Lasell College, 1844 Commonwealth Avenue, Newton, MA 02466, USA
Superiority phenomena in multiple WH-questions are analyzed as an effect of a range of sentencelevel but discourse-related contingencies centering around question focus and concomitant constraints on question structure. It is argued that multiple interrogation formations in English are restricted by constraints ensuring the informativeness of the question focus: These constraints interact with intonational and logophoric characteristics of the sentence to determine felicity of multiple interrogation structures. In particular, subject-object and argument-adjunct asymmetries in focus projection from a lexical item receiving primary stress translate into similar asymmetries in multiple interrogation structures. The proposed model accounts for systematic amelioration and degradation effects by invoking principled contingencies and by appeal to a contextual dependency spectrum allowing amelioration via targeted manipulation, such that question-internal structural adjustments affecting focus projection and question-external contextual priming can satisfy the informativeness requirements of multiple interrogation structures.
Keywords: WH-questions; Multiple interrogation; Logophoricity; Superiority; Focus; Discourse
Consider the following contrast:
(1) a. Who bought what?
b. What did who buy?
The contrast between (1a) and (1b) has been adduced frequently as a superiority effect: In a multiple interrogation with two possible candidates for WH-fronting, the structurally Journal of Pragmatics 37 (2005) 120378-2166/see front matter 2004 Elsevier B.V. All rights reserved.
* Tel.: ?-617-243-2224.
E-mail address: firstname.lastname@example.org (B. Kennedy).
superior WH-term is selected for WH-movement, and the structurally inferior WH-term remains in situ: in its base position.1 The paradigm (1a, b) and a range of other WH in situformations have been examined in the literature under different models. Of these, certainstructural accounts (e.g., Pesetsky, 1987) include a discourse-theoretic component; other explanations (e.g., Kuno and Takami, 1993) distinguish functional and syntactic factors.
Judgments on certain sentences are disputed, but there has been general agreement on examples like (1a, b).
The analysis presented here reconsiders examples (1a, b), as well as other well-studied multiple interrogation examples; it also adduces some less well-studied structures. The approach, however, diverges from earlier approaches in that it reassesses the sources of grammaticality contrast. Although this study, like many others, examines decontextualized, artificial sentences as its primary data, it builds on the observation that the
grammaticality status of certain multiple interrogation structures appears to improve if a favorable context is provided. Question-external manipulations are therefore examined; for instance, (1b) becomes more acceptable (if not fully grammatical) if it is embedded in discourses like (2a, b):
(2) a. I know that each book was bought by some buyer, but what did who buy?
b. Each of the gifts is lovely, but some of the cards have been separated from the gifts, and I want to thank each of you individually! So you will all have to tell me: What did who buy?
(1a), in contrast, does not require this type of contextualization or priming context in order to be judged grammatical.
Question-internal manipulations can improve acceptability as well; thus (1b) is also improved when a second in situ WH-term is added (What did who buy where?). It is argued that both external and internal variations that improve acceptability do so by altering or mitigating contingencies of question focus. Structural modifications within the question itself alter the locus or extent of focused material in the question (internal variation);
alternatively, the effects of a paucity of focused material in the question can be mitigated by enriched prior context that activates crucial material (external variation). In the latter case, the question acceptability is contingent on the type of contextual support that appears in (2a, b). Decontextualized examples like (1a) and (1b), therefore, differing with respect to syntactic well-formedness, also represent two endpoints on a spectrum of contextual dependence. Because (1a) does not require a priming context, it can be uttered in a range of
1 A reviewer has pointed to the difficulty in defining the notion in situ with reference to WH-movement, illustrating the problem with examples (ia, b):
(i) a. Who took a picture of whom?
b. Who took whose picture?
Both whom in (ia) and whose in (ib) qualify as WH-terms that have remained in situ while who has been
selected for WH-movement, yet either (ia) or (ib) (depending on one抯 derivational assumptions) has responded to another type of movement rule in the syntax. AWH-term that remains in situ, therefore, as that notion is used in the present discussion, must be a term that retains its position when the syntactic rule of WH-fronting applies to move another WH-term in the sentence.
2 B. Kennedy / Journal of Pragmatics 37 (2005) 12 discourses. (1b), however, in the absence of internal manipulation, requires that specialized information be activated in its immediate discourse context and is therefore uttered under relative contextual dependence.
This analysis does not, of course, preclude the possibility of other sources of ungrammaticality in multiple interrogations. However, focus occurrence within the question plays a peculiar role in superiority phenomena in English, where grammaticality contrasts appear to relate directly to the fronting of one of multiple WH-phrases. This (it is argued here) is due to the relationship between the fronted WH-term and the question focus. The English single WH-questioner fronts a WH-phrase in a request to identify an unknown (WH) quantity, and the subsequent open statement serves as a semantic predicate that holds of (is true of) the fronted WH-term. Thus, the open statement that constitutes the body of the question is informative semantically, with respect to the WH-term. More is expected by the listener, however; typically, the question is also informative pragmatically, insofar as its key addition to the discourse is focused via phonological highlighting (stress). This focus of the question represents key information that the questioner pairs with the WH-term: not merely what is true of the WH-term, but what the questioner wants to say about the WHterm. Material in the body of the question that is not highlighted in this way will still be part of the semantic predicate but must have been activated in immediately prior discourse; question material that is highlighted by focusing prosody, however, is thereby activated or
introduced within the question itself and will instruct following discourse turns. The same expectations regarding question focus hold for the English multiple WH-question, but the presence of two or more WH-terms to be identified renders more complex the satisfaction of this focusing requirement. And a multiple interrogation that does not focus key information will once again be more dependent on discourse context for information activation. This focusing expectation and its effect on contextual dependency are explored here, and I develop ranking metrics that predict the acceptability of decontextualized WHquestions by projecting relative contextual dependence. The metrics rate WH-question structures by considering informativeness: Those questions that utilize conventional focusing mechanisms to highlight information will be less dependent on specific context,
because (like (1a) and unlike (1b)) they will not require contextual priming. Thesemetrics are developed in an account of both orderly and problematic data; they are then tested against new structures.
As I present this analysis of the prosody-mediated relationship in WH-questions between structure and function, I utilize certain familiar notions: focus, stress, and focus projection. In addition, I introduce new terminology: contextual dependence, focus sequence, and contextual priming. Because focus has been analyzed under many different frameworks, I briefly introduce my use of that term and related terms in Section 2. I continue in that section to discuss the ways in which prosody, structure, and context interact in the formulation of the single English WH-question; and I develop the notion of the focus sequence, pairing a fronted WH-term with the question抯 focus. Section 3 moves to the multiple WH-question, in a review of previous superiority accounts. In Section 4, an account of multiple interrogation is developed along the lines of the Section 2 principles for single WH-questions, and ranking metrics that guide the distribution of multipleWH-terms in English are proposed. In Section 5, the metrics are tested against new or problematic B. Kennedy / Journal of Pragmatics 37 (2005) 1?2 3
structures and are amended to account for variations in focus projection and logophoricity. Section 6 summarizes the findings of this study.
2. The interpretation of WH-questions: issues of focus
It was stated in the Section 1 that conventions of English WH-questioning call for material to be encoded in the body of the question as question focus via prosodic and structural mechanisms. These mechanisms involve focal stress and focus projection, concepts that are addressed briefly before it is argued that focus in the English WHquestion?BR>like focus in other constructions is an information-highlighting mechanism signaled by stress and not typically residing in the fronted WH-phrase. The study of focus has been both enriched and complicated by various controversies regarding its definition, taxonomy, position in the language model, and cross-linguistic
universality (cf. Roberts, 1998 for discussion of some of these issues). A focus constituent often conveys new, newly activated, or contrasted information. In addition, focus is used to designate the target of logical operators such as only or intensifying operators (cf. Kennedy, 1999a). One commonly accepted view of focus (concentrating on the case of sentence focus) analyzes propositions informationally as comprising common ground material presupposed by discourse participants and focus material that is new or of central interest
and is highlighted by intonation and/or structure. This type of analysis was formalized by Jackendoff (1972), who identified focus as the new informational, F(ocus)-marked member of a pair that also included a presupposition derived via l-abstraction on the focus constituent. Problems regarding the characterization of focus material strictly as new information have been addressed by scholars like Chafe (e.g., 1994) and Lambrecht (1994).
Chafe recast the old杗ew information dichotomy in terms of information activation: Information may be given, accessible, or new, with intonational prominence fundamental in the activation of both new and accessible (but inactive) segments. Lambrecht characterized focus as relational, representing that segment of content that distinguishes an assertion from its presupposition, so that it is the selection of focal information rather than a
focal referent that is unpredictable or novel. There has been general concurrence on the association between the point of maximal intonational prominence and the sentence focus, but scholars have disagreed on the relationship of that position of intonational prominence to structure. Thus, the Nuclear Stress Rule (NSR) of Chomsky and Halle (1968) related focus-associated prosody directly to surface syntactic structure; Selkirk (1983) abandoned the NSR itself but went on to map the phonological (prosodic) manifestations of focus onto sentence syntax in a more complex fashion, accommodating contemporaneous views on the structure of the grammar. Bolinger (1972), representing an opposite view, saw focus as a highlighting mechanism that signaled special informational value of an element and was placed in accordance with communicative intent rather than grammar. Under a more formal but similarly pragmatic model, Roberts (1998) differentiated prosodic and syntactic structures, suggesting that prosody indicating focus is not annotated directly onto syntax but is instead a feature of an 4 B. Kennedy / Journal of Pragmatics 37 (2005) 12 independent grammar component. Conventional presuppositions permit prosody, for instance, to relate the denotation of an expression to its information role within discourse. Equally controversial have been questions of taxonomy, with many studies (cf. for instance, Rooth, 1992) distinguishing typologically between two kinds of focus: information focus, signifying new or asserted information; and operator focus, which targets an entity from a domain and bears contrastive connotation. Roberts, on the other hand, analyzed operator focus as a subcase of the more general pragmatic category of information focus by which a distinguished constituent is selected via a conventional stress
mechanism. The conception of focus utilized here derives most directly from information structural characterizations such as those of Lambrecht and Chafe and from the work of Erteschik-Shir (e.g., 1973, 1986, 1997), who defined focus in terms of speaker intent: Focus is assigned to a constituent to which a speaker wishes to draw the hearer抯 attention (cf. also Kennedy, 1993, 1999a). Because the focus under consideration in this study appears in questions, however, its conceptualization brings special issues. First, there is controversy
regarding the association of focus with the WH-element in a WH-question; second, the discourse role of the WH-question complicates the definition of focus in terms of an assertion resupposition relation. Focus has often been invoked in an account of WHquestions: It is frequently argued that the WH-phrase in a WH-question is the focus of the question (cf. for instance, Culicover and Rochemont, 1983; Lambrecht, 1994; Rochemont, 1986). Because an English WH-phrase does not generally receive primary sentence stress, however, inherent focus on a WH-phrase (as in Rochemont model) would preclude a reliable association between stress and focus in English. Not only does the English WHphrase not attract primary stress, but the association between WH-words and stress also varies cross-linguistically, challenging further an across-the-board statement about the
relationship between WH-words and focus in questions (cf. Culicover and Rochemont, 1983; Ladd, 1996 for treatment of this issue). Concentrating on the English case alone, dissociation of focus in a WH-question from prosodic prominence (so that the WH-phrase is termed focus) adheres to previously cited notions of focus as that constituent to which the speaker wishes to draw the hearer抯 attention and whose choice is unpredicted or novel in the discourse content.
One reason not to analyze the EnglishWH-term as inherent focus, however, derives from the observation that stress sometimes does fall on a WH-term. Collapsing stressed and unstressed WH-words with respect to focus (such that in both cases the WH-term would be considered focus) would sacrifice insight into informational and discourse distinctions between examples like (3a) and (3b) (underscore indicates stress):
(3) a. Who bought the book?
b. Who bought the book?
Notice that (3a) but not (3b) would be an appropriate sequitur to Everybody appears to have pencils and notebooks, but . . .. On the other hand, (3b) would follow I know that someone here has purchased Dilbert, but . . .. (3b), that is, presumes that the information that someone has bought the book in question has just been foregrounded or activated in the preceding discourse context, and therefore that material need not be focused in the question B. Kennedy / Journal of Pragmatics 37 (2005) 12 5 itself. (3a), in contrast, permits no such assumption; in (3a), the question itself foregrounds the notion that someone bought the book. The differential positioning of stress in the two examples is consistent with the suggestion that this information is focused in (3a) but presupposed in (3b), with focus occurring instead on the stressed WH-term in (3b). This analysis of a WH-phrase as a focus in (3b) but not in (3a) permits us to maintain the association between prosody and focus, proposed by Roberts (1998) as a linguistic universal; that analysis also suggests that the functional role of focus in a WH-question is one of foregrounding. In (3a), focus foregrounds information that the questioner wishes
to activate: The questioner associates the WH-term who informationally with focused the book (or bought the book; cf. discussion of focus projection in Section 2.2). Retraction of focus to the WH-phrase in (3b) can occur only in the case in which this type of important information about the entity being questioned has just been foregrounded (cf. Erteschik-Shir, 1986, 1997, where a WH-phrase is analyzed as focus under limited circumstances).
Focus in the English WH-question, like that in other structures, can thus be related to discourse agenda and information flow. Focus is therefore seen here as an information-highlighting mechanism not only in the more typical case of declarative sentence focus but also in other constructions.Drawing on Chafe work, I emphasize the role of focus as an information activation mechanism; drawing on Lambrecht insight, I emphasize the relational aspect of focus, insofar as focus directs us to selection of information to be activated, relative to presupposed material. As a combined effect of perceived duration, amplitude, and pitch, moreover, stress designates an English focus constituent. Language conventions and, therefore, interlocutor expectations about the informational role of focus in a given structural context will determine the way in which focus is interpreted; in thisway, the approach might again resemble that of Lambrecht, who made crucial distinctions between focus construction types. In the present approach, however, construction-particular occurrences of focus are related directly to the communicative function of the construction. In addition, structural contingencies interact crucially with focus and constrain its domain: The relationship between locus of stress and broader focus constituency is mediated systematically by focus projection.
2.2. Focus projection
Prosody, with its manifestation in stress, is a consistent but at times indirect focus marker mfor English. The indirect nature of the relation between stress and focus can be attributed in part to the phenomenon of focus projection. Focus can be characterized as discourseassociated; but an articulated, sentence-level prosodic structure (that need not coincide strictly with surface syntax) has often been argued to play a role in focus projection from the locus of primary stress to a broader focused constituent.2 Thus, for instance, Kennedy
(1999a), Rochemont (1986), and Selkirk (1983) discussed data like (4a, b) (see alsoCinque, 1993).
2 Lambrecht (1994) accounted in a different way for the focus interpretations corresponding to the various
systematic stress patterns. In his analysis, distinct prosodic patterns are associated with three informationstructure categories, each conveying a particular focus interpretation. 6 B. Kennedy / Journal of Pragmatics 37 (2005) 1?2
(4) a. John bought Dilbert.
b. John bought Dilbert.
A WH-question set is often used to diagnose focus constituency, under the assumption
that it is the focus constituent in the answer that identifies the WH-phrase in the
question (cf. Paul, 1880; Sgall et al., 1986, for instance). In an active, agentive sentence
like (4a), a stressed object NP can signify either narrow focus on that NP alone or
broad focus projected to VP or IP; thus (4a) can be used to answer What did John buy?
with focus on the object NP, What did John do? with focus on the VP, or What
happened? with focus projected to the entire sentence. (4b), in contrast, is unambiguous
with respect to focus interpretation; stress on the subject NP can signify only
narrow focus on that NP. Thus, (4b) can be used as an answer only to a
WH-question like Who bought Dilbert? In sum, focus in (4a) can project from a stressed lexical
ite (referred to here as the nuclear term in the focus constituent) to a broader constituent,
but focus does not project in (4b) beyond the subject NP.3 More generally, focus can
project from a stressed verb or (a term representing) a stressed object argument to the
VP or IP level, but focus does not typically project from the subject position; thus
(5) a. John recognized Bill.
b. John recognized Bill.
c. His company recognized Bill.
Focus can (although need not) project from the stressed object NP in (5a) to VP (or IP)
level; thus (5a) might answer either Who did John recognize at the awards ceremony? (no
focus projection, and narrowly focused object NP) or Did John know anyone at the awards
ceremony? (focus projection, and focused VP). (5c) represents the case in which focus can
project from V: When recognize signifies knowledge formally? the verb typically
receives stress within VP. Focus in (5c) can project from V to VP or IP level; thus
(5c might answer What happened at the awards ceremony? (focused IP). (5b), however,
permits only narrow focus on the subject NP, answering a question like Did anyone
recognize Bill at the awards ceremony? Projection from an adjunct position is more
complex, as is projection within NP (cf. Selkirk, 1983).
2.3. Focal stress in WH-questions
Focus projection in a WH-question, like projection in a statement, interacts with
argument structure; as in a statement, focus will introduce or activate special information.
In the WH-question, we see that this focused information is linked in a focus pair with the
fronted WH-phrase. The focused information is generally distinguished from semantically
predicated information in the question in that it represents what (regarding the questioned
entity) is being highlighted for discourse purposes, rather than simply what holds of the
fronted WH-term; it therefore reflects previous conversation and presages future con-
3 See Faber (1987) and Kennedy (1999a) for discussion of focus projection from a stressed subject NP.
B. Kennedy / Journal of Pragmatics 37 (2005) 1?2 7 versational turns. Most important to the present discussion, its presence or absence will
influence the degree to which the question is dependent on specific information preactivation
in prior discourse for felicity.
Notice first that just as in a statement, stress on an object NP in aWH-question like (3a),
repeated here, can project focus to VP, whereas stress on a subject NP in both the question
(6a) and the statement (6b) signifies only narrow focus. In (3b), likewise, stressed who is
(3) a. Who bought the book?
b. Who bought the book?
(6) a. What did John buy?
b. John bought the book.
Note, second, that (as observed in Section 2.1) differential positioning of stress in (3a)
and (3b) accompanies differences in focused/presupposed status of information in the
question. Although it has often been argued that the abstract resulting from l-abstraction of
the WH-phrase (cf. Jackendoff, 1972) represents the presupposed material in a question,
consider again the distinction between (3a) and (3b). In both cases, l-abstraction of the
WH-phrase yields the abstract bought the book (x); and in (3b), it does indeed appear to be
presupposed that someone bought the book. In (3a), however, focal highlighting on NP (or,
under focus projection, on VP) gives that constituent抯 denotation a different role in the
question. Whether the focused information in the body of the question (3a) is new to the
discourse or old and accessible but inactive material (cf. Chafe, 1994), occurrence of focal
highlighting signifies a discourse need for its activation, in contradistinction to the case in
(3b). 4 Stated in converse, a WH-question like (3b) with no focus-activated information in
the body of the question will require appropriate preactivation of material in its prior
In addition, focus highlighting reflects the questioner抯 agenda in posing the question:
The questioner exploits prosody systematically to highlight information he or she wishes to
link with the WH-phrase. This function of prosody can be illuminated if we distinguish the
notions of focus and (semantic) predication. Lambrecht (1994), citing the traditional
definition of (semantic) predicate as that property attributed in a sentence to its subject,
differentiated semantic and pragmatic predicates: The semantic predicate expresses what
is true梬hat holds梠f a subject, whereas the pragmatic predicate expresses what is said
of a subject. In a declarative like (6b), the semantic predicate would be bought the book,
because this property holds of the subject John. From a pragmatic perspective, however, the
subject of the utterance (6b) is bought the book, and the pragmatic predicate梬hat the
speaker would like to say about buying the book, highlighted by focal stress梬ould be [is]
John. Extending Lambrecht抯 distinction to a WH-question like (6a), the semantic
predication relation associates what and John bought, insofar as John bought will hold
4 Compare Lambrecht抯 (1994) analysis, however, whereby a fronted WH-phrase is a main sentence focus,
with the open proposition constituting the body of the question as presupposed material. Lambrecht analyzed the
class of accent falling on BOOK in (3a) as an activation accent that demarcates topical material.
8 B. Kennedy / Journal of Pragmatics 37 (2005) 1?2 of the entity questioned by what. This relation, however, is not one of pragmatic
predication, which would relate bought [something] as subject and [was] John as predicate.
Focus intonation directs us to the pragmatic predicate and often dissociates semantic and
pragmatic predicates, which can but need not converge. This in turn is because focus,
reflecting speaker goals, directs us not simply to what is true but to what is being said of
contextual topics: to a highlighted addition to the discourse that will guide future
Lambrecht抯 distinction between semantic and pragmatic predication in declaratives
can be drawn on to illuminate the discourse contribution of prosodic features of theWHquestion.
As noted, his semantic predication relation can be compared to the relation
between the WH-phrase and the l-abstract that introduces those properties that characterize?BR>that are true of梩he questioned entity. The relation between theWH-term and
the question抯 focus, however, involves Lambrecht抯 pragmatic predication. Just as
Lambrecht抯 pragmatic predicate represented not simply what was true of a subject
but what was being expressed by a speaker about it, the question focus reflects the
questioner抯 agenda, representing crucial discourse information that the questioner wants
to pair with theWH-term. In (6a), semantic predication tells us that John bought [it] must
hold of the constant value replacing what in the answer. Focus, however, links that value
with a crucial aspect of this semantic predicate: The speaker asks what it was that it was
John who bought it.
5 Because the question抯 prosody must be representative in this way of the questioner抯agenda but also appropriate to the discourse (must be faithful to novelty and/or accessibility
status of information), we see that focus placement in the question serves two
functions: It (a) ensures that the question is relevant with respect to embedding discourse
and (b) guides future discourse by highlighting a key addition to the discourse in the form
of a pairing of WH-term and focus. (6a), for example, is linked by intonation to discourse
information or discourse set: Absence of stress on buy indicates that the fact that there were
a number of buyers of things is common ground.6 But the narrow focus on the subject NP in
that example also reflects the questioner抯 special interest in that segment of question, an
interest that guides future discourse. To see how a question抯 focus performs these two
functions, compare (6a) and (7):
(6) a. What did John buy?
(7) What did John buy?
Exploiting the relation of semantic predication, the speaker poses a question such that the
l-abstract in the body of the question will hold of the constant replacing the WH-phrase in
the answer. (6a) and (7), therefore, sharing the abstract John bought (x) although differing
5 A reviewer has commented that the question translation that reflects speaker focus and identification
agenda will involve this type of clefted utterance; as Lambrecht pointed out, a cleft is indeed one constructional
strategy whereby a speaker clarifies the structuring of information.
6 This statement oversimplifies the relation between presupposition and discourse common ground; in many
cases the hearer will amend the discourse context to assimilate information that is introduced as ostensible
presupposition but has not yet achieved the status of common ground, via the process of presupposition
accommodation (cf. for instance, Heim, 1983).
B. Kennedy / Journal of Pragmatics 37 (2005) 1?2 9 in focus, can be answered in the same way: The constant value Dilbert (for instance) can
instantiate what in both cases.We have also seen, however, that focus distinguishes relation
to previous discourse; thus (6a) presumes that individuals bought things and foregrounds
John as novel or inactive content, whereas (7) presumes that John is preactivated and
foregrounds buy (unless focus projects beyond buy to IP). In addition, notice that the two
questions will move the discourse in different directions; thus (8a) is a more appropriate
response to (6a), whereas (8b) would be more appropriate to (7).
(8) a. Dilbert, and Pete bought The Far Side.
b. Dilbert, and now he抯 saving up for Garfield.
Each question, that is, motivates the respondent抯 addition of different information to the
discourse. The focus in a WH-question therefore reflects not only distinctive discourse
history but also a distinctive discourse contribution.
(3a) and (6a) also showus that intonation supports the linking of focusedmaterial to the
WH-phrase. We have seen that each question has distinctive primary stress that
designates focus, either directly (as narrow focus) or indirectly (via focus projection).
In addition, however, there is a secondary stress in each example that falls on the WHterm:
who in (3a) and what in (6a). Although it was argued in Section 2.1 that the WHterm
is not inherently the question focus, it participates in a pairing, effected by
(secondary and primary) stress; the question focus represents material that the speaker
links with the WH-term. This linking derives a focus sequence of WH-term and question
focus; the focus sequence for who in (3a) is given in (9), and the focus sequence for what
in (6a) is given in (10).
(9) hwho, bought the booki (10) hwhat, Johni
The focus in the sequence in (10) is only an NP (John), because the focus in the WHquestion
(6a) can project no further than narrowly focused John. The sequence in (9)
presumes that focus has projected to VP (bought the book); under an alternative
interpretation, the sequence for that example would be hwho, the booki, with narrowly
focused object NP. As noted, differential focus assignment will both reflect distinctions
in prior discourse context and affect future information flow. Thus, for instance,
question (3a) with narrowly focused object NP (the book) could appear in the following
(11) A: The neighbors bought up almost everything at the yard sale.
B: Who bought the book?
A: Mary did. And I bet you want to know who bought the tape set. It was Bill.
In discourse (11), narrow focus on the object NP in (11B) (?3a)) reflects the fact that the
previous utterance has just activated people were buying things. The WH-question activates
the book, the focus of the question. The paired sequence hwho, the booki, moreover, steers
the discourse toward discussion of buyer-object pairs.
10 B. Kennedy / Journal of Pragmatics 37 (2005) 1?2
Consider now an example in which the WH-word appears in a which-NP, as in (12):
(12) Which boy bought the book?
Stress on the object NP in (12) identifies the question focus as the book or, under focus
projection, bought the book. There are also two candidate WH-phrases: which and
which boy. Assuming focus projection to VP, so that bought the book is the question
focus, (12) is still associated with two possible focus sequences, (13a) and (13b),
depending on the intended WH-phrase; each sequence pairs the question focus with a
distinct questioned term. Prominence within NP will determine which WH-term
option is intended; this issue is discussed in greater detail in Section 4, when focus
projection within a which-NP is addressed. Prominence on which represents narrow
focus and indicates that which alone is the questioned term, whereas prominence on
boy permits focus projection to NP and indicates that the full WH-NP is the questioned
term; it is this (secondary) stress within the which-NP that indicates the intended WHterm
and therefore selects between (13a) and (13b) as focus sequences associated with
(13) a. hwhich, bought the booki b. hwhich boy, bought the booki
The sequence in (13a) is associated with a question in which which is prominent in the WHNP
as the questioned term, a question occurring in a context in which some boys are
already under discussion as active discourse entities. The sequence in (13b) is associated
with a question in which boy is prominent in the WH-NP, receiving secondary sentence
stress. This second stress activates boys as the buyers of interest and identifies which boy as
the relevant WH-term.
Notice that in the examples we have considered, the WH-question (and the focus
sequence notation displayed in (9), (10), and (13)) links a WH-term with a non-WH
focus. As initially noted in Section 2.1, however, there are cases like (3b) where stress
falls on the WH-term. We observed that (3b) requires that certain information has just
been activated in the discourse, as in I know that someone here purchased Dilbert. But
who bought the book? A semantic predication (bought the book) appears, so that the
listener is given a semantic basis on which to evaluate who. Indeed, earlier comparison of
(3a) and (3b), repeated here, revealed that the two questions shared the l-abstract bought
the book (x), so that in both cases the semantic predicate bought the book holds of the
(3) a. Who bought the book?
b. Who bought the book?
Focus activation of information within the two questions differs, however: In (3a), material
within the semantic predicate is activated, whereas no material within the semantic
predicate is activated in (3b). The focus sequences (9) (for (3a)) and (14) (for (3b)) reflect
B. Kennedy / Journal of Pragmatics 37 (2005) 1?2 11
(9) hwho, bought the booki (14) hwhoi
The representation (14) tells us that who is bothWH-term and focus in (3b); it indicates that
who is associated with no focused material in the body of the question. The absence of
focus within that non-WH material, however, forces a special dependence on the previous
sentence: (3b) must be preceded by discourse that activates predicate content. Apparently
semantic predication content alone is not adequate in a question: That content will establish
contingencies that the constant replacing the WH-term in the answer must satisfy, yet the
content must also be newly activated information in the discourse. This activation may
occur via (a) activation in prior context or (b) activation via focus in the question itself.
Utterance of (3b) involves the first activation mode and entails a specific relation to the
discourse, due to a mediating priming context: a preceding discourse in which non-WH
question content has been activated. When no material other than a WH-term is highlighted
prosodically (as in (3b)), only material that has already been activated梞aterial already in
the speaker抯 focal consciousness, in Chafe抯 (1994) terms梐ppears in the question itself;
prior discourse must have brought this material into the speaker抯 focal consciousness. In
the discourse suggested for (3b), I know that someone here purchased Dilbert. But who
bought the book?, the boldface material in the sentence preceding the WH-question serves
as priming context, activating information that is not highlighted in the question itself.
The preceding discussion suggests a contextual dependence hierarchy, where !
signifies 慽nvolves less contextual independence? and hBi ! hAi signifies that sequence
B occurs under less contextual independence than does sequence A:
(15) hWHi ! hWH, . . .i (16) AWH-question occurs under heightened contextual dependence if its felicitous
utterance requires specific priming context.
(15) and (16) state that a sequence comprising a WH-term alone and including no non-WH
focus signals a question that is more dependent on specific priming context than does a
sequence including a non-WH focus. The WH-question whose focus sequence comprises
only a WH-term, insofar as it requires a priming context to activate material that later
appears (unfocused) in the question, occurs in a more restricted set of contexts: only those
contexts that are enriched as priming contexts.
To summarize Section 2, I considered focus, its associated prosody, and the informational
import of the prosodic contour in single WH-questions. A general conceptualization
of focus as a functional highlighting mechanism was established; the association of stress
with focus in English was maintained, but the association between focus and the English
WH-phrase was not. Instead, it was proposed that stress is used to highlight the WHquestion抯
focus, paired informationally with the fronted WH-term in a focus sequence.
Although the body of the question introduces a semantic predicate whose content holds of
the constant replacing the WH-term in the answer, activation of that content must occur by
(a) focus highlighting of all or part of the predicate in the question itself, or (b) activation in
a priming context. In the first case, prosody (secondary and primary stress) in a single WHquestion
links the WH-term and focused non-WH material within the question. In the
12 B. Kennedy / Journal of Pragmatics 37 (2005) 1?2 second case, content activation occurs externally (prior to the question); in this case, a
focus sequence comprising only a focused WH-term is associated with a question that is
felicitous in a restricted set of contexts. In the next section, we move from single to multiple
WH-questions; we begin consideration of multiple interrogations by reviewing superiority
3. Superiority effects in multiple interrogations
3.1. Syntactic accounts
(1) a. Who bought what?
b. What did who buy?
The contrast in (1a, b), repeated here, is often given as a central exemplar of the
superiority effect. A multipleWH-question in English, like a singleWH-question, contains
a fronted WH-term in [Spec, CP] (specifier of CP) position; it also includes one or more
WH-terms in situ: terms that have not been moved by WH-fronting in the syntax and
therefore occupy base positions at surface structure (cf. note 1). The superiority effect has
been explained syntactically and pragmatically, as well as in accounts that invoke both
structure and function. As a seminal structural statement of the constraint governing the
superiority effect, consider (17), in which the relative S-structure positions of the WHterms
are limited by Chomsky抯 (1973) Superiority Condition:
(17) Superiority Condition: No rule can involve X, Y in the structure . . . X . . . [. . . Z
. . . WYV . . .] . . ., where the rule applies ambiguously to Z and Y, and Z is
superior to Y. The category A is superior to the category B if every major
category dominating A dominates B as well but not conversely.
The Superiority Condition (17) looks at a D-structure with two sources of WH-movement
and requires that a rule like WH-movement select the superior source. Because who in
subject position is superior in the base to what in object position, the rule of WHmovement,
applying ambiguously to who and what, must apply to who, yielding (1a). (1b)
is excluded because WH-movement did not involve the superior term in that example.
(Vacuous movement for the subject WH-term is assumed here, but cf. Chomsky, 1986.)
Later accounts of superiority invoked the Empty Category Principle (ECP), a general
constraint (cf. for instance, Chomsky, 1986) requiring that an empty category be licensed:
(18) Empty Category Principle: A nonpronominal empty category must be properly
governed, where A properly governs B iff A theta-governs or
antecedent governs B.
(19a), as the LF-representation for (1a), shows both empty categories properly governed;
but (19b), as the LF representation for (1b), is ill formed because the subject empty
B. Kennedy / Journal of Pragmatics 37 (2005) 1?2 13 category is neither theta governed nor antecedent-governed (cf. Lasnik and Saito, 1984; LF
stands for 憀ogical form?.
Problematic to the ECP account of (1a, b), however, was an example like (20):
(20) Who knows what who bought?
The ECP does not distinguish the embedded question in the multiple interrogation (20)
from the matrix question (1b); yet (20) is grammatical. In addition, (21a, b) from Pesetsky
(1987) display the same superiority effect as do (1a, b), although the LF representations
(22a, b) show the traces of LF-movement properly governed in both cases:
(21) a. Whomi did you persuade ti [PRO to read whatj]?
b. ??Whatj did you persuade whomi [PRO to read tj]?
Lasnik and Saito (1992) therefore abandoned an ECP approach to superiority phenomena
and proposed an Operator Disjointness Condition (ODC), reformulated by Epstein (1998)
as the Scope-Marking Condition (SMC):
(23) Scope-Marking Condition: In the LF component, a Wh in situ Y can adjoin to a
WH-chain X only if X c-commanded Y at S-structure.
The SMC assumed LF-adjunction of a WH in situ to a scope-marking WH-chain rather
than to a WH-phrase (chain head), stipulating that the WH-chain must have c-commanded
WH in situ at S-structure, where a chain c-commands an item iff all members of
the chain c-command that item. An important consequence of (23) is an account not only
for the grammatical status of (20) but also for its restricted interpretation. Notice that in
situ who in (20) must take wide scope; that is, it contrasts with what in Baker抯 (1970)
(24) Who knows where we bought what?
(24) is ambiguous, in that in situ what can be associated with the WH-phrase in either the
embedded or the matrix [Spec, CP]: (24) can be answered by either (25a) or (25b):
(25) a. John knows where we bought what.
b. John knows where we bought biscuits, and Mary knows where we
14 B. Kennedy / Journal of Pragmatics 37 (2005) 1?2
In (20), in contrast, in situ who can be associated only with the WH-phrase in the matrix
[Spec, CP], so that (26b) is an acceptable answer but (26a) is not:
(26) a. John knows what who bought.
b. John knows what Bill bought, and Mary knows what Ralph bought.
In order to capture this type of contrast, Baker proposed a Q morpheme in interrogative
Comp; scope of a WH-phrase might be represented via coindexation with matrix or
embedded Q, so that what in (24) would be coindexed with either matrix or embedded Q,
whereas who in (20) would be coindexed only with matrix Q. By the SMC, in situ who in
(20) cannot adjoin to what in the embedded [Spec, CP] for the same reason that a similar
adjunction cannot occur in (1b): All members of theWH-chain hwhat, ti do not c-command
who at S-structure. The WH-chain headed by who in matrix [Spec, CP] does, however, ccommand
in situ who, so that the latter term can be adjoined at the matrix level. Scope of in
situ who is therefore matrix level, yielding an interpretation associated with response (26b).
In contrast, the SMC will permit scope marking for what in (24) at both the embedded and
the matrix levels, because all members of theWH-chains headed by both who and where ccommand
what at S-structure. Returning to constraint on movement, we see a locality metric in Chomsky抯 (1995)
Minimal Link Condition (MLC):
(27) Minimal Link Condition: K attracts a only if there is no b, b closer to K than a,
such that K attracts b.
Under Chomsky抯 Minimalist framework, syntactic movement (actually feature movement),
is recast as a feature-checking attraction between two elements. Like the earlier
Superiority Condition, the MLC will preclude fronting of aWH-term a to a position where
its WH features can be checked if (as in a multiple interrogation) it is in competition with a
more local WH-term b, where greater locality of b with respect to K can be (informally)
characterized as structural superiority: b will be closer to c-commanding K if b asymmetrically
c-commands a. The more local WH-term will move in the syntax to [Spec, CP] in
order to check the strong [H] feature of C. (See, for instance, Boskovic and Lasnik,
1999; Marantz, 1995; Tanaka, 1999 for applications of the MLC to superiority phenomena.)
We have therefore seen two types of structural accounts for superiority phenomena:
Chomsky抯 movement-constraining Superiority and (later) Minimal Link Conditions, and
the representation-constraining ECP and ODC/SMC conventions. Another representational
approach, developed in Pesetsky (1987) and May (1985), utilized a Path Containment
(28) Path Containment Condition: Intersecting A0-categorial paths must embed, not
Under this approach, each path created by syntactic movement comprises a sequence of
nodes connecting a moved category and its trace. If two paths intersect (share a segment),
B. Kennedy / Journal of Pragmatics 37 (2005) 1?2 15
(28) requires that one path properly contain all the category members of the other. Again
assuming adjunction at LF of WH in situ to a WH-phrase in [Spec, CP], the empirical
consequences of the PCC are similar to those of the SMC on examples like (1a, b); but the
PCC fails on (20), because the A0-chain headed by what will overlap the chain headed by in
situ, LF-moved who.
There is another set of examples, however, on which the PCC is more successful than
either the SMC or the MLC: examples in which the addition of a second WH-phrase in situameliorates a structure like (1b). Thus, consider grammatical (29):
(29) What did who buy where?
Although the addition of where to (1b) yields grammatical (29), adjunction at LF of who to
the scope-marking chain headed by what will still be illicit by the SMC, because one
member of the chain hwhat, ti does not c-command who at S-Structure. The MLC,
moreover, will block syntactic movement of what, because shorter movement of who
would be an option. The PCC, in contrast, accounted for (29) by utilizing path union. Thus,
May (1985) proposed that in the case of twoWH-terms in situ, the PCC should be evaluated
with respect to the union of those paths. This approach correctly predicts that (29) will be
well-formed. A further type of example, however, is explained by neither the PCC, the SMC, nor the
MLC: biclausal examples like (30a), structured as in (30b).
(30) a. What did John learn to do where?
In the ambiguous string (30a), where can modify either do or learn. (30b) represents the
latter interpretation, under which where is a matrix-level adjunct, modifying learn.
Under this interpretation, LF-movement paths for where and what will overlap, so that
the PCC will incorrectly predict (30a) to be ungrammatical under the desired interpretation.
Given matrix-level where, the SMC and the MLC will likewise rule out (30a),
because the trace of the WH-phrase in [Spec, CP] does not c-command where at Sstructure.
To summarize, I have considered syntactic accounts of superiority phenomena. We
reviewed an early movement account, Chomsky抯 (1973) Superiority Condition; various
constraints on representation were then discussed. An ECP approach was challenged by
examples (20) and (21b); the Operator Disjointness Condition of Lasnik and Saito (1992)
and its reformulation as Epstein抯 (1998) Scope-Marking Condition were better motivated
empirically, with the SMC also providing an explanation for scopal interpretation ofWHin
situ. Chomsky抯 (1995) Minimal Link Condition bore the same liabilities as the earlier
Superiority Condition when confronted by crucial examples like (20) and (29); the Path
Containment Condition (May, 1985; Pesetsky, 1987), although similar at some points to the
SMC, did not, crucially, explain the status of an example like (20) but did predict the
ameliorative effect of the secondWH in situ in (29). But neither the SMC, the MLC, nor the
PCC accounted for biclausal example (30a), where the fronted WH-phrase originates in the
embedded clause and the WH in situ is matrix level.
16 B. Kennedy / Journal of Pragmatics 37 (2005) 1?2
3.2. Pragmatic approaches
To review the approaches considered thus far, consider the following examples, against
which the syntactic accounts discussed in Section 3.1 can be assessed:
(31) a. ??Which boy was which book bought by?
b. Which book was bought by which boy?
Because (31a) bypasses the structurally superior (and therefore more local to matrix [Spec,
CP]) WH-phrase which book to front the prepositional object NP which boy, its reduced
grammaticality is predicted by the Superiority Condition as well as by the MLC; (31b)
conforms to structural superiority/locality constraints and is correctly predicted to be
grammatical by both conditions. The SMC will also distinguish between (31a) and (31b): It
rules out (31a) because the trace of the moved WH-NP which boy does not c-command in
situ which book at S-structure, but it rules in (31b) because both the fronted WH-phrase
which book and itsWH-trace in [Spec, IP] c-command in situ which boy. The PCC likewise
excludes (31a), because movement paths for the two which-NPs will overlap rather than
Although (31a) and (31b) differ in acceptability, however, it is significant that (31a) is
not as bad as its structural counterpart (32):
(32) Who was what bought by?
The ameliorative effect of the occurrence of which-NPs in an example like (31a) is even
more pronounced in an example like (33b):
(33) a. Which boy bought which book?
b. Which book did which boy buy?
This effect was discussed at length in Pesetsky (1987), one of a number of accounts of
multiple interrogation phenomena that have not relied solely on structural distinctions
between grammatical and ungrammatical questions. Although syntactic in approach,
Pesetsky抯 (1987) account of examples like (33a, b) drew on discourse facts. Other factors
being equal, examples like (33a, b) ought to behave like their structural counterparts in
(1a, b); yet (33b), unlike (1b), is grammatical. Pesetsky attributed this difference to the
7 The NP-trace of which book c-commands which boy as well in (31b), but Epstein (1998) distinguished in
his SMC analysis between A0- and A-chains, such that it is the A0-chain but not the linked A-chain in an example
like (31b) that is relevant to SMC application. It is of interest that the PCC, if it considered both strands of the
linked A0- and A-chains in examples like (31a) and (31b), could also distinguish between preposition-stranding
(31a) and its pied-piping counterpart (i):
(i) ?By which boy was which book bought?
(i), although not fully grammatical, is preferable to (31b), a distinction that could be accounted for by the PCC if its
path computation combined A0- and A-movement paths.
B. Kennedy / Journal of Pragmatics 37 (2005) 1?2 17 occurrence of which boy rather than who as questioned subject NP in (33b) and
characterized the which-phrase as discourse-linked or D-linked. The introduction of
D-linked phrases like which book and which boy presupposes that speaker and hearer share
knowledge of contextual sets of books and boys; Pesetsky proposed that D-linked phrases
were not quantified phrases and therefore did not move at LF, obviating any PCC
Erteschik-Shir (1997), although sharing with Pesetsky the conclusion that discourse
linking is crucial to the grammaticality of an example like (33b), questioned the reference
to discourse-theoretic factors in the context of an explicitly structural account, pointing out
that an LF-movement rule would then have to refer to extrasentential context for its
application. She cited the following example of Pesetsky抯, in which D-linking is a
consequence of the textual context itself rather than the WH-phrase structure but permits
the subject WH-phrase in situ that is disallowed in (1b).
(34) I know that we need to install transistor A, transistor B, and transistor C, and I
know that these three holes are for transistors, but I抣l be damned if I can figure
out from the instructions where what goes!
Kuno and Takami (1993), like Pesetsky, identified discourse-theoretic considerations;
the following functional constraints selected between multiple interrogation structures,
however, rather than feeding syntactic rules:
(35) a. Sorting Key Hypothesis: In a multiple WH-question, the leftmost WHword
represents the key for sorting relevant pieces of information in the
b. List Requirement Hypothesis: Why and how acquire the property of for
what reason and in what manner in contexts that make it clear that there
is a predetermined list of items that corresponds to each of the multiple
WH-expressions in the sentence.
In (35a), Kuno and Takami proposed that the relative position of the WH-phrases in a
multiple interrogation like (1a) or (1b) will reflect the speaker抯 arrangement of information.
The occurrence of which-phrases in (33b) suggests available lists of possible referents
(purchasers and purchased items), so that either piece of information may be the sorting key
and either which-phrase can be fronted.
Kuno and Takami utilized (35b) as well as the syntactic constraint (36) to explain the
contrast between (37a), on the one hand, and (37b, c), on the other:
(36) S00-WH Hypothesis: Why and how are base generated as S00 adjuncts, whereas
other WH-adjuncts are base generated as daughters of VP.
(37) a. Tell me who came why.
b. Tell me who bought what why.
c. Tell me who came for what reason.
18 B. Kennedy / Journal of Pragmatics 37 (2005) 1?2
Independently motivated in Kuno and Takami (1993) by facts involving a range of
fronted elements, (36) rules out (37a) because sentence-final why would need to be
dominated by VP, in violation of (36). In (37c), however, sentence-final for what reason,
unconstrained by (36), can occur legitimately as a VP-level adjunct. In (37b), the
occurrence of three questioned terms insinuates three discourse-established lists: a list
of purchasers, a list of purchased items, and a list of reasons for purchase. By (35b), the
contextual presence of these predetermined lists permits why to acquire the property of
for what reason and therefore to appear under VP. Unaccounted for, however, is the
contrast in (38):
(38) a. Who bought what why?
b. What did who buy why?
By (35b), (38b) should, like (37b) and (38a), be an ameliorated example.
In Section 3, I have reviewed syntactic and discourse-theoretic accounts of multiple
interrogation. A residue of data will challenge a purely syntactic account; a discoursetheoretic
account is more flexible but is also challenged to establish a consistent model. The
model introduced in the next section is a pragmatic one; it is argued that superiority effects
in multiple interrogations are driven by discourse requirements based on the same
conventions (discussed in Section 2) that serve single WH-questions.
4. The interpretation of multiple interrogations
We now return to the findings regarding single WH-questions in Section 2 and consider
their extension to multiple interrogations; we see that multiple interrogations, like single
WH-questions, are sensitive to contingencies of focus and vary with respect to context
dependency. As we study prosody and focus sequences within multiple interrogations, we
see that (a) exclusion of non-WH material in the question focus is again associated with
context dependence, and (b) prosodic prominence (nuclear stress in the question focus) is
associated with WH in situ.
Recall first that the prosodic pattern in (3b) (repeated here) was associated with relative
contextual dependence and thus a need for contextual priming, as was reflected in (15) and
(16) (also repeated):
(3b) Who bought the book?
(15) hWHi ! hWH, . . .i (16) AWH-question occurs under heightened contextual dependence if its felicitous
utterance requires specific priming context.
Although contextual dependence was associated in (3b) with a stressed WH-term, we now
consider multiple interrogation examples in which a stressed WH-term is typical and bears
no special association with contextual dependence. Consider (39):
(39) Who bought what?
B. Kennedy / Journal of Pragmatics 37 (2005) 1?2 19
In the multiple interrogation (39), two items are questioned; it appears, moreover, that
stress must fall on the WH in situ.8 Thus, (40) is prosodically ill-formed, although the
corresponding single WH-question with a stressed verb, as in (41), is accepted.9
8 Notice that the in-situ questioned term is also stressed in a concealed question like (i):
(i) Who bought a book?
One possible response for (i) interprets a book as an interrogative phrase and (i), therefore, as a multiple
interrogation with a family-of-answers interpretation, warranting a response like the following:
(ii) John bought Dilbert, and Mary bought The Far Side.
Under a contour like that in (iii), however, an answer like (ii) is inappropriate:
(iii) Who bought a book?
We see that an indefinite interpreted as a WH in situ must, like a morphological WH in situ, receive stress. Now
notice the following contrast:
(iv) a. Who bought a book?
b. What did some kid buy?
(ivb), unlike (iva), does not generate a family-of-answers response. Thus, a concealed multiple interrogation is
constrained both intonationally and structurally like a multiple WH-question: The indefinite interpreted as WH in
situ must be stressed and cannot occur in subject position (cf. What did who buy?). (An example like (ivb), with an
indefinite in subject position, can be used under only a limited set of circumstances. Thus, a mother, anxious about
her child抯 purchase, might run into a store after her child has come and gone, asking, What did some kid [come in
9 It is not the case that the unacceptability of (40) is due to the fact that stress is not sentence-final in that
example; an example like the following is just as bad:
(i) Who gave what to Mary?It is true, however, as pointed out by Peter Culicover, that an example like (40) could occur in a repair context like
(ii) (No, no, no,) I didn抰 say who bought what; I said, who sought what?