— or in cases where the term answer is constituted by a lexical projection, cf. (12). and is, therefore, the theoretically preferred option. Hence, I will assume from now on that the constituents corresponding to a wh-phrase are in fact F-marked. Then, obviously, the question emerges, why ‘standard approaches’ to F-marking do not permit this F-marker, and whether there is any straightforward and natural way of modifying (one of) them in such a way that they do. 2.1 The Problem within Projection Approaches First of all, let’s have a look at so-called ‘projection theories,’ the most prominent representative of which is presumably Selkirk (1984, 1996). Selkirk (1996) assumes that F-marking is controlled by the set of rules given in (13) and (14). (13) Basic Focus Rule An accented word is F-marked. (14) Focus Projection a. F-marking of the head of a phrase licenses the F-marking of the phrase. b. F-marking of an internal argument of a head licenses the F-marking of the head. Now reconsider Schwarzschild’s example (9) in the light of (13) and (14). The prenominal adjective blue is accented; hence it is F-marked by the Basic Focus Rule (13). However, being an adjunct, it cannot license F-marking of the non-accented head of the DP, cf. (14b). Since there is no other candidate that could license F-marking of the head, it must be concluded that the head is not F-marked. But since the head is not F-marked, F-marking of the DP isn’t licensed either. Is there a straightforward way of modifying this approach? As far as I can see, the answer is no. The crucial problem is that any mechanism that allows F-markers to project from prenominal adjectives to the DP containing them cannot prevent the Fmarker from projecting to VP if the DP is an internal argument of the verbal head; i.e., the Q/A-sequence in (15) would be predicted to be well formed in general, especially in an out-of-the-blue utterance. (15) a. What did John do? b. *He [drove Mary’s [RED] convertible]. 2.2 The Problem within Information-Theoretical Approaches The other prominent approach, which can be traced back to the work of Arnim von Stechow (cf. von Stechow 1981) but became well known with the work of Schwarzschild (1999), assumes a more direct connection between the information-theoretical notion of being ‘given’ and F-marking. Schwarzschild (1999) provides us with two basic information-theoretical principles, the first stating that non-F-marked constituents must be GIVEN, cf. (16), the second taking the form of an instruction to F-mark as little as possible, cf. (17). 9 To my knowledge, the only approach to F-marking and accent placement that, in principle, allows for F-structures of the required kind is the one proposed in Jacobs (1988, 1991). However, since this approach crucially relies on the assumption that syntactic F-marking is dependent on semantic focusing, it is certainly not an option in a framework in which syntax is assumed to precede semantic interpretation, like, e.g., the Chomskyan generative framework presupposed here. (16) GIVENness If a constituent is not F-marked, it must be GIVEN. (17) AVOIDF Do not F-mark. Contrary to Selkirk’s conception, the existence of an F-marker is not due to a constituent being accented; rather accenting is a consequence of F-marking. This is ensured by a constraint called FOC, cf. (18). The distinction between Foc-marked and F-marked phrases, however, is not important for our purposes, since in all the relevant examples discussed so far each F-marked constituent is at the same time a Foc-marked constituent. (18) FOC A Foc-marked phrase contains an accent. There are two more things to say. First, it must be determined precisely what it means for a constituent to be GIVEN, cf. (19). (19) Definition of GIVEN (partial, informal version) a. An utterance U counts as GIVEN iff it has a salient antecedent A and, modulo existential type-shifting, A entails the existential F-closure of U. b. Existential F-closure of U := the result of replacing F-marked phrases in U with variables and existentially closing the result, modulo existential type-shifting. Second, it must be emphasized that the constraints GIVENness, AVOIDF and FOC are organized in an optimality theoretical manner, i.e., one is allowed to violate constraints according to the partial order given in (20). (20) Ranking ‘>>’ (‘overrules’) of constraints a. GIVENness >> AVOIDF b. FOC >> AVOIDF Having introduced the most basic assumptions of Schwarzschild’s approach to F-marking, I can now show why in the convertible example (9) the DP her [BLUE]F convertible must not be F-marked: as Schwarzschild (1999:161) himself shows, the DP in question is GIVEN in the sense specified in (19): existential type-shifting of the DP results in the propositional expression ∃P[P(her [BLUE]F convertible)]; F-closure, in turn, results in the propositional expression ∃X∃P[P(her X convertible)] that is entailed by previous discourse, cf. (21). Consequently, F-marking of the DP is optional; and since F-marking is optional, it is ruled out by AVOIDF. Following exactly the same line of reasoning, it can be shown that VP and S do not carry an F-marker either, cf. (21b,c). (21) John drove Mary’s red convertible ENTAILS a. ∃X∃P[P(her X convertible)], therefore DP is GIVEN. b. ∃X∃y[y drove her X convertible)], therefore VP is GIVEN. c. ∃X[He drove her X convertible], therefore S is GIVEN. Again, the question to be answered is whether there is a straightforward way to modify this approach. This time the answer is ‘yes, in principle.’ The only reason why the DP must not be F-marked is a violation of AVOIDF. However, as is clear from (20), the constraint AVOIDF can be violated if there is another constraint that is ranked higher. Since neither GIVENness nor FOC will force F-marking on the DP, there must exist another, independently needed constraint allowing for violation of AVOIDF. In the following two sections, it will be argued that there is in fact good evidence for the existence of a constraint with this property, a constraint allowing for the presence of (focus-sensitive) rhetorical relations. 3 A Slightly Modified Hamblin Approach: Functional wh-Phrases 3.1 Questions and Answers Since it will turn out that one of the rhetorical relations to be licensed by this constraint is the Q/A-relation, the semantics of focus and the semantics of wh-interrogatives I am assuming need to be outlined. To this end, consider the well-formed Q/A-sequence in (22). (22) a. What did John drive? b. John drove [Mary’s red conVERtible]F. Without offering an argument, I will adopt the structured meaning approach to F/Bstructures as developed in von Stechow (1981) and Cresswell & von Stechow (1982), i.e., the F/B-structure in (22b), repeated as (23a), is represented as a structured proposition consisting of the focus ‘Mary’s red convertible’ and the property ‘being driven by John,’ cf. (23b). (23) a. John drove [Mary’s red conVERtible]F b. 〈Mary’s red convertible, λx.John drove x〉 Following Hamblin’s (1973) dictum that ‘‘a question sets up a choice-situation between a set of propositions, namely, those propositions that count as answers to it,’’ and taking the insight into account that F/B-structures are at the heart of the Q/A-relation, it is perfectly straightforward to construe a question like (22a), repeated here as (24a), as denoting a set of structured propositions, cf. (24b), and more precisely (24c). (24) a. What did John drive? b. {〈Mary’s red convertible, λx.John drove x〉, 10 Note that the assumption that the whole DP is F-marked does not influence the realization of the accent within the DP. This is simply because this assumption results in one Foc-phrase being embedded within another. 11 Cf. e.g. von Stechow (1991) for relevant discussion. 12 Being ‘hybrid’ in nature, i.e., a combination of the structured meaning approach to F/B-structures and the propositional approach to the semantics of (wh-)questions, this approach inherits both the structural information encoded by the structured meaning approach to F/B-structures (or its relative, the categorial approach to (wh-)questions) and —given a suitable type-system as developed in Reich (2001)— the typeuniformity of the propositional approach. Presupposing a theory of propositional attitudes as developed e.g. in Cresswell & von Stechow (1982), this approach obviously allows for a theory of question embeddings in the spirit of Karttunen (1977). In the following, the variable p is intended to range over structured propositions (as well as unstructured ones). 〈Peter’s Porsche, λx.John drove x〉, ... } c. λp∃x[thing’(x) & p = 〈x, λy.John drove y〉] Thus, wh-interrogatives are still taken to denote sets of possible answers; the notion of being a possible answer, however, has now been relativized to possible F/B-structures. 3.2 Wh-Phrases as Functional Expressions Of course, the propositions contained in the denotation of a wh-interrogative must be structured independently. This is precisely what I take to be the task of wh-phrases. Concretely, I propose to analyze wh-phrases not as a (type-shifted) variant of indefinites like something, cf. (25a), but as primarily functional expressions with an indefinite core that shape the F/B-structure of possible answers, cf. (25b). (25) a. ‘Traditional’: (what)’ = λQλp∃x[thing’(x) & Q(x)(p)] b. Proposal: (what)’ = λQλp∃P∃x[thing’(x) & Q(P) & p = 〈x, P〉] Given this, the well-formedness condition imposed on Q/A-sequences can be reduced to the simplest condition one can think of, namely the ∈-relation, cf. (26). (26) A is a direct/congruent answer to Q iff [ [A] ] ∈ [ [Q] ]. As will be clear from the discussion in section 4, the generalization given in (1) is a direct consequence of the well-formedness condition (26) imposed on Q/A-sequences. As far as the logical form and the interpretation of wh-interrogatives are concerned, the functional view on wh-phrases is in essence consistent with the ‘traditional analysis’ of wh-interrogatives within the generative framework (cf. e.g. von Stechow 1993), i.e., a wh-interrogative like (27a) is analyzed on the level of logical form as indicated in (27b). (27) a. What did John drive? b. CP